1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 */ 85 86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <asm/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/init.h> 111 #include <linux/highmem.h> 112 #include <linux/user_namespace.h> 113 #include <linux/static_key.h> 114 #include <linux/memcontrol.h> 115 #include <linux/prefetch.h> 116 #include <linux/compat.h> 117 #include <linux/mroute.h> 118 #include <linux/mroute6.h> 119 #include <linux/icmpv6.h> 120 121 #include <linux/uaccess.h> 122 123 #include <linux/netdevice.h> 124 #include <net/protocol.h> 125 #include <linux/skbuff.h> 126 #include <net/net_namespace.h> 127 #include <net/request_sock.h> 128 #include <net/sock.h> 129 #include <linux/net_tstamp.h> 130 #include <net/xfrm.h> 131 #include <linux/ipsec.h> 132 #include <net/cls_cgroup.h> 133 #include <net/netprio_cgroup.h> 134 #include <linux/sock_diag.h> 135 136 #include <linux/filter.h> 137 #include <net/sock_reuseport.h> 138 #include <net/bpf_sk_storage.h> 139 140 #include <trace/events/sock.h> 141 142 #include <net/tcp.h> 143 #include <net/busy_poll.h> 144 #include <net/phonet/phonet.h> 145 146 #include <linux/ethtool.h> 147 148 #include "dev.h" 149 150 static DEFINE_MUTEX(proto_list_mutex); 151 static LIST_HEAD(proto_list); 152 153 static void sock_def_write_space_wfree(struct sock *sk); 154 static void sock_def_write_space(struct sock *sk); 155 156 /** 157 * sk_ns_capable - General socket capability test 158 * @sk: Socket to use a capability on or through 159 * @user_ns: The user namespace of the capability to use 160 * @cap: The capability to use 161 * 162 * Test to see if the opener of the socket had when the socket was 163 * created and the current process has the capability @cap in the user 164 * namespace @user_ns. 165 */ 166 bool sk_ns_capable(const struct sock *sk, 167 struct user_namespace *user_ns, int cap) 168 { 169 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 170 ns_capable(user_ns, cap); 171 } 172 EXPORT_SYMBOL(sk_ns_capable); 173 174 /** 175 * sk_capable - Socket global capability test 176 * @sk: Socket to use a capability on or through 177 * @cap: The global capability to use 178 * 179 * Test to see if the opener of the socket had when the socket was 180 * created and the current process has the capability @cap in all user 181 * namespaces. 182 */ 183 bool sk_capable(const struct sock *sk, int cap) 184 { 185 return sk_ns_capable(sk, &init_user_ns, cap); 186 } 187 EXPORT_SYMBOL(sk_capable); 188 189 /** 190 * sk_net_capable - Network namespace socket capability test 191 * @sk: Socket to use a capability on or through 192 * @cap: The capability to use 193 * 194 * Test to see if the opener of the socket had when the socket was created 195 * and the current process has the capability @cap over the network namespace 196 * the socket is a member of. 197 */ 198 bool sk_net_capable(const struct sock *sk, int cap) 199 { 200 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 201 } 202 EXPORT_SYMBOL(sk_net_capable); 203 204 /* 205 * Each address family might have different locking rules, so we have 206 * one slock key per address family and separate keys for internal and 207 * userspace sockets. 208 */ 209 static struct lock_class_key af_family_keys[AF_MAX]; 210 static struct lock_class_key af_family_kern_keys[AF_MAX]; 211 static struct lock_class_key af_family_slock_keys[AF_MAX]; 212 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 213 214 /* 215 * Make lock validator output more readable. (we pre-construct these 216 * strings build-time, so that runtime initialization of socket 217 * locks is fast): 218 */ 219 220 #define _sock_locks(x) \ 221 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 222 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 223 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 224 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 225 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 226 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 227 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 228 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 229 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 230 x "27" , x "28" , x "AF_CAN" , \ 231 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 232 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 233 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 234 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 235 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 236 x "AF_MCTP" , \ 237 x "AF_MAX" 238 239 static const char *const af_family_key_strings[AF_MAX+1] = { 240 _sock_locks("sk_lock-") 241 }; 242 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 243 _sock_locks("slock-") 244 }; 245 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 246 _sock_locks("clock-") 247 }; 248 249 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 250 _sock_locks("k-sk_lock-") 251 }; 252 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 253 _sock_locks("k-slock-") 254 }; 255 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 256 _sock_locks("k-clock-") 257 }; 258 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 259 _sock_locks("rlock-") 260 }; 261 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 262 _sock_locks("wlock-") 263 }; 264 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 265 _sock_locks("elock-") 266 }; 267 268 /* 269 * sk_callback_lock and sk queues locking rules are per-address-family, 270 * so split the lock classes by using a per-AF key: 271 */ 272 static struct lock_class_key af_callback_keys[AF_MAX]; 273 static struct lock_class_key af_rlock_keys[AF_MAX]; 274 static struct lock_class_key af_wlock_keys[AF_MAX]; 275 static struct lock_class_key af_elock_keys[AF_MAX]; 276 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 277 278 /* Run time adjustable parameters. */ 279 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 280 EXPORT_SYMBOL(sysctl_wmem_max); 281 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 282 EXPORT_SYMBOL(sysctl_rmem_max); 283 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 284 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 285 286 /* Maximal space eaten by iovec or ancillary data plus some space */ 287 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 288 EXPORT_SYMBOL(sysctl_optmem_max); 289 290 int sysctl_tstamp_allow_data __read_mostly = 1; 291 292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 293 EXPORT_SYMBOL_GPL(memalloc_socks_key); 294 295 /** 296 * sk_set_memalloc - sets %SOCK_MEMALLOC 297 * @sk: socket to set it on 298 * 299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 300 * It's the responsibility of the admin to adjust min_free_kbytes 301 * to meet the requirements 302 */ 303 void sk_set_memalloc(struct sock *sk) 304 { 305 sock_set_flag(sk, SOCK_MEMALLOC); 306 sk->sk_allocation |= __GFP_MEMALLOC; 307 static_branch_inc(&memalloc_socks_key); 308 } 309 EXPORT_SYMBOL_GPL(sk_set_memalloc); 310 311 void sk_clear_memalloc(struct sock *sk) 312 { 313 sock_reset_flag(sk, SOCK_MEMALLOC); 314 sk->sk_allocation &= ~__GFP_MEMALLOC; 315 static_branch_dec(&memalloc_socks_key); 316 317 /* 318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 319 * progress of swapping. SOCK_MEMALLOC may be cleared while 320 * it has rmem allocations due to the last swapfile being deactivated 321 * but there is a risk that the socket is unusable due to exceeding 322 * the rmem limits. Reclaim the reserves and obey rmem limits again. 323 */ 324 sk_mem_reclaim(sk); 325 } 326 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 327 328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 329 { 330 int ret; 331 unsigned int noreclaim_flag; 332 333 /* these should have been dropped before queueing */ 334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 335 336 noreclaim_flag = memalloc_noreclaim_save(); 337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 338 tcp_v6_do_rcv, 339 tcp_v4_do_rcv, 340 sk, skb); 341 memalloc_noreclaim_restore(noreclaim_flag); 342 343 return ret; 344 } 345 EXPORT_SYMBOL(__sk_backlog_rcv); 346 347 void sk_error_report(struct sock *sk) 348 { 349 sk->sk_error_report(sk); 350 351 switch (sk->sk_family) { 352 case AF_INET: 353 fallthrough; 354 case AF_INET6: 355 trace_inet_sk_error_report(sk); 356 break; 357 default: 358 break; 359 } 360 } 361 EXPORT_SYMBOL(sk_error_report); 362 363 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 364 { 365 struct __kernel_sock_timeval tv; 366 367 if (timeo == MAX_SCHEDULE_TIMEOUT) { 368 tv.tv_sec = 0; 369 tv.tv_usec = 0; 370 } else { 371 tv.tv_sec = timeo / HZ; 372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 373 } 374 375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 377 *(struct old_timeval32 *)optval = tv32; 378 return sizeof(tv32); 379 } 380 381 if (old_timeval) { 382 struct __kernel_old_timeval old_tv; 383 old_tv.tv_sec = tv.tv_sec; 384 old_tv.tv_usec = tv.tv_usec; 385 *(struct __kernel_old_timeval *)optval = old_tv; 386 return sizeof(old_tv); 387 } 388 389 *(struct __kernel_sock_timeval *)optval = tv; 390 return sizeof(tv); 391 } 392 EXPORT_SYMBOL(sock_get_timeout); 393 394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 395 sockptr_t optval, int optlen, bool old_timeval) 396 { 397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 398 struct old_timeval32 tv32; 399 400 if (optlen < sizeof(tv32)) 401 return -EINVAL; 402 403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 404 return -EFAULT; 405 tv->tv_sec = tv32.tv_sec; 406 tv->tv_usec = tv32.tv_usec; 407 } else if (old_timeval) { 408 struct __kernel_old_timeval old_tv; 409 410 if (optlen < sizeof(old_tv)) 411 return -EINVAL; 412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 413 return -EFAULT; 414 tv->tv_sec = old_tv.tv_sec; 415 tv->tv_usec = old_tv.tv_usec; 416 } else { 417 if (optlen < sizeof(*tv)) 418 return -EINVAL; 419 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 420 return -EFAULT; 421 } 422 423 return 0; 424 } 425 EXPORT_SYMBOL(sock_copy_user_timeval); 426 427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 428 bool old_timeval) 429 { 430 struct __kernel_sock_timeval tv; 431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 432 433 if (err) 434 return err; 435 436 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 437 return -EDOM; 438 439 if (tv.tv_sec < 0) { 440 static int warned __read_mostly; 441 442 *timeo_p = 0; 443 if (warned < 10 && net_ratelimit()) { 444 warned++; 445 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 446 __func__, current->comm, task_pid_nr(current)); 447 } 448 return 0; 449 } 450 *timeo_p = MAX_SCHEDULE_TIMEOUT; 451 if (tv.tv_sec == 0 && tv.tv_usec == 0) 452 return 0; 453 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 454 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 455 return 0; 456 } 457 458 static bool sock_needs_netstamp(const struct sock *sk) 459 { 460 switch (sk->sk_family) { 461 case AF_UNSPEC: 462 case AF_UNIX: 463 return false; 464 default: 465 return true; 466 } 467 } 468 469 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 470 { 471 if (sk->sk_flags & flags) { 472 sk->sk_flags &= ~flags; 473 if (sock_needs_netstamp(sk) && 474 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 475 net_disable_timestamp(); 476 } 477 } 478 479 480 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 481 { 482 unsigned long flags; 483 struct sk_buff_head *list = &sk->sk_receive_queue; 484 485 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 486 atomic_inc(&sk->sk_drops); 487 trace_sock_rcvqueue_full(sk, skb); 488 return -ENOMEM; 489 } 490 491 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 492 atomic_inc(&sk->sk_drops); 493 return -ENOBUFS; 494 } 495 496 skb->dev = NULL; 497 skb_set_owner_r(skb, sk); 498 499 /* we escape from rcu protected region, make sure we dont leak 500 * a norefcounted dst 501 */ 502 skb_dst_force(skb); 503 504 spin_lock_irqsave(&list->lock, flags); 505 sock_skb_set_dropcount(sk, skb); 506 __skb_queue_tail(list, skb); 507 spin_unlock_irqrestore(&list->lock, flags); 508 509 if (!sock_flag(sk, SOCK_DEAD)) 510 sk->sk_data_ready(sk); 511 return 0; 512 } 513 EXPORT_SYMBOL(__sock_queue_rcv_skb); 514 515 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 516 enum skb_drop_reason *reason) 517 { 518 enum skb_drop_reason drop_reason; 519 int err; 520 521 err = sk_filter(sk, skb); 522 if (err) { 523 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 524 goto out; 525 } 526 err = __sock_queue_rcv_skb(sk, skb); 527 switch (err) { 528 case -ENOMEM: 529 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 530 break; 531 case -ENOBUFS: 532 drop_reason = SKB_DROP_REASON_PROTO_MEM; 533 break; 534 default: 535 drop_reason = SKB_NOT_DROPPED_YET; 536 break; 537 } 538 out: 539 if (reason) 540 *reason = drop_reason; 541 return err; 542 } 543 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 544 545 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 546 const int nested, unsigned int trim_cap, bool refcounted) 547 { 548 int rc = NET_RX_SUCCESS; 549 550 if (sk_filter_trim_cap(sk, skb, trim_cap)) 551 goto discard_and_relse; 552 553 skb->dev = NULL; 554 555 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 556 atomic_inc(&sk->sk_drops); 557 goto discard_and_relse; 558 } 559 if (nested) 560 bh_lock_sock_nested(sk); 561 else 562 bh_lock_sock(sk); 563 if (!sock_owned_by_user(sk)) { 564 /* 565 * trylock + unlock semantics: 566 */ 567 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 568 569 rc = sk_backlog_rcv(sk, skb); 570 571 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 572 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 573 bh_unlock_sock(sk); 574 atomic_inc(&sk->sk_drops); 575 goto discard_and_relse; 576 } 577 578 bh_unlock_sock(sk); 579 out: 580 if (refcounted) 581 sock_put(sk); 582 return rc; 583 discard_and_relse: 584 kfree_skb(skb); 585 goto out; 586 } 587 EXPORT_SYMBOL(__sk_receive_skb); 588 589 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 590 u32)); 591 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 592 u32)); 593 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 594 { 595 struct dst_entry *dst = __sk_dst_get(sk); 596 597 if (dst && dst->obsolete && 598 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 599 dst, cookie) == NULL) { 600 sk_tx_queue_clear(sk); 601 sk->sk_dst_pending_confirm = 0; 602 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 603 dst_release(dst); 604 return NULL; 605 } 606 607 return dst; 608 } 609 EXPORT_SYMBOL(__sk_dst_check); 610 611 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 612 { 613 struct dst_entry *dst = sk_dst_get(sk); 614 615 if (dst && dst->obsolete && 616 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 617 dst, cookie) == NULL) { 618 sk_dst_reset(sk); 619 dst_release(dst); 620 return NULL; 621 } 622 623 return dst; 624 } 625 EXPORT_SYMBOL(sk_dst_check); 626 627 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 628 { 629 int ret = -ENOPROTOOPT; 630 #ifdef CONFIG_NETDEVICES 631 struct net *net = sock_net(sk); 632 633 /* Sorry... */ 634 ret = -EPERM; 635 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 636 goto out; 637 638 ret = -EINVAL; 639 if (ifindex < 0) 640 goto out; 641 642 /* Paired with all READ_ONCE() done locklessly. */ 643 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 644 645 if (sk->sk_prot->rehash) 646 sk->sk_prot->rehash(sk); 647 sk_dst_reset(sk); 648 649 ret = 0; 650 651 out: 652 #endif 653 654 return ret; 655 } 656 657 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 658 { 659 int ret; 660 661 if (lock_sk) 662 lock_sock(sk); 663 ret = sock_bindtoindex_locked(sk, ifindex); 664 if (lock_sk) 665 release_sock(sk); 666 667 return ret; 668 } 669 EXPORT_SYMBOL(sock_bindtoindex); 670 671 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 672 { 673 int ret = -ENOPROTOOPT; 674 #ifdef CONFIG_NETDEVICES 675 struct net *net = sock_net(sk); 676 char devname[IFNAMSIZ]; 677 int index; 678 679 ret = -EINVAL; 680 if (optlen < 0) 681 goto out; 682 683 /* Bind this socket to a particular device like "eth0", 684 * as specified in the passed interface name. If the 685 * name is "" or the option length is zero the socket 686 * is not bound. 687 */ 688 if (optlen > IFNAMSIZ - 1) 689 optlen = IFNAMSIZ - 1; 690 memset(devname, 0, sizeof(devname)); 691 692 ret = -EFAULT; 693 if (copy_from_sockptr(devname, optval, optlen)) 694 goto out; 695 696 index = 0; 697 if (devname[0] != '\0') { 698 struct net_device *dev; 699 700 rcu_read_lock(); 701 dev = dev_get_by_name_rcu(net, devname); 702 if (dev) 703 index = dev->ifindex; 704 rcu_read_unlock(); 705 ret = -ENODEV; 706 if (!dev) 707 goto out; 708 } 709 710 sockopt_lock_sock(sk); 711 ret = sock_bindtoindex_locked(sk, index); 712 sockopt_release_sock(sk); 713 out: 714 #endif 715 716 return ret; 717 } 718 719 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 720 sockptr_t optlen, int len) 721 { 722 int ret = -ENOPROTOOPT; 723 #ifdef CONFIG_NETDEVICES 724 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 725 struct net *net = sock_net(sk); 726 char devname[IFNAMSIZ]; 727 728 if (bound_dev_if == 0) { 729 len = 0; 730 goto zero; 731 } 732 733 ret = -EINVAL; 734 if (len < IFNAMSIZ) 735 goto out; 736 737 ret = netdev_get_name(net, devname, bound_dev_if); 738 if (ret) 739 goto out; 740 741 len = strlen(devname) + 1; 742 743 ret = -EFAULT; 744 if (copy_to_sockptr(optval, devname, len)) 745 goto out; 746 747 zero: 748 ret = -EFAULT; 749 if (copy_to_sockptr(optlen, &len, sizeof(int))) 750 goto out; 751 752 ret = 0; 753 754 out: 755 #endif 756 757 return ret; 758 } 759 760 bool sk_mc_loop(struct sock *sk) 761 { 762 if (dev_recursion_level()) 763 return false; 764 if (!sk) 765 return true; 766 switch (sk->sk_family) { 767 case AF_INET: 768 return inet_sk(sk)->mc_loop; 769 #if IS_ENABLED(CONFIG_IPV6) 770 case AF_INET6: 771 return inet6_sk(sk)->mc_loop; 772 #endif 773 } 774 WARN_ON_ONCE(1); 775 return true; 776 } 777 EXPORT_SYMBOL(sk_mc_loop); 778 779 void sock_set_reuseaddr(struct sock *sk) 780 { 781 lock_sock(sk); 782 sk->sk_reuse = SK_CAN_REUSE; 783 release_sock(sk); 784 } 785 EXPORT_SYMBOL(sock_set_reuseaddr); 786 787 void sock_set_reuseport(struct sock *sk) 788 { 789 lock_sock(sk); 790 sk->sk_reuseport = true; 791 release_sock(sk); 792 } 793 EXPORT_SYMBOL(sock_set_reuseport); 794 795 void sock_no_linger(struct sock *sk) 796 { 797 lock_sock(sk); 798 sk->sk_lingertime = 0; 799 sock_set_flag(sk, SOCK_LINGER); 800 release_sock(sk); 801 } 802 EXPORT_SYMBOL(sock_no_linger); 803 804 void sock_set_priority(struct sock *sk, u32 priority) 805 { 806 lock_sock(sk); 807 sk->sk_priority = priority; 808 release_sock(sk); 809 } 810 EXPORT_SYMBOL(sock_set_priority); 811 812 void sock_set_sndtimeo(struct sock *sk, s64 secs) 813 { 814 lock_sock(sk); 815 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 816 sk->sk_sndtimeo = secs * HZ; 817 else 818 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 819 release_sock(sk); 820 } 821 EXPORT_SYMBOL(sock_set_sndtimeo); 822 823 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 824 { 825 if (val) { 826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 827 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 828 sock_set_flag(sk, SOCK_RCVTSTAMP); 829 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 830 } else { 831 sock_reset_flag(sk, SOCK_RCVTSTAMP); 832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 833 } 834 } 835 836 void sock_enable_timestamps(struct sock *sk) 837 { 838 lock_sock(sk); 839 __sock_set_timestamps(sk, true, false, true); 840 release_sock(sk); 841 } 842 EXPORT_SYMBOL(sock_enable_timestamps); 843 844 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 845 { 846 switch (optname) { 847 case SO_TIMESTAMP_OLD: 848 __sock_set_timestamps(sk, valbool, false, false); 849 break; 850 case SO_TIMESTAMP_NEW: 851 __sock_set_timestamps(sk, valbool, true, false); 852 break; 853 case SO_TIMESTAMPNS_OLD: 854 __sock_set_timestamps(sk, valbool, false, true); 855 break; 856 case SO_TIMESTAMPNS_NEW: 857 __sock_set_timestamps(sk, valbool, true, true); 858 break; 859 } 860 } 861 862 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 863 { 864 struct net *net = sock_net(sk); 865 struct net_device *dev = NULL; 866 bool match = false; 867 int *vclock_index; 868 int i, num; 869 870 if (sk->sk_bound_dev_if) 871 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 872 873 if (!dev) { 874 pr_err("%s: sock not bind to device\n", __func__); 875 return -EOPNOTSUPP; 876 } 877 878 num = ethtool_get_phc_vclocks(dev, &vclock_index); 879 dev_put(dev); 880 881 for (i = 0; i < num; i++) { 882 if (*(vclock_index + i) == phc_index) { 883 match = true; 884 break; 885 } 886 } 887 888 if (num > 0) 889 kfree(vclock_index); 890 891 if (!match) 892 return -EINVAL; 893 894 sk->sk_bind_phc = phc_index; 895 896 return 0; 897 } 898 899 int sock_set_timestamping(struct sock *sk, int optname, 900 struct so_timestamping timestamping) 901 { 902 int val = timestamping.flags; 903 int ret; 904 905 if (val & ~SOF_TIMESTAMPING_MASK) 906 return -EINVAL; 907 908 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 909 !(val & SOF_TIMESTAMPING_OPT_ID)) 910 return -EINVAL; 911 912 if (val & SOF_TIMESTAMPING_OPT_ID && 913 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 914 if (sk_is_tcp(sk)) { 915 if ((1 << sk->sk_state) & 916 (TCPF_CLOSE | TCPF_LISTEN)) 917 return -EINVAL; 918 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 919 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 920 else 921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 922 } else { 923 atomic_set(&sk->sk_tskey, 0); 924 } 925 } 926 927 if (val & SOF_TIMESTAMPING_OPT_STATS && 928 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 929 return -EINVAL; 930 931 if (val & SOF_TIMESTAMPING_BIND_PHC) { 932 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 933 if (ret) 934 return ret; 935 } 936 937 sk->sk_tsflags = val; 938 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 939 940 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 941 sock_enable_timestamp(sk, 942 SOCK_TIMESTAMPING_RX_SOFTWARE); 943 else 944 sock_disable_timestamp(sk, 945 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 946 return 0; 947 } 948 949 void sock_set_keepalive(struct sock *sk) 950 { 951 lock_sock(sk); 952 if (sk->sk_prot->keepalive) 953 sk->sk_prot->keepalive(sk, true); 954 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 955 release_sock(sk); 956 } 957 EXPORT_SYMBOL(sock_set_keepalive); 958 959 static void __sock_set_rcvbuf(struct sock *sk, int val) 960 { 961 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 962 * as a negative value. 963 */ 964 val = min_t(int, val, INT_MAX / 2); 965 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 966 967 /* We double it on the way in to account for "struct sk_buff" etc. 968 * overhead. Applications assume that the SO_RCVBUF setting they make 969 * will allow that much actual data to be received on that socket. 970 * 971 * Applications are unaware that "struct sk_buff" and other overheads 972 * allocate from the receive buffer during socket buffer allocation. 973 * 974 * And after considering the possible alternatives, returning the value 975 * we actually used in getsockopt is the most desirable behavior. 976 */ 977 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 978 } 979 980 void sock_set_rcvbuf(struct sock *sk, int val) 981 { 982 lock_sock(sk); 983 __sock_set_rcvbuf(sk, val); 984 release_sock(sk); 985 } 986 EXPORT_SYMBOL(sock_set_rcvbuf); 987 988 static void __sock_set_mark(struct sock *sk, u32 val) 989 { 990 if (val != sk->sk_mark) { 991 sk->sk_mark = val; 992 sk_dst_reset(sk); 993 } 994 } 995 996 void sock_set_mark(struct sock *sk, u32 val) 997 { 998 lock_sock(sk); 999 __sock_set_mark(sk, val); 1000 release_sock(sk); 1001 } 1002 EXPORT_SYMBOL(sock_set_mark); 1003 1004 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1005 { 1006 /* Round down bytes to multiple of pages */ 1007 bytes = round_down(bytes, PAGE_SIZE); 1008 1009 WARN_ON(bytes > sk->sk_reserved_mem); 1010 sk->sk_reserved_mem -= bytes; 1011 sk_mem_reclaim(sk); 1012 } 1013 1014 static int sock_reserve_memory(struct sock *sk, int bytes) 1015 { 1016 long allocated; 1017 bool charged; 1018 int pages; 1019 1020 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1021 return -EOPNOTSUPP; 1022 1023 if (!bytes) 1024 return 0; 1025 1026 pages = sk_mem_pages(bytes); 1027 1028 /* pre-charge to memcg */ 1029 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1030 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1031 if (!charged) 1032 return -ENOMEM; 1033 1034 /* pre-charge to forward_alloc */ 1035 sk_memory_allocated_add(sk, pages); 1036 allocated = sk_memory_allocated(sk); 1037 /* If the system goes into memory pressure with this 1038 * precharge, give up and return error. 1039 */ 1040 if (allocated > sk_prot_mem_limits(sk, 1)) { 1041 sk_memory_allocated_sub(sk, pages); 1042 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1043 return -ENOMEM; 1044 } 1045 sk->sk_forward_alloc += pages << PAGE_SHIFT; 1046 1047 sk->sk_reserved_mem += pages << PAGE_SHIFT; 1048 1049 return 0; 1050 } 1051 1052 void sockopt_lock_sock(struct sock *sk) 1053 { 1054 /* When current->bpf_ctx is set, the setsockopt is called from 1055 * a bpf prog. bpf has ensured the sk lock has been 1056 * acquired before calling setsockopt(). 1057 */ 1058 if (has_current_bpf_ctx()) 1059 return; 1060 1061 lock_sock(sk); 1062 } 1063 EXPORT_SYMBOL(sockopt_lock_sock); 1064 1065 void sockopt_release_sock(struct sock *sk) 1066 { 1067 if (has_current_bpf_ctx()) 1068 return; 1069 1070 release_sock(sk); 1071 } 1072 EXPORT_SYMBOL(sockopt_release_sock); 1073 1074 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1075 { 1076 return has_current_bpf_ctx() || ns_capable(ns, cap); 1077 } 1078 EXPORT_SYMBOL(sockopt_ns_capable); 1079 1080 bool sockopt_capable(int cap) 1081 { 1082 return has_current_bpf_ctx() || capable(cap); 1083 } 1084 EXPORT_SYMBOL(sockopt_capable); 1085 1086 /* 1087 * This is meant for all protocols to use and covers goings on 1088 * at the socket level. Everything here is generic. 1089 */ 1090 1091 int sk_setsockopt(struct sock *sk, int level, int optname, 1092 sockptr_t optval, unsigned int optlen) 1093 { 1094 struct so_timestamping timestamping; 1095 struct socket *sock = sk->sk_socket; 1096 struct sock_txtime sk_txtime; 1097 int val; 1098 int valbool; 1099 struct linger ling; 1100 int ret = 0; 1101 1102 /* 1103 * Options without arguments 1104 */ 1105 1106 if (optname == SO_BINDTODEVICE) 1107 return sock_setbindtodevice(sk, optval, optlen); 1108 1109 if (optlen < sizeof(int)) 1110 return -EINVAL; 1111 1112 if (copy_from_sockptr(&val, optval, sizeof(val))) 1113 return -EFAULT; 1114 1115 valbool = val ? 1 : 0; 1116 1117 sockopt_lock_sock(sk); 1118 1119 switch (optname) { 1120 case SO_DEBUG: 1121 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1122 ret = -EACCES; 1123 else 1124 sock_valbool_flag(sk, SOCK_DBG, valbool); 1125 break; 1126 case SO_REUSEADDR: 1127 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1128 break; 1129 case SO_REUSEPORT: 1130 sk->sk_reuseport = valbool; 1131 break; 1132 case SO_TYPE: 1133 case SO_PROTOCOL: 1134 case SO_DOMAIN: 1135 case SO_ERROR: 1136 ret = -ENOPROTOOPT; 1137 break; 1138 case SO_DONTROUTE: 1139 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1140 sk_dst_reset(sk); 1141 break; 1142 case SO_BROADCAST: 1143 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1144 break; 1145 case SO_SNDBUF: 1146 /* Don't error on this BSD doesn't and if you think 1147 * about it this is right. Otherwise apps have to 1148 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1149 * are treated in BSD as hints 1150 */ 1151 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1152 set_sndbuf: 1153 /* Ensure val * 2 fits into an int, to prevent max_t() 1154 * from treating it as a negative value. 1155 */ 1156 val = min_t(int, val, INT_MAX / 2); 1157 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1158 WRITE_ONCE(sk->sk_sndbuf, 1159 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1160 /* Wake up sending tasks if we upped the value. */ 1161 sk->sk_write_space(sk); 1162 break; 1163 1164 case SO_SNDBUFFORCE: 1165 if (!sockopt_capable(CAP_NET_ADMIN)) { 1166 ret = -EPERM; 1167 break; 1168 } 1169 1170 /* No negative values (to prevent underflow, as val will be 1171 * multiplied by 2). 1172 */ 1173 if (val < 0) 1174 val = 0; 1175 goto set_sndbuf; 1176 1177 case SO_RCVBUF: 1178 /* Don't error on this BSD doesn't and if you think 1179 * about it this is right. Otherwise apps have to 1180 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1181 * are treated in BSD as hints 1182 */ 1183 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1184 break; 1185 1186 case SO_RCVBUFFORCE: 1187 if (!sockopt_capable(CAP_NET_ADMIN)) { 1188 ret = -EPERM; 1189 break; 1190 } 1191 1192 /* No negative values (to prevent underflow, as val will be 1193 * multiplied by 2). 1194 */ 1195 __sock_set_rcvbuf(sk, max(val, 0)); 1196 break; 1197 1198 case SO_KEEPALIVE: 1199 if (sk->sk_prot->keepalive) 1200 sk->sk_prot->keepalive(sk, valbool); 1201 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1202 break; 1203 1204 case SO_OOBINLINE: 1205 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1206 break; 1207 1208 case SO_NO_CHECK: 1209 sk->sk_no_check_tx = valbool; 1210 break; 1211 1212 case SO_PRIORITY: 1213 if ((val >= 0 && val <= 6) || 1214 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1215 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1216 sk->sk_priority = val; 1217 else 1218 ret = -EPERM; 1219 break; 1220 1221 case SO_LINGER: 1222 if (optlen < sizeof(ling)) { 1223 ret = -EINVAL; /* 1003.1g */ 1224 break; 1225 } 1226 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1227 ret = -EFAULT; 1228 break; 1229 } 1230 if (!ling.l_onoff) 1231 sock_reset_flag(sk, SOCK_LINGER); 1232 else { 1233 #if (BITS_PER_LONG == 32) 1234 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1235 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1236 else 1237 #endif 1238 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1239 sock_set_flag(sk, SOCK_LINGER); 1240 } 1241 break; 1242 1243 case SO_BSDCOMPAT: 1244 break; 1245 1246 case SO_PASSCRED: 1247 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1248 break; 1249 1250 case SO_PASSPIDFD: 1251 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1252 break; 1253 1254 case SO_TIMESTAMP_OLD: 1255 case SO_TIMESTAMP_NEW: 1256 case SO_TIMESTAMPNS_OLD: 1257 case SO_TIMESTAMPNS_NEW: 1258 sock_set_timestamp(sk, optname, valbool); 1259 break; 1260 1261 case SO_TIMESTAMPING_NEW: 1262 case SO_TIMESTAMPING_OLD: 1263 if (optlen == sizeof(timestamping)) { 1264 if (copy_from_sockptr(×tamping, optval, 1265 sizeof(timestamping))) { 1266 ret = -EFAULT; 1267 break; 1268 } 1269 } else { 1270 memset(×tamping, 0, sizeof(timestamping)); 1271 timestamping.flags = val; 1272 } 1273 ret = sock_set_timestamping(sk, optname, timestamping); 1274 break; 1275 1276 case SO_RCVLOWAT: 1277 if (val < 0) 1278 val = INT_MAX; 1279 if (sock && sock->ops->set_rcvlowat) 1280 ret = sock->ops->set_rcvlowat(sk, val); 1281 else 1282 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1283 break; 1284 1285 case SO_RCVTIMEO_OLD: 1286 case SO_RCVTIMEO_NEW: 1287 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1288 optlen, optname == SO_RCVTIMEO_OLD); 1289 break; 1290 1291 case SO_SNDTIMEO_OLD: 1292 case SO_SNDTIMEO_NEW: 1293 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1294 optlen, optname == SO_SNDTIMEO_OLD); 1295 break; 1296 1297 case SO_ATTACH_FILTER: { 1298 struct sock_fprog fprog; 1299 1300 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1301 if (!ret) 1302 ret = sk_attach_filter(&fprog, sk); 1303 break; 1304 } 1305 case SO_ATTACH_BPF: 1306 ret = -EINVAL; 1307 if (optlen == sizeof(u32)) { 1308 u32 ufd; 1309 1310 ret = -EFAULT; 1311 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1312 break; 1313 1314 ret = sk_attach_bpf(ufd, sk); 1315 } 1316 break; 1317 1318 case SO_ATTACH_REUSEPORT_CBPF: { 1319 struct sock_fprog fprog; 1320 1321 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1322 if (!ret) 1323 ret = sk_reuseport_attach_filter(&fprog, sk); 1324 break; 1325 } 1326 case SO_ATTACH_REUSEPORT_EBPF: 1327 ret = -EINVAL; 1328 if (optlen == sizeof(u32)) { 1329 u32 ufd; 1330 1331 ret = -EFAULT; 1332 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1333 break; 1334 1335 ret = sk_reuseport_attach_bpf(ufd, sk); 1336 } 1337 break; 1338 1339 case SO_DETACH_REUSEPORT_BPF: 1340 ret = reuseport_detach_prog(sk); 1341 break; 1342 1343 case SO_DETACH_FILTER: 1344 ret = sk_detach_filter(sk); 1345 break; 1346 1347 case SO_LOCK_FILTER: 1348 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1349 ret = -EPERM; 1350 else 1351 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1352 break; 1353 1354 case SO_PASSSEC: 1355 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1356 break; 1357 case SO_MARK: 1358 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1359 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1360 ret = -EPERM; 1361 break; 1362 } 1363 1364 __sock_set_mark(sk, val); 1365 break; 1366 case SO_RCVMARK: 1367 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1368 break; 1369 1370 case SO_RXQ_OVFL: 1371 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1372 break; 1373 1374 case SO_WIFI_STATUS: 1375 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1376 break; 1377 1378 case SO_PEEK_OFF: 1379 if (sock->ops->set_peek_off) 1380 ret = sock->ops->set_peek_off(sk, val); 1381 else 1382 ret = -EOPNOTSUPP; 1383 break; 1384 1385 case SO_NOFCS: 1386 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1387 break; 1388 1389 case SO_SELECT_ERR_QUEUE: 1390 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1391 break; 1392 1393 #ifdef CONFIG_NET_RX_BUSY_POLL 1394 case SO_BUSY_POLL: 1395 if (val < 0) 1396 ret = -EINVAL; 1397 else 1398 WRITE_ONCE(sk->sk_ll_usec, val); 1399 break; 1400 case SO_PREFER_BUSY_POLL: 1401 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1402 ret = -EPERM; 1403 else 1404 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1405 break; 1406 case SO_BUSY_POLL_BUDGET: 1407 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1408 ret = -EPERM; 1409 } else { 1410 if (val < 0 || val > U16_MAX) 1411 ret = -EINVAL; 1412 else 1413 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1414 } 1415 break; 1416 #endif 1417 1418 case SO_MAX_PACING_RATE: 1419 { 1420 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1421 1422 if (sizeof(ulval) != sizeof(val) && 1423 optlen >= sizeof(ulval) && 1424 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1425 ret = -EFAULT; 1426 break; 1427 } 1428 if (ulval != ~0UL) 1429 cmpxchg(&sk->sk_pacing_status, 1430 SK_PACING_NONE, 1431 SK_PACING_NEEDED); 1432 sk->sk_max_pacing_rate = ulval; 1433 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1434 break; 1435 } 1436 case SO_INCOMING_CPU: 1437 reuseport_update_incoming_cpu(sk, val); 1438 break; 1439 1440 case SO_CNX_ADVICE: 1441 if (val == 1) 1442 dst_negative_advice(sk); 1443 break; 1444 1445 case SO_ZEROCOPY: 1446 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1447 if (!(sk_is_tcp(sk) || 1448 (sk->sk_type == SOCK_DGRAM && 1449 sk->sk_protocol == IPPROTO_UDP))) 1450 ret = -EOPNOTSUPP; 1451 } else if (sk->sk_family != PF_RDS) { 1452 ret = -EOPNOTSUPP; 1453 } 1454 if (!ret) { 1455 if (val < 0 || val > 1) 1456 ret = -EINVAL; 1457 else 1458 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1459 } 1460 break; 1461 1462 case SO_TXTIME: 1463 if (optlen != sizeof(struct sock_txtime)) { 1464 ret = -EINVAL; 1465 break; 1466 } else if (copy_from_sockptr(&sk_txtime, optval, 1467 sizeof(struct sock_txtime))) { 1468 ret = -EFAULT; 1469 break; 1470 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1471 ret = -EINVAL; 1472 break; 1473 } 1474 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1475 * scheduler has enough safe guards. 1476 */ 1477 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1478 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1479 ret = -EPERM; 1480 break; 1481 } 1482 sock_valbool_flag(sk, SOCK_TXTIME, true); 1483 sk->sk_clockid = sk_txtime.clockid; 1484 sk->sk_txtime_deadline_mode = 1485 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1486 sk->sk_txtime_report_errors = 1487 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1488 break; 1489 1490 case SO_BINDTOIFINDEX: 1491 ret = sock_bindtoindex_locked(sk, val); 1492 break; 1493 1494 case SO_BUF_LOCK: 1495 if (val & ~SOCK_BUF_LOCK_MASK) { 1496 ret = -EINVAL; 1497 break; 1498 } 1499 sk->sk_userlocks = val | (sk->sk_userlocks & 1500 ~SOCK_BUF_LOCK_MASK); 1501 break; 1502 1503 case SO_RESERVE_MEM: 1504 { 1505 int delta; 1506 1507 if (val < 0) { 1508 ret = -EINVAL; 1509 break; 1510 } 1511 1512 delta = val - sk->sk_reserved_mem; 1513 if (delta < 0) 1514 sock_release_reserved_memory(sk, -delta); 1515 else 1516 ret = sock_reserve_memory(sk, delta); 1517 break; 1518 } 1519 1520 case SO_TXREHASH: 1521 if (val < -1 || val > 1) { 1522 ret = -EINVAL; 1523 break; 1524 } 1525 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1526 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1527 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1528 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1529 break; 1530 1531 default: 1532 ret = -ENOPROTOOPT; 1533 break; 1534 } 1535 sockopt_release_sock(sk); 1536 return ret; 1537 } 1538 1539 int sock_setsockopt(struct socket *sock, int level, int optname, 1540 sockptr_t optval, unsigned int optlen) 1541 { 1542 return sk_setsockopt(sock->sk, level, optname, 1543 optval, optlen); 1544 } 1545 EXPORT_SYMBOL(sock_setsockopt); 1546 1547 static const struct cred *sk_get_peer_cred(struct sock *sk) 1548 { 1549 const struct cred *cred; 1550 1551 spin_lock(&sk->sk_peer_lock); 1552 cred = get_cred(sk->sk_peer_cred); 1553 spin_unlock(&sk->sk_peer_lock); 1554 1555 return cred; 1556 } 1557 1558 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1559 struct ucred *ucred) 1560 { 1561 ucred->pid = pid_vnr(pid); 1562 ucred->uid = ucred->gid = -1; 1563 if (cred) { 1564 struct user_namespace *current_ns = current_user_ns(); 1565 1566 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1567 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1568 } 1569 } 1570 1571 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1572 { 1573 struct user_namespace *user_ns = current_user_ns(); 1574 int i; 1575 1576 for (i = 0; i < src->ngroups; i++) { 1577 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1578 1579 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1580 return -EFAULT; 1581 } 1582 1583 return 0; 1584 } 1585 1586 int sk_getsockopt(struct sock *sk, int level, int optname, 1587 sockptr_t optval, sockptr_t optlen) 1588 { 1589 struct socket *sock = sk->sk_socket; 1590 1591 union { 1592 int val; 1593 u64 val64; 1594 unsigned long ulval; 1595 struct linger ling; 1596 struct old_timeval32 tm32; 1597 struct __kernel_old_timeval tm; 1598 struct __kernel_sock_timeval stm; 1599 struct sock_txtime txtime; 1600 struct so_timestamping timestamping; 1601 } v; 1602 1603 int lv = sizeof(int); 1604 int len; 1605 1606 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1607 return -EFAULT; 1608 if (len < 0) 1609 return -EINVAL; 1610 1611 memset(&v, 0, sizeof(v)); 1612 1613 switch (optname) { 1614 case SO_DEBUG: 1615 v.val = sock_flag(sk, SOCK_DBG); 1616 break; 1617 1618 case SO_DONTROUTE: 1619 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1620 break; 1621 1622 case SO_BROADCAST: 1623 v.val = sock_flag(sk, SOCK_BROADCAST); 1624 break; 1625 1626 case SO_SNDBUF: 1627 v.val = sk->sk_sndbuf; 1628 break; 1629 1630 case SO_RCVBUF: 1631 v.val = sk->sk_rcvbuf; 1632 break; 1633 1634 case SO_REUSEADDR: 1635 v.val = sk->sk_reuse; 1636 break; 1637 1638 case SO_REUSEPORT: 1639 v.val = sk->sk_reuseport; 1640 break; 1641 1642 case SO_KEEPALIVE: 1643 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1644 break; 1645 1646 case SO_TYPE: 1647 v.val = sk->sk_type; 1648 break; 1649 1650 case SO_PROTOCOL: 1651 v.val = sk->sk_protocol; 1652 break; 1653 1654 case SO_DOMAIN: 1655 v.val = sk->sk_family; 1656 break; 1657 1658 case SO_ERROR: 1659 v.val = -sock_error(sk); 1660 if (v.val == 0) 1661 v.val = xchg(&sk->sk_err_soft, 0); 1662 break; 1663 1664 case SO_OOBINLINE: 1665 v.val = sock_flag(sk, SOCK_URGINLINE); 1666 break; 1667 1668 case SO_NO_CHECK: 1669 v.val = sk->sk_no_check_tx; 1670 break; 1671 1672 case SO_PRIORITY: 1673 v.val = sk->sk_priority; 1674 break; 1675 1676 case SO_LINGER: 1677 lv = sizeof(v.ling); 1678 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1679 v.ling.l_linger = sk->sk_lingertime / HZ; 1680 break; 1681 1682 case SO_BSDCOMPAT: 1683 break; 1684 1685 case SO_TIMESTAMP_OLD: 1686 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1687 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1688 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1689 break; 1690 1691 case SO_TIMESTAMPNS_OLD: 1692 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1693 break; 1694 1695 case SO_TIMESTAMP_NEW: 1696 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1697 break; 1698 1699 case SO_TIMESTAMPNS_NEW: 1700 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1701 break; 1702 1703 case SO_TIMESTAMPING_OLD: 1704 lv = sizeof(v.timestamping); 1705 v.timestamping.flags = sk->sk_tsflags; 1706 v.timestamping.bind_phc = sk->sk_bind_phc; 1707 break; 1708 1709 case SO_RCVTIMEO_OLD: 1710 case SO_RCVTIMEO_NEW: 1711 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1712 break; 1713 1714 case SO_SNDTIMEO_OLD: 1715 case SO_SNDTIMEO_NEW: 1716 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1717 break; 1718 1719 case SO_RCVLOWAT: 1720 v.val = sk->sk_rcvlowat; 1721 break; 1722 1723 case SO_SNDLOWAT: 1724 v.val = 1; 1725 break; 1726 1727 case SO_PASSCRED: 1728 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1729 break; 1730 1731 case SO_PASSPIDFD: 1732 v.val = !!test_bit(SOCK_PASSPIDFD, &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_PEERPIDFD: 1751 { 1752 struct pid *peer_pid; 1753 struct file *pidfd_file = NULL; 1754 int pidfd; 1755 1756 if (len > sizeof(pidfd)) 1757 len = sizeof(pidfd); 1758 1759 spin_lock(&sk->sk_peer_lock); 1760 peer_pid = get_pid(sk->sk_peer_pid); 1761 spin_unlock(&sk->sk_peer_lock); 1762 1763 if (!peer_pid) 1764 return -ESRCH; 1765 1766 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1767 put_pid(peer_pid); 1768 if (pidfd < 0) 1769 return pidfd; 1770 1771 if (copy_to_sockptr(optval, &pidfd, len) || 1772 copy_to_sockptr(optlen, &len, sizeof(int))) { 1773 put_unused_fd(pidfd); 1774 fput(pidfd_file); 1775 1776 return -EFAULT; 1777 } 1778 1779 fd_install(pidfd, pidfd_file); 1780 return 0; 1781 } 1782 1783 case SO_PEERGROUPS: 1784 { 1785 const struct cred *cred; 1786 int ret, n; 1787 1788 cred = sk_get_peer_cred(sk); 1789 if (!cred) 1790 return -ENODATA; 1791 1792 n = cred->group_info->ngroups; 1793 if (len < n * sizeof(gid_t)) { 1794 len = n * sizeof(gid_t); 1795 put_cred(cred); 1796 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1797 } 1798 len = n * sizeof(gid_t); 1799 1800 ret = groups_to_user(optval, cred->group_info); 1801 put_cred(cred); 1802 if (ret) 1803 return ret; 1804 goto lenout; 1805 } 1806 1807 case SO_PEERNAME: 1808 { 1809 struct sockaddr_storage address; 1810 1811 lv = sock->ops->getname(sock, (struct sockaddr *)&address, 2); 1812 if (lv < 0) 1813 return -ENOTCONN; 1814 if (lv < len) 1815 return -EINVAL; 1816 if (copy_to_sockptr(optval, &address, len)) 1817 return -EFAULT; 1818 goto lenout; 1819 } 1820 1821 /* Dubious BSD thing... Probably nobody even uses it, but 1822 * the UNIX standard wants it for whatever reason... -DaveM 1823 */ 1824 case SO_ACCEPTCONN: 1825 v.val = sk->sk_state == TCP_LISTEN; 1826 break; 1827 1828 case SO_PASSSEC: 1829 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1830 break; 1831 1832 case SO_PEERSEC: 1833 return security_socket_getpeersec_stream(sock, 1834 optval, optlen, len); 1835 1836 case SO_MARK: 1837 v.val = sk->sk_mark; 1838 break; 1839 1840 case SO_RCVMARK: 1841 v.val = sock_flag(sk, SOCK_RCVMARK); 1842 break; 1843 1844 case SO_RXQ_OVFL: 1845 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1846 break; 1847 1848 case SO_WIFI_STATUS: 1849 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1850 break; 1851 1852 case SO_PEEK_OFF: 1853 if (!sock->ops->set_peek_off) 1854 return -EOPNOTSUPP; 1855 1856 v.val = sk->sk_peek_off; 1857 break; 1858 case SO_NOFCS: 1859 v.val = sock_flag(sk, SOCK_NOFCS); 1860 break; 1861 1862 case SO_BINDTODEVICE: 1863 return sock_getbindtodevice(sk, optval, optlen, len); 1864 1865 case SO_GET_FILTER: 1866 len = sk_get_filter(sk, optval, len); 1867 if (len < 0) 1868 return len; 1869 1870 goto lenout; 1871 1872 case SO_LOCK_FILTER: 1873 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1874 break; 1875 1876 case SO_BPF_EXTENSIONS: 1877 v.val = bpf_tell_extensions(); 1878 break; 1879 1880 case SO_SELECT_ERR_QUEUE: 1881 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1882 break; 1883 1884 #ifdef CONFIG_NET_RX_BUSY_POLL 1885 case SO_BUSY_POLL: 1886 v.val = sk->sk_ll_usec; 1887 break; 1888 case SO_PREFER_BUSY_POLL: 1889 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1890 break; 1891 #endif 1892 1893 case SO_MAX_PACING_RATE: 1894 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1895 lv = sizeof(v.ulval); 1896 v.ulval = sk->sk_max_pacing_rate; 1897 } else { 1898 /* 32bit version */ 1899 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1900 } 1901 break; 1902 1903 case SO_INCOMING_CPU: 1904 v.val = READ_ONCE(sk->sk_incoming_cpu); 1905 break; 1906 1907 case SO_MEMINFO: 1908 { 1909 u32 meminfo[SK_MEMINFO_VARS]; 1910 1911 sk_get_meminfo(sk, meminfo); 1912 1913 len = min_t(unsigned int, len, sizeof(meminfo)); 1914 if (copy_to_sockptr(optval, &meminfo, len)) 1915 return -EFAULT; 1916 1917 goto lenout; 1918 } 1919 1920 #ifdef CONFIG_NET_RX_BUSY_POLL 1921 case SO_INCOMING_NAPI_ID: 1922 v.val = READ_ONCE(sk->sk_napi_id); 1923 1924 /* aggregate non-NAPI IDs down to 0 */ 1925 if (v.val < MIN_NAPI_ID) 1926 v.val = 0; 1927 1928 break; 1929 #endif 1930 1931 case SO_COOKIE: 1932 lv = sizeof(u64); 1933 if (len < lv) 1934 return -EINVAL; 1935 v.val64 = sock_gen_cookie(sk); 1936 break; 1937 1938 case SO_ZEROCOPY: 1939 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1940 break; 1941 1942 case SO_TXTIME: 1943 lv = sizeof(v.txtime); 1944 v.txtime.clockid = sk->sk_clockid; 1945 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1946 SOF_TXTIME_DEADLINE_MODE : 0; 1947 v.txtime.flags |= sk->sk_txtime_report_errors ? 1948 SOF_TXTIME_REPORT_ERRORS : 0; 1949 break; 1950 1951 case SO_BINDTOIFINDEX: 1952 v.val = READ_ONCE(sk->sk_bound_dev_if); 1953 break; 1954 1955 case SO_NETNS_COOKIE: 1956 lv = sizeof(u64); 1957 if (len != lv) 1958 return -EINVAL; 1959 v.val64 = sock_net(sk)->net_cookie; 1960 break; 1961 1962 case SO_BUF_LOCK: 1963 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1964 break; 1965 1966 case SO_RESERVE_MEM: 1967 v.val = sk->sk_reserved_mem; 1968 break; 1969 1970 case SO_TXREHASH: 1971 v.val = sk->sk_txrehash; 1972 break; 1973 1974 default: 1975 /* We implement the SO_SNDLOWAT etc to not be settable 1976 * (1003.1g 7). 1977 */ 1978 return -ENOPROTOOPT; 1979 } 1980 1981 if (len > lv) 1982 len = lv; 1983 if (copy_to_sockptr(optval, &v, len)) 1984 return -EFAULT; 1985 lenout: 1986 if (copy_to_sockptr(optlen, &len, sizeof(int))) 1987 return -EFAULT; 1988 return 0; 1989 } 1990 1991 int sock_getsockopt(struct socket *sock, int level, int optname, 1992 char __user *optval, int __user *optlen) 1993 { 1994 return sk_getsockopt(sock->sk, level, optname, 1995 USER_SOCKPTR(optval), 1996 USER_SOCKPTR(optlen)); 1997 } 1998 1999 /* 2000 * Initialize an sk_lock. 2001 * 2002 * (We also register the sk_lock with the lock validator.) 2003 */ 2004 static inline void sock_lock_init(struct sock *sk) 2005 { 2006 if (sk->sk_kern_sock) 2007 sock_lock_init_class_and_name( 2008 sk, 2009 af_family_kern_slock_key_strings[sk->sk_family], 2010 af_family_kern_slock_keys + sk->sk_family, 2011 af_family_kern_key_strings[sk->sk_family], 2012 af_family_kern_keys + sk->sk_family); 2013 else 2014 sock_lock_init_class_and_name( 2015 sk, 2016 af_family_slock_key_strings[sk->sk_family], 2017 af_family_slock_keys + sk->sk_family, 2018 af_family_key_strings[sk->sk_family], 2019 af_family_keys + sk->sk_family); 2020 } 2021 2022 /* 2023 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2024 * even temporarly, because of RCU lookups. sk_node should also be left as is. 2025 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2026 */ 2027 static void sock_copy(struct sock *nsk, const struct sock *osk) 2028 { 2029 const struct proto *prot = READ_ONCE(osk->sk_prot); 2030 #ifdef CONFIG_SECURITY_NETWORK 2031 void *sptr = nsk->sk_security; 2032 #endif 2033 2034 /* If we move sk_tx_queue_mapping out of the private section, 2035 * we must check if sk_tx_queue_clear() is called after 2036 * sock_copy() in sk_clone_lock(). 2037 */ 2038 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2039 offsetof(struct sock, sk_dontcopy_begin) || 2040 offsetof(struct sock, sk_tx_queue_mapping) >= 2041 offsetof(struct sock, sk_dontcopy_end)); 2042 2043 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2044 2045 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2046 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2047 2048 #ifdef CONFIG_SECURITY_NETWORK 2049 nsk->sk_security = sptr; 2050 security_sk_clone(osk, nsk); 2051 #endif 2052 } 2053 2054 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2055 int family) 2056 { 2057 struct sock *sk; 2058 struct kmem_cache *slab; 2059 2060 slab = prot->slab; 2061 if (slab != NULL) { 2062 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2063 if (!sk) 2064 return sk; 2065 if (want_init_on_alloc(priority)) 2066 sk_prot_clear_nulls(sk, prot->obj_size); 2067 } else 2068 sk = kmalloc(prot->obj_size, priority); 2069 2070 if (sk != NULL) { 2071 if (security_sk_alloc(sk, family, priority)) 2072 goto out_free; 2073 2074 if (!try_module_get(prot->owner)) 2075 goto out_free_sec; 2076 } 2077 2078 return sk; 2079 2080 out_free_sec: 2081 security_sk_free(sk); 2082 out_free: 2083 if (slab != NULL) 2084 kmem_cache_free(slab, sk); 2085 else 2086 kfree(sk); 2087 return NULL; 2088 } 2089 2090 static void sk_prot_free(struct proto *prot, struct sock *sk) 2091 { 2092 struct kmem_cache *slab; 2093 struct module *owner; 2094 2095 owner = prot->owner; 2096 slab = prot->slab; 2097 2098 cgroup_sk_free(&sk->sk_cgrp_data); 2099 mem_cgroup_sk_free(sk); 2100 security_sk_free(sk); 2101 if (slab != NULL) 2102 kmem_cache_free(slab, sk); 2103 else 2104 kfree(sk); 2105 module_put(owner); 2106 } 2107 2108 /** 2109 * sk_alloc - All socket objects are allocated here 2110 * @net: the applicable net namespace 2111 * @family: protocol family 2112 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2113 * @prot: struct proto associated with this new sock instance 2114 * @kern: is this to be a kernel socket? 2115 */ 2116 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2117 struct proto *prot, int kern) 2118 { 2119 struct sock *sk; 2120 2121 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2122 if (sk) { 2123 sk->sk_family = family; 2124 /* 2125 * See comment in struct sock definition to understand 2126 * why we need sk_prot_creator -acme 2127 */ 2128 sk->sk_prot = sk->sk_prot_creator = prot; 2129 sk->sk_kern_sock = kern; 2130 sock_lock_init(sk); 2131 sk->sk_net_refcnt = kern ? 0 : 1; 2132 if (likely(sk->sk_net_refcnt)) { 2133 get_net_track(net, &sk->ns_tracker, priority); 2134 sock_inuse_add(net, 1); 2135 } else { 2136 __netns_tracker_alloc(net, &sk->ns_tracker, 2137 false, priority); 2138 } 2139 2140 sock_net_set(sk, net); 2141 refcount_set(&sk->sk_wmem_alloc, 1); 2142 2143 mem_cgroup_sk_alloc(sk); 2144 cgroup_sk_alloc(&sk->sk_cgrp_data); 2145 sock_update_classid(&sk->sk_cgrp_data); 2146 sock_update_netprioidx(&sk->sk_cgrp_data); 2147 sk_tx_queue_clear(sk); 2148 } 2149 2150 return sk; 2151 } 2152 EXPORT_SYMBOL(sk_alloc); 2153 2154 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2155 * grace period. This is the case for UDP sockets and TCP listeners. 2156 */ 2157 static void __sk_destruct(struct rcu_head *head) 2158 { 2159 struct sock *sk = container_of(head, struct sock, sk_rcu); 2160 struct sk_filter *filter; 2161 2162 if (sk->sk_destruct) 2163 sk->sk_destruct(sk); 2164 2165 filter = rcu_dereference_check(sk->sk_filter, 2166 refcount_read(&sk->sk_wmem_alloc) == 0); 2167 if (filter) { 2168 sk_filter_uncharge(sk, filter); 2169 RCU_INIT_POINTER(sk->sk_filter, NULL); 2170 } 2171 2172 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2173 2174 #ifdef CONFIG_BPF_SYSCALL 2175 bpf_sk_storage_free(sk); 2176 #endif 2177 2178 if (atomic_read(&sk->sk_omem_alloc)) 2179 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2180 __func__, atomic_read(&sk->sk_omem_alloc)); 2181 2182 if (sk->sk_frag.page) { 2183 put_page(sk->sk_frag.page); 2184 sk->sk_frag.page = NULL; 2185 } 2186 2187 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2188 put_cred(sk->sk_peer_cred); 2189 put_pid(sk->sk_peer_pid); 2190 2191 if (likely(sk->sk_net_refcnt)) 2192 put_net_track(sock_net(sk), &sk->ns_tracker); 2193 else 2194 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2195 2196 sk_prot_free(sk->sk_prot_creator, sk); 2197 } 2198 2199 void sk_destruct(struct sock *sk) 2200 { 2201 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2202 2203 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2204 reuseport_detach_sock(sk); 2205 use_call_rcu = true; 2206 } 2207 2208 if (use_call_rcu) 2209 call_rcu(&sk->sk_rcu, __sk_destruct); 2210 else 2211 __sk_destruct(&sk->sk_rcu); 2212 } 2213 2214 static void __sk_free(struct sock *sk) 2215 { 2216 if (likely(sk->sk_net_refcnt)) 2217 sock_inuse_add(sock_net(sk), -1); 2218 2219 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2220 sock_diag_broadcast_destroy(sk); 2221 else 2222 sk_destruct(sk); 2223 } 2224 2225 void sk_free(struct sock *sk) 2226 { 2227 /* 2228 * We subtract one from sk_wmem_alloc and can know if 2229 * some packets are still in some tx queue. 2230 * If not null, sock_wfree() will call __sk_free(sk) later 2231 */ 2232 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2233 __sk_free(sk); 2234 } 2235 EXPORT_SYMBOL(sk_free); 2236 2237 static void sk_init_common(struct sock *sk) 2238 { 2239 skb_queue_head_init(&sk->sk_receive_queue); 2240 skb_queue_head_init(&sk->sk_write_queue); 2241 skb_queue_head_init(&sk->sk_error_queue); 2242 2243 rwlock_init(&sk->sk_callback_lock); 2244 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2245 af_rlock_keys + sk->sk_family, 2246 af_family_rlock_key_strings[sk->sk_family]); 2247 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2248 af_wlock_keys + sk->sk_family, 2249 af_family_wlock_key_strings[sk->sk_family]); 2250 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2251 af_elock_keys + sk->sk_family, 2252 af_family_elock_key_strings[sk->sk_family]); 2253 lockdep_set_class_and_name(&sk->sk_callback_lock, 2254 af_callback_keys + sk->sk_family, 2255 af_family_clock_key_strings[sk->sk_family]); 2256 } 2257 2258 /** 2259 * sk_clone_lock - clone a socket, and lock its clone 2260 * @sk: the socket to clone 2261 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2262 * 2263 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2264 */ 2265 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2266 { 2267 struct proto *prot = READ_ONCE(sk->sk_prot); 2268 struct sk_filter *filter; 2269 bool is_charged = true; 2270 struct sock *newsk; 2271 2272 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2273 if (!newsk) 2274 goto out; 2275 2276 sock_copy(newsk, sk); 2277 2278 newsk->sk_prot_creator = prot; 2279 2280 /* SANITY */ 2281 if (likely(newsk->sk_net_refcnt)) { 2282 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2283 sock_inuse_add(sock_net(newsk), 1); 2284 } else { 2285 /* Kernel sockets are not elevating the struct net refcount. 2286 * Instead, use a tracker to more easily detect if a layer 2287 * is not properly dismantling its kernel sockets at netns 2288 * destroy time. 2289 */ 2290 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2291 false, priority); 2292 } 2293 sk_node_init(&newsk->sk_node); 2294 sock_lock_init(newsk); 2295 bh_lock_sock(newsk); 2296 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2297 newsk->sk_backlog.len = 0; 2298 2299 atomic_set(&newsk->sk_rmem_alloc, 0); 2300 2301 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2302 refcount_set(&newsk->sk_wmem_alloc, 1); 2303 2304 atomic_set(&newsk->sk_omem_alloc, 0); 2305 sk_init_common(newsk); 2306 2307 newsk->sk_dst_cache = NULL; 2308 newsk->sk_dst_pending_confirm = 0; 2309 newsk->sk_wmem_queued = 0; 2310 newsk->sk_forward_alloc = 0; 2311 newsk->sk_reserved_mem = 0; 2312 atomic_set(&newsk->sk_drops, 0); 2313 newsk->sk_send_head = NULL; 2314 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2315 atomic_set(&newsk->sk_zckey, 0); 2316 2317 sock_reset_flag(newsk, SOCK_DONE); 2318 2319 /* sk->sk_memcg will be populated at accept() time */ 2320 newsk->sk_memcg = NULL; 2321 2322 cgroup_sk_clone(&newsk->sk_cgrp_data); 2323 2324 rcu_read_lock(); 2325 filter = rcu_dereference(sk->sk_filter); 2326 if (filter != NULL) 2327 /* though it's an empty new sock, the charging may fail 2328 * if sysctl_optmem_max was changed between creation of 2329 * original socket and cloning 2330 */ 2331 is_charged = sk_filter_charge(newsk, filter); 2332 RCU_INIT_POINTER(newsk->sk_filter, filter); 2333 rcu_read_unlock(); 2334 2335 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2336 /* We need to make sure that we don't uncharge the new 2337 * socket if we couldn't charge it in the first place 2338 * as otherwise we uncharge the parent's filter. 2339 */ 2340 if (!is_charged) 2341 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2342 sk_free_unlock_clone(newsk); 2343 newsk = NULL; 2344 goto out; 2345 } 2346 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2347 2348 if (bpf_sk_storage_clone(sk, newsk)) { 2349 sk_free_unlock_clone(newsk); 2350 newsk = NULL; 2351 goto out; 2352 } 2353 2354 /* Clear sk_user_data if parent had the pointer tagged 2355 * as not suitable for copying when cloning. 2356 */ 2357 if (sk_user_data_is_nocopy(newsk)) 2358 newsk->sk_user_data = NULL; 2359 2360 newsk->sk_err = 0; 2361 newsk->sk_err_soft = 0; 2362 newsk->sk_priority = 0; 2363 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2364 2365 /* Before updating sk_refcnt, we must commit prior changes to memory 2366 * (Documentation/RCU/rculist_nulls.rst for details) 2367 */ 2368 smp_wmb(); 2369 refcount_set(&newsk->sk_refcnt, 2); 2370 2371 sk_set_socket(newsk, NULL); 2372 sk_tx_queue_clear(newsk); 2373 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2374 2375 if (newsk->sk_prot->sockets_allocated) 2376 sk_sockets_allocated_inc(newsk); 2377 2378 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2379 net_enable_timestamp(); 2380 out: 2381 return newsk; 2382 } 2383 EXPORT_SYMBOL_GPL(sk_clone_lock); 2384 2385 void sk_free_unlock_clone(struct sock *sk) 2386 { 2387 /* It is still raw copy of parent, so invalidate 2388 * destructor and make plain sk_free() */ 2389 sk->sk_destruct = NULL; 2390 bh_unlock_sock(sk); 2391 sk_free(sk); 2392 } 2393 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2394 2395 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2396 { 2397 bool is_ipv6 = false; 2398 u32 max_size; 2399 2400 #if IS_ENABLED(CONFIG_IPV6) 2401 is_ipv6 = (sk->sk_family == AF_INET6 && 2402 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2403 #endif 2404 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2405 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2406 READ_ONCE(dst->dev->gso_ipv4_max_size); 2407 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2408 max_size = GSO_LEGACY_MAX_SIZE; 2409 2410 return max_size - (MAX_TCP_HEADER + 1); 2411 } 2412 2413 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2414 { 2415 u32 max_segs = 1; 2416 2417 sk->sk_route_caps = dst->dev->features; 2418 if (sk_is_tcp(sk)) 2419 sk->sk_route_caps |= NETIF_F_GSO; 2420 if (sk->sk_route_caps & NETIF_F_GSO) 2421 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2422 if (unlikely(sk->sk_gso_disabled)) 2423 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2424 if (sk_can_gso(sk)) { 2425 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2426 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2427 } else { 2428 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2429 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2430 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2431 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2432 } 2433 } 2434 sk->sk_gso_max_segs = max_segs; 2435 sk_dst_set(sk, dst); 2436 } 2437 EXPORT_SYMBOL_GPL(sk_setup_caps); 2438 2439 /* 2440 * Simple resource managers for sockets. 2441 */ 2442 2443 2444 /* 2445 * Write buffer destructor automatically called from kfree_skb. 2446 */ 2447 void sock_wfree(struct sk_buff *skb) 2448 { 2449 struct sock *sk = skb->sk; 2450 unsigned int len = skb->truesize; 2451 bool free; 2452 2453 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2454 if (sock_flag(sk, SOCK_RCU_FREE) && 2455 sk->sk_write_space == sock_def_write_space) { 2456 rcu_read_lock(); 2457 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2458 sock_def_write_space_wfree(sk); 2459 rcu_read_unlock(); 2460 if (unlikely(free)) 2461 __sk_free(sk); 2462 return; 2463 } 2464 2465 /* 2466 * Keep a reference on sk_wmem_alloc, this will be released 2467 * after sk_write_space() call 2468 */ 2469 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2470 sk->sk_write_space(sk); 2471 len = 1; 2472 } 2473 /* 2474 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2475 * could not do because of in-flight packets 2476 */ 2477 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2478 __sk_free(sk); 2479 } 2480 EXPORT_SYMBOL(sock_wfree); 2481 2482 /* This variant of sock_wfree() is used by TCP, 2483 * since it sets SOCK_USE_WRITE_QUEUE. 2484 */ 2485 void __sock_wfree(struct sk_buff *skb) 2486 { 2487 struct sock *sk = skb->sk; 2488 2489 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2490 __sk_free(sk); 2491 } 2492 2493 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2494 { 2495 skb_orphan(skb); 2496 skb->sk = sk; 2497 #ifdef CONFIG_INET 2498 if (unlikely(!sk_fullsock(sk))) { 2499 skb->destructor = sock_edemux; 2500 sock_hold(sk); 2501 return; 2502 } 2503 #endif 2504 skb->destructor = sock_wfree; 2505 skb_set_hash_from_sk(skb, sk); 2506 /* 2507 * We used to take a refcount on sk, but following operation 2508 * is enough to guarantee sk_free() wont free this sock until 2509 * all in-flight packets are completed 2510 */ 2511 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2512 } 2513 EXPORT_SYMBOL(skb_set_owner_w); 2514 2515 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2516 { 2517 #ifdef CONFIG_TLS_DEVICE 2518 /* Drivers depend on in-order delivery for crypto offload, 2519 * partial orphan breaks out-of-order-OK logic. 2520 */ 2521 if (skb->decrypted) 2522 return false; 2523 #endif 2524 return (skb->destructor == sock_wfree || 2525 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2526 } 2527 2528 /* This helper is used by netem, as it can hold packets in its 2529 * delay queue. We want to allow the owner socket to send more 2530 * packets, as if they were already TX completed by a typical driver. 2531 * But we also want to keep skb->sk set because some packet schedulers 2532 * rely on it (sch_fq for example). 2533 */ 2534 void skb_orphan_partial(struct sk_buff *skb) 2535 { 2536 if (skb_is_tcp_pure_ack(skb)) 2537 return; 2538 2539 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2540 return; 2541 2542 skb_orphan(skb); 2543 } 2544 EXPORT_SYMBOL(skb_orphan_partial); 2545 2546 /* 2547 * Read buffer destructor automatically called from kfree_skb. 2548 */ 2549 void sock_rfree(struct sk_buff *skb) 2550 { 2551 struct sock *sk = skb->sk; 2552 unsigned int len = skb->truesize; 2553 2554 atomic_sub(len, &sk->sk_rmem_alloc); 2555 sk_mem_uncharge(sk, len); 2556 } 2557 EXPORT_SYMBOL(sock_rfree); 2558 2559 /* 2560 * Buffer destructor for skbs that are not used directly in read or write 2561 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2562 */ 2563 void sock_efree(struct sk_buff *skb) 2564 { 2565 sock_put(skb->sk); 2566 } 2567 EXPORT_SYMBOL(sock_efree); 2568 2569 /* Buffer destructor for prefetch/receive path where reference count may 2570 * not be held, e.g. for listen sockets. 2571 */ 2572 #ifdef CONFIG_INET 2573 void sock_pfree(struct sk_buff *skb) 2574 { 2575 if (sk_is_refcounted(skb->sk)) 2576 sock_gen_put(skb->sk); 2577 } 2578 EXPORT_SYMBOL(sock_pfree); 2579 #endif /* CONFIG_INET */ 2580 2581 kuid_t sock_i_uid(struct sock *sk) 2582 { 2583 kuid_t uid; 2584 2585 read_lock_bh(&sk->sk_callback_lock); 2586 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2587 read_unlock_bh(&sk->sk_callback_lock); 2588 return uid; 2589 } 2590 EXPORT_SYMBOL(sock_i_uid); 2591 2592 unsigned long __sock_i_ino(struct sock *sk) 2593 { 2594 unsigned long ino; 2595 2596 read_lock(&sk->sk_callback_lock); 2597 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2598 read_unlock(&sk->sk_callback_lock); 2599 return ino; 2600 } 2601 EXPORT_SYMBOL(__sock_i_ino); 2602 2603 unsigned long sock_i_ino(struct sock *sk) 2604 { 2605 unsigned long ino; 2606 2607 local_bh_disable(); 2608 ino = __sock_i_ino(sk); 2609 local_bh_enable(); 2610 return ino; 2611 } 2612 EXPORT_SYMBOL(sock_i_ino); 2613 2614 /* 2615 * Allocate a skb from the socket's send buffer. 2616 */ 2617 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2618 gfp_t priority) 2619 { 2620 if (force || 2621 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2622 struct sk_buff *skb = alloc_skb(size, priority); 2623 2624 if (skb) { 2625 skb_set_owner_w(skb, sk); 2626 return skb; 2627 } 2628 } 2629 return NULL; 2630 } 2631 EXPORT_SYMBOL(sock_wmalloc); 2632 2633 static void sock_ofree(struct sk_buff *skb) 2634 { 2635 struct sock *sk = skb->sk; 2636 2637 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2638 } 2639 2640 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2641 gfp_t priority) 2642 { 2643 struct sk_buff *skb; 2644 2645 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2646 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2647 READ_ONCE(sysctl_optmem_max)) 2648 return NULL; 2649 2650 skb = alloc_skb(size, priority); 2651 if (!skb) 2652 return NULL; 2653 2654 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2655 skb->sk = sk; 2656 skb->destructor = sock_ofree; 2657 return skb; 2658 } 2659 2660 /* 2661 * Allocate a memory block from the socket's option memory buffer. 2662 */ 2663 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2664 { 2665 int optmem_max = READ_ONCE(sysctl_optmem_max); 2666 2667 if ((unsigned int)size <= optmem_max && 2668 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2669 void *mem; 2670 /* First do the add, to avoid the race if kmalloc 2671 * might sleep. 2672 */ 2673 atomic_add(size, &sk->sk_omem_alloc); 2674 mem = kmalloc(size, priority); 2675 if (mem) 2676 return mem; 2677 atomic_sub(size, &sk->sk_omem_alloc); 2678 } 2679 return NULL; 2680 } 2681 EXPORT_SYMBOL(sock_kmalloc); 2682 2683 /* Free an option memory block. Note, we actually want the inline 2684 * here as this allows gcc to detect the nullify and fold away the 2685 * condition entirely. 2686 */ 2687 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2688 const bool nullify) 2689 { 2690 if (WARN_ON_ONCE(!mem)) 2691 return; 2692 if (nullify) 2693 kfree_sensitive(mem); 2694 else 2695 kfree(mem); 2696 atomic_sub(size, &sk->sk_omem_alloc); 2697 } 2698 2699 void sock_kfree_s(struct sock *sk, void *mem, int size) 2700 { 2701 __sock_kfree_s(sk, mem, size, false); 2702 } 2703 EXPORT_SYMBOL(sock_kfree_s); 2704 2705 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2706 { 2707 __sock_kfree_s(sk, mem, size, true); 2708 } 2709 EXPORT_SYMBOL(sock_kzfree_s); 2710 2711 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2712 I think, these locks should be removed for datagram sockets. 2713 */ 2714 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2715 { 2716 DEFINE_WAIT(wait); 2717 2718 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2719 for (;;) { 2720 if (!timeo) 2721 break; 2722 if (signal_pending(current)) 2723 break; 2724 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2725 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2726 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2727 break; 2728 if (sk->sk_shutdown & SEND_SHUTDOWN) 2729 break; 2730 if (sk->sk_err) 2731 break; 2732 timeo = schedule_timeout(timeo); 2733 } 2734 finish_wait(sk_sleep(sk), &wait); 2735 return timeo; 2736 } 2737 2738 2739 /* 2740 * Generic send/receive buffer handlers 2741 */ 2742 2743 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2744 unsigned long data_len, int noblock, 2745 int *errcode, int max_page_order) 2746 { 2747 struct sk_buff *skb; 2748 long timeo; 2749 int err; 2750 2751 timeo = sock_sndtimeo(sk, noblock); 2752 for (;;) { 2753 err = sock_error(sk); 2754 if (err != 0) 2755 goto failure; 2756 2757 err = -EPIPE; 2758 if (sk->sk_shutdown & SEND_SHUTDOWN) 2759 goto failure; 2760 2761 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2762 break; 2763 2764 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2765 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2766 err = -EAGAIN; 2767 if (!timeo) 2768 goto failure; 2769 if (signal_pending(current)) 2770 goto interrupted; 2771 timeo = sock_wait_for_wmem(sk, timeo); 2772 } 2773 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2774 errcode, sk->sk_allocation); 2775 if (skb) 2776 skb_set_owner_w(skb, sk); 2777 return skb; 2778 2779 interrupted: 2780 err = sock_intr_errno(timeo); 2781 failure: 2782 *errcode = err; 2783 return NULL; 2784 } 2785 EXPORT_SYMBOL(sock_alloc_send_pskb); 2786 2787 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2788 struct sockcm_cookie *sockc) 2789 { 2790 u32 tsflags; 2791 2792 switch (cmsg->cmsg_type) { 2793 case SO_MARK: 2794 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2795 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2796 return -EPERM; 2797 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2798 return -EINVAL; 2799 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2800 break; 2801 case SO_TIMESTAMPING_OLD: 2802 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2803 return -EINVAL; 2804 2805 tsflags = *(u32 *)CMSG_DATA(cmsg); 2806 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2807 return -EINVAL; 2808 2809 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2810 sockc->tsflags |= tsflags; 2811 break; 2812 case SCM_TXTIME: 2813 if (!sock_flag(sk, SOCK_TXTIME)) 2814 return -EINVAL; 2815 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2816 return -EINVAL; 2817 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2818 break; 2819 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2820 case SCM_RIGHTS: 2821 case SCM_CREDENTIALS: 2822 break; 2823 default: 2824 return -EINVAL; 2825 } 2826 return 0; 2827 } 2828 EXPORT_SYMBOL(__sock_cmsg_send); 2829 2830 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2831 struct sockcm_cookie *sockc) 2832 { 2833 struct cmsghdr *cmsg; 2834 int ret; 2835 2836 for_each_cmsghdr(cmsg, msg) { 2837 if (!CMSG_OK(msg, cmsg)) 2838 return -EINVAL; 2839 if (cmsg->cmsg_level != SOL_SOCKET) 2840 continue; 2841 ret = __sock_cmsg_send(sk, cmsg, sockc); 2842 if (ret) 2843 return ret; 2844 } 2845 return 0; 2846 } 2847 EXPORT_SYMBOL(sock_cmsg_send); 2848 2849 static void sk_enter_memory_pressure(struct sock *sk) 2850 { 2851 if (!sk->sk_prot->enter_memory_pressure) 2852 return; 2853 2854 sk->sk_prot->enter_memory_pressure(sk); 2855 } 2856 2857 static void sk_leave_memory_pressure(struct sock *sk) 2858 { 2859 if (sk->sk_prot->leave_memory_pressure) { 2860 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2861 tcp_leave_memory_pressure, sk); 2862 } else { 2863 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2864 2865 if (memory_pressure && READ_ONCE(*memory_pressure)) 2866 WRITE_ONCE(*memory_pressure, 0); 2867 } 2868 } 2869 2870 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2871 2872 /** 2873 * skb_page_frag_refill - check that a page_frag contains enough room 2874 * @sz: minimum size of the fragment we want to get 2875 * @pfrag: pointer to page_frag 2876 * @gfp: priority for memory allocation 2877 * 2878 * Note: While this allocator tries to use high order pages, there is 2879 * no guarantee that allocations succeed. Therefore, @sz MUST be 2880 * less or equal than PAGE_SIZE. 2881 */ 2882 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2883 { 2884 if (pfrag->page) { 2885 if (page_ref_count(pfrag->page) == 1) { 2886 pfrag->offset = 0; 2887 return true; 2888 } 2889 if (pfrag->offset + sz <= pfrag->size) 2890 return true; 2891 put_page(pfrag->page); 2892 } 2893 2894 pfrag->offset = 0; 2895 if (SKB_FRAG_PAGE_ORDER && 2896 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2897 /* Avoid direct reclaim but allow kswapd to wake */ 2898 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2899 __GFP_COMP | __GFP_NOWARN | 2900 __GFP_NORETRY, 2901 SKB_FRAG_PAGE_ORDER); 2902 if (likely(pfrag->page)) { 2903 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2904 return true; 2905 } 2906 } 2907 pfrag->page = alloc_page(gfp); 2908 if (likely(pfrag->page)) { 2909 pfrag->size = PAGE_SIZE; 2910 return true; 2911 } 2912 return false; 2913 } 2914 EXPORT_SYMBOL(skb_page_frag_refill); 2915 2916 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2917 { 2918 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2919 return true; 2920 2921 sk_enter_memory_pressure(sk); 2922 sk_stream_moderate_sndbuf(sk); 2923 return false; 2924 } 2925 EXPORT_SYMBOL(sk_page_frag_refill); 2926 2927 void __lock_sock(struct sock *sk) 2928 __releases(&sk->sk_lock.slock) 2929 __acquires(&sk->sk_lock.slock) 2930 { 2931 DEFINE_WAIT(wait); 2932 2933 for (;;) { 2934 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2935 TASK_UNINTERRUPTIBLE); 2936 spin_unlock_bh(&sk->sk_lock.slock); 2937 schedule(); 2938 spin_lock_bh(&sk->sk_lock.slock); 2939 if (!sock_owned_by_user(sk)) 2940 break; 2941 } 2942 finish_wait(&sk->sk_lock.wq, &wait); 2943 } 2944 2945 void __release_sock(struct sock *sk) 2946 __releases(&sk->sk_lock.slock) 2947 __acquires(&sk->sk_lock.slock) 2948 { 2949 struct sk_buff *skb, *next; 2950 2951 while ((skb = sk->sk_backlog.head) != NULL) { 2952 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2953 2954 spin_unlock_bh(&sk->sk_lock.slock); 2955 2956 do { 2957 next = skb->next; 2958 prefetch(next); 2959 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2960 skb_mark_not_on_list(skb); 2961 sk_backlog_rcv(sk, skb); 2962 2963 cond_resched(); 2964 2965 skb = next; 2966 } while (skb != NULL); 2967 2968 spin_lock_bh(&sk->sk_lock.slock); 2969 } 2970 2971 /* 2972 * Doing the zeroing here guarantee we can not loop forever 2973 * while a wild producer attempts to flood us. 2974 */ 2975 sk->sk_backlog.len = 0; 2976 } 2977 2978 void __sk_flush_backlog(struct sock *sk) 2979 { 2980 spin_lock_bh(&sk->sk_lock.slock); 2981 __release_sock(sk); 2982 spin_unlock_bh(&sk->sk_lock.slock); 2983 } 2984 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2985 2986 /** 2987 * sk_wait_data - wait for data to arrive at sk_receive_queue 2988 * @sk: sock to wait on 2989 * @timeo: for how long 2990 * @skb: last skb seen on sk_receive_queue 2991 * 2992 * Now socket state including sk->sk_err is changed only under lock, 2993 * hence we may omit checks after joining wait queue. 2994 * We check receive queue before schedule() only as optimization; 2995 * it is very likely that release_sock() added new data. 2996 */ 2997 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2998 { 2999 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3000 int rc; 3001 3002 add_wait_queue(sk_sleep(sk), &wait); 3003 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3004 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3005 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3006 remove_wait_queue(sk_sleep(sk), &wait); 3007 return rc; 3008 } 3009 EXPORT_SYMBOL(sk_wait_data); 3010 3011 /** 3012 * __sk_mem_raise_allocated - increase memory_allocated 3013 * @sk: socket 3014 * @size: memory size to allocate 3015 * @amt: pages to allocate 3016 * @kind: allocation type 3017 * 3018 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 3019 */ 3020 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3021 { 3022 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 3023 struct proto *prot = sk->sk_prot; 3024 bool charged = true; 3025 long allocated; 3026 3027 sk_memory_allocated_add(sk, amt); 3028 allocated = sk_memory_allocated(sk); 3029 if (memcg_charge && 3030 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3031 gfp_memcg_charge()))) 3032 goto suppress_allocation; 3033 3034 /* Under limit. */ 3035 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3036 sk_leave_memory_pressure(sk); 3037 return 1; 3038 } 3039 3040 /* Under pressure. */ 3041 if (allocated > sk_prot_mem_limits(sk, 1)) 3042 sk_enter_memory_pressure(sk); 3043 3044 /* Over hard limit. */ 3045 if (allocated > sk_prot_mem_limits(sk, 2)) 3046 goto suppress_allocation; 3047 3048 /* guarantee minimum buffer size under pressure */ 3049 if (kind == SK_MEM_RECV) { 3050 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3051 return 1; 3052 3053 } else { /* SK_MEM_SEND */ 3054 int wmem0 = sk_get_wmem0(sk, prot); 3055 3056 if (sk->sk_type == SOCK_STREAM) { 3057 if (sk->sk_wmem_queued < wmem0) 3058 return 1; 3059 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3060 return 1; 3061 } 3062 } 3063 3064 if (sk_has_memory_pressure(sk)) { 3065 u64 alloc; 3066 3067 if (!sk_under_memory_pressure(sk)) 3068 return 1; 3069 alloc = sk_sockets_allocated_read_positive(sk); 3070 if (sk_prot_mem_limits(sk, 2) > alloc * 3071 sk_mem_pages(sk->sk_wmem_queued + 3072 atomic_read(&sk->sk_rmem_alloc) + 3073 sk->sk_forward_alloc)) 3074 return 1; 3075 } 3076 3077 suppress_allocation: 3078 3079 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3080 sk_stream_moderate_sndbuf(sk); 3081 3082 /* Fail only if socket is _under_ its sndbuf. 3083 * In this case we cannot block, so that we have to fail. 3084 */ 3085 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3086 /* Force charge with __GFP_NOFAIL */ 3087 if (memcg_charge && !charged) { 3088 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3089 gfp_memcg_charge() | __GFP_NOFAIL); 3090 } 3091 return 1; 3092 } 3093 } 3094 3095 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3096 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3097 3098 sk_memory_allocated_sub(sk, amt); 3099 3100 if (memcg_charge && charged) 3101 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3102 3103 return 0; 3104 } 3105 3106 /** 3107 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3108 * @sk: socket 3109 * @size: memory size to allocate 3110 * @kind: allocation type 3111 * 3112 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3113 * rmem allocation. This function assumes that protocols which have 3114 * memory_pressure use sk_wmem_queued as write buffer accounting. 3115 */ 3116 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3117 { 3118 int ret, amt = sk_mem_pages(size); 3119 3120 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3121 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3122 if (!ret) 3123 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3124 return ret; 3125 } 3126 EXPORT_SYMBOL(__sk_mem_schedule); 3127 3128 /** 3129 * __sk_mem_reduce_allocated - reclaim memory_allocated 3130 * @sk: socket 3131 * @amount: number of quanta 3132 * 3133 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3134 */ 3135 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3136 { 3137 sk_memory_allocated_sub(sk, amount); 3138 3139 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3140 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3141 3142 if (sk_under_memory_pressure(sk) && 3143 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3144 sk_leave_memory_pressure(sk); 3145 } 3146 3147 /** 3148 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3149 * @sk: socket 3150 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3151 */ 3152 void __sk_mem_reclaim(struct sock *sk, int amount) 3153 { 3154 amount >>= PAGE_SHIFT; 3155 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3156 __sk_mem_reduce_allocated(sk, amount); 3157 } 3158 EXPORT_SYMBOL(__sk_mem_reclaim); 3159 3160 int sk_set_peek_off(struct sock *sk, int val) 3161 { 3162 sk->sk_peek_off = val; 3163 return 0; 3164 } 3165 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3166 3167 /* 3168 * Set of default routines for initialising struct proto_ops when 3169 * the protocol does not support a particular function. In certain 3170 * cases where it makes no sense for a protocol to have a "do nothing" 3171 * function, some default processing is provided. 3172 */ 3173 3174 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3175 { 3176 return -EOPNOTSUPP; 3177 } 3178 EXPORT_SYMBOL(sock_no_bind); 3179 3180 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3181 int len, int flags) 3182 { 3183 return -EOPNOTSUPP; 3184 } 3185 EXPORT_SYMBOL(sock_no_connect); 3186 3187 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3188 { 3189 return -EOPNOTSUPP; 3190 } 3191 EXPORT_SYMBOL(sock_no_socketpair); 3192 3193 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3194 bool kern) 3195 { 3196 return -EOPNOTSUPP; 3197 } 3198 EXPORT_SYMBOL(sock_no_accept); 3199 3200 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3201 int peer) 3202 { 3203 return -EOPNOTSUPP; 3204 } 3205 EXPORT_SYMBOL(sock_no_getname); 3206 3207 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3208 { 3209 return -EOPNOTSUPP; 3210 } 3211 EXPORT_SYMBOL(sock_no_ioctl); 3212 3213 int sock_no_listen(struct socket *sock, int backlog) 3214 { 3215 return -EOPNOTSUPP; 3216 } 3217 EXPORT_SYMBOL(sock_no_listen); 3218 3219 int sock_no_shutdown(struct socket *sock, int how) 3220 { 3221 return -EOPNOTSUPP; 3222 } 3223 EXPORT_SYMBOL(sock_no_shutdown); 3224 3225 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3226 { 3227 return -EOPNOTSUPP; 3228 } 3229 EXPORT_SYMBOL(sock_no_sendmsg); 3230 3231 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3232 { 3233 return -EOPNOTSUPP; 3234 } 3235 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3236 3237 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3238 int flags) 3239 { 3240 return -EOPNOTSUPP; 3241 } 3242 EXPORT_SYMBOL(sock_no_recvmsg); 3243 3244 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3245 { 3246 /* Mirror missing mmap method error code */ 3247 return -ENODEV; 3248 } 3249 EXPORT_SYMBOL(sock_no_mmap); 3250 3251 /* 3252 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3253 * various sock-based usage counts. 3254 */ 3255 void __receive_sock(struct file *file) 3256 { 3257 struct socket *sock; 3258 3259 sock = sock_from_file(file); 3260 if (sock) { 3261 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3262 sock_update_classid(&sock->sk->sk_cgrp_data); 3263 } 3264 } 3265 3266 /* 3267 * Default Socket Callbacks 3268 */ 3269 3270 static void sock_def_wakeup(struct sock *sk) 3271 { 3272 struct socket_wq *wq; 3273 3274 rcu_read_lock(); 3275 wq = rcu_dereference(sk->sk_wq); 3276 if (skwq_has_sleeper(wq)) 3277 wake_up_interruptible_all(&wq->wait); 3278 rcu_read_unlock(); 3279 } 3280 3281 static void sock_def_error_report(struct sock *sk) 3282 { 3283 struct socket_wq *wq; 3284 3285 rcu_read_lock(); 3286 wq = rcu_dereference(sk->sk_wq); 3287 if (skwq_has_sleeper(wq)) 3288 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3289 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3290 rcu_read_unlock(); 3291 } 3292 3293 void sock_def_readable(struct sock *sk) 3294 { 3295 struct socket_wq *wq; 3296 3297 trace_sk_data_ready(sk); 3298 3299 rcu_read_lock(); 3300 wq = rcu_dereference(sk->sk_wq); 3301 if (skwq_has_sleeper(wq)) 3302 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3303 EPOLLRDNORM | EPOLLRDBAND); 3304 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3305 rcu_read_unlock(); 3306 } 3307 3308 static void sock_def_write_space(struct sock *sk) 3309 { 3310 struct socket_wq *wq; 3311 3312 rcu_read_lock(); 3313 3314 /* Do not wake up a writer until he can make "significant" 3315 * progress. --DaveM 3316 */ 3317 if (sock_writeable(sk)) { 3318 wq = rcu_dereference(sk->sk_wq); 3319 if (skwq_has_sleeper(wq)) 3320 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3321 EPOLLWRNORM | EPOLLWRBAND); 3322 3323 /* Should agree with poll, otherwise some programs break */ 3324 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3325 } 3326 3327 rcu_read_unlock(); 3328 } 3329 3330 /* An optimised version of sock_def_write_space(), should only be called 3331 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3332 * ->sk_wmem_alloc. 3333 */ 3334 static void sock_def_write_space_wfree(struct sock *sk) 3335 { 3336 /* Do not wake up a writer until he can make "significant" 3337 * progress. --DaveM 3338 */ 3339 if (sock_writeable(sk)) { 3340 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3341 3342 /* rely on refcount_sub from sock_wfree() */ 3343 smp_mb__after_atomic(); 3344 if (wq && waitqueue_active(&wq->wait)) 3345 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3346 EPOLLWRNORM | EPOLLWRBAND); 3347 3348 /* Should agree with poll, otherwise some programs break */ 3349 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3350 } 3351 } 3352 3353 static void sock_def_destruct(struct sock *sk) 3354 { 3355 } 3356 3357 void sk_send_sigurg(struct sock *sk) 3358 { 3359 if (sk->sk_socket && sk->sk_socket->file) 3360 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3361 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3362 } 3363 EXPORT_SYMBOL(sk_send_sigurg); 3364 3365 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3366 unsigned long expires) 3367 { 3368 if (!mod_timer(timer, expires)) 3369 sock_hold(sk); 3370 } 3371 EXPORT_SYMBOL(sk_reset_timer); 3372 3373 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3374 { 3375 if (del_timer(timer)) 3376 __sock_put(sk); 3377 } 3378 EXPORT_SYMBOL(sk_stop_timer); 3379 3380 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3381 { 3382 if (del_timer_sync(timer)) 3383 __sock_put(sk); 3384 } 3385 EXPORT_SYMBOL(sk_stop_timer_sync); 3386 3387 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3388 { 3389 sk_init_common(sk); 3390 sk->sk_send_head = NULL; 3391 3392 timer_setup(&sk->sk_timer, NULL, 0); 3393 3394 sk->sk_allocation = GFP_KERNEL; 3395 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3396 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3397 sk->sk_state = TCP_CLOSE; 3398 sk->sk_use_task_frag = true; 3399 sk_set_socket(sk, sock); 3400 3401 sock_set_flag(sk, SOCK_ZAPPED); 3402 3403 if (sock) { 3404 sk->sk_type = sock->type; 3405 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3406 sock->sk = sk; 3407 } else { 3408 RCU_INIT_POINTER(sk->sk_wq, NULL); 3409 } 3410 sk->sk_uid = uid; 3411 3412 rwlock_init(&sk->sk_callback_lock); 3413 if (sk->sk_kern_sock) 3414 lockdep_set_class_and_name( 3415 &sk->sk_callback_lock, 3416 af_kern_callback_keys + sk->sk_family, 3417 af_family_kern_clock_key_strings[sk->sk_family]); 3418 else 3419 lockdep_set_class_and_name( 3420 &sk->sk_callback_lock, 3421 af_callback_keys + sk->sk_family, 3422 af_family_clock_key_strings[sk->sk_family]); 3423 3424 sk->sk_state_change = sock_def_wakeup; 3425 sk->sk_data_ready = sock_def_readable; 3426 sk->sk_write_space = sock_def_write_space; 3427 sk->sk_error_report = sock_def_error_report; 3428 sk->sk_destruct = sock_def_destruct; 3429 3430 sk->sk_frag.page = NULL; 3431 sk->sk_frag.offset = 0; 3432 sk->sk_peek_off = -1; 3433 3434 sk->sk_peer_pid = NULL; 3435 sk->sk_peer_cred = NULL; 3436 spin_lock_init(&sk->sk_peer_lock); 3437 3438 sk->sk_write_pending = 0; 3439 sk->sk_rcvlowat = 1; 3440 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3441 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3442 3443 sk->sk_stamp = SK_DEFAULT_STAMP; 3444 #if BITS_PER_LONG==32 3445 seqlock_init(&sk->sk_stamp_seq); 3446 #endif 3447 atomic_set(&sk->sk_zckey, 0); 3448 3449 #ifdef CONFIG_NET_RX_BUSY_POLL 3450 sk->sk_napi_id = 0; 3451 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3452 #endif 3453 3454 sk->sk_max_pacing_rate = ~0UL; 3455 sk->sk_pacing_rate = ~0UL; 3456 WRITE_ONCE(sk->sk_pacing_shift, 10); 3457 sk->sk_incoming_cpu = -1; 3458 3459 sk_rx_queue_clear(sk); 3460 /* 3461 * Before updating sk_refcnt, we must commit prior changes to memory 3462 * (Documentation/RCU/rculist_nulls.rst for details) 3463 */ 3464 smp_wmb(); 3465 refcount_set(&sk->sk_refcnt, 1); 3466 atomic_set(&sk->sk_drops, 0); 3467 } 3468 EXPORT_SYMBOL(sock_init_data_uid); 3469 3470 void sock_init_data(struct socket *sock, struct sock *sk) 3471 { 3472 kuid_t uid = sock ? 3473 SOCK_INODE(sock)->i_uid : 3474 make_kuid(sock_net(sk)->user_ns, 0); 3475 3476 sock_init_data_uid(sock, sk, uid); 3477 } 3478 EXPORT_SYMBOL(sock_init_data); 3479 3480 void lock_sock_nested(struct sock *sk, int subclass) 3481 { 3482 /* The sk_lock has mutex_lock() semantics here. */ 3483 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3484 3485 might_sleep(); 3486 spin_lock_bh(&sk->sk_lock.slock); 3487 if (sock_owned_by_user_nocheck(sk)) 3488 __lock_sock(sk); 3489 sk->sk_lock.owned = 1; 3490 spin_unlock_bh(&sk->sk_lock.slock); 3491 } 3492 EXPORT_SYMBOL(lock_sock_nested); 3493 3494 void release_sock(struct sock *sk) 3495 { 3496 spin_lock_bh(&sk->sk_lock.slock); 3497 if (sk->sk_backlog.tail) 3498 __release_sock(sk); 3499 3500 /* Warning : release_cb() might need to release sk ownership, 3501 * ie call sock_release_ownership(sk) before us. 3502 */ 3503 if (sk->sk_prot->release_cb) 3504 sk->sk_prot->release_cb(sk); 3505 3506 sock_release_ownership(sk); 3507 if (waitqueue_active(&sk->sk_lock.wq)) 3508 wake_up(&sk->sk_lock.wq); 3509 spin_unlock_bh(&sk->sk_lock.slock); 3510 } 3511 EXPORT_SYMBOL(release_sock); 3512 3513 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3514 { 3515 might_sleep(); 3516 spin_lock_bh(&sk->sk_lock.slock); 3517 3518 if (!sock_owned_by_user_nocheck(sk)) { 3519 /* 3520 * Fast path return with bottom halves disabled and 3521 * sock::sk_lock.slock held. 3522 * 3523 * The 'mutex' is not contended and holding 3524 * sock::sk_lock.slock prevents all other lockers to 3525 * proceed so the corresponding unlock_sock_fast() can 3526 * avoid the slow path of release_sock() completely and 3527 * just release slock. 3528 * 3529 * From a semantical POV this is equivalent to 'acquiring' 3530 * the 'mutex', hence the corresponding lockdep 3531 * mutex_release() has to happen in the fast path of 3532 * unlock_sock_fast(). 3533 */ 3534 return false; 3535 } 3536 3537 __lock_sock(sk); 3538 sk->sk_lock.owned = 1; 3539 __acquire(&sk->sk_lock.slock); 3540 spin_unlock_bh(&sk->sk_lock.slock); 3541 return true; 3542 } 3543 EXPORT_SYMBOL(__lock_sock_fast); 3544 3545 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3546 bool timeval, bool time32) 3547 { 3548 struct sock *sk = sock->sk; 3549 struct timespec64 ts; 3550 3551 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3552 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3553 if (ts.tv_sec == -1) 3554 return -ENOENT; 3555 if (ts.tv_sec == 0) { 3556 ktime_t kt = ktime_get_real(); 3557 sock_write_timestamp(sk, kt); 3558 ts = ktime_to_timespec64(kt); 3559 } 3560 3561 if (timeval) 3562 ts.tv_nsec /= 1000; 3563 3564 #ifdef CONFIG_COMPAT_32BIT_TIME 3565 if (time32) 3566 return put_old_timespec32(&ts, userstamp); 3567 #endif 3568 #ifdef CONFIG_SPARC64 3569 /* beware of padding in sparc64 timeval */ 3570 if (timeval && !in_compat_syscall()) { 3571 struct __kernel_old_timeval __user tv = { 3572 .tv_sec = ts.tv_sec, 3573 .tv_usec = ts.tv_nsec, 3574 }; 3575 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3576 return -EFAULT; 3577 return 0; 3578 } 3579 #endif 3580 return put_timespec64(&ts, userstamp); 3581 } 3582 EXPORT_SYMBOL(sock_gettstamp); 3583 3584 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3585 { 3586 if (!sock_flag(sk, flag)) { 3587 unsigned long previous_flags = sk->sk_flags; 3588 3589 sock_set_flag(sk, flag); 3590 /* 3591 * we just set one of the two flags which require net 3592 * time stamping, but time stamping might have been on 3593 * already because of the other one 3594 */ 3595 if (sock_needs_netstamp(sk) && 3596 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3597 net_enable_timestamp(); 3598 } 3599 } 3600 3601 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3602 int level, int type) 3603 { 3604 struct sock_exterr_skb *serr; 3605 struct sk_buff *skb; 3606 int copied, err; 3607 3608 err = -EAGAIN; 3609 skb = sock_dequeue_err_skb(sk); 3610 if (skb == NULL) 3611 goto out; 3612 3613 copied = skb->len; 3614 if (copied > len) { 3615 msg->msg_flags |= MSG_TRUNC; 3616 copied = len; 3617 } 3618 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3619 if (err) 3620 goto out_free_skb; 3621 3622 sock_recv_timestamp(msg, sk, skb); 3623 3624 serr = SKB_EXT_ERR(skb); 3625 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3626 3627 msg->msg_flags |= MSG_ERRQUEUE; 3628 err = copied; 3629 3630 out_free_skb: 3631 kfree_skb(skb); 3632 out: 3633 return err; 3634 } 3635 EXPORT_SYMBOL(sock_recv_errqueue); 3636 3637 /* 3638 * Get a socket option on an socket. 3639 * 3640 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3641 * asynchronous errors should be reported by getsockopt. We assume 3642 * this means if you specify SO_ERROR (otherwise whats the point of it). 3643 */ 3644 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3645 char __user *optval, int __user *optlen) 3646 { 3647 struct sock *sk = sock->sk; 3648 3649 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3650 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3651 } 3652 EXPORT_SYMBOL(sock_common_getsockopt); 3653 3654 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3655 int flags) 3656 { 3657 struct sock *sk = sock->sk; 3658 int addr_len = 0; 3659 int err; 3660 3661 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3662 if (err >= 0) 3663 msg->msg_namelen = addr_len; 3664 return err; 3665 } 3666 EXPORT_SYMBOL(sock_common_recvmsg); 3667 3668 /* 3669 * Set socket options on an inet socket. 3670 */ 3671 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3672 sockptr_t optval, unsigned int optlen) 3673 { 3674 struct sock *sk = sock->sk; 3675 3676 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3677 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3678 } 3679 EXPORT_SYMBOL(sock_common_setsockopt); 3680 3681 void sk_common_release(struct sock *sk) 3682 { 3683 if (sk->sk_prot->destroy) 3684 sk->sk_prot->destroy(sk); 3685 3686 /* 3687 * Observation: when sk_common_release is called, processes have 3688 * no access to socket. But net still has. 3689 * Step one, detach it from networking: 3690 * 3691 * A. Remove from hash tables. 3692 */ 3693 3694 sk->sk_prot->unhash(sk); 3695 3696 /* 3697 * In this point socket cannot receive new packets, but it is possible 3698 * that some packets are in flight because some CPU runs receiver and 3699 * did hash table lookup before we unhashed socket. They will achieve 3700 * receive queue and will be purged by socket destructor. 3701 * 3702 * Also we still have packets pending on receive queue and probably, 3703 * our own packets waiting in device queues. sock_destroy will drain 3704 * receive queue, but transmitted packets will delay socket destruction 3705 * until the last reference will be released. 3706 */ 3707 3708 sock_orphan(sk); 3709 3710 xfrm_sk_free_policy(sk); 3711 3712 sock_put(sk); 3713 } 3714 EXPORT_SYMBOL(sk_common_release); 3715 3716 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3717 { 3718 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3719 3720 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3721 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3722 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3723 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3724 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3725 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3726 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3727 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3728 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3729 } 3730 3731 #ifdef CONFIG_PROC_FS 3732 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3733 3734 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3735 { 3736 int cpu, idx = prot->inuse_idx; 3737 int res = 0; 3738 3739 for_each_possible_cpu(cpu) 3740 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3741 3742 return res >= 0 ? res : 0; 3743 } 3744 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3745 3746 int sock_inuse_get(struct net *net) 3747 { 3748 int cpu, res = 0; 3749 3750 for_each_possible_cpu(cpu) 3751 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3752 3753 return res; 3754 } 3755 3756 EXPORT_SYMBOL_GPL(sock_inuse_get); 3757 3758 static int __net_init sock_inuse_init_net(struct net *net) 3759 { 3760 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3761 if (net->core.prot_inuse == NULL) 3762 return -ENOMEM; 3763 return 0; 3764 } 3765 3766 static void __net_exit sock_inuse_exit_net(struct net *net) 3767 { 3768 free_percpu(net->core.prot_inuse); 3769 } 3770 3771 static struct pernet_operations net_inuse_ops = { 3772 .init = sock_inuse_init_net, 3773 .exit = sock_inuse_exit_net, 3774 }; 3775 3776 static __init int net_inuse_init(void) 3777 { 3778 if (register_pernet_subsys(&net_inuse_ops)) 3779 panic("Cannot initialize net inuse counters"); 3780 3781 return 0; 3782 } 3783 3784 core_initcall(net_inuse_init); 3785 3786 static int assign_proto_idx(struct proto *prot) 3787 { 3788 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3789 3790 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3791 pr_err("PROTO_INUSE_NR exhausted\n"); 3792 return -ENOSPC; 3793 } 3794 3795 set_bit(prot->inuse_idx, proto_inuse_idx); 3796 return 0; 3797 } 3798 3799 static void release_proto_idx(struct proto *prot) 3800 { 3801 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3802 clear_bit(prot->inuse_idx, proto_inuse_idx); 3803 } 3804 #else 3805 static inline int assign_proto_idx(struct proto *prot) 3806 { 3807 return 0; 3808 } 3809 3810 static inline void release_proto_idx(struct proto *prot) 3811 { 3812 } 3813 3814 #endif 3815 3816 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3817 { 3818 if (!twsk_prot) 3819 return; 3820 kfree(twsk_prot->twsk_slab_name); 3821 twsk_prot->twsk_slab_name = NULL; 3822 kmem_cache_destroy(twsk_prot->twsk_slab); 3823 twsk_prot->twsk_slab = NULL; 3824 } 3825 3826 static int tw_prot_init(const struct proto *prot) 3827 { 3828 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3829 3830 if (!twsk_prot) 3831 return 0; 3832 3833 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3834 prot->name); 3835 if (!twsk_prot->twsk_slab_name) 3836 return -ENOMEM; 3837 3838 twsk_prot->twsk_slab = 3839 kmem_cache_create(twsk_prot->twsk_slab_name, 3840 twsk_prot->twsk_obj_size, 0, 3841 SLAB_ACCOUNT | prot->slab_flags, 3842 NULL); 3843 if (!twsk_prot->twsk_slab) { 3844 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3845 prot->name); 3846 return -ENOMEM; 3847 } 3848 3849 return 0; 3850 } 3851 3852 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3853 { 3854 if (!rsk_prot) 3855 return; 3856 kfree(rsk_prot->slab_name); 3857 rsk_prot->slab_name = NULL; 3858 kmem_cache_destroy(rsk_prot->slab); 3859 rsk_prot->slab = NULL; 3860 } 3861 3862 static int req_prot_init(const struct proto *prot) 3863 { 3864 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3865 3866 if (!rsk_prot) 3867 return 0; 3868 3869 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3870 prot->name); 3871 if (!rsk_prot->slab_name) 3872 return -ENOMEM; 3873 3874 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3875 rsk_prot->obj_size, 0, 3876 SLAB_ACCOUNT | prot->slab_flags, 3877 NULL); 3878 3879 if (!rsk_prot->slab) { 3880 pr_crit("%s: Can't create request sock SLAB cache!\n", 3881 prot->name); 3882 return -ENOMEM; 3883 } 3884 return 0; 3885 } 3886 3887 int proto_register(struct proto *prot, int alloc_slab) 3888 { 3889 int ret = -ENOBUFS; 3890 3891 if (prot->memory_allocated && !prot->sysctl_mem) { 3892 pr_err("%s: missing sysctl_mem\n", prot->name); 3893 return -EINVAL; 3894 } 3895 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3896 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3897 return -EINVAL; 3898 } 3899 if (alloc_slab) { 3900 prot->slab = kmem_cache_create_usercopy(prot->name, 3901 prot->obj_size, 0, 3902 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3903 prot->slab_flags, 3904 prot->useroffset, prot->usersize, 3905 NULL); 3906 3907 if (prot->slab == NULL) { 3908 pr_crit("%s: Can't create sock SLAB cache!\n", 3909 prot->name); 3910 goto out; 3911 } 3912 3913 if (req_prot_init(prot)) 3914 goto out_free_request_sock_slab; 3915 3916 if (tw_prot_init(prot)) 3917 goto out_free_timewait_sock_slab; 3918 } 3919 3920 mutex_lock(&proto_list_mutex); 3921 ret = assign_proto_idx(prot); 3922 if (ret) { 3923 mutex_unlock(&proto_list_mutex); 3924 goto out_free_timewait_sock_slab; 3925 } 3926 list_add(&prot->node, &proto_list); 3927 mutex_unlock(&proto_list_mutex); 3928 return ret; 3929 3930 out_free_timewait_sock_slab: 3931 if (alloc_slab) 3932 tw_prot_cleanup(prot->twsk_prot); 3933 out_free_request_sock_slab: 3934 if (alloc_slab) { 3935 req_prot_cleanup(prot->rsk_prot); 3936 3937 kmem_cache_destroy(prot->slab); 3938 prot->slab = NULL; 3939 } 3940 out: 3941 return ret; 3942 } 3943 EXPORT_SYMBOL(proto_register); 3944 3945 void proto_unregister(struct proto *prot) 3946 { 3947 mutex_lock(&proto_list_mutex); 3948 release_proto_idx(prot); 3949 list_del(&prot->node); 3950 mutex_unlock(&proto_list_mutex); 3951 3952 kmem_cache_destroy(prot->slab); 3953 prot->slab = NULL; 3954 3955 req_prot_cleanup(prot->rsk_prot); 3956 tw_prot_cleanup(prot->twsk_prot); 3957 } 3958 EXPORT_SYMBOL(proto_unregister); 3959 3960 int sock_load_diag_module(int family, int protocol) 3961 { 3962 if (!protocol) { 3963 if (!sock_is_registered(family)) 3964 return -ENOENT; 3965 3966 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3967 NETLINK_SOCK_DIAG, family); 3968 } 3969 3970 #ifdef CONFIG_INET 3971 if (family == AF_INET && 3972 protocol != IPPROTO_RAW && 3973 protocol < MAX_INET_PROTOS && 3974 !rcu_access_pointer(inet_protos[protocol])) 3975 return -ENOENT; 3976 #endif 3977 3978 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3979 NETLINK_SOCK_DIAG, family, protocol); 3980 } 3981 EXPORT_SYMBOL(sock_load_diag_module); 3982 3983 #ifdef CONFIG_PROC_FS 3984 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3985 __acquires(proto_list_mutex) 3986 { 3987 mutex_lock(&proto_list_mutex); 3988 return seq_list_start_head(&proto_list, *pos); 3989 } 3990 3991 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3992 { 3993 return seq_list_next(v, &proto_list, pos); 3994 } 3995 3996 static void proto_seq_stop(struct seq_file *seq, void *v) 3997 __releases(proto_list_mutex) 3998 { 3999 mutex_unlock(&proto_list_mutex); 4000 } 4001 4002 static char proto_method_implemented(const void *method) 4003 { 4004 return method == NULL ? 'n' : 'y'; 4005 } 4006 static long sock_prot_memory_allocated(struct proto *proto) 4007 { 4008 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4009 } 4010 4011 static const char *sock_prot_memory_pressure(struct proto *proto) 4012 { 4013 return proto->memory_pressure != NULL ? 4014 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4015 } 4016 4017 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4018 { 4019 4020 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4021 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4022 proto->name, 4023 proto->obj_size, 4024 sock_prot_inuse_get(seq_file_net(seq), proto), 4025 sock_prot_memory_allocated(proto), 4026 sock_prot_memory_pressure(proto), 4027 proto->max_header, 4028 proto->slab == NULL ? "no" : "yes", 4029 module_name(proto->owner), 4030 proto_method_implemented(proto->close), 4031 proto_method_implemented(proto->connect), 4032 proto_method_implemented(proto->disconnect), 4033 proto_method_implemented(proto->accept), 4034 proto_method_implemented(proto->ioctl), 4035 proto_method_implemented(proto->init), 4036 proto_method_implemented(proto->destroy), 4037 proto_method_implemented(proto->shutdown), 4038 proto_method_implemented(proto->setsockopt), 4039 proto_method_implemented(proto->getsockopt), 4040 proto_method_implemented(proto->sendmsg), 4041 proto_method_implemented(proto->recvmsg), 4042 proto_method_implemented(proto->bind), 4043 proto_method_implemented(proto->backlog_rcv), 4044 proto_method_implemented(proto->hash), 4045 proto_method_implemented(proto->unhash), 4046 proto_method_implemented(proto->get_port), 4047 proto_method_implemented(proto->enter_memory_pressure)); 4048 } 4049 4050 static int proto_seq_show(struct seq_file *seq, void *v) 4051 { 4052 if (v == &proto_list) 4053 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4054 "protocol", 4055 "size", 4056 "sockets", 4057 "memory", 4058 "press", 4059 "maxhdr", 4060 "slab", 4061 "module", 4062 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4063 else 4064 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4065 return 0; 4066 } 4067 4068 static const struct seq_operations proto_seq_ops = { 4069 .start = proto_seq_start, 4070 .next = proto_seq_next, 4071 .stop = proto_seq_stop, 4072 .show = proto_seq_show, 4073 }; 4074 4075 static __net_init int proto_init_net(struct net *net) 4076 { 4077 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4078 sizeof(struct seq_net_private))) 4079 return -ENOMEM; 4080 4081 return 0; 4082 } 4083 4084 static __net_exit void proto_exit_net(struct net *net) 4085 { 4086 remove_proc_entry("protocols", net->proc_net); 4087 } 4088 4089 4090 static __net_initdata struct pernet_operations proto_net_ops = { 4091 .init = proto_init_net, 4092 .exit = proto_exit_net, 4093 }; 4094 4095 static int __init proto_init(void) 4096 { 4097 return register_pernet_subsys(&proto_net_ops); 4098 } 4099 4100 subsys_initcall(proto_init); 4101 4102 #endif /* PROC_FS */ 4103 4104 #ifdef CONFIG_NET_RX_BUSY_POLL 4105 bool sk_busy_loop_end(void *p, unsigned long start_time) 4106 { 4107 struct sock *sk = p; 4108 4109 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4110 sk_busy_loop_timeout(sk, start_time); 4111 } 4112 EXPORT_SYMBOL(sk_busy_loop_end); 4113 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4114 4115 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4116 { 4117 if (!sk->sk_prot->bind_add) 4118 return -EOPNOTSUPP; 4119 return sk->sk_prot->bind_add(sk, addr, addr_len); 4120 } 4121 EXPORT_SYMBOL(sock_bind_add); 4122 4123 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4124 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4125 void __user *arg, void *karg, size_t size) 4126 { 4127 int ret; 4128 4129 if (copy_from_user(karg, arg, size)) 4130 return -EFAULT; 4131 4132 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4133 if (ret) 4134 return ret; 4135 4136 if (copy_to_user(arg, karg, size)) 4137 return -EFAULT; 4138 4139 return 0; 4140 } 4141 EXPORT_SYMBOL(sock_ioctl_inout); 4142 4143 /* This is the most common ioctl prep function, where the result (4 bytes) is 4144 * copied back to userspace if the ioctl() returns successfully. No input is 4145 * copied from userspace as input argument. 4146 */ 4147 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4148 { 4149 int ret, karg = 0; 4150 4151 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4152 if (ret) 4153 return ret; 4154 4155 return put_user(karg, (int __user *)arg); 4156 } 4157 4158 /* A wrapper around sock ioctls, which copies the data from userspace 4159 * (depending on the protocol/ioctl), and copies back the result to userspace. 4160 * The main motivation for this function is to pass kernel memory to the 4161 * protocol ioctl callbacks, instead of userspace memory. 4162 */ 4163 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4164 { 4165 int rc = 1; 4166 4167 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4168 rc = ipmr_sk_ioctl(sk, cmd, arg); 4169 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4170 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4171 else if (sk_is_phonet(sk)) 4172 rc = phonet_sk_ioctl(sk, cmd, arg); 4173 4174 /* If ioctl was processed, returns its value */ 4175 if (rc <= 0) 4176 return rc; 4177 4178 /* Otherwise call the default handler */ 4179 return sock_ioctl_out(sk, cmd, arg); 4180 } 4181 EXPORT_SYMBOL(sk_ioctl); 4182