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 if (valbool) 1248 set_bit(SOCK_PASSCRED, &sock->flags); 1249 else 1250 clear_bit(SOCK_PASSCRED, &sock->flags); 1251 break; 1252 1253 case SO_PASSPIDFD: 1254 if (valbool) 1255 set_bit(SOCK_PASSPIDFD, &sock->flags); 1256 else 1257 clear_bit(SOCK_PASSPIDFD, &sock->flags); 1258 break; 1259 1260 case SO_TIMESTAMP_OLD: 1261 case SO_TIMESTAMP_NEW: 1262 case SO_TIMESTAMPNS_OLD: 1263 case SO_TIMESTAMPNS_NEW: 1264 sock_set_timestamp(sk, optname, valbool); 1265 break; 1266 1267 case SO_TIMESTAMPING_NEW: 1268 case SO_TIMESTAMPING_OLD: 1269 if (optlen == sizeof(timestamping)) { 1270 if (copy_from_sockptr(×tamping, optval, 1271 sizeof(timestamping))) { 1272 ret = -EFAULT; 1273 break; 1274 } 1275 } else { 1276 memset(×tamping, 0, sizeof(timestamping)); 1277 timestamping.flags = val; 1278 } 1279 ret = sock_set_timestamping(sk, optname, timestamping); 1280 break; 1281 1282 case SO_RCVLOWAT: 1283 if (val < 0) 1284 val = INT_MAX; 1285 if (sock && sock->ops->set_rcvlowat) 1286 ret = sock->ops->set_rcvlowat(sk, val); 1287 else 1288 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1289 break; 1290 1291 case SO_RCVTIMEO_OLD: 1292 case SO_RCVTIMEO_NEW: 1293 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1294 optlen, optname == SO_RCVTIMEO_OLD); 1295 break; 1296 1297 case SO_SNDTIMEO_OLD: 1298 case SO_SNDTIMEO_NEW: 1299 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1300 optlen, optname == SO_SNDTIMEO_OLD); 1301 break; 1302 1303 case SO_ATTACH_FILTER: { 1304 struct sock_fprog fprog; 1305 1306 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1307 if (!ret) 1308 ret = sk_attach_filter(&fprog, sk); 1309 break; 1310 } 1311 case SO_ATTACH_BPF: 1312 ret = -EINVAL; 1313 if (optlen == sizeof(u32)) { 1314 u32 ufd; 1315 1316 ret = -EFAULT; 1317 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1318 break; 1319 1320 ret = sk_attach_bpf(ufd, sk); 1321 } 1322 break; 1323 1324 case SO_ATTACH_REUSEPORT_CBPF: { 1325 struct sock_fprog fprog; 1326 1327 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1328 if (!ret) 1329 ret = sk_reuseport_attach_filter(&fprog, sk); 1330 break; 1331 } 1332 case SO_ATTACH_REUSEPORT_EBPF: 1333 ret = -EINVAL; 1334 if (optlen == sizeof(u32)) { 1335 u32 ufd; 1336 1337 ret = -EFAULT; 1338 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1339 break; 1340 1341 ret = sk_reuseport_attach_bpf(ufd, sk); 1342 } 1343 break; 1344 1345 case SO_DETACH_REUSEPORT_BPF: 1346 ret = reuseport_detach_prog(sk); 1347 break; 1348 1349 case SO_DETACH_FILTER: 1350 ret = sk_detach_filter(sk); 1351 break; 1352 1353 case SO_LOCK_FILTER: 1354 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1355 ret = -EPERM; 1356 else 1357 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1358 break; 1359 1360 case SO_PASSSEC: 1361 if (valbool) 1362 set_bit(SOCK_PASSSEC, &sock->flags); 1363 else 1364 clear_bit(SOCK_PASSSEC, &sock->flags); 1365 break; 1366 case SO_MARK: 1367 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1368 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1369 ret = -EPERM; 1370 break; 1371 } 1372 1373 __sock_set_mark(sk, val); 1374 break; 1375 case SO_RCVMARK: 1376 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1377 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1378 ret = -EPERM; 1379 break; 1380 } 1381 1382 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1383 break; 1384 1385 case SO_RXQ_OVFL: 1386 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1387 break; 1388 1389 case SO_WIFI_STATUS: 1390 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1391 break; 1392 1393 case SO_PEEK_OFF: 1394 if (sock->ops->set_peek_off) 1395 ret = sock->ops->set_peek_off(sk, val); 1396 else 1397 ret = -EOPNOTSUPP; 1398 break; 1399 1400 case SO_NOFCS: 1401 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1402 break; 1403 1404 case SO_SELECT_ERR_QUEUE: 1405 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1406 break; 1407 1408 #ifdef CONFIG_NET_RX_BUSY_POLL 1409 case SO_BUSY_POLL: 1410 if (val < 0) 1411 ret = -EINVAL; 1412 else 1413 WRITE_ONCE(sk->sk_ll_usec, val); 1414 break; 1415 case SO_PREFER_BUSY_POLL: 1416 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1417 ret = -EPERM; 1418 else 1419 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1420 break; 1421 case SO_BUSY_POLL_BUDGET: 1422 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1423 ret = -EPERM; 1424 } else { 1425 if (val < 0 || val > U16_MAX) 1426 ret = -EINVAL; 1427 else 1428 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1429 } 1430 break; 1431 #endif 1432 1433 case SO_MAX_PACING_RATE: 1434 { 1435 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1436 1437 if (sizeof(ulval) != sizeof(val) && 1438 optlen >= sizeof(ulval) && 1439 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1440 ret = -EFAULT; 1441 break; 1442 } 1443 if (ulval != ~0UL) 1444 cmpxchg(&sk->sk_pacing_status, 1445 SK_PACING_NONE, 1446 SK_PACING_NEEDED); 1447 sk->sk_max_pacing_rate = ulval; 1448 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1449 break; 1450 } 1451 case SO_INCOMING_CPU: 1452 reuseport_update_incoming_cpu(sk, val); 1453 break; 1454 1455 case SO_CNX_ADVICE: 1456 if (val == 1) 1457 dst_negative_advice(sk); 1458 break; 1459 1460 case SO_ZEROCOPY: 1461 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1462 if (!(sk_is_tcp(sk) || 1463 (sk->sk_type == SOCK_DGRAM && 1464 sk->sk_protocol == IPPROTO_UDP))) 1465 ret = -EOPNOTSUPP; 1466 } else if (sk->sk_family != PF_RDS) { 1467 ret = -EOPNOTSUPP; 1468 } 1469 if (!ret) { 1470 if (val < 0 || val > 1) 1471 ret = -EINVAL; 1472 else 1473 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1474 } 1475 break; 1476 1477 case SO_TXTIME: 1478 if (optlen != sizeof(struct sock_txtime)) { 1479 ret = -EINVAL; 1480 break; 1481 } else if (copy_from_sockptr(&sk_txtime, optval, 1482 sizeof(struct sock_txtime))) { 1483 ret = -EFAULT; 1484 break; 1485 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1486 ret = -EINVAL; 1487 break; 1488 } 1489 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1490 * scheduler has enough safe guards. 1491 */ 1492 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1493 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1494 ret = -EPERM; 1495 break; 1496 } 1497 sock_valbool_flag(sk, SOCK_TXTIME, true); 1498 sk->sk_clockid = sk_txtime.clockid; 1499 sk->sk_txtime_deadline_mode = 1500 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1501 sk->sk_txtime_report_errors = 1502 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1503 break; 1504 1505 case SO_BINDTOIFINDEX: 1506 ret = sock_bindtoindex_locked(sk, val); 1507 break; 1508 1509 case SO_BUF_LOCK: 1510 if (val & ~SOCK_BUF_LOCK_MASK) { 1511 ret = -EINVAL; 1512 break; 1513 } 1514 sk->sk_userlocks = val | (sk->sk_userlocks & 1515 ~SOCK_BUF_LOCK_MASK); 1516 break; 1517 1518 case SO_RESERVE_MEM: 1519 { 1520 int delta; 1521 1522 if (val < 0) { 1523 ret = -EINVAL; 1524 break; 1525 } 1526 1527 delta = val - sk->sk_reserved_mem; 1528 if (delta < 0) 1529 sock_release_reserved_memory(sk, -delta); 1530 else 1531 ret = sock_reserve_memory(sk, delta); 1532 break; 1533 } 1534 1535 case SO_TXREHASH: 1536 if (val < -1 || val > 1) { 1537 ret = -EINVAL; 1538 break; 1539 } 1540 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1541 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1542 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1543 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1544 break; 1545 1546 default: 1547 ret = -ENOPROTOOPT; 1548 break; 1549 } 1550 sockopt_release_sock(sk); 1551 return ret; 1552 } 1553 1554 int sock_setsockopt(struct socket *sock, int level, int optname, 1555 sockptr_t optval, unsigned int optlen) 1556 { 1557 return sk_setsockopt(sock->sk, level, optname, 1558 optval, optlen); 1559 } 1560 EXPORT_SYMBOL(sock_setsockopt); 1561 1562 static const struct cred *sk_get_peer_cred(struct sock *sk) 1563 { 1564 const struct cred *cred; 1565 1566 spin_lock(&sk->sk_peer_lock); 1567 cred = get_cred(sk->sk_peer_cred); 1568 spin_unlock(&sk->sk_peer_lock); 1569 1570 return cred; 1571 } 1572 1573 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1574 struct ucred *ucred) 1575 { 1576 ucred->pid = pid_vnr(pid); 1577 ucred->uid = ucred->gid = -1; 1578 if (cred) { 1579 struct user_namespace *current_ns = current_user_ns(); 1580 1581 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1582 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1583 } 1584 } 1585 1586 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1587 { 1588 struct user_namespace *user_ns = current_user_ns(); 1589 int i; 1590 1591 for (i = 0; i < src->ngroups; i++) { 1592 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1593 1594 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1595 return -EFAULT; 1596 } 1597 1598 return 0; 1599 } 1600 1601 int sk_getsockopt(struct sock *sk, int level, int optname, 1602 sockptr_t optval, sockptr_t optlen) 1603 { 1604 struct socket *sock = sk->sk_socket; 1605 1606 union { 1607 int val; 1608 u64 val64; 1609 unsigned long ulval; 1610 struct linger ling; 1611 struct old_timeval32 tm32; 1612 struct __kernel_old_timeval tm; 1613 struct __kernel_sock_timeval stm; 1614 struct sock_txtime txtime; 1615 struct so_timestamping timestamping; 1616 } v; 1617 1618 int lv = sizeof(int); 1619 int len; 1620 1621 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1622 return -EFAULT; 1623 if (len < 0) 1624 return -EINVAL; 1625 1626 memset(&v, 0, sizeof(v)); 1627 1628 switch (optname) { 1629 case SO_DEBUG: 1630 v.val = sock_flag(sk, SOCK_DBG); 1631 break; 1632 1633 case SO_DONTROUTE: 1634 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1635 break; 1636 1637 case SO_BROADCAST: 1638 v.val = sock_flag(sk, SOCK_BROADCAST); 1639 break; 1640 1641 case SO_SNDBUF: 1642 v.val = sk->sk_sndbuf; 1643 break; 1644 1645 case SO_RCVBUF: 1646 v.val = sk->sk_rcvbuf; 1647 break; 1648 1649 case SO_REUSEADDR: 1650 v.val = sk->sk_reuse; 1651 break; 1652 1653 case SO_REUSEPORT: 1654 v.val = sk->sk_reuseport; 1655 break; 1656 1657 case SO_KEEPALIVE: 1658 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1659 break; 1660 1661 case SO_TYPE: 1662 v.val = sk->sk_type; 1663 break; 1664 1665 case SO_PROTOCOL: 1666 v.val = sk->sk_protocol; 1667 break; 1668 1669 case SO_DOMAIN: 1670 v.val = sk->sk_family; 1671 break; 1672 1673 case SO_ERROR: 1674 v.val = -sock_error(sk); 1675 if (v.val == 0) 1676 v.val = xchg(&sk->sk_err_soft, 0); 1677 break; 1678 1679 case SO_OOBINLINE: 1680 v.val = sock_flag(sk, SOCK_URGINLINE); 1681 break; 1682 1683 case SO_NO_CHECK: 1684 v.val = sk->sk_no_check_tx; 1685 break; 1686 1687 case SO_PRIORITY: 1688 v.val = sk->sk_priority; 1689 break; 1690 1691 case SO_LINGER: 1692 lv = sizeof(v.ling); 1693 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1694 v.ling.l_linger = sk->sk_lingertime / HZ; 1695 break; 1696 1697 case SO_BSDCOMPAT: 1698 break; 1699 1700 case SO_TIMESTAMP_OLD: 1701 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1702 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1703 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1704 break; 1705 1706 case SO_TIMESTAMPNS_OLD: 1707 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1708 break; 1709 1710 case SO_TIMESTAMP_NEW: 1711 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1712 break; 1713 1714 case SO_TIMESTAMPNS_NEW: 1715 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1716 break; 1717 1718 case SO_TIMESTAMPING_OLD: 1719 lv = sizeof(v.timestamping); 1720 v.timestamping.flags = sk->sk_tsflags; 1721 v.timestamping.bind_phc = sk->sk_bind_phc; 1722 break; 1723 1724 case SO_RCVTIMEO_OLD: 1725 case SO_RCVTIMEO_NEW: 1726 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1727 break; 1728 1729 case SO_SNDTIMEO_OLD: 1730 case SO_SNDTIMEO_NEW: 1731 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1732 break; 1733 1734 case SO_RCVLOWAT: 1735 v.val = sk->sk_rcvlowat; 1736 break; 1737 1738 case SO_SNDLOWAT: 1739 v.val = 1; 1740 break; 1741 1742 case SO_PASSCRED: 1743 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1744 break; 1745 1746 case SO_PASSPIDFD: 1747 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1748 break; 1749 1750 case SO_PEERCRED: 1751 { 1752 struct ucred peercred; 1753 if (len > sizeof(peercred)) 1754 len = sizeof(peercred); 1755 1756 spin_lock(&sk->sk_peer_lock); 1757 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1758 spin_unlock(&sk->sk_peer_lock); 1759 1760 if (copy_to_sockptr(optval, &peercred, len)) 1761 return -EFAULT; 1762 goto lenout; 1763 } 1764 1765 case SO_PEERPIDFD: 1766 { 1767 struct pid *peer_pid; 1768 struct file *pidfd_file = NULL; 1769 int pidfd; 1770 1771 if (len > sizeof(pidfd)) 1772 len = sizeof(pidfd); 1773 1774 spin_lock(&sk->sk_peer_lock); 1775 peer_pid = get_pid(sk->sk_peer_pid); 1776 spin_unlock(&sk->sk_peer_lock); 1777 1778 if (!peer_pid) 1779 return -ESRCH; 1780 1781 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1782 put_pid(peer_pid); 1783 if (pidfd < 0) 1784 return pidfd; 1785 1786 if (copy_to_sockptr(optval, &pidfd, len) || 1787 copy_to_sockptr(optlen, &len, sizeof(int))) { 1788 put_unused_fd(pidfd); 1789 fput(pidfd_file); 1790 1791 return -EFAULT; 1792 } 1793 1794 fd_install(pidfd, pidfd_file); 1795 return 0; 1796 } 1797 1798 case SO_PEERGROUPS: 1799 { 1800 const struct cred *cred; 1801 int ret, n; 1802 1803 cred = sk_get_peer_cred(sk); 1804 if (!cred) 1805 return -ENODATA; 1806 1807 n = cred->group_info->ngroups; 1808 if (len < n * sizeof(gid_t)) { 1809 len = n * sizeof(gid_t); 1810 put_cred(cred); 1811 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1812 } 1813 len = n * sizeof(gid_t); 1814 1815 ret = groups_to_user(optval, cred->group_info); 1816 put_cred(cred); 1817 if (ret) 1818 return ret; 1819 goto lenout; 1820 } 1821 1822 case SO_PEERNAME: 1823 { 1824 char address[128]; 1825 1826 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1827 if (lv < 0) 1828 return -ENOTCONN; 1829 if (lv < len) 1830 return -EINVAL; 1831 if (copy_to_sockptr(optval, address, len)) 1832 return -EFAULT; 1833 goto lenout; 1834 } 1835 1836 /* Dubious BSD thing... Probably nobody even uses it, but 1837 * the UNIX standard wants it for whatever reason... -DaveM 1838 */ 1839 case SO_ACCEPTCONN: 1840 v.val = sk->sk_state == TCP_LISTEN; 1841 break; 1842 1843 case SO_PASSSEC: 1844 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1845 break; 1846 1847 case SO_PEERSEC: 1848 return security_socket_getpeersec_stream(sock, 1849 optval, optlen, len); 1850 1851 case SO_MARK: 1852 v.val = sk->sk_mark; 1853 break; 1854 1855 case SO_RCVMARK: 1856 v.val = sock_flag(sk, SOCK_RCVMARK); 1857 break; 1858 1859 case SO_RXQ_OVFL: 1860 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1861 break; 1862 1863 case SO_WIFI_STATUS: 1864 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1865 break; 1866 1867 case SO_PEEK_OFF: 1868 if (!sock->ops->set_peek_off) 1869 return -EOPNOTSUPP; 1870 1871 v.val = sk->sk_peek_off; 1872 break; 1873 case SO_NOFCS: 1874 v.val = sock_flag(sk, SOCK_NOFCS); 1875 break; 1876 1877 case SO_BINDTODEVICE: 1878 return sock_getbindtodevice(sk, optval, optlen, len); 1879 1880 case SO_GET_FILTER: 1881 len = sk_get_filter(sk, optval, len); 1882 if (len < 0) 1883 return len; 1884 1885 goto lenout; 1886 1887 case SO_LOCK_FILTER: 1888 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1889 break; 1890 1891 case SO_BPF_EXTENSIONS: 1892 v.val = bpf_tell_extensions(); 1893 break; 1894 1895 case SO_SELECT_ERR_QUEUE: 1896 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1897 break; 1898 1899 #ifdef CONFIG_NET_RX_BUSY_POLL 1900 case SO_BUSY_POLL: 1901 v.val = sk->sk_ll_usec; 1902 break; 1903 case SO_PREFER_BUSY_POLL: 1904 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1905 break; 1906 #endif 1907 1908 case SO_MAX_PACING_RATE: 1909 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1910 lv = sizeof(v.ulval); 1911 v.ulval = sk->sk_max_pacing_rate; 1912 } else { 1913 /* 32bit version */ 1914 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1915 } 1916 break; 1917 1918 case SO_INCOMING_CPU: 1919 v.val = READ_ONCE(sk->sk_incoming_cpu); 1920 break; 1921 1922 case SO_MEMINFO: 1923 { 1924 u32 meminfo[SK_MEMINFO_VARS]; 1925 1926 sk_get_meminfo(sk, meminfo); 1927 1928 len = min_t(unsigned int, len, sizeof(meminfo)); 1929 if (copy_to_sockptr(optval, &meminfo, len)) 1930 return -EFAULT; 1931 1932 goto lenout; 1933 } 1934 1935 #ifdef CONFIG_NET_RX_BUSY_POLL 1936 case SO_INCOMING_NAPI_ID: 1937 v.val = READ_ONCE(sk->sk_napi_id); 1938 1939 /* aggregate non-NAPI IDs down to 0 */ 1940 if (v.val < MIN_NAPI_ID) 1941 v.val = 0; 1942 1943 break; 1944 #endif 1945 1946 case SO_COOKIE: 1947 lv = sizeof(u64); 1948 if (len < lv) 1949 return -EINVAL; 1950 v.val64 = sock_gen_cookie(sk); 1951 break; 1952 1953 case SO_ZEROCOPY: 1954 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1955 break; 1956 1957 case SO_TXTIME: 1958 lv = sizeof(v.txtime); 1959 v.txtime.clockid = sk->sk_clockid; 1960 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1961 SOF_TXTIME_DEADLINE_MODE : 0; 1962 v.txtime.flags |= sk->sk_txtime_report_errors ? 1963 SOF_TXTIME_REPORT_ERRORS : 0; 1964 break; 1965 1966 case SO_BINDTOIFINDEX: 1967 v.val = READ_ONCE(sk->sk_bound_dev_if); 1968 break; 1969 1970 case SO_NETNS_COOKIE: 1971 lv = sizeof(u64); 1972 if (len != lv) 1973 return -EINVAL; 1974 v.val64 = sock_net(sk)->net_cookie; 1975 break; 1976 1977 case SO_BUF_LOCK: 1978 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1979 break; 1980 1981 case SO_RESERVE_MEM: 1982 v.val = sk->sk_reserved_mem; 1983 break; 1984 1985 case SO_TXREHASH: 1986 v.val = sk->sk_txrehash; 1987 break; 1988 1989 default: 1990 /* We implement the SO_SNDLOWAT etc to not be settable 1991 * (1003.1g 7). 1992 */ 1993 return -ENOPROTOOPT; 1994 } 1995 1996 if (len > lv) 1997 len = lv; 1998 if (copy_to_sockptr(optval, &v, len)) 1999 return -EFAULT; 2000 lenout: 2001 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2002 return -EFAULT; 2003 return 0; 2004 } 2005 2006 int sock_getsockopt(struct socket *sock, int level, int optname, 2007 char __user *optval, int __user *optlen) 2008 { 2009 return sk_getsockopt(sock->sk, level, optname, 2010 USER_SOCKPTR(optval), 2011 USER_SOCKPTR(optlen)); 2012 } 2013 2014 /* 2015 * Initialize an sk_lock. 2016 * 2017 * (We also register the sk_lock with the lock validator.) 2018 */ 2019 static inline void sock_lock_init(struct sock *sk) 2020 { 2021 if (sk->sk_kern_sock) 2022 sock_lock_init_class_and_name( 2023 sk, 2024 af_family_kern_slock_key_strings[sk->sk_family], 2025 af_family_kern_slock_keys + sk->sk_family, 2026 af_family_kern_key_strings[sk->sk_family], 2027 af_family_kern_keys + sk->sk_family); 2028 else 2029 sock_lock_init_class_and_name( 2030 sk, 2031 af_family_slock_key_strings[sk->sk_family], 2032 af_family_slock_keys + sk->sk_family, 2033 af_family_key_strings[sk->sk_family], 2034 af_family_keys + sk->sk_family); 2035 } 2036 2037 /* 2038 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2039 * even temporarly, because of RCU lookups. sk_node should also be left as is. 2040 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2041 */ 2042 static void sock_copy(struct sock *nsk, const struct sock *osk) 2043 { 2044 const struct proto *prot = READ_ONCE(osk->sk_prot); 2045 #ifdef CONFIG_SECURITY_NETWORK 2046 void *sptr = nsk->sk_security; 2047 #endif 2048 2049 /* If we move sk_tx_queue_mapping out of the private section, 2050 * we must check if sk_tx_queue_clear() is called after 2051 * sock_copy() in sk_clone_lock(). 2052 */ 2053 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2054 offsetof(struct sock, sk_dontcopy_begin) || 2055 offsetof(struct sock, sk_tx_queue_mapping) >= 2056 offsetof(struct sock, sk_dontcopy_end)); 2057 2058 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2059 2060 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2061 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 2062 2063 #ifdef CONFIG_SECURITY_NETWORK 2064 nsk->sk_security = sptr; 2065 security_sk_clone(osk, nsk); 2066 #endif 2067 } 2068 2069 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2070 int family) 2071 { 2072 struct sock *sk; 2073 struct kmem_cache *slab; 2074 2075 slab = prot->slab; 2076 if (slab != NULL) { 2077 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2078 if (!sk) 2079 return sk; 2080 if (want_init_on_alloc(priority)) 2081 sk_prot_clear_nulls(sk, prot->obj_size); 2082 } else 2083 sk = kmalloc(prot->obj_size, priority); 2084 2085 if (sk != NULL) { 2086 if (security_sk_alloc(sk, family, priority)) 2087 goto out_free; 2088 2089 if (!try_module_get(prot->owner)) 2090 goto out_free_sec; 2091 } 2092 2093 return sk; 2094 2095 out_free_sec: 2096 security_sk_free(sk); 2097 out_free: 2098 if (slab != NULL) 2099 kmem_cache_free(slab, sk); 2100 else 2101 kfree(sk); 2102 return NULL; 2103 } 2104 2105 static void sk_prot_free(struct proto *prot, struct sock *sk) 2106 { 2107 struct kmem_cache *slab; 2108 struct module *owner; 2109 2110 owner = prot->owner; 2111 slab = prot->slab; 2112 2113 cgroup_sk_free(&sk->sk_cgrp_data); 2114 mem_cgroup_sk_free(sk); 2115 security_sk_free(sk); 2116 if (slab != NULL) 2117 kmem_cache_free(slab, sk); 2118 else 2119 kfree(sk); 2120 module_put(owner); 2121 } 2122 2123 /** 2124 * sk_alloc - All socket objects are allocated here 2125 * @net: the applicable net namespace 2126 * @family: protocol family 2127 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2128 * @prot: struct proto associated with this new sock instance 2129 * @kern: is this to be a kernel socket? 2130 */ 2131 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2132 struct proto *prot, int kern) 2133 { 2134 struct sock *sk; 2135 2136 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2137 if (sk) { 2138 sk->sk_family = family; 2139 /* 2140 * See comment in struct sock definition to understand 2141 * why we need sk_prot_creator -acme 2142 */ 2143 sk->sk_prot = sk->sk_prot_creator = prot; 2144 sk->sk_kern_sock = kern; 2145 sock_lock_init(sk); 2146 sk->sk_net_refcnt = kern ? 0 : 1; 2147 if (likely(sk->sk_net_refcnt)) { 2148 get_net_track(net, &sk->ns_tracker, priority); 2149 sock_inuse_add(net, 1); 2150 } else { 2151 __netns_tracker_alloc(net, &sk->ns_tracker, 2152 false, priority); 2153 } 2154 2155 sock_net_set(sk, net); 2156 refcount_set(&sk->sk_wmem_alloc, 1); 2157 2158 mem_cgroup_sk_alloc(sk); 2159 cgroup_sk_alloc(&sk->sk_cgrp_data); 2160 sock_update_classid(&sk->sk_cgrp_data); 2161 sock_update_netprioidx(&sk->sk_cgrp_data); 2162 sk_tx_queue_clear(sk); 2163 } 2164 2165 return sk; 2166 } 2167 EXPORT_SYMBOL(sk_alloc); 2168 2169 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2170 * grace period. This is the case for UDP sockets and TCP listeners. 2171 */ 2172 static void __sk_destruct(struct rcu_head *head) 2173 { 2174 struct sock *sk = container_of(head, struct sock, sk_rcu); 2175 struct sk_filter *filter; 2176 2177 if (sk->sk_destruct) 2178 sk->sk_destruct(sk); 2179 2180 filter = rcu_dereference_check(sk->sk_filter, 2181 refcount_read(&sk->sk_wmem_alloc) == 0); 2182 if (filter) { 2183 sk_filter_uncharge(sk, filter); 2184 RCU_INIT_POINTER(sk->sk_filter, NULL); 2185 } 2186 2187 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2188 2189 #ifdef CONFIG_BPF_SYSCALL 2190 bpf_sk_storage_free(sk); 2191 #endif 2192 2193 if (atomic_read(&sk->sk_omem_alloc)) 2194 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2195 __func__, atomic_read(&sk->sk_omem_alloc)); 2196 2197 if (sk->sk_frag.page) { 2198 put_page(sk->sk_frag.page); 2199 sk->sk_frag.page = NULL; 2200 } 2201 2202 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2203 put_cred(sk->sk_peer_cred); 2204 put_pid(sk->sk_peer_pid); 2205 2206 if (likely(sk->sk_net_refcnt)) 2207 put_net_track(sock_net(sk), &sk->ns_tracker); 2208 else 2209 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2210 2211 sk_prot_free(sk->sk_prot_creator, sk); 2212 } 2213 2214 void sk_destruct(struct sock *sk) 2215 { 2216 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2217 2218 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2219 reuseport_detach_sock(sk); 2220 use_call_rcu = true; 2221 } 2222 2223 if (use_call_rcu) 2224 call_rcu(&sk->sk_rcu, __sk_destruct); 2225 else 2226 __sk_destruct(&sk->sk_rcu); 2227 } 2228 2229 static void __sk_free(struct sock *sk) 2230 { 2231 if (likely(sk->sk_net_refcnt)) 2232 sock_inuse_add(sock_net(sk), -1); 2233 2234 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2235 sock_diag_broadcast_destroy(sk); 2236 else 2237 sk_destruct(sk); 2238 } 2239 2240 void sk_free(struct sock *sk) 2241 { 2242 /* 2243 * We subtract one from sk_wmem_alloc and can know if 2244 * some packets are still in some tx queue. 2245 * If not null, sock_wfree() will call __sk_free(sk) later 2246 */ 2247 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2248 __sk_free(sk); 2249 } 2250 EXPORT_SYMBOL(sk_free); 2251 2252 static void sk_init_common(struct sock *sk) 2253 { 2254 skb_queue_head_init(&sk->sk_receive_queue); 2255 skb_queue_head_init(&sk->sk_write_queue); 2256 skb_queue_head_init(&sk->sk_error_queue); 2257 2258 rwlock_init(&sk->sk_callback_lock); 2259 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2260 af_rlock_keys + sk->sk_family, 2261 af_family_rlock_key_strings[sk->sk_family]); 2262 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2263 af_wlock_keys + sk->sk_family, 2264 af_family_wlock_key_strings[sk->sk_family]); 2265 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2266 af_elock_keys + sk->sk_family, 2267 af_family_elock_key_strings[sk->sk_family]); 2268 lockdep_set_class_and_name(&sk->sk_callback_lock, 2269 af_callback_keys + sk->sk_family, 2270 af_family_clock_key_strings[sk->sk_family]); 2271 } 2272 2273 /** 2274 * sk_clone_lock - clone a socket, and lock its clone 2275 * @sk: the socket to clone 2276 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2277 * 2278 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2279 */ 2280 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2281 { 2282 struct proto *prot = READ_ONCE(sk->sk_prot); 2283 struct sk_filter *filter; 2284 bool is_charged = true; 2285 struct sock *newsk; 2286 2287 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2288 if (!newsk) 2289 goto out; 2290 2291 sock_copy(newsk, sk); 2292 2293 newsk->sk_prot_creator = prot; 2294 2295 /* SANITY */ 2296 if (likely(newsk->sk_net_refcnt)) { 2297 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2298 sock_inuse_add(sock_net(newsk), 1); 2299 } else { 2300 /* Kernel sockets are not elevating the struct net refcount. 2301 * Instead, use a tracker to more easily detect if a layer 2302 * is not properly dismantling its kernel sockets at netns 2303 * destroy time. 2304 */ 2305 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2306 false, priority); 2307 } 2308 sk_node_init(&newsk->sk_node); 2309 sock_lock_init(newsk); 2310 bh_lock_sock(newsk); 2311 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2312 newsk->sk_backlog.len = 0; 2313 2314 atomic_set(&newsk->sk_rmem_alloc, 0); 2315 2316 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2317 refcount_set(&newsk->sk_wmem_alloc, 1); 2318 2319 atomic_set(&newsk->sk_omem_alloc, 0); 2320 sk_init_common(newsk); 2321 2322 newsk->sk_dst_cache = NULL; 2323 newsk->sk_dst_pending_confirm = 0; 2324 newsk->sk_wmem_queued = 0; 2325 newsk->sk_forward_alloc = 0; 2326 newsk->sk_reserved_mem = 0; 2327 atomic_set(&newsk->sk_drops, 0); 2328 newsk->sk_send_head = NULL; 2329 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2330 atomic_set(&newsk->sk_zckey, 0); 2331 2332 sock_reset_flag(newsk, SOCK_DONE); 2333 2334 /* sk->sk_memcg will be populated at accept() time */ 2335 newsk->sk_memcg = NULL; 2336 2337 cgroup_sk_clone(&newsk->sk_cgrp_data); 2338 2339 rcu_read_lock(); 2340 filter = rcu_dereference(sk->sk_filter); 2341 if (filter != NULL) 2342 /* though it's an empty new sock, the charging may fail 2343 * if sysctl_optmem_max was changed between creation of 2344 * original socket and cloning 2345 */ 2346 is_charged = sk_filter_charge(newsk, filter); 2347 RCU_INIT_POINTER(newsk->sk_filter, filter); 2348 rcu_read_unlock(); 2349 2350 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2351 /* We need to make sure that we don't uncharge the new 2352 * socket if we couldn't charge it in the first place 2353 * as otherwise we uncharge the parent's filter. 2354 */ 2355 if (!is_charged) 2356 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2357 sk_free_unlock_clone(newsk); 2358 newsk = NULL; 2359 goto out; 2360 } 2361 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2362 2363 if (bpf_sk_storage_clone(sk, newsk)) { 2364 sk_free_unlock_clone(newsk); 2365 newsk = NULL; 2366 goto out; 2367 } 2368 2369 /* Clear sk_user_data if parent had the pointer tagged 2370 * as not suitable for copying when cloning. 2371 */ 2372 if (sk_user_data_is_nocopy(newsk)) 2373 newsk->sk_user_data = NULL; 2374 2375 newsk->sk_err = 0; 2376 newsk->sk_err_soft = 0; 2377 newsk->sk_priority = 0; 2378 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2379 2380 /* Before updating sk_refcnt, we must commit prior changes to memory 2381 * (Documentation/RCU/rculist_nulls.rst for details) 2382 */ 2383 smp_wmb(); 2384 refcount_set(&newsk->sk_refcnt, 2); 2385 2386 sk_set_socket(newsk, NULL); 2387 sk_tx_queue_clear(newsk); 2388 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2389 2390 if (newsk->sk_prot->sockets_allocated) 2391 sk_sockets_allocated_inc(newsk); 2392 2393 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2394 net_enable_timestamp(); 2395 out: 2396 return newsk; 2397 } 2398 EXPORT_SYMBOL_GPL(sk_clone_lock); 2399 2400 void sk_free_unlock_clone(struct sock *sk) 2401 { 2402 /* It is still raw copy of parent, so invalidate 2403 * destructor and make plain sk_free() */ 2404 sk->sk_destruct = NULL; 2405 bh_unlock_sock(sk); 2406 sk_free(sk); 2407 } 2408 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2409 2410 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2411 { 2412 bool is_ipv6 = false; 2413 u32 max_size; 2414 2415 #if IS_ENABLED(CONFIG_IPV6) 2416 is_ipv6 = (sk->sk_family == AF_INET6 && 2417 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2418 #endif 2419 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2420 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2421 READ_ONCE(dst->dev->gso_ipv4_max_size); 2422 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2423 max_size = GSO_LEGACY_MAX_SIZE; 2424 2425 return max_size - (MAX_TCP_HEADER + 1); 2426 } 2427 2428 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2429 { 2430 u32 max_segs = 1; 2431 2432 sk->sk_route_caps = dst->dev->features; 2433 if (sk_is_tcp(sk)) 2434 sk->sk_route_caps |= NETIF_F_GSO; 2435 if (sk->sk_route_caps & NETIF_F_GSO) 2436 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2437 if (unlikely(sk->sk_gso_disabled)) 2438 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2439 if (sk_can_gso(sk)) { 2440 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2441 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2442 } else { 2443 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2444 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2445 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2446 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2447 } 2448 } 2449 sk->sk_gso_max_segs = max_segs; 2450 sk_dst_set(sk, dst); 2451 } 2452 EXPORT_SYMBOL_GPL(sk_setup_caps); 2453 2454 /* 2455 * Simple resource managers for sockets. 2456 */ 2457 2458 2459 /* 2460 * Write buffer destructor automatically called from kfree_skb. 2461 */ 2462 void sock_wfree(struct sk_buff *skb) 2463 { 2464 struct sock *sk = skb->sk; 2465 unsigned int len = skb->truesize; 2466 bool free; 2467 2468 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2469 if (sock_flag(sk, SOCK_RCU_FREE) && 2470 sk->sk_write_space == sock_def_write_space) { 2471 rcu_read_lock(); 2472 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2473 sock_def_write_space_wfree(sk); 2474 rcu_read_unlock(); 2475 if (unlikely(free)) 2476 __sk_free(sk); 2477 return; 2478 } 2479 2480 /* 2481 * Keep a reference on sk_wmem_alloc, this will be released 2482 * after sk_write_space() call 2483 */ 2484 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2485 sk->sk_write_space(sk); 2486 len = 1; 2487 } 2488 /* 2489 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2490 * could not do because of in-flight packets 2491 */ 2492 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2493 __sk_free(sk); 2494 } 2495 EXPORT_SYMBOL(sock_wfree); 2496 2497 /* This variant of sock_wfree() is used by TCP, 2498 * since it sets SOCK_USE_WRITE_QUEUE. 2499 */ 2500 void __sock_wfree(struct sk_buff *skb) 2501 { 2502 struct sock *sk = skb->sk; 2503 2504 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2505 __sk_free(sk); 2506 } 2507 2508 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2509 { 2510 skb_orphan(skb); 2511 skb->sk = sk; 2512 #ifdef CONFIG_INET 2513 if (unlikely(!sk_fullsock(sk))) { 2514 skb->destructor = sock_edemux; 2515 sock_hold(sk); 2516 return; 2517 } 2518 #endif 2519 skb->destructor = sock_wfree; 2520 skb_set_hash_from_sk(skb, sk); 2521 /* 2522 * We used to take a refcount on sk, but following operation 2523 * is enough to guarantee sk_free() wont free this sock until 2524 * all in-flight packets are completed 2525 */ 2526 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2527 } 2528 EXPORT_SYMBOL(skb_set_owner_w); 2529 2530 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2531 { 2532 #ifdef CONFIG_TLS_DEVICE 2533 /* Drivers depend on in-order delivery for crypto offload, 2534 * partial orphan breaks out-of-order-OK logic. 2535 */ 2536 if (skb->decrypted) 2537 return false; 2538 #endif 2539 return (skb->destructor == sock_wfree || 2540 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2541 } 2542 2543 /* This helper is used by netem, as it can hold packets in its 2544 * delay queue. We want to allow the owner socket to send more 2545 * packets, as if they were already TX completed by a typical driver. 2546 * But we also want to keep skb->sk set because some packet schedulers 2547 * rely on it (sch_fq for example). 2548 */ 2549 void skb_orphan_partial(struct sk_buff *skb) 2550 { 2551 if (skb_is_tcp_pure_ack(skb)) 2552 return; 2553 2554 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2555 return; 2556 2557 skb_orphan(skb); 2558 } 2559 EXPORT_SYMBOL(skb_orphan_partial); 2560 2561 /* 2562 * Read buffer destructor automatically called from kfree_skb. 2563 */ 2564 void sock_rfree(struct sk_buff *skb) 2565 { 2566 struct sock *sk = skb->sk; 2567 unsigned int len = skb->truesize; 2568 2569 atomic_sub(len, &sk->sk_rmem_alloc); 2570 sk_mem_uncharge(sk, len); 2571 } 2572 EXPORT_SYMBOL(sock_rfree); 2573 2574 /* 2575 * Buffer destructor for skbs that are not used directly in read or write 2576 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2577 */ 2578 void sock_efree(struct sk_buff *skb) 2579 { 2580 sock_put(skb->sk); 2581 } 2582 EXPORT_SYMBOL(sock_efree); 2583 2584 /* Buffer destructor for prefetch/receive path where reference count may 2585 * not be held, e.g. for listen sockets. 2586 */ 2587 #ifdef CONFIG_INET 2588 void sock_pfree(struct sk_buff *skb) 2589 { 2590 if (sk_is_refcounted(skb->sk)) 2591 sock_gen_put(skb->sk); 2592 } 2593 EXPORT_SYMBOL(sock_pfree); 2594 #endif /* CONFIG_INET */ 2595 2596 kuid_t sock_i_uid(struct sock *sk) 2597 { 2598 kuid_t uid; 2599 2600 read_lock_bh(&sk->sk_callback_lock); 2601 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2602 read_unlock_bh(&sk->sk_callback_lock); 2603 return uid; 2604 } 2605 EXPORT_SYMBOL(sock_i_uid); 2606 2607 unsigned long sock_i_ino(struct sock *sk) 2608 { 2609 unsigned long ino; 2610 2611 read_lock_bh(&sk->sk_callback_lock); 2612 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2613 read_unlock_bh(&sk->sk_callback_lock); 2614 return ino; 2615 } 2616 EXPORT_SYMBOL(sock_i_ino); 2617 2618 /* 2619 * Allocate a skb from the socket's send buffer. 2620 */ 2621 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2622 gfp_t priority) 2623 { 2624 if (force || 2625 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2626 struct sk_buff *skb = alloc_skb(size, priority); 2627 2628 if (skb) { 2629 skb_set_owner_w(skb, sk); 2630 return skb; 2631 } 2632 } 2633 return NULL; 2634 } 2635 EXPORT_SYMBOL(sock_wmalloc); 2636 2637 static void sock_ofree(struct sk_buff *skb) 2638 { 2639 struct sock *sk = skb->sk; 2640 2641 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2642 } 2643 2644 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2645 gfp_t priority) 2646 { 2647 struct sk_buff *skb; 2648 2649 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2650 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2651 READ_ONCE(sysctl_optmem_max)) 2652 return NULL; 2653 2654 skb = alloc_skb(size, priority); 2655 if (!skb) 2656 return NULL; 2657 2658 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2659 skb->sk = sk; 2660 skb->destructor = sock_ofree; 2661 return skb; 2662 } 2663 2664 /* 2665 * Allocate a memory block from the socket's option memory buffer. 2666 */ 2667 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2668 { 2669 int optmem_max = READ_ONCE(sysctl_optmem_max); 2670 2671 if ((unsigned int)size <= optmem_max && 2672 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2673 void *mem; 2674 /* First do the add, to avoid the race if kmalloc 2675 * might sleep. 2676 */ 2677 atomic_add(size, &sk->sk_omem_alloc); 2678 mem = kmalloc(size, priority); 2679 if (mem) 2680 return mem; 2681 atomic_sub(size, &sk->sk_omem_alloc); 2682 } 2683 return NULL; 2684 } 2685 EXPORT_SYMBOL(sock_kmalloc); 2686 2687 /* Free an option memory block. Note, we actually want the inline 2688 * here as this allows gcc to detect the nullify and fold away the 2689 * condition entirely. 2690 */ 2691 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2692 const bool nullify) 2693 { 2694 if (WARN_ON_ONCE(!mem)) 2695 return; 2696 if (nullify) 2697 kfree_sensitive(mem); 2698 else 2699 kfree(mem); 2700 atomic_sub(size, &sk->sk_omem_alloc); 2701 } 2702 2703 void sock_kfree_s(struct sock *sk, void *mem, int size) 2704 { 2705 __sock_kfree_s(sk, mem, size, false); 2706 } 2707 EXPORT_SYMBOL(sock_kfree_s); 2708 2709 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2710 { 2711 __sock_kfree_s(sk, mem, size, true); 2712 } 2713 EXPORT_SYMBOL(sock_kzfree_s); 2714 2715 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2716 I think, these locks should be removed for datagram sockets. 2717 */ 2718 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2719 { 2720 DEFINE_WAIT(wait); 2721 2722 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2723 for (;;) { 2724 if (!timeo) 2725 break; 2726 if (signal_pending(current)) 2727 break; 2728 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2729 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2730 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2731 break; 2732 if (sk->sk_shutdown & SEND_SHUTDOWN) 2733 break; 2734 if (sk->sk_err) 2735 break; 2736 timeo = schedule_timeout(timeo); 2737 } 2738 finish_wait(sk_sleep(sk), &wait); 2739 return timeo; 2740 } 2741 2742 2743 /* 2744 * Generic send/receive buffer handlers 2745 */ 2746 2747 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2748 unsigned long data_len, int noblock, 2749 int *errcode, int max_page_order) 2750 { 2751 struct sk_buff *skb; 2752 long timeo; 2753 int err; 2754 2755 timeo = sock_sndtimeo(sk, noblock); 2756 for (;;) { 2757 err = sock_error(sk); 2758 if (err != 0) 2759 goto failure; 2760 2761 err = -EPIPE; 2762 if (sk->sk_shutdown & SEND_SHUTDOWN) 2763 goto failure; 2764 2765 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2766 break; 2767 2768 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2769 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2770 err = -EAGAIN; 2771 if (!timeo) 2772 goto failure; 2773 if (signal_pending(current)) 2774 goto interrupted; 2775 timeo = sock_wait_for_wmem(sk, timeo); 2776 } 2777 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2778 errcode, sk->sk_allocation); 2779 if (skb) 2780 skb_set_owner_w(skb, sk); 2781 return skb; 2782 2783 interrupted: 2784 err = sock_intr_errno(timeo); 2785 failure: 2786 *errcode = err; 2787 return NULL; 2788 } 2789 EXPORT_SYMBOL(sock_alloc_send_pskb); 2790 2791 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2792 struct sockcm_cookie *sockc) 2793 { 2794 u32 tsflags; 2795 2796 switch (cmsg->cmsg_type) { 2797 case SO_MARK: 2798 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2799 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2800 return -EPERM; 2801 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2802 return -EINVAL; 2803 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2804 break; 2805 case SO_TIMESTAMPING_OLD: 2806 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2807 return -EINVAL; 2808 2809 tsflags = *(u32 *)CMSG_DATA(cmsg); 2810 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2811 return -EINVAL; 2812 2813 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2814 sockc->tsflags |= tsflags; 2815 break; 2816 case SCM_TXTIME: 2817 if (!sock_flag(sk, SOCK_TXTIME)) 2818 return -EINVAL; 2819 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2820 return -EINVAL; 2821 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2822 break; 2823 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2824 case SCM_RIGHTS: 2825 case SCM_CREDENTIALS: 2826 break; 2827 default: 2828 return -EINVAL; 2829 } 2830 return 0; 2831 } 2832 EXPORT_SYMBOL(__sock_cmsg_send); 2833 2834 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2835 struct sockcm_cookie *sockc) 2836 { 2837 struct cmsghdr *cmsg; 2838 int ret; 2839 2840 for_each_cmsghdr(cmsg, msg) { 2841 if (!CMSG_OK(msg, cmsg)) 2842 return -EINVAL; 2843 if (cmsg->cmsg_level != SOL_SOCKET) 2844 continue; 2845 ret = __sock_cmsg_send(sk, cmsg, sockc); 2846 if (ret) 2847 return ret; 2848 } 2849 return 0; 2850 } 2851 EXPORT_SYMBOL(sock_cmsg_send); 2852 2853 static void sk_enter_memory_pressure(struct sock *sk) 2854 { 2855 if (!sk->sk_prot->enter_memory_pressure) 2856 return; 2857 2858 sk->sk_prot->enter_memory_pressure(sk); 2859 } 2860 2861 static void sk_leave_memory_pressure(struct sock *sk) 2862 { 2863 if (sk->sk_prot->leave_memory_pressure) { 2864 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2865 tcp_leave_memory_pressure, sk); 2866 } else { 2867 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2868 2869 if (memory_pressure && READ_ONCE(*memory_pressure)) 2870 WRITE_ONCE(*memory_pressure, 0); 2871 } 2872 } 2873 2874 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2875 2876 /** 2877 * skb_page_frag_refill - check that a page_frag contains enough room 2878 * @sz: minimum size of the fragment we want to get 2879 * @pfrag: pointer to page_frag 2880 * @gfp: priority for memory allocation 2881 * 2882 * Note: While this allocator tries to use high order pages, there is 2883 * no guarantee that allocations succeed. Therefore, @sz MUST be 2884 * less or equal than PAGE_SIZE. 2885 */ 2886 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2887 { 2888 if (pfrag->page) { 2889 if (page_ref_count(pfrag->page) == 1) { 2890 pfrag->offset = 0; 2891 return true; 2892 } 2893 if (pfrag->offset + sz <= pfrag->size) 2894 return true; 2895 put_page(pfrag->page); 2896 } 2897 2898 pfrag->offset = 0; 2899 if (SKB_FRAG_PAGE_ORDER && 2900 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2901 /* Avoid direct reclaim but allow kswapd to wake */ 2902 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2903 __GFP_COMP | __GFP_NOWARN | 2904 __GFP_NORETRY, 2905 SKB_FRAG_PAGE_ORDER); 2906 if (likely(pfrag->page)) { 2907 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2908 return true; 2909 } 2910 } 2911 pfrag->page = alloc_page(gfp); 2912 if (likely(pfrag->page)) { 2913 pfrag->size = PAGE_SIZE; 2914 return true; 2915 } 2916 return false; 2917 } 2918 EXPORT_SYMBOL(skb_page_frag_refill); 2919 2920 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2921 { 2922 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2923 return true; 2924 2925 sk_enter_memory_pressure(sk); 2926 sk_stream_moderate_sndbuf(sk); 2927 return false; 2928 } 2929 EXPORT_SYMBOL(sk_page_frag_refill); 2930 2931 void __lock_sock(struct sock *sk) 2932 __releases(&sk->sk_lock.slock) 2933 __acquires(&sk->sk_lock.slock) 2934 { 2935 DEFINE_WAIT(wait); 2936 2937 for (;;) { 2938 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2939 TASK_UNINTERRUPTIBLE); 2940 spin_unlock_bh(&sk->sk_lock.slock); 2941 schedule(); 2942 spin_lock_bh(&sk->sk_lock.slock); 2943 if (!sock_owned_by_user(sk)) 2944 break; 2945 } 2946 finish_wait(&sk->sk_lock.wq, &wait); 2947 } 2948 2949 void __release_sock(struct sock *sk) 2950 __releases(&sk->sk_lock.slock) 2951 __acquires(&sk->sk_lock.slock) 2952 { 2953 struct sk_buff *skb, *next; 2954 2955 while ((skb = sk->sk_backlog.head) != NULL) { 2956 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2957 2958 spin_unlock_bh(&sk->sk_lock.slock); 2959 2960 do { 2961 next = skb->next; 2962 prefetch(next); 2963 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2964 skb_mark_not_on_list(skb); 2965 sk_backlog_rcv(sk, skb); 2966 2967 cond_resched(); 2968 2969 skb = next; 2970 } while (skb != NULL); 2971 2972 spin_lock_bh(&sk->sk_lock.slock); 2973 } 2974 2975 /* 2976 * Doing the zeroing here guarantee we can not loop forever 2977 * while a wild producer attempts to flood us. 2978 */ 2979 sk->sk_backlog.len = 0; 2980 } 2981 2982 void __sk_flush_backlog(struct sock *sk) 2983 { 2984 spin_lock_bh(&sk->sk_lock.slock); 2985 __release_sock(sk); 2986 spin_unlock_bh(&sk->sk_lock.slock); 2987 } 2988 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2989 2990 /** 2991 * sk_wait_data - wait for data to arrive at sk_receive_queue 2992 * @sk: sock to wait on 2993 * @timeo: for how long 2994 * @skb: last skb seen on sk_receive_queue 2995 * 2996 * Now socket state including sk->sk_err is changed only under lock, 2997 * hence we may omit checks after joining wait queue. 2998 * We check receive queue before schedule() only as optimization; 2999 * it is very likely that release_sock() added new data. 3000 */ 3001 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3002 { 3003 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3004 int rc; 3005 3006 add_wait_queue(sk_sleep(sk), &wait); 3007 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3008 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3009 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3010 remove_wait_queue(sk_sleep(sk), &wait); 3011 return rc; 3012 } 3013 EXPORT_SYMBOL(sk_wait_data); 3014 3015 /** 3016 * __sk_mem_raise_allocated - increase memory_allocated 3017 * @sk: socket 3018 * @size: memory size to allocate 3019 * @amt: pages to allocate 3020 * @kind: allocation type 3021 * 3022 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 3023 */ 3024 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3025 { 3026 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 3027 struct proto *prot = sk->sk_prot; 3028 bool charged = true; 3029 long allocated; 3030 3031 sk_memory_allocated_add(sk, amt); 3032 allocated = sk_memory_allocated(sk); 3033 if (memcg_charge && 3034 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3035 gfp_memcg_charge()))) 3036 goto suppress_allocation; 3037 3038 /* Under limit. */ 3039 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3040 sk_leave_memory_pressure(sk); 3041 return 1; 3042 } 3043 3044 /* Under pressure. */ 3045 if (allocated > sk_prot_mem_limits(sk, 1)) 3046 sk_enter_memory_pressure(sk); 3047 3048 /* Over hard limit. */ 3049 if (allocated > sk_prot_mem_limits(sk, 2)) 3050 goto suppress_allocation; 3051 3052 /* guarantee minimum buffer size under pressure */ 3053 if (kind == SK_MEM_RECV) { 3054 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3055 return 1; 3056 3057 } else { /* SK_MEM_SEND */ 3058 int wmem0 = sk_get_wmem0(sk, prot); 3059 3060 if (sk->sk_type == SOCK_STREAM) { 3061 if (sk->sk_wmem_queued < wmem0) 3062 return 1; 3063 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3064 return 1; 3065 } 3066 } 3067 3068 if (sk_has_memory_pressure(sk)) { 3069 u64 alloc; 3070 3071 if (!sk_under_memory_pressure(sk)) 3072 return 1; 3073 alloc = sk_sockets_allocated_read_positive(sk); 3074 if (sk_prot_mem_limits(sk, 2) > alloc * 3075 sk_mem_pages(sk->sk_wmem_queued + 3076 atomic_read(&sk->sk_rmem_alloc) + 3077 sk->sk_forward_alloc)) 3078 return 1; 3079 } 3080 3081 suppress_allocation: 3082 3083 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3084 sk_stream_moderate_sndbuf(sk); 3085 3086 /* Fail only if socket is _under_ its sndbuf. 3087 * In this case we cannot block, so that we have to fail. 3088 */ 3089 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3090 /* Force charge with __GFP_NOFAIL */ 3091 if (memcg_charge && !charged) { 3092 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3093 gfp_memcg_charge() | __GFP_NOFAIL); 3094 } 3095 return 1; 3096 } 3097 } 3098 3099 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3100 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3101 3102 sk_memory_allocated_sub(sk, amt); 3103 3104 if (memcg_charge && charged) 3105 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3106 3107 return 0; 3108 } 3109 3110 /** 3111 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3112 * @sk: socket 3113 * @size: memory size to allocate 3114 * @kind: allocation type 3115 * 3116 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3117 * rmem allocation. This function assumes that protocols which have 3118 * memory_pressure use sk_wmem_queued as write buffer accounting. 3119 */ 3120 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3121 { 3122 int ret, amt = sk_mem_pages(size); 3123 3124 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3125 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3126 if (!ret) 3127 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3128 return ret; 3129 } 3130 EXPORT_SYMBOL(__sk_mem_schedule); 3131 3132 /** 3133 * __sk_mem_reduce_allocated - reclaim memory_allocated 3134 * @sk: socket 3135 * @amount: number of quanta 3136 * 3137 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3138 */ 3139 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3140 { 3141 sk_memory_allocated_sub(sk, amount); 3142 3143 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3144 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3145 3146 if (sk_under_memory_pressure(sk) && 3147 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3148 sk_leave_memory_pressure(sk); 3149 } 3150 3151 /** 3152 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3153 * @sk: socket 3154 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3155 */ 3156 void __sk_mem_reclaim(struct sock *sk, int amount) 3157 { 3158 amount >>= PAGE_SHIFT; 3159 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3160 __sk_mem_reduce_allocated(sk, amount); 3161 } 3162 EXPORT_SYMBOL(__sk_mem_reclaim); 3163 3164 int sk_set_peek_off(struct sock *sk, int val) 3165 { 3166 sk->sk_peek_off = val; 3167 return 0; 3168 } 3169 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3170 3171 /* 3172 * Set of default routines for initialising struct proto_ops when 3173 * the protocol does not support a particular function. In certain 3174 * cases where it makes no sense for a protocol to have a "do nothing" 3175 * function, some default processing is provided. 3176 */ 3177 3178 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3179 { 3180 return -EOPNOTSUPP; 3181 } 3182 EXPORT_SYMBOL(sock_no_bind); 3183 3184 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3185 int len, int flags) 3186 { 3187 return -EOPNOTSUPP; 3188 } 3189 EXPORT_SYMBOL(sock_no_connect); 3190 3191 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3192 { 3193 return -EOPNOTSUPP; 3194 } 3195 EXPORT_SYMBOL(sock_no_socketpair); 3196 3197 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3198 bool kern) 3199 { 3200 return -EOPNOTSUPP; 3201 } 3202 EXPORT_SYMBOL(sock_no_accept); 3203 3204 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3205 int peer) 3206 { 3207 return -EOPNOTSUPP; 3208 } 3209 EXPORT_SYMBOL(sock_no_getname); 3210 3211 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3212 { 3213 return -EOPNOTSUPP; 3214 } 3215 EXPORT_SYMBOL(sock_no_ioctl); 3216 3217 int sock_no_listen(struct socket *sock, int backlog) 3218 { 3219 return -EOPNOTSUPP; 3220 } 3221 EXPORT_SYMBOL(sock_no_listen); 3222 3223 int sock_no_shutdown(struct socket *sock, int how) 3224 { 3225 return -EOPNOTSUPP; 3226 } 3227 EXPORT_SYMBOL(sock_no_shutdown); 3228 3229 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3230 { 3231 return -EOPNOTSUPP; 3232 } 3233 EXPORT_SYMBOL(sock_no_sendmsg); 3234 3235 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3236 { 3237 return -EOPNOTSUPP; 3238 } 3239 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3240 3241 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3242 int flags) 3243 { 3244 return -EOPNOTSUPP; 3245 } 3246 EXPORT_SYMBOL(sock_no_recvmsg); 3247 3248 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3249 { 3250 /* Mirror missing mmap method error code */ 3251 return -ENODEV; 3252 } 3253 EXPORT_SYMBOL(sock_no_mmap); 3254 3255 /* 3256 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3257 * various sock-based usage counts. 3258 */ 3259 void __receive_sock(struct file *file) 3260 { 3261 struct socket *sock; 3262 3263 sock = sock_from_file(file); 3264 if (sock) { 3265 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3266 sock_update_classid(&sock->sk->sk_cgrp_data); 3267 } 3268 } 3269 3270 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3271 { 3272 ssize_t res; 3273 struct msghdr msg = {.msg_flags = flags}; 3274 struct kvec iov; 3275 char *kaddr = kmap(page); 3276 iov.iov_base = kaddr + offset; 3277 iov.iov_len = size; 3278 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3279 kunmap(page); 3280 return res; 3281 } 3282 EXPORT_SYMBOL(sock_no_sendpage); 3283 3284 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3285 int offset, size_t size, int flags) 3286 { 3287 ssize_t res; 3288 struct msghdr msg = {.msg_flags = flags}; 3289 struct kvec iov; 3290 char *kaddr = kmap(page); 3291 3292 iov.iov_base = kaddr + offset; 3293 iov.iov_len = size; 3294 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3295 kunmap(page); 3296 return res; 3297 } 3298 EXPORT_SYMBOL(sock_no_sendpage_locked); 3299 3300 /* 3301 * Default Socket Callbacks 3302 */ 3303 3304 static void sock_def_wakeup(struct sock *sk) 3305 { 3306 struct socket_wq *wq; 3307 3308 rcu_read_lock(); 3309 wq = rcu_dereference(sk->sk_wq); 3310 if (skwq_has_sleeper(wq)) 3311 wake_up_interruptible_all(&wq->wait); 3312 rcu_read_unlock(); 3313 } 3314 3315 static void sock_def_error_report(struct sock *sk) 3316 { 3317 struct socket_wq *wq; 3318 3319 rcu_read_lock(); 3320 wq = rcu_dereference(sk->sk_wq); 3321 if (skwq_has_sleeper(wq)) 3322 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3323 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3324 rcu_read_unlock(); 3325 } 3326 3327 void sock_def_readable(struct sock *sk) 3328 { 3329 struct socket_wq *wq; 3330 3331 trace_sk_data_ready(sk); 3332 3333 rcu_read_lock(); 3334 wq = rcu_dereference(sk->sk_wq); 3335 if (skwq_has_sleeper(wq)) 3336 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3337 EPOLLRDNORM | EPOLLRDBAND); 3338 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3339 rcu_read_unlock(); 3340 } 3341 3342 static void sock_def_write_space(struct sock *sk) 3343 { 3344 struct socket_wq *wq; 3345 3346 rcu_read_lock(); 3347 3348 /* Do not wake up a writer until he can make "significant" 3349 * progress. --DaveM 3350 */ 3351 if (sock_writeable(sk)) { 3352 wq = rcu_dereference(sk->sk_wq); 3353 if (skwq_has_sleeper(wq)) 3354 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3355 EPOLLWRNORM | EPOLLWRBAND); 3356 3357 /* Should agree with poll, otherwise some programs break */ 3358 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3359 } 3360 3361 rcu_read_unlock(); 3362 } 3363 3364 /* An optimised version of sock_def_write_space(), should only be called 3365 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3366 * ->sk_wmem_alloc. 3367 */ 3368 static void sock_def_write_space_wfree(struct sock *sk) 3369 { 3370 /* Do not wake up a writer until he can make "significant" 3371 * progress. --DaveM 3372 */ 3373 if (sock_writeable(sk)) { 3374 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3375 3376 /* rely on refcount_sub from sock_wfree() */ 3377 smp_mb__after_atomic(); 3378 if (wq && waitqueue_active(&wq->wait)) 3379 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3380 EPOLLWRNORM | EPOLLWRBAND); 3381 3382 /* Should agree with poll, otherwise some programs break */ 3383 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3384 } 3385 } 3386 3387 static void sock_def_destruct(struct sock *sk) 3388 { 3389 } 3390 3391 void sk_send_sigurg(struct sock *sk) 3392 { 3393 if (sk->sk_socket && sk->sk_socket->file) 3394 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3395 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3396 } 3397 EXPORT_SYMBOL(sk_send_sigurg); 3398 3399 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3400 unsigned long expires) 3401 { 3402 if (!mod_timer(timer, expires)) 3403 sock_hold(sk); 3404 } 3405 EXPORT_SYMBOL(sk_reset_timer); 3406 3407 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3408 { 3409 if (del_timer(timer)) 3410 __sock_put(sk); 3411 } 3412 EXPORT_SYMBOL(sk_stop_timer); 3413 3414 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3415 { 3416 if (del_timer_sync(timer)) 3417 __sock_put(sk); 3418 } 3419 EXPORT_SYMBOL(sk_stop_timer_sync); 3420 3421 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3422 { 3423 sk_init_common(sk); 3424 sk->sk_send_head = NULL; 3425 3426 timer_setup(&sk->sk_timer, NULL, 0); 3427 3428 sk->sk_allocation = GFP_KERNEL; 3429 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3430 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3431 sk->sk_state = TCP_CLOSE; 3432 sk->sk_use_task_frag = true; 3433 sk_set_socket(sk, sock); 3434 3435 sock_set_flag(sk, SOCK_ZAPPED); 3436 3437 if (sock) { 3438 sk->sk_type = sock->type; 3439 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3440 sock->sk = sk; 3441 } else { 3442 RCU_INIT_POINTER(sk->sk_wq, NULL); 3443 } 3444 sk->sk_uid = uid; 3445 3446 rwlock_init(&sk->sk_callback_lock); 3447 if (sk->sk_kern_sock) 3448 lockdep_set_class_and_name( 3449 &sk->sk_callback_lock, 3450 af_kern_callback_keys + sk->sk_family, 3451 af_family_kern_clock_key_strings[sk->sk_family]); 3452 else 3453 lockdep_set_class_and_name( 3454 &sk->sk_callback_lock, 3455 af_callback_keys + sk->sk_family, 3456 af_family_clock_key_strings[sk->sk_family]); 3457 3458 sk->sk_state_change = sock_def_wakeup; 3459 sk->sk_data_ready = sock_def_readable; 3460 sk->sk_write_space = sock_def_write_space; 3461 sk->sk_error_report = sock_def_error_report; 3462 sk->sk_destruct = sock_def_destruct; 3463 3464 sk->sk_frag.page = NULL; 3465 sk->sk_frag.offset = 0; 3466 sk->sk_peek_off = -1; 3467 3468 sk->sk_peer_pid = NULL; 3469 sk->sk_peer_cred = NULL; 3470 spin_lock_init(&sk->sk_peer_lock); 3471 3472 sk->sk_write_pending = 0; 3473 sk->sk_rcvlowat = 1; 3474 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3475 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3476 3477 sk->sk_stamp = SK_DEFAULT_STAMP; 3478 #if BITS_PER_LONG==32 3479 seqlock_init(&sk->sk_stamp_seq); 3480 #endif 3481 atomic_set(&sk->sk_zckey, 0); 3482 3483 #ifdef CONFIG_NET_RX_BUSY_POLL 3484 sk->sk_napi_id = 0; 3485 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3486 #endif 3487 3488 sk->sk_max_pacing_rate = ~0UL; 3489 sk->sk_pacing_rate = ~0UL; 3490 WRITE_ONCE(sk->sk_pacing_shift, 10); 3491 sk->sk_incoming_cpu = -1; 3492 3493 sk_rx_queue_clear(sk); 3494 /* 3495 * Before updating sk_refcnt, we must commit prior changes to memory 3496 * (Documentation/RCU/rculist_nulls.rst for details) 3497 */ 3498 smp_wmb(); 3499 refcount_set(&sk->sk_refcnt, 1); 3500 atomic_set(&sk->sk_drops, 0); 3501 } 3502 EXPORT_SYMBOL(sock_init_data_uid); 3503 3504 void sock_init_data(struct socket *sock, struct sock *sk) 3505 { 3506 kuid_t uid = sock ? 3507 SOCK_INODE(sock)->i_uid : 3508 make_kuid(sock_net(sk)->user_ns, 0); 3509 3510 sock_init_data_uid(sock, sk, uid); 3511 } 3512 EXPORT_SYMBOL(sock_init_data); 3513 3514 void lock_sock_nested(struct sock *sk, int subclass) 3515 { 3516 /* The sk_lock has mutex_lock() semantics here. */ 3517 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3518 3519 might_sleep(); 3520 spin_lock_bh(&sk->sk_lock.slock); 3521 if (sock_owned_by_user_nocheck(sk)) 3522 __lock_sock(sk); 3523 sk->sk_lock.owned = 1; 3524 spin_unlock_bh(&sk->sk_lock.slock); 3525 } 3526 EXPORT_SYMBOL(lock_sock_nested); 3527 3528 void release_sock(struct sock *sk) 3529 { 3530 spin_lock_bh(&sk->sk_lock.slock); 3531 if (sk->sk_backlog.tail) 3532 __release_sock(sk); 3533 3534 /* Warning : release_cb() might need to release sk ownership, 3535 * ie call sock_release_ownership(sk) before us. 3536 */ 3537 if (sk->sk_prot->release_cb) 3538 sk->sk_prot->release_cb(sk); 3539 3540 sock_release_ownership(sk); 3541 if (waitqueue_active(&sk->sk_lock.wq)) 3542 wake_up(&sk->sk_lock.wq); 3543 spin_unlock_bh(&sk->sk_lock.slock); 3544 } 3545 EXPORT_SYMBOL(release_sock); 3546 3547 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3548 { 3549 might_sleep(); 3550 spin_lock_bh(&sk->sk_lock.slock); 3551 3552 if (!sock_owned_by_user_nocheck(sk)) { 3553 /* 3554 * Fast path return with bottom halves disabled and 3555 * sock::sk_lock.slock held. 3556 * 3557 * The 'mutex' is not contended and holding 3558 * sock::sk_lock.slock prevents all other lockers to 3559 * proceed so the corresponding unlock_sock_fast() can 3560 * avoid the slow path of release_sock() completely and 3561 * just release slock. 3562 * 3563 * From a semantical POV this is equivalent to 'acquiring' 3564 * the 'mutex', hence the corresponding lockdep 3565 * mutex_release() has to happen in the fast path of 3566 * unlock_sock_fast(). 3567 */ 3568 return false; 3569 } 3570 3571 __lock_sock(sk); 3572 sk->sk_lock.owned = 1; 3573 __acquire(&sk->sk_lock.slock); 3574 spin_unlock_bh(&sk->sk_lock.slock); 3575 return true; 3576 } 3577 EXPORT_SYMBOL(__lock_sock_fast); 3578 3579 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3580 bool timeval, bool time32) 3581 { 3582 struct sock *sk = sock->sk; 3583 struct timespec64 ts; 3584 3585 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3586 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3587 if (ts.tv_sec == -1) 3588 return -ENOENT; 3589 if (ts.tv_sec == 0) { 3590 ktime_t kt = ktime_get_real(); 3591 sock_write_timestamp(sk, kt); 3592 ts = ktime_to_timespec64(kt); 3593 } 3594 3595 if (timeval) 3596 ts.tv_nsec /= 1000; 3597 3598 #ifdef CONFIG_COMPAT_32BIT_TIME 3599 if (time32) 3600 return put_old_timespec32(&ts, userstamp); 3601 #endif 3602 #ifdef CONFIG_SPARC64 3603 /* beware of padding in sparc64 timeval */ 3604 if (timeval && !in_compat_syscall()) { 3605 struct __kernel_old_timeval __user tv = { 3606 .tv_sec = ts.tv_sec, 3607 .tv_usec = ts.tv_nsec, 3608 }; 3609 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3610 return -EFAULT; 3611 return 0; 3612 } 3613 #endif 3614 return put_timespec64(&ts, userstamp); 3615 } 3616 EXPORT_SYMBOL(sock_gettstamp); 3617 3618 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3619 { 3620 if (!sock_flag(sk, flag)) { 3621 unsigned long previous_flags = sk->sk_flags; 3622 3623 sock_set_flag(sk, flag); 3624 /* 3625 * we just set one of the two flags which require net 3626 * time stamping, but time stamping might have been on 3627 * already because of the other one 3628 */ 3629 if (sock_needs_netstamp(sk) && 3630 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3631 net_enable_timestamp(); 3632 } 3633 } 3634 3635 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3636 int level, int type) 3637 { 3638 struct sock_exterr_skb *serr; 3639 struct sk_buff *skb; 3640 int copied, err; 3641 3642 err = -EAGAIN; 3643 skb = sock_dequeue_err_skb(sk); 3644 if (skb == NULL) 3645 goto out; 3646 3647 copied = skb->len; 3648 if (copied > len) { 3649 msg->msg_flags |= MSG_TRUNC; 3650 copied = len; 3651 } 3652 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3653 if (err) 3654 goto out_free_skb; 3655 3656 sock_recv_timestamp(msg, sk, skb); 3657 3658 serr = SKB_EXT_ERR(skb); 3659 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3660 3661 msg->msg_flags |= MSG_ERRQUEUE; 3662 err = copied; 3663 3664 out_free_skb: 3665 kfree_skb(skb); 3666 out: 3667 return err; 3668 } 3669 EXPORT_SYMBOL(sock_recv_errqueue); 3670 3671 /* 3672 * Get a socket option on an socket. 3673 * 3674 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3675 * asynchronous errors should be reported by getsockopt. We assume 3676 * this means if you specify SO_ERROR (otherwise whats the point of it). 3677 */ 3678 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3679 char __user *optval, int __user *optlen) 3680 { 3681 struct sock *sk = sock->sk; 3682 3683 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3684 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3685 } 3686 EXPORT_SYMBOL(sock_common_getsockopt); 3687 3688 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3689 int flags) 3690 { 3691 struct sock *sk = sock->sk; 3692 int addr_len = 0; 3693 int err; 3694 3695 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3696 if (err >= 0) 3697 msg->msg_namelen = addr_len; 3698 return err; 3699 } 3700 EXPORT_SYMBOL(sock_common_recvmsg); 3701 3702 /* 3703 * Set socket options on an inet socket. 3704 */ 3705 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3706 sockptr_t optval, unsigned int optlen) 3707 { 3708 struct sock *sk = sock->sk; 3709 3710 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3711 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3712 } 3713 EXPORT_SYMBOL(sock_common_setsockopt); 3714 3715 void sk_common_release(struct sock *sk) 3716 { 3717 if (sk->sk_prot->destroy) 3718 sk->sk_prot->destroy(sk); 3719 3720 /* 3721 * Observation: when sk_common_release is called, processes have 3722 * no access to socket. But net still has. 3723 * Step one, detach it from networking: 3724 * 3725 * A. Remove from hash tables. 3726 */ 3727 3728 sk->sk_prot->unhash(sk); 3729 3730 /* 3731 * In this point socket cannot receive new packets, but it is possible 3732 * that some packets are in flight because some CPU runs receiver and 3733 * did hash table lookup before we unhashed socket. They will achieve 3734 * receive queue and will be purged by socket destructor. 3735 * 3736 * Also we still have packets pending on receive queue and probably, 3737 * our own packets waiting in device queues. sock_destroy will drain 3738 * receive queue, but transmitted packets will delay socket destruction 3739 * until the last reference will be released. 3740 */ 3741 3742 sock_orphan(sk); 3743 3744 xfrm_sk_free_policy(sk); 3745 3746 sock_put(sk); 3747 } 3748 EXPORT_SYMBOL(sk_common_release); 3749 3750 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3751 { 3752 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3753 3754 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3755 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3756 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3757 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3758 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3759 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3760 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3761 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3762 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3763 } 3764 3765 #ifdef CONFIG_PROC_FS 3766 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3767 3768 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3769 { 3770 int cpu, idx = prot->inuse_idx; 3771 int res = 0; 3772 3773 for_each_possible_cpu(cpu) 3774 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3775 3776 return res >= 0 ? res : 0; 3777 } 3778 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3779 3780 int sock_inuse_get(struct net *net) 3781 { 3782 int cpu, res = 0; 3783 3784 for_each_possible_cpu(cpu) 3785 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3786 3787 return res; 3788 } 3789 3790 EXPORT_SYMBOL_GPL(sock_inuse_get); 3791 3792 static int __net_init sock_inuse_init_net(struct net *net) 3793 { 3794 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3795 if (net->core.prot_inuse == NULL) 3796 return -ENOMEM; 3797 return 0; 3798 } 3799 3800 static void __net_exit sock_inuse_exit_net(struct net *net) 3801 { 3802 free_percpu(net->core.prot_inuse); 3803 } 3804 3805 static struct pernet_operations net_inuse_ops = { 3806 .init = sock_inuse_init_net, 3807 .exit = sock_inuse_exit_net, 3808 }; 3809 3810 static __init int net_inuse_init(void) 3811 { 3812 if (register_pernet_subsys(&net_inuse_ops)) 3813 panic("Cannot initialize net inuse counters"); 3814 3815 return 0; 3816 } 3817 3818 core_initcall(net_inuse_init); 3819 3820 static int assign_proto_idx(struct proto *prot) 3821 { 3822 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3823 3824 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3825 pr_err("PROTO_INUSE_NR exhausted\n"); 3826 return -ENOSPC; 3827 } 3828 3829 set_bit(prot->inuse_idx, proto_inuse_idx); 3830 return 0; 3831 } 3832 3833 static void release_proto_idx(struct proto *prot) 3834 { 3835 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3836 clear_bit(prot->inuse_idx, proto_inuse_idx); 3837 } 3838 #else 3839 static inline int assign_proto_idx(struct proto *prot) 3840 { 3841 return 0; 3842 } 3843 3844 static inline void release_proto_idx(struct proto *prot) 3845 { 3846 } 3847 3848 #endif 3849 3850 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3851 { 3852 if (!twsk_prot) 3853 return; 3854 kfree(twsk_prot->twsk_slab_name); 3855 twsk_prot->twsk_slab_name = NULL; 3856 kmem_cache_destroy(twsk_prot->twsk_slab); 3857 twsk_prot->twsk_slab = NULL; 3858 } 3859 3860 static int tw_prot_init(const struct proto *prot) 3861 { 3862 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3863 3864 if (!twsk_prot) 3865 return 0; 3866 3867 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3868 prot->name); 3869 if (!twsk_prot->twsk_slab_name) 3870 return -ENOMEM; 3871 3872 twsk_prot->twsk_slab = 3873 kmem_cache_create(twsk_prot->twsk_slab_name, 3874 twsk_prot->twsk_obj_size, 0, 3875 SLAB_ACCOUNT | prot->slab_flags, 3876 NULL); 3877 if (!twsk_prot->twsk_slab) { 3878 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3879 prot->name); 3880 return -ENOMEM; 3881 } 3882 3883 return 0; 3884 } 3885 3886 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3887 { 3888 if (!rsk_prot) 3889 return; 3890 kfree(rsk_prot->slab_name); 3891 rsk_prot->slab_name = NULL; 3892 kmem_cache_destroy(rsk_prot->slab); 3893 rsk_prot->slab = NULL; 3894 } 3895 3896 static int req_prot_init(const struct proto *prot) 3897 { 3898 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3899 3900 if (!rsk_prot) 3901 return 0; 3902 3903 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3904 prot->name); 3905 if (!rsk_prot->slab_name) 3906 return -ENOMEM; 3907 3908 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3909 rsk_prot->obj_size, 0, 3910 SLAB_ACCOUNT | prot->slab_flags, 3911 NULL); 3912 3913 if (!rsk_prot->slab) { 3914 pr_crit("%s: Can't create request sock SLAB cache!\n", 3915 prot->name); 3916 return -ENOMEM; 3917 } 3918 return 0; 3919 } 3920 3921 int proto_register(struct proto *prot, int alloc_slab) 3922 { 3923 int ret = -ENOBUFS; 3924 3925 if (prot->memory_allocated && !prot->sysctl_mem) { 3926 pr_err("%s: missing sysctl_mem\n", prot->name); 3927 return -EINVAL; 3928 } 3929 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3930 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3931 return -EINVAL; 3932 } 3933 if (alloc_slab) { 3934 prot->slab = kmem_cache_create_usercopy(prot->name, 3935 prot->obj_size, 0, 3936 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3937 prot->slab_flags, 3938 prot->useroffset, prot->usersize, 3939 NULL); 3940 3941 if (prot->slab == NULL) { 3942 pr_crit("%s: Can't create sock SLAB cache!\n", 3943 prot->name); 3944 goto out; 3945 } 3946 3947 if (req_prot_init(prot)) 3948 goto out_free_request_sock_slab; 3949 3950 if (tw_prot_init(prot)) 3951 goto out_free_timewait_sock_slab; 3952 } 3953 3954 mutex_lock(&proto_list_mutex); 3955 ret = assign_proto_idx(prot); 3956 if (ret) { 3957 mutex_unlock(&proto_list_mutex); 3958 goto out_free_timewait_sock_slab; 3959 } 3960 list_add(&prot->node, &proto_list); 3961 mutex_unlock(&proto_list_mutex); 3962 return ret; 3963 3964 out_free_timewait_sock_slab: 3965 if (alloc_slab) 3966 tw_prot_cleanup(prot->twsk_prot); 3967 out_free_request_sock_slab: 3968 if (alloc_slab) { 3969 req_prot_cleanup(prot->rsk_prot); 3970 3971 kmem_cache_destroy(prot->slab); 3972 prot->slab = NULL; 3973 } 3974 out: 3975 return ret; 3976 } 3977 EXPORT_SYMBOL(proto_register); 3978 3979 void proto_unregister(struct proto *prot) 3980 { 3981 mutex_lock(&proto_list_mutex); 3982 release_proto_idx(prot); 3983 list_del(&prot->node); 3984 mutex_unlock(&proto_list_mutex); 3985 3986 kmem_cache_destroy(prot->slab); 3987 prot->slab = NULL; 3988 3989 req_prot_cleanup(prot->rsk_prot); 3990 tw_prot_cleanup(prot->twsk_prot); 3991 } 3992 EXPORT_SYMBOL(proto_unregister); 3993 3994 int sock_load_diag_module(int family, int protocol) 3995 { 3996 if (!protocol) { 3997 if (!sock_is_registered(family)) 3998 return -ENOENT; 3999 4000 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4001 NETLINK_SOCK_DIAG, family); 4002 } 4003 4004 #ifdef CONFIG_INET 4005 if (family == AF_INET && 4006 protocol != IPPROTO_RAW && 4007 protocol < MAX_INET_PROTOS && 4008 !rcu_access_pointer(inet_protos[protocol])) 4009 return -ENOENT; 4010 #endif 4011 4012 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4013 NETLINK_SOCK_DIAG, family, protocol); 4014 } 4015 EXPORT_SYMBOL(sock_load_diag_module); 4016 4017 #ifdef CONFIG_PROC_FS 4018 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4019 __acquires(proto_list_mutex) 4020 { 4021 mutex_lock(&proto_list_mutex); 4022 return seq_list_start_head(&proto_list, *pos); 4023 } 4024 4025 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4026 { 4027 return seq_list_next(v, &proto_list, pos); 4028 } 4029 4030 static void proto_seq_stop(struct seq_file *seq, void *v) 4031 __releases(proto_list_mutex) 4032 { 4033 mutex_unlock(&proto_list_mutex); 4034 } 4035 4036 static char proto_method_implemented(const void *method) 4037 { 4038 return method == NULL ? 'n' : 'y'; 4039 } 4040 static long sock_prot_memory_allocated(struct proto *proto) 4041 { 4042 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4043 } 4044 4045 static const char *sock_prot_memory_pressure(struct proto *proto) 4046 { 4047 return proto->memory_pressure != NULL ? 4048 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4049 } 4050 4051 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4052 { 4053 4054 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4055 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4056 proto->name, 4057 proto->obj_size, 4058 sock_prot_inuse_get(seq_file_net(seq), proto), 4059 sock_prot_memory_allocated(proto), 4060 sock_prot_memory_pressure(proto), 4061 proto->max_header, 4062 proto->slab == NULL ? "no" : "yes", 4063 module_name(proto->owner), 4064 proto_method_implemented(proto->close), 4065 proto_method_implemented(proto->connect), 4066 proto_method_implemented(proto->disconnect), 4067 proto_method_implemented(proto->accept), 4068 proto_method_implemented(proto->ioctl), 4069 proto_method_implemented(proto->init), 4070 proto_method_implemented(proto->destroy), 4071 proto_method_implemented(proto->shutdown), 4072 proto_method_implemented(proto->setsockopt), 4073 proto_method_implemented(proto->getsockopt), 4074 proto_method_implemented(proto->sendmsg), 4075 proto_method_implemented(proto->recvmsg), 4076 proto_method_implemented(proto->sendpage), 4077 proto_method_implemented(proto->bind), 4078 proto_method_implemented(proto->backlog_rcv), 4079 proto_method_implemented(proto->hash), 4080 proto_method_implemented(proto->unhash), 4081 proto_method_implemented(proto->get_port), 4082 proto_method_implemented(proto->enter_memory_pressure)); 4083 } 4084 4085 static int proto_seq_show(struct seq_file *seq, void *v) 4086 { 4087 if (v == &proto_list) 4088 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4089 "protocol", 4090 "size", 4091 "sockets", 4092 "memory", 4093 "press", 4094 "maxhdr", 4095 "slab", 4096 "module", 4097 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4098 else 4099 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4100 return 0; 4101 } 4102 4103 static const struct seq_operations proto_seq_ops = { 4104 .start = proto_seq_start, 4105 .next = proto_seq_next, 4106 .stop = proto_seq_stop, 4107 .show = proto_seq_show, 4108 }; 4109 4110 static __net_init int proto_init_net(struct net *net) 4111 { 4112 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4113 sizeof(struct seq_net_private))) 4114 return -ENOMEM; 4115 4116 return 0; 4117 } 4118 4119 static __net_exit void proto_exit_net(struct net *net) 4120 { 4121 remove_proc_entry("protocols", net->proc_net); 4122 } 4123 4124 4125 static __net_initdata struct pernet_operations proto_net_ops = { 4126 .init = proto_init_net, 4127 .exit = proto_exit_net, 4128 }; 4129 4130 static int __init proto_init(void) 4131 { 4132 return register_pernet_subsys(&proto_net_ops); 4133 } 4134 4135 subsys_initcall(proto_init); 4136 4137 #endif /* PROC_FS */ 4138 4139 #ifdef CONFIG_NET_RX_BUSY_POLL 4140 bool sk_busy_loop_end(void *p, unsigned long start_time) 4141 { 4142 struct sock *sk = p; 4143 4144 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4145 sk_busy_loop_timeout(sk, start_time); 4146 } 4147 EXPORT_SYMBOL(sk_busy_loop_end); 4148 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4149 4150 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4151 { 4152 if (!sk->sk_prot->bind_add) 4153 return -EOPNOTSUPP; 4154 return sk->sk_prot->bind_add(sk, addr, addr_len); 4155 } 4156 EXPORT_SYMBOL(sock_bind_add); 4157 4158 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4159 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4160 void __user *arg, void *karg, size_t size) 4161 { 4162 int ret; 4163 4164 if (copy_from_user(karg, arg, size)) 4165 return -EFAULT; 4166 4167 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4168 if (ret) 4169 return ret; 4170 4171 if (copy_to_user(arg, karg, size)) 4172 return -EFAULT; 4173 4174 return 0; 4175 } 4176 EXPORT_SYMBOL(sock_ioctl_inout); 4177 4178 /* This is the most common ioctl prep function, where the result (4 bytes) is 4179 * copied back to userspace if the ioctl() returns successfully. No input is 4180 * copied from userspace as input argument. 4181 */ 4182 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4183 { 4184 int ret, karg = 0; 4185 4186 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4187 if (ret) 4188 return ret; 4189 4190 return put_user(karg, (int __user *)arg); 4191 } 4192 4193 /* A wrapper around sock ioctls, which copies the data from userspace 4194 * (depending on the protocol/ioctl), and copies back the result to userspace. 4195 * The main motivation for this function is to pass kernel memory to the 4196 * protocol ioctl callbacks, instead of userspace memory. 4197 */ 4198 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4199 { 4200 int rc = 1; 4201 4202 if (sk_is_ipmr(sk)) 4203 rc = ipmr_sk_ioctl(sk, cmd, arg); 4204 else if (sk_is_icmpv6(sk)) 4205 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4206 else if (sk_is_phonet(sk)) 4207 rc = phonet_sk_ioctl(sk, cmd, arg); 4208 4209 /* If ioctl was processed, returns its value */ 4210 if (rc <= 0) 4211 return rc; 4212 4213 /* Otherwise call the default handler */ 4214 return sock_ioctl_out(sk, cmd, arg); 4215 } 4216 EXPORT_SYMBOL(sk_ioctl); 4217