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 <linux/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/udp.h> 111 #include <linux/init.h> 112 #include <linux/highmem.h> 113 #include <linux/user_namespace.h> 114 #include <linux/static_key.h> 115 #include <linux/memcontrol.h> 116 #include <linux/prefetch.h> 117 #include <linux/compat.h> 118 #include <linux/mroute.h> 119 #include <linux/mroute6.h> 120 #include <linux/icmpv6.h> 121 122 #include <linux/uaccess.h> 123 124 #include <linux/netdevice.h> 125 #include <net/protocol.h> 126 #include <linux/skbuff.h> 127 #include <linux/skbuff_ref.h> 128 #include <net/net_namespace.h> 129 #include <net/request_sock.h> 130 #include <net/sock.h> 131 #include <net/proto_memory.h> 132 #include <linux/net_tstamp.h> 133 #include <net/xfrm.h> 134 #include <linux/ipsec.h> 135 #include <net/cls_cgroup.h> 136 #include <net/netprio_cgroup.h> 137 #include <linux/sock_diag.h> 138 139 #include <linux/filter.h> 140 #include <net/sock_reuseport.h> 141 #include <net/bpf_sk_storage.h> 142 143 #include <trace/events/sock.h> 144 145 #include <net/tcp.h> 146 #include <net/busy_poll.h> 147 #include <net/phonet/phonet.h> 148 149 #include <linux/ethtool.h> 150 151 #include "dev.h" 152 153 static DEFINE_MUTEX(proto_list_mutex); 154 static LIST_HEAD(proto_list); 155 156 static void sock_def_write_space_wfree(struct sock *sk); 157 static void sock_def_write_space(struct sock *sk); 158 159 /** 160 * sk_ns_capable - General socket capability test 161 * @sk: Socket to use a capability on or through 162 * @user_ns: The user namespace of the capability to use 163 * @cap: The capability to use 164 * 165 * Test to see if the opener of the socket had when the socket was 166 * created and the current process has the capability @cap in the user 167 * namespace @user_ns. 168 */ 169 bool sk_ns_capable(const struct sock *sk, 170 struct user_namespace *user_ns, int cap) 171 { 172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 173 ns_capable(user_ns, cap); 174 } 175 EXPORT_SYMBOL(sk_ns_capable); 176 177 /** 178 * sk_capable - Socket global capability test 179 * @sk: Socket to use a capability on or through 180 * @cap: The global capability to use 181 * 182 * Test to see if the opener of the socket had when the socket was 183 * created and the current process has the capability @cap in all user 184 * namespaces. 185 */ 186 bool sk_capable(const struct sock *sk, int cap) 187 { 188 return sk_ns_capable(sk, &init_user_ns, cap); 189 } 190 EXPORT_SYMBOL(sk_capable); 191 192 /** 193 * sk_net_capable - Network namespace socket capability test 194 * @sk: Socket to use a capability on or through 195 * @cap: The capability to use 196 * 197 * Test to see if the opener of the socket had when the socket was created 198 * and the current process has the capability @cap over the network namespace 199 * the socket is a member of. 200 */ 201 bool sk_net_capable(const struct sock *sk, int cap) 202 { 203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 204 } 205 EXPORT_SYMBOL(sk_net_capable); 206 207 /* 208 * Each address family might have different locking rules, so we have 209 * one slock key per address family and separate keys for internal and 210 * userspace sockets. 211 */ 212 static struct lock_class_key af_family_keys[AF_MAX]; 213 static struct lock_class_key af_family_kern_keys[AF_MAX]; 214 static struct lock_class_key af_family_slock_keys[AF_MAX]; 215 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 216 217 /* 218 * Make lock validator output more readable. (we pre-construct these 219 * strings build-time, so that runtime initialization of socket 220 * locks is fast): 221 */ 222 223 #define _sock_locks(x) \ 224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 233 x "27" , x "28" , x "AF_CAN" , \ 234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 239 x "AF_MCTP" , \ 240 x "AF_MAX" 241 242 static const char *const af_family_key_strings[AF_MAX+1] = { 243 _sock_locks("sk_lock-") 244 }; 245 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 246 _sock_locks("slock-") 247 }; 248 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 249 _sock_locks("clock-") 250 }; 251 252 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 253 _sock_locks("k-sk_lock-") 254 }; 255 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 256 _sock_locks("k-slock-") 257 }; 258 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 259 _sock_locks("k-clock-") 260 }; 261 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 262 _sock_locks("rlock-") 263 }; 264 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 265 _sock_locks("wlock-") 266 }; 267 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 268 _sock_locks("elock-") 269 }; 270 271 /* 272 * sk_callback_lock and sk queues locking rules are per-address-family, 273 * so split the lock classes by using a per-AF key: 274 */ 275 static struct lock_class_key af_callback_keys[AF_MAX]; 276 static struct lock_class_key af_rlock_keys[AF_MAX]; 277 static struct lock_class_key af_wlock_keys[AF_MAX]; 278 static struct lock_class_key af_elock_keys[AF_MAX]; 279 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 280 281 /* Run time adjustable parameters. */ 282 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 283 EXPORT_SYMBOL(sysctl_wmem_max); 284 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 285 EXPORT_SYMBOL(sysctl_rmem_max); 286 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 287 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 288 289 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 290 EXPORT_SYMBOL_GPL(memalloc_socks_key); 291 292 /** 293 * sk_set_memalloc - sets %SOCK_MEMALLOC 294 * @sk: socket to set it on 295 * 296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 297 * It's the responsibility of the admin to adjust min_free_kbytes 298 * to meet the requirements 299 */ 300 void sk_set_memalloc(struct sock *sk) 301 { 302 sock_set_flag(sk, SOCK_MEMALLOC); 303 sk->sk_allocation |= __GFP_MEMALLOC; 304 static_branch_inc(&memalloc_socks_key); 305 } 306 EXPORT_SYMBOL_GPL(sk_set_memalloc); 307 308 void sk_clear_memalloc(struct sock *sk) 309 { 310 sock_reset_flag(sk, SOCK_MEMALLOC); 311 sk->sk_allocation &= ~__GFP_MEMALLOC; 312 static_branch_dec(&memalloc_socks_key); 313 314 /* 315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 316 * progress of swapping. SOCK_MEMALLOC may be cleared while 317 * it has rmem allocations due to the last swapfile being deactivated 318 * but there is a risk that the socket is unusable due to exceeding 319 * the rmem limits. Reclaim the reserves and obey rmem limits again. 320 */ 321 sk_mem_reclaim(sk); 322 } 323 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 324 325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 326 { 327 int ret; 328 unsigned int noreclaim_flag; 329 330 /* these should have been dropped before queueing */ 331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 332 333 noreclaim_flag = memalloc_noreclaim_save(); 334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 335 tcp_v6_do_rcv, 336 tcp_v4_do_rcv, 337 sk, skb); 338 memalloc_noreclaim_restore(noreclaim_flag); 339 340 return ret; 341 } 342 EXPORT_SYMBOL(__sk_backlog_rcv); 343 344 void sk_error_report(struct sock *sk) 345 { 346 sk->sk_error_report(sk); 347 348 switch (sk->sk_family) { 349 case AF_INET: 350 fallthrough; 351 case AF_INET6: 352 trace_inet_sk_error_report(sk); 353 break; 354 default: 355 break; 356 } 357 } 358 EXPORT_SYMBOL(sk_error_report); 359 360 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 361 { 362 struct __kernel_sock_timeval tv; 363 364 if (timeo == MAX_SCHEDULE_TIMEOUT) { 365 tv.tv_sec = 0; 366 tv.tv_usec = 0; 367 } else { 368 tv.tv_sec = timeo / HZ; 369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 370 } 371 372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 374 *(struct old_timeval32 *)optval = tv32; 375 return sizeof(tv32); 376 } 377 378 if (old_timeval) { 379 struct __kernel_old_timeval old_tv; 380 old_tv.tv_sec = tv.tv_sec; 381 old_tv.tv_usec = tv.tv_usec; 382 *(struct __kernel_old_timeval *)optval = old_tv; 383 return sizeof(old_tv); 384 } 385 386 *(struct __kernel_sock_timeval *)optval = tv; 387 return sizeof(tv); 388 } 389 EXPORT_SYMBOL(sock_get_timeout); 390 391 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 392 sockptr_t optval, int optlen, bool old_timeval) 393 { 394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 395 struct old_timeval32 tv32; 396 397 if (optlen < sizeof(tv32)) 398 return -EINVAL; 399 400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 401 return -EFAULT; 402 tv->tv_sec = tv32.tv_sec; 403 tv->tv_usec = tv32.tv_usec; 404 } else if (old_timeval) { 405 struct __kernel_old_timeval old_tv; 406 407 if (optlen < sizeof(old_tv)) 408 return -EINVAL; 409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 410 return -EFAULT; 411 tv->tv_sec = old_tv.tv_sec; 412 tv->tv_usec = old_tv.tv_usec; 413 } else { 414 if (optlen < sizeof(*tv)) 415 return -EINVAL; 416 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 417 return -EFAULT; 418 } 419 420 return 0; 421 } 422 EXPORT_SYMBOL(sock_copy_user_timeval); 423 424 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 425 bool old_timeval) 426 { 427 struct __kernel_sock_timeval tv; 428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 429 long val; 430 431 if (err) 432 return err; 433 434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 435 return -EDOM; 436 437 if (tv.tv_sec < 0) { 438 static int warned __read_mostly; 439 440 WRITE_ONCE(*timeo_p, 0); 441 if (warned < 10 && net_ratelimit()) { 442 warned++; 443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 444 __func__, current->comm, task_pid_nr(current)); 445 } 446 return 0; 447 } 448 val = MAX_SCHEDULE_TIMEOUT; 449 if ((tv.tv_sec || tv.tv_usec) && 450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 452 USEC_PER_SEC / HZ); 453 WRITE_ONCE(*timeo_p, val); 454 return 0; 455 } 456 457 static bool sock_needs_netstamp(const struct sock *sk) 458 { 459 switch (sk->sk_family) { 460 case AF_UNSPEC: 461 case AF_UNIX: 462 return false; 463 default: 464 return true; 465 } 466 } 467 468 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 469 { 470 if (sk->sk_flags & flags) { 471 sk->sk_flags &= ~flags; 472 if (sock_needs_netstamp(sk) && 473 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 474 net_disable_timestamp(); 475 } 476 } 477 478 479 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 480 { 481 unsigned long flags; 482 struct sk_buff_head *list = &sk->sk_receive_queue; 483 484 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) { 485 atomic_inc(&sk->sk_drops); 486 trace_sock_rcvqueue_full(sk, skb); 487 return -ENOMEM; 488 } 489 490 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 491 atomic_inc(&sk->sk_drops); 492 return -ENOBUFS; 493 } 494 495 skb->dev = NULL; 496 skb_set_owner_r(skb, sk); 497 498 /* we escape from rcu protected region, make sure we dont leak 499 * a norefcounted dst 500 */ 501 skb_dst_force(skb); 502 503 spin_lock_irqsave(&list->lock, flags); 504 sock_skb_set_dropcount(sk, skb); 505 __skb_queue_tail(list, skb); 506 spin_unlock_irqrestore(&list->lock, flags); 507 508 if (!sock_flag(sk, SOCK_DEAD)) 509 sk->sk_data_ready(sk); 510 return 0; 511 } 512 EXPORT_SYMBOL(__sock_queue_rcv_skb); 513 514 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 515 enum skb_drop_reason *reason) 516 { 517 enum skb_drop_reason drop_reason; 518 int err; 519 520 err = sk_filter(sk, skb); 521 if (err) { 522 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 523 goto out; 524 } 525 err = __sock_queue_rcv_skb(sk, skb); 526 switch (err) { 527 case -ENOMEM: 528 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 529 break; 530 case -ENOBUFS: 531 drop_reason = SKB_DROP_REASON_PROTO_MEM; 532 break; 533 default: 534 drop_reason = SKB_NOT_DROPPED_YET; 535 break; 536 } 537 out: 538 if (reason) 539 *reason = drop_reason; 540 return err; 541 } 542 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 543 544 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 545 const int nested, unsigned int trim_cap, bool refcounted) 546 { 547 int rc = NET_RX_SUCCESS; 548 549 if (sk_filter_trim_cap(sk, skb, trim_cap)) 550 goto discard_and_relse; 551 552 skb->dev = NULL; 553 554 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) { 555 atomic_inc(&sk->sk_drops); 556 goto discard_and_relse; 557 } 558 if (nested) 559 bh_lock_sock_nested(sk); 560 else 561 bh_lock_sock(sk); 562 if (!sock_owned_by_user(sk)) { 563 /* 564 * trylock + unlock semantics: 565 */ 566 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 567 568 rc = sk_backlog_rcv(sk, skb); 569 570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 571 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 572 bh_unlock_sock(sk); 573 atomic_inc(&sk->sk_drops); 574 goto discard_and_relse; 575 } 576 577 bh_unlock_sock(sk); 578 out: 579 if (refcounted) 580 sock_put(sk); 581 return rc; 582 discard_and_relse: 583 kfree_skb(skb); 584 goto out; 585 } 586 EXPORT_SYMBOL(__sk_receive_skb); 587 588 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 589 u32)); 590 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 591 u32)); 592 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 593 { 594 struct dst_entry *dst = __sk_dst_get(sk); 595 596 if (dst && dst->obsolete && 597 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 598 dst, cookie) == NULL) { 599 sk_tx_queue_clear(sk); 600 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 601 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 602 dst_release(dst); 603 return NULL; 604 } 605 606 return dst; 607 } 608 EXPORT_SYMBOL(__sk_dst_check); 609 610 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 611 { 612 struct dst_entry *dst = sk_dst_get(sk); 613 614 if (dst && dst->obsolete && 615 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 616 dst, cookie) == NULL) { 617 sk_dst_reset(sk); 618 dst_release(dst); 619 return NULL; 620 } 621 622 return dst; 623 } 624 EXPORT_SYMBOL(sk_dst_check); 625 626 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 627 { 628 int ret = -ENOPROTOOPT; 629 #ifdef CONFIG_NETDEVICES 630 struct net *net = sock_net(sk); 631 632 /* Sorry... */ 633 ret = -EPERM; 634 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 635 goto out; 636 637 ret = -EINVAL; 638 if (ifindex < 0) 639 goto out; 640 641 /* Paired with all READ_ONCE() done locklessly. */ 642 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 643 644 if (sk->sk_prot->rehash) 645 sk->sk_prot->rehash(sk); 646 sk_dst_reset(sk); 647 648 ret = 0; 649 650 out: 651 #endif 652 653 return ret; 654 } 655 656 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 657 { 658 int ret; 659 660 if (lock_sk) 661 lock_sock(sk); 662 ret = sock_bindtoindex_locked(sk, ifindex); 663 if (lock_sk) 664 release_sock(sk); 665 666 return ret; 667 } 668 EXPORT_SYMBOL(sock_bindtoindex); 669 670 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 671 { 672 int ret = -ENOPROTOOPT; 673 #ifdef CONFIG_NETDEVICES 674 struct net *net = sock_net(sk); 675 char devname[IFNAMSIZ]; 676 int index; 677 678 ret = -EINVAL; 679 if (optlen < 0) 680 goto out; 681 682 /* Bind this socket to a particular device like "eth0", 683 * as specified in the passed interface name. If the 684 * name is "" or the option length is zero the socket 685 * is not bound. 686 */ 687 if (optlen > IFNAMSIZ - 1) 688 optlen = IFNAMSIZ - 1; 689 memset(devname, 0, sizeof(devname)); 690 691 ret = -EFAULT; 692 if (copy_from_sockptr(devname, optval, optlen)) 693 goto out; 694 695 index = 0; 696 if (devname[0] != '\0') { 697 struct net_device *dev; 698 699 rcu_read_lock(); 700 dev = dev_get_by_name_rcu(net, devname); 701 if (dev) 702 index = dev->ifindex; 703 rcu_read_unlock(); 704 ret = -ENODEV; 705 if (!dev) 706 goto out; 707 } 708 709 sockopt_lock_sock(sk); 710 ret = sock_bindtoindex_locked(sk, index); 711 sockopt_release_sock(sk); 712 out: 713 #endif 714 715 return ret; 716 } 717 718 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 719 sockptr_t optlen, int len) 720 { 721 int ret = -ENOPROTOOPT; 722 #ifdef CONFIG_NETDEVICES 723 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 724 struct net *net = sock_net(sk); 725 char devname[IFNAMSIZ]; 726 727 if (bound_dev_if == 0) { 728 len = 0; 729 goto zero; 730 } 731 732 ret = -EINVAL; 733 if (len < IFNAMSIZ) 734 goto out; 735 736 ret = netdev_get_name(net, devname, bound_dev_if); 737 if (ret) 738 goto out; 739 740 len = strlen(devname) + 1; 741 742 ret = -EFAULT; 743 if (copy_to_sockptr(optval, devname, len)) 744 goto out; 745 746 zero: 747 ret = -EFAULT; 748 if (copy_to_sockptr(optlen, &len, sizeof(int))) 749 goto out; 750 751 ret = 0; 752 753 out: 754 #endif 755 756 return ret; 757 } 758 759 bool sk_mc_loop(const struct sock *sk) 760 { 761 if (dev_recursion_level()) 762 return false; 763 if (!sk) 764 return true; 765 /* IPV6_ADDRFORM can change sk->sk_family under us. */ 766 switch (READ_ONCE(sk->sk_family)) { 767 case AF_INET: 768 return inet_test_bit(MC_LOOP, sk); 769 #if IS_ENABLED(CONFIG_IPV6) 770 case AF_INET6: 771 return inet6_test_bit(MC6_LOOP, sk); 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 WRITE_ONCE(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 WRITE_ONCE(sk->sk_priority, priority); 807 } 808 EXPORT_SYMBOL(sock_set_priority); 809 810 void sock_set_sndtimeo(struct sock *sk, s64 secs) 811 { 812 lock_sock(sk); 813 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 814 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 815 else 816 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 817 release_sock(sk); 818 } 819 EXPORT_SYMBOL(sock_set_sndtimeo); 820 821 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 822 { 823 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val); 824 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns); 825 if (val) { 826 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 827 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 828 } 829 } 830 831 void sock_enable_timestamps(struct sock *sk) 832 { 833 lock_sock(sk); 834 __sock_set_timestamps(sk, true, false, true); 835 release_sock(sk); 836 } 837 EXPORT_SYMBOL(sock_enable_timestamps); 838 839 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 840 { 841 switch (optname) { 842 case SO_TIMESTAMP_OLD: 843 __sock_set_timestamps(sk, valbool, false, false); 844 break; 845 case SO_TIMESTAMP_NEW: 846 __sock_set_timestamps(sk, valbool, true, false); 847 break; 848 case SO_TIMESTAMPNS_OLD: 849 __sock_set_timestamps(sk, valbool, false, true); 850 break; 851 case SO_TIMESTAMPNS_NEW: 852 __sock_set_timestamps(sk, valbool, true, true); 853 break; 854 } 855 } 856 857 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 858 { 859 struct net *net = sock_net(sk); 860 struct net_device *dev = NULL; 861 bool match = false; 862 int *vclock_index; 863 int i, num; 864 865 if (sk->sk_bound_dev_if) 866 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 867 868 if (!dev) { 869 pr_err("%s: sock not bind to device\n", __func__); 870 return -EOPNOTSUPP; 871 } 872 873 num = ethtool_get_phc_vclocks(dev, &vclock_index); 874 dev_put(dev); 875 876 for (i = 0; i < num; i++) { 877 if (*(vclock_index + i) == phc_index) { 878 match = true; 879 break; 880 } 881 } 882 883 if (num > 0) 884 kfree(vclock_index); 885 886 if (!match) 887 return -EINVAL; 888 889 WRITE_ONCE(sk->sk_bind_phc, phc_index); 890 891 return 0; 892 } 893 894 int sock_set_timestamping(struct sock *sk, int optname, 895 struct so_timestamping timestamping) 896 { 897 int val = timestamping.flags; 898 int ret; 899 900 if (val & ~SOF_TIMESTAMPING_MASK) 901 return -EINVAL; 902 903 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 904 !(val & SOF_TIMESTAMPING_OPT_ID)) 905 return -EINVAL; 906 907 if (val & SOF_TIMESTAMPING_OPT_ID && 908 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 909 if (sk_is_tcp(sk)) { 910 if ((1 << sk->sk_state) & 911 (TCPF_CLOSE | TCPF_LISTEN)) 912 return -EINVAL; 913 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 915 else 916 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 917 } else { 918 atomic_set(&sk->sk_tskey, 0); 919 } 920 } 921 922 if (val & SOF_TIMESTAMPING_OPT_STATS && 923 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 924 return -EINVAL; 925 926 if (val & SOF_TIMESTAMPING_BIND_PHC) { 927 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 928 if (ret) 929 return ret; 930 } 931 932 WRITE_ONCE(sk->sk_tsflags, val); 933 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 934 935 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 936 sock_enable_timestamp(sk, 937 SOCK_TIMESTAMPING_RX_SOFTWARE); 938 else 939 sock_disable_timestamp(sk, 940 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 941 return 0; 942 } 943 944 void sock_set_keepalive(struct sock *sk) 945 { 946 lock_sock(sk); 947 if (sk->sk_prot->keepalive) 948 sk->sk_prot->keepalive(sk, true); 949 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 950 release_sock(sk); 951 } 952 EXPORT_SYMBOL(sock_set_keepalive); 953 954 static void __sock_set_rcvbuf(struct sock *sk, int val) 955 { 956 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 957 * as a negative value. 958 */ 959 val = min_t(int, val, INT_MAX / 2); 960 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 961 962 /* We double it on the way in to account for "struct sk_buff" etc. 963 * overhead. Applications assume that the SO_RCVBUF setting they make 964 * will allow that much actual data to be received on that socket. 965 * 966 * Applications are unaware that "struct sk_buff" and other overheads 967 * allocate from the receive buffer during socket buffer allocation. 968 * 969 * And after considering the possible alternatives, returning the value 970 * we actually used in getsockopt is the most desirable behavior. 971 */ 972 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 973 } 974 975 void sock_set_rcvbuf(struct sock *sk, int val) 976 { 977 lock_sock(sk); 978 __sock_set_rcvbuf(sk, val); 979 release_sock(sk); 980 } 981 EXPORT_SYMBOL(sock_set_rcvbuf); 982 983 static void __sock_set_mark(struct sock *sk, u32 val) 984 { 985 if (val != sk->sk_mark) { 986 WRITE_ONCE(sk->sk_mark, val); 987 sk_dst_reset(sk); 988 } 989 } 990 991 void sock_set_mark(struct sock *sk, u32 val) 992 { 993 lock_sock(sk); 994 __sock_set_mark(sk, val); 995 release_sock(sk); 996 } 997 EXPORT_SYMBOL(sock_set_mark); 998 999 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1000 { 1001 /* Round down bytes to multiple of pages */ 1002 bytes = round_down(bytes, PAGE_SIZE); 1003 1004 WARN_ON(bytes > sk->sk_reserved_mem); 1005 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1006 sk_mem_reclaim(sk); 1007 } 1008 1009 static int sock_reserve_memory(struct sock *sk, int bytes) 1010 { 1011 long allocated; 1012 bool charged; 1013 int pages; 1014 1015 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1016 return -EOPNOTSUPP; 1017 1018 if (!bytes) 1019 return 0; 1020 1021 pages = sk_mem_pages(bytes); 1022 1023 /* pre-charge to memcg */ 1024 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1025 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1026 if (!charged) 1027 return -ENOMEM; 1028 1029 /* pre-charge to forward_alloc */ 1030 sk_memory_allocated_add(sk, pages); 1031 allocated = sk_memory_allocated(sk); 1032 /* If the system goes into memory pressure with this 1033 * precharge, give up and return error. 1034 */ 1035 if (allocated > sk_prot_mem_limits(sk, 1)) { 1036 sk_memory_allocated_sub(sk, pages); 1037 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1038 return -ENOMEM; 1039 } 1040 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1041 1042 WRITE_ONCE(sk->sk_reserved_mem, 1043 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1044 1045 return 0; 1046 } 1047 1048 #ifdef CONFIG_PAGE_POOL 1049 1050 /* This is the number of tokens that the user can SO_DEVMEM_DONTNEED in 1051 * 1 syscall. The limit exists to limit the amount of memory the kernel 1052 * allocates to copy these tokens. 1053 */ 1054 #define MAX_DONTNEED_TOKENS 128 1055 1056 static noinline_for_stack int 1057 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1058 { 1059 unsigned int num_tokens, i, j, k, netmem_num = 0; 1060 struct dmabuf_token *tokens; 1061 netmem_ref netmems[16]; 1062 int ret = 0; 1063 1064 if (!sk_is_tcp(sk)) 1065 return -EBADF; 1066 1067 if (optlen % sizeof(struct dmabuf_token) || 1068 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1069 return -EINVAL; 1070 1071 tokens = kvmalloc_array(optlen, sizeof(*tokens), GFP_KERNEL); 1072 if (!tokens) 1073 return -ENOMEM; 1074 1075 num_tokens = optlen / sizeof(struct dmabuf_token); 1076 if (copy_from_sockptr(tokens, optval, optlen)) { 1077 kvfree(tokens); 1078 return -EFAULT; 1079 } 1080 1081 xa_lock_bh(&sk->sk_user_frags); 1082 for (i = 0; i < num_tokens; i++) { 1083 for (j = 0; j < tokens[i].token_count; j++) { 1084 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1085 &sk->sk_user_frags, tokens[i].token_start + j); 1086 1087 if (netmem && 1088 !WARN_ON_ONCE(!netmem_is_net_iov(netmem))) { 1089 netmems[netmem_num++] = netmem; 1090 if (netmem_num == ARRAY_SIZE(netmems)) { 1091 xa_unlock_bh(&sk->sk_user_frags); 1092 for (k = 0; k < netmem_num; k++) 1093 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1094 netmem_num = 0; 1095 xa_lock_bh(&sk->sk_user_frags); 1096 } 1097 ret++; 1098 } 1099 } 1100 } 1101 1102 xa_unlock_bh(&sk->sk_user_frags); 1103 for (k = 0; k < netmem_num; k++) 1104 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1105 1106 kvfree(tokens); 1107 return ret; 1108 } 1109 #endif 1110 1111 void sockopt_lock_sock(struct sock *sk) 1112 { 1113 /* When current->bpf_ctx is set, the setsockopt is called from 1114 * a bpf prog. bpf has ensured the sk lock has been 1115 * acquired before calling setsockopt(). 1116 */ 1117 if (has_current_bpf_ctx()) 1118 return; 1119 1120 lock_sock(sk); 1121 } 1122 EXPORT_SYMBOL(sockopt_lock_sock); 1123 1124 void sockopt_release_sock(struct sock *sk) 1125 { 1126 if (has_current_bpf_ctx()) 1127 return; 1128 1129 release_sock(sk); 1130 } 1131 EXPORT_SYMBOL(sockopt_release_sock); 1132 1133 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1134 { 1135 return has_current_bpf_ctx() || ns_capable(ns, cap); 1136 } 1137 EXPORT_SYMBOL(sockopt_ns_capable); 1138 1139 bool sockopt_capable(int cap) 1140 { 1141 return has_current_bpf_ctx() || capable(cap); 1142 } 1143 EXPORT_SYMBOL(sockopt_capable); 1144 1145 static int sockopt_validate_clockid(__kernel_clockid_t value) 1146 { 1147 switch (value) { 1148 case CLOCK_REALTIME: 1149 case CLOCK_MONOTONIC: 1150 case CLOCK_TAI: 1151 return 0; 1152 } 1153 return -EINVAL; 1154 } 1155 1156 /* 1157 * This is meant for all protocols to use and covers goings on 1158 * at the socket level. Everything here is generic. 1159 */ 1160 1161 int sk_setsockopt(struct sock *sk, int level, int optname, 1162 sockptr_t optval, unsigned int optlen) 1163 { 1164 struct so_timestamping timestamping; 1165 struct socket *sock = sk->sk_socket; 1166 struct sock_txtime sk_txtime; 1167 int val; 1168 int valbool; 1169 struct linger ling; 1170 int ret = 0; 1171 1172 /* 1173 * Options without arguments 1174 */ 1175 1176 if (optname == SO_BINDTODEVICE) 1177 return sock_setbindtodevice(sk, optval, optlen); 1178 1179 if (optlen < sizeof(int)) 1180 return -EINVAL; 1181 1182 if (copy_from_sockptr(&val, optval, sizeof(val))) 1183 return -EFAULT; 1184 1185 valbool = val ? 1 : 0; 1186 1187 /* handle options which do not require locking the socket. */ 1188 switch (optname) { 1189 case SO_PRIORITY: 1190 if ((val >= 0 && val <= 6) || 1191 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1192 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1193 sock_set_priority(sk, val); 1194 return 0; 1195 } 1196 return -EPERM; 1197 case SO_PASSSEC: 1198 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1199 return 0; 1200 case SO_PASSCRED: 1201 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1202 return 0; 1203 case SO_PASSPIDFD: 1204 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1205 return 0; 1206 case SO_TYPE: 1207 case SO_PROTOCOL: 1208 case SO_DOMAIN: 1209 case SO_ERROR: 1210 return -ENOPROTOOPT; 1211 #ifdef CONFIG_NET_RX_BUSY_POLL 1212 case SO_BUSY_POLL: 1213 if (val < 0) 1214 return -EINVAL; 1215 WRITE_ONCE(sk->sk_ll_usec, val); 1216 return 0; 1217 case SO_PREFER_BUSY_POLL: 1218 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1219 return -EPERM; 1220 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1221 return 0; 1222 case SO_BUSY_POLL_BUDGET: 1223 if (val > READ_ONCE(sk->sk_busy_poll_budget) && 1224 !sockopt_capable(CAP_NET_ADMIN)) 1225 return -EPERM; 1226 if (val < 0 || val > U16_MAX) 1227 return -EINVAL; 1228 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1229 return 0; 1230 #endif 1231 case SO_MAX_PACING_RATE: 1232 { 1233 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1234 unsigned long pacing_rate; 1235 1236 if (sizeof(ulval) != sizeof(val) && 1237 optlen >= sizeof(ulval) && 1238 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1239 return -EFAULT; 1240 } 1241 if (ulval != ~0UL) 1242 cmpxchg(&sk->sk_pacing_status, 1243 SK_PACING_NONE, 1244 SK_PACING_NEEDED); 1245 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1246 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1247 pacing_rate = READ_ONCE(sk->sk_pacing_rate); 1248 if (ulval < pacing_rate) 1249 WRITE_ONCE(sk->sk_pacing_rate, ulval); 1250 return 0; 1251 } 1252 case SO_TXREHASH: 1253 if (val < -1 || val > 1) 1254 return -EINVAL; 1255 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1256 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1257 /* Paired with READ_ONCE() in tcp_rtx_synack() 1258 * and sk_getsockopt(). 1259 */ 1260 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1261 return 0; 1262 case SO_PEEK_OFF: 1263 { 1264 int (*set_peek_off)(struct sock *sk, int val); 1265 1266 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1267 if (set_peek_off) 1268 ret = set_peek_off(sk, val); 1269 else 1270 ret = -EOPNOTSUPP; 1271 return ret; 1272 } 1273 #ifdef CONFIG_PAGE_POOL 1274 case SO_DEVMEM_DONTNEED: 1275 return sock_devmem_dontneed(sk, optval, optlen); 1276 #endif 1277 } 1278 1279 sockopt_lock_sock(sk); 1280 1281 switch (optname) { 1282 case SO_DEBUG: 1283 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1284 ret = -EACCES; 1285 else 1286 sock_valbool_flag(sk, SOCK_DBG, valbool); 1287 break; 1288 case SO_REUSEADDR: 1289 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1290 break; 1291 case SO_REUSEPORT: 1292 sk->sk_reuseport = valbool; 1293 break; 1294 case SO_DONTROUTE: 1295 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1296 sk_dst_reset(sk); 1297 break; 1298 case SO_BROADCAST: 1299 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1300 break; 1301 case SO_SNDBUF: 1302 /* Don't error on this BSD doesn't and if you think 1303 * about it this is right. Otherwise apps have to 1304 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1305 * are treated in BSD as hints 1306 */ 1307 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1308 set_sndbuf: 1309 /* Ensure val * 2 fits into an int, to prevent max_t() 1310 * from treating it as a negative value. 1311 */ 1312 val = min_t(int, val, INT_MAX / 2); 1313 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1314 WRITE_ONCE(sk->sk_sndbuf, 1315 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1316 /* Wake up sending tasks if we upped the value. */ 1317 sk->sk_write_space(sk); 1318 break; 1319 1320 case SO_SNDBUFFORCE: 1321 if (!sockopt_capable(CAP_NET_ADMIN)) { 1322 ret = -EPERM; 1323 break; 1324 } 1325 1326 /* No negative values (to prevent underflow, as val will be 1327 * multiplied by 2). 1328 */ 1329 if (val < 0) 1330 val = 0; 1331 goto set_sndbuf; 1332 1333 case SO_RCVBUF: 1334 /* Don't error on this BSD doesn't and if you think 1335 * about it this is right. Otherwise apps have to 1336 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1337 * are treated in BSD as hints 1338 */ 1339 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1340 break; 1341 1342 case SO_RCVBUFFORCE: 1343 if (!sockopt_capable(CAP_NET_ADMIN)) { 1344 ret = -EPERM; 1345 break; 1346 } 1347 1348 /* No negative values (to prevent underflow, as val will be 1349 * multiplied by 2). 1350 */ 1351 __sock_set_rcvbuf(sk, max(val, 0)); 1352 break; 1353 1354 case SO_KEEPALIVE: 1355 if (sk->sk_prot->keepalive) 1356 sk->sk_prot->keepalive(sk, valbool); 1357 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1358 break; 1359 1360 case SO_OOBINLINE: 1361 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1362 break; 1363 1364 case SO_NO_CHECK: 1365 sk->sk_no_check_tx = valbool; 1366 break; 1367 1368 case SO_LINGER: 1369 if (optlen < sizeof(ling)) { 1370 ret = -EINVAL; /* 1003.1g */ 1371 break; 1372 } 1373 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1374 ret = -EFAULT; 1375 break; 1376 } 1377 if (!ling.l_onoff) { 1378 sock_reset_flag(sk, SOCK_LINGER); 1379 } else { 1380 unsigned long t_sec = ling.l_linger; 1381 1382 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1383 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1384 else 1385 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1386 sock_set_flag(sk, SOCK_LINGER); 1387 } 1388 break; 1389 1390 case SO_BSDCOMPAT: 1391 break; 1392 1393 case SO_TIMESTAMP_OLD: 1394 case SO_TIMESTAMP_NEW: 1395 case SO_TIMESTAMPNS_OLD: 1396 case SO_TIMESTAMPNS_NEW: 1397 sock_set_timestamp(sk, optname, valbool); 1398 break; 1399 1400 case SO_TIMESTAMPING_NEW: 1401 case SO_TIMESTAMPING_OLD: 1402 if (optlen == sizeof(timestamping)) { 1403 if (copy_from_sockptr(×tamping, optval, 1404 sizeof(timestamping))) { 1405 ret = -EFAULT; 1406 break; 1407 } 1408 } else { 1409 memset(×tamping, 0, sizeof(timestamping)); 1410 timestamping.flags = val; 1411 } 1412 ret = sock_set_timestamping(sk, optname, timestamping); 1413 break; 1414 1415 case SO_RCVLOWAT: 1416 { 1417 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1418 1419 if (val < 0) 1420 val = INT_MAX; 1421 if (sock) 1422 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1423 if (set_rcvlowat) 1424 ret = set_rcvlowat(sk, val); 1425 else 1426 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1427 break; 1428 } 1429 case SO_RCVTIMEO_OLD: 1430 case SO_RCVTIMEO_NEW: 1431 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1432 optlen, optname == SO_RCVTIMEO_OLD); 1433 break; 1434 1435 case SO_SNDTIMEO_OLD: 1436 case SO_SNDTIMEO_NEW: 1437 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1438 optlen, optname == SO_SNDTIMEO_OLD); 1439 break; 1440 1441 case SO_ATTACH_FILTER: { 1442 struct sock_fprog fprog; 1443 1444 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1445 if (!ret) 1446 ret = sk_attach_filter(&fprog, sk); 1447 break; 1448 } 1449 case SO_ATTACH_BPF: 1450 ret = -EINVAL; 1451 if (optlen == sizeof(u32)) { 1452 u32 ufd; 1453 1454 ret = -EFAULT; 1455 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1456 break; 1457 1458 ret = sk_attach_bpf(ufd, sk); 1459 } 1460 break; 1461 1462 case SO_ATTACH_REUSEPORT_CBPF: { 1463 struct sock_fprog fprog; 1464 1465 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1466 if (!ret) 1467 ret = sk_reuseport_attach_filter(&fprog, sk); 1468 break; 1469 } 1470 case SO_ATTACH_REUSEPORT_EBPF: 1471 ret = -EINVAL; 1472 if (optlen == sizeof(u32)) { 1473 u32 ufd; 1474 1475 ret = -EFAULT; 1476 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1477 break; 1478 1479 ret = sk_reuseport_attach_bpf(ufd, sk); 1480 } 1481 break; 1482 1483 case SO_DETACH_REUSEPORT_BPF: 1484 ret = reuseport_detach_prog(sk); 1485 break; 1486 1487 case SO_DETACH_FILTER: 1488 ret = sk_detach_filter(sk); 1489 break; 1490 1491 case SO_LOCK_FILTER: 1492 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1493 ret = -EPERM; 1494 else 1495 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1496 break; 1497 1498 case SO_MARK: 1499 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1500 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1501 ret = -EPERM; 1502 break; 1503 } 1504 1505 __sock_set_mark(sk, val); 1506 break; 1507 case SO_RCVMARK: 1508 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1509 break; 1510 1511 case SO_RXQ_OVFL: 1512 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1513 break; 1514 1515 case SO_WIFI_STATUS: 1516 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1517 break; 1518 1519 case SO_NOFCS: 1520 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1521 break; 1522 1523 case SO_SELECT_ERR_QUEUE: 1524 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1525 break; 1526 1527 1528 case SO_INCOMING_CPU: 1529 reuseport_update_incoming_cpu(sk, val); 1530 break; 1531 1532 case SO_CNX_ADVICE: 1533 if (val == 1) 1534 dst_negative_advice(sk); 1535 break; 1536 1537 case SO_ZEROCOPY: 1538 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1539 if (!(sk_is_tcp(sk) || 1540 (sk->sk_type == SOCK_DGRAM && 1541 sk->sk_protocol == IPPROTO_UDP))) 1542 ret = -EOPNOTSUPP; 1543 } else if (sk->sk_family != PF_RDS) { 1544 ret = -EOPNOTSUPP; 1545 } 1546 if (!ret) { 1547 if (val < 0 || val > 1) 1548 ret = -EINVAL; 1549 else 1550 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1551 } 1552 break; 1553 1554 case SO_TXTIME: 1555 if (optlen != sizeof(struct sock_txtime)) { 1556 ret = -EINVAL; 1557 break; 1558 } else if (copy_from_sockptr(&sk_txtime, optval, 1559 sizeof(struct sock_txtime))) { 1560 ret = -EFAULT; 1561 break; 1562 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1563 ret = -EINVAL; 1564 break; 1565 } 1566 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1567 * scheduler has enough safe guards. 1568 */ 1569 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1570 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1571 ret = -EPERM; 1572 break; 1573 } 1574 1575 ret = sockopt_validate_clockid(sk_txtime.clockid); 1576 if (ret) 1577 break; 1578 1579 sock_valbool_flag(sk, SOCK_TXTIME, true); 1580 sk->sk_clockid = sk_txtime.clockid; 1581 sk->sk_txtime_deadline_mode = 1582 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1583 sk->sk_txtime_report_errors = 1584 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1585 break; 1586 1587 case SO_BINDTOIFINDEX: 1588 ret = sock_bindtoindex_locked(sk, val); 1589 break; 1590 1591 case SO_BUF_LOCK: 1592 if (val & ~SOCK_BUF_LOCK_MASK) { 1593 ret = -EINVAL; 1594 break; 1595 } 1596 sk->sk_userlocks = val | (sk->sk_userlocks & 1597 ~SOCK_BUF_LOCK_MASK); 1598 break; 1599 1600 case SO_RESERVE_MEM: 1601 { 1602 int delta; 1603 1604 if (val < 0) { 1605 ret = -EINVAL; 1606 break; 1607 } 1608 1609 delta = val - sk->sk_reserved_mem; 1610 if (delta < 0) 1611 sock_release_reserved_memory(sk, -delta); 1612 else 1613 ret = sock_reserve_memory(sk, delta); 1614 break; 1615 } 1616 1617 default: 1618 ret = -ENOPROTOOPT; 1619 break; 1620 } 1621 sockopt_release_sock(sk); 1622 return ret; 1623 } 1624 1625 int sock_setsockopt(struct socket *sock, int level, int optname, 1626 sockptr_t optval, unsigned int optlen) 1627 { 1628 return sk_setsockopt(sock->sk, level, optname, 1629 optval, optlen); 1630 } 1631 EXPORT_SYMBOL(sock_setsockopt); 1632 1633 static const struct cred *sk_get_peer_cred(struct sock *sk) 1634 { 1635 const struct cred *cred; 1636 1637 spin_lock(&sk->sk_peer_lock); 1638 cred = get_cred(sk->sk_peer_cred); 1639 spin_unlock(&sk->sk_peer_lock); 1640 1641 return cred; 1642 } 1643 1644 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1645 struct ucred *ucred) 1646 { 1647 ucred->pid = pid_vnr(pid); 1648 ucred->uid = ucred->gid = -1; 1649 if (cred) { 1650 struct user_namespace *current_ns = current_user_ns(); 1651 1652 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1653 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1654 } 1655 } 1656 1657 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1658 { 1659 struct user_namespace *user_ns = current_user_ns(); 1660 int i; 1661 1662 for (i = 0; i < src->ngroups; i++) { 1663 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1664 1665 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1666 return -EFAULT; 1667 } 1668 1669 return 0; 1670 } 1671 1672 int sk_getsockopt(struct sock *sk, int level, int optname, 1673 sockptr_t optval, sockptr_t optlen) 1674 { 1675 struct socket *sock = sk->sk_socket; 1676 1677 union { 1678 int val; 1679 u64 val64; 1680 unsigned long ulval; 1681 struct linger ling; 1682 struct old_timeval32 tm32; 1683 struct __kernel_old_timeval tm; 1684 struct __kernel_sock_timeval stm; 1685 struct sock_txtime txtime; 1686 struct so_timestamping timestamping; 1687 } v; 1688 1689 int lv = sizeof(int); 1690 int len; 1691 1692 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1693 return -EFAULT; 1694 if (len < 0) 1695 return -EINVAL; 1696 1697 memset(&v, 0, sizeof(v)); 1698 1699 switch (optname) { 1700 case SO_DEBUG: 1701 v.val = sock_flag(sk, SOCK_DBG); 1702 break; 1703 1704 case SO_DONTROUTE: 1705 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1706 break; 1707 1708 case SO_BROADCAST: 1709 v.val = sock_flag(sk, SOCK_BROADCAST); 1710 break; 1711 1712 case SO_SNDBUF: 1713 v.val = READ_ONCE(sk->sk_sndbuf); 1714 break; 1715 1716 case SO_RCVBUF: 1717 v.val = READ_ONCE(sk->sk_rcvbuf); 1718 break; 1719 1720 case SO_REUSEADDR: 1721 v.val = sk->sk_reuse; 1722 break; 1723 1724 case SO_REUSEPORT: 1725 v.val = sk->sk_reuseport; 1726 break; 1727 1728 case SO_KEEPALIVE: 1729 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1730 break; 1731 1732 case SO_TYPE: 1733 v.val = sk->sk_type; 1734 break; 1735 1736 case SO_PROTOCOL: 1737 v.val = sk->sk_protocol; 1738 break; 1739 1740 case SO_DOMAIN: 1741 v.val = sk->sk_family; 1742 break; 1743 1744 case SO_ERROR: 1745 v.val = -sock_error(sk); 1746 if (v.val == 0) 1747 v.val = xchg(&sk->sk_err_soft, 0); 1748 break; 1749 1750 case SO_OOBINLINE: 1751 v.val = sock_flag(sk, SOCK_URGINLINE); 1752 break; 1753 1754 case SO_NO_CHECK: 1755 v.val = sk->sk_no_check_tx; 1756 break; 1757 1758 case SO_PRIORITY: 1759 v.val = READ_ONCE(sk->sk_priority); 1760 break; 1761 1762 case SO_LINGER: 1763 lv = sizeof(v.ling); 1764 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1765 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1766 break; 1767 1768 case SO_BSDCOMPAT: 1769 break; 1770 1771 case SO_TIMESTAMP_OLD: 1772 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1773 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1774 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1775 break; 1776 1777 case SO_TIMESTAMPNS_OLD: 1778 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1779 break; 1780 1781 case SO_TIMESTAMP_NEW: 1782 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1783 break; 1784 1785 case SO_TIMESTAMPNS_NEW: 1786 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1787 break; 1788 1789 case SO_TIMESTAMPING_OLD: 1790 case SO_TIMESTAMPING_NEW: 1791 lv = sizeof(v.timestamping); 1792 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1793 * returning the flags when they were set through the same option. 1794 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1795 */ 1796 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1797 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1798 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1799 } 1800 break; 1801 1802 case SO_RCVTIMEO_OLD: 1803 case SO_RCVTIMEO_NEW: 1804 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1805 SO_RCVTIMEO_OLD == optname); 1806 break; 1807 1808 case SO_SNDTIMEO_OLD: 1809 case SO_SNDTIMEO_NEW: 1810 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1811 SO_SNDTIMEO_OLD == optname); 1812 break; 1813 1814 case SO_RCVLOWAT: 1815 v.val = READ_ONCE(sk->sk_rcvlowat); 1816 break; 1817 1818 case SO_SNDLOWAT: 1819 v.val = 1; 1820 break; 1821 1822 case SO_PASSCRED: 1823 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1824 break; 1825 1826 case SO_PASSPIDFD: 1827 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1828 break; 1829 1830 case SO_PEERCRED: 1831 { 1832 struct ucred peercred; 1833 if (len > sizeof(peercred)) 1834 len = sizeof(peercred); 1835 1836 spin_lock(&sk->sk_peer_lock); 1837 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1838 spin_unlock(&sk->sk_peer_lock); 1839 1840 if (copy_to_sockptr(optval, &peercred, len)) 1841 return -EFAULT; 1842 goto lenout; 1843 } 1844 1845 case SO_PEERPIDFD: 1846 { 1847 struct pid *peer_pid; 1848 struct file *pidfd_file = NULL; 1849 int pidfd; 1850 1851 if (len > sizeof(pidfd)) 1852 len = sizeof(pidfd); 1853 1854 spin_lock(&sk->sk_peer_lock); 1855 peer_pid = get_pid(sk->sk_peer_pid); 1856 spin_unlock(&sk->sk_peer_lock); 1857 1858 if (!peer_pid) 1859 return -ENODATA; 1860 1861 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1862 put_pid(peer_pid); 1863 if (pidfd < 0) 1864 return pidfd; 1865 1866 if (copy_to_sockptr(optval, &pidfd, len) || 1867 copy_to_sockptr(optlen, &len, sizeof(int))) { 1868 put_unused_fd(pidfd); 1869 fput(pidfd_file); 1870 1871 return -EFAULT; 1872 } 1873 1874 fd_install(pidfd, pidfd_file); 1875 return 0; 1876 } 1877 1878 case SO_PEERGROUPS: 1879 { 1880 const struct cred *cred; 1881 int ret, n; 1882 1883 cred = sk_get_peer_cred(sk); 1884 if (!cred) 1885 return -ENODATA; 1886 1887 n = cred->group_info->ngroups; 1888 if (len < n * sizeof(gid_t)) { 1889 len = n * sizeof(gid_t); 1890 put_cred(cred); 1891 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1892 } 1893 len = n * sizeof(gid_t); 1894 1895 ret = groups_to_user(optval, cred->group_info); 1896 put_cred(cred); 1897 if (ret) 1898 return ret; 1899 goto lenout; 1900 } 1901 1902 case SO_PEERNAME: 1903 { 1904 struct sockaddr_storage address; 1905 1906 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 1907 if (lv < 0) 1908 return -ENOTCONN; 1909 if (lv < len) 1910 return -EINVAL; 1911 if (copy_to_sockptr(optval, &address, len)) 1912 return -EFAULT; 1913 goto lenout; 1914 } 1915 1916 /* Dubious BSD thing... Probably nobody even uses it, but 1917 * the UNIX standard wants it for whatever reason... -DaveM 1918 */ 1919 case SO_ACCEPTCONN: 1920 v.val = sk->sk_state == TCP_LISTEN; 1921 break; 1922 1923 case SO_PASSSEC: 1924 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1925 break; 1926 1927 case SO_PEERSEC: 1928 return security_socket_getpeersec_stream(sock, 1929 optval, optlen, len); 1930 1931 case SO_MARK: 1932 v.val = READ_ONCE(sk->sk_mark); 1933 break; 1934 1935 case SO_RCVMARK: 1936 v.val = sock_flag(sk, SOCK_RCVMARK); 1937 break; 1938 1939 case SO_RXQ_OVFL: 1940 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1941 break; 1942 1943 case SO_WIFI_STATUS: 1944 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1945 break; 1946 1947 case SO_PEEK_OFF: 1948 if (!READ_ONCE(sock->ops)->set_peek_off) 1949 return -EOPNOTSUPP; 1950 1951 v.val = READ_ONCE(sk->sk_peek_off); 1952 break; 1953 case SO_NOFCS: 1954 v.val = sock_flag(sk, SOCK_NOFCS); 1955 break; 1956 1957 case SO_BINDTODEVICE: 1958 return sock_getbindtodevice(sk, optval, optlen, len); 1959 1960 case SO_GET_FILTER: 1961 len = sk_get_filter(sk, optval, len); 1962 if (len < 0) 1963 return len; 1964 1965 goto lenout; 1966 1967 case SO_LOCK_FILTER: 1968 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1969 break; 1970 1971 case SO_BPF_EXTENSIONS: 1972 v.val = bpf_tell_extensions(); 1973 break; 1974 1975 case SO_SELECT_ERR_QUEUE: 1976 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1977 break; 1978 1979 #ifdef CONFIG_NET_RX_BUSY_POLL 1980 case SO_BUSY_POLL: 1981 v.val = READ_ONCE(sk->sk_ll_usec); 1982 break; 1983 case SO_PREFER_BUSY_POLL: 1984 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1985 break; 1986 #endif 1987 1988 case SO_MAX_PACING_RATE: 1989 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 1990 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1991 lv = sizeof(v.ulval); 1992 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 1993 } else { 1994 /* 32bit version */ 1995 v.val = min_t(unsigned long, ~0U, 1996 READ_ONCE(sk->sk_max_pacing_rate)); 1997 } 1998 break; 1999 2000 case SO_INCOMING_CPU: 2001 v.val = READ_ONCE(sk->sk_incoming_cpu); 2002 break; 2003 2004 case SO_MEMINFO: 2005 { 2006 u32 meminfo[SK_MEMINFO_VARS]; 2007 2008 sk_get_meminfo(sk, meminfo); 2009 2010 len = min_t(unsigned int, len, sizeof(meminfo)); 2011 if (copy_to_sockptr(optval, &meminfo, len)) 2012 return -EFAULT; 2013 2014 goto lenout; 2015 } 2016 2017 #ifdef CONFIG_NET_RX_BUSY_POLL 2018 case SO_INCOMING_NAPI_ID: 2019 v.val = READ_ONCE(sk->sk_napi_id); 2020 2021 /* aggregate non-NAPI IDs down to 0 */ 2022 if (v.val < MIN_NAPI_ID) 2023 v.val = 0; 2024 2025 break; 2026 #endif 2027 2028 case SO_COOKIE: 2029 lv = sizeof(u64); 2030 if (len < lv) 2031 return -EINVAL; 2032 v.val64 = sock_gen_cookie(sk); 2033 break; 2034 2035 case SO_ZEROCOPY: 2036 v.val = sock_flag(sk, SOCK_ZEROCOPY); 2037 break; 2038 2039 case SO_TXTIME: 2040 lv = sizeof(v.txtime); 2041 v.txtime.clockid = sk->sk_clockid; 2042 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 2043 SOF_TXTIME_DEADLINE_MODE : 0; 2044 v.txtime.flags |= sk->sk_txtime_report_errors ? 2045 SOF_TXTIME_REPORT_ERRORS : 0; 2046 break; 2047 2048 case SO_BINDTOIFINDEX: 2049 v.val = READ_ONCE(sk->sk_bound_dev_if); 2050 break; 2051 2052 case SO_NETNS_COOKIE: 2053 lv = sizeof(u64); 2054 if (len != lv) 2055 return -EINVAL; 2056 v.val64 = sock_net(sk)->net_cookie; 2057 break; 2058 2059 case SO_BUF_LOCK: 2060 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 2061 break; 2062 2063 case SO_RESERVE_MEM: 2064 v.val = READ_ONCE(sk->sk_reserved_mem); 2065 break; 2066 2067 case SO_TXREHASH: 2068 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2069 v.val = READ_ONCE(sk->sk_txrehash); 2070 break; 2071 2072 default: 2073 /* We implement the SO_SNDLOWAT etc to not be settable 2074 * (1003.1g 7). 2075 */ 2076 return -ENOPROTOOPT; 2077 } 2078 2079 if (len > lv) 2080 len = lv; 2081 if (copy_to_sockptr(optval, &v, len)) 2082 return -EFAULT; 2083 lenout: 2084 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2085 return -EFAULT; 2086 return 0; 2087 } 2088 2089 /* 2090 * Initialize an sk_lock. 2091 * 2092 * (We also register the sk_lock with the lock validator.) 2093 */ 2094 static inline void sock_lock_init(struct sock *sk) 2095 { 2096 if (sk->sk_kern_sock) 2097 sock_lock_init_class_and_name( 2098 sk, 2099 af_family_kern_slock_key_strings[sk->sk_family], 2100 af_family_kern_slock_keys + sk->sk_family, 2101 af_family_kern_key_strings[sk->sk_family], 2102 af_family_kern_keys + sk->sk_family); 2103 else 2104 sock_lock_init_class_and_name( 2105 sk, 2106 af_family_slock_key_strings[sk->sk_family], 2107 af_family_slock_keys + sk->sk_family, 2108 af_family_key_strings[sk->sk_family], 2109 af_family_keys + sk->sk_family); 2110 } 2111 2112 /* 2113 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2114 * even temporarily, because of RCU lookups. sk_node should also be left as is. 2115 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2116 */ 2117 static void sock_copy(struct sock *nsk, const struct sock *osk) 2118 { 2119 const struct proto *prot = READ_ONCE(osk->sk_prot); 2120 #ifdef CONFIG_SECURITY_NETWORK 2121 void *sptr = nsk->sk_security; 2122 #endif 2123 2124 /* If we move sk_tx_queue_mapping out of the private section, 2125 * we must check if sk_tx_queue_clear() is called after 2126 * sock_copy() in sk_clone_lock(). 2127 */ 2128 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2129 offsetof(struct sock, sk_dontcopy_begin) || 2130 offsetof(struct sock, sk_tx_queue_mapping) >= 2131 offsetof(struct sock, sk_dontcopy_end)); 2132 2133 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2134 2135 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2136 prot->obj_size - offsetof(struct sock, sk_dontcopy_end), 2137 /* alloc is larger than struct, see sk_prot_alloc() */); 2138 2139 #ifdef CONFIG_SECURITY_NETWORK 2140 nsk->sk_security = sptr; 2141 security_sk_clone(osk, nsk); 2142 #endif 2143 } 2144 2145 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2146 int family) 2147 { 2148 struct sock *sk; 2149 struct kmem_cache *slab; 2150 2151 slab = prot->slab; 2152 if (slab != NULL) { 2153 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2154 if (!sk) 2155 return sk; 2156 if (want_init_on_alloc(priority)) 2157 sk_prot_clear_nulls(sk, prot->obj_size); 2158 } else 2159 sk = kmalloc(prot->obj_size, priority); 2160 2161 if (sk != NULL) { 2162 if (security_sk_alloc(sk, family, priority)) 2163 goto out_free; 2164 2165 if (!try_module_get(prot->owner)) 2166 goto out_free_sec; 2167 } 2168 2169 return sk; 2170 2171 out_free_sec: 2172 security_sk_free(sk); 2173 out_free: 2174 if (slab != NULL) 2175 kmem_cache_free(slab, sk); 2176 else 2177 kfree(sk); 2178 return NULL; 2179 } 2180 2181 static void sk_prot_free(struct proto *prot, struct sock *sk) 2182 { 2183 struct kmem_cache *slab; 2184 struct module *owner; 2185 2186 owner = prot->owner; 2187 slab = prot->slab; 2188 2189 cgroup_sk_free(&sk->sk_cgrp_data); 2190 mem_cgroup_sk_free(sk); 2191 security_sk_free(sk); 2192 if (slab != NULL) 2193 kmem_cache_free(slab, sk); 2194 else 2195 kfree(sk); 2196 module_put(owner); 2197 } 2198 2199 /** 2200 * sk_alloc - All socket objects are allocated here 2201 * @net: the applicable net namespace 2202 * @family: protocol family 2203 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2204 * @prot: struct proto associated with this new sock instance 2205 * @kern: is this to be a kernel socket? 2206 */ 2207 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2208 struct proto *prot, int kern) 2209 { 2210 struct sock *sk; 2211 2212 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2213 if (sk) { 2214 sk->sk_family = family; 2215 /* 2216 * See comment in struct sock definition to understand 2217 * why we need sk_prot_creator -acme 2218 */ 2219 sk->sk_prot = sk->sk_prot_creator = prot; 2220 sk->sk_kern_sock = kern; 2221 sock_lock_init(sk); 2222 sk->sk_net_refcnt = kern ? 0 : 1; 2223 if (likely(sk->sk_net_refcnt)) { 2224 get_net_track(net, &sk->ns_tracker, priority); 2225 sock_inuse_add(net, 1); 2226 } else { 2227 __netns_tracker_alloc(net, &sk->ns_tracker, 2228 false, priority); 2229 } 2230 2231 sock_net_set(sk, net); 2232 refcount_set(&sk->sk_wmem_alloc, 1); 2233 2234 mem_cgroup_sk_alloc(sk); 2235 cgroup_sk_alloc(&sk->sk_cgrp_data); 2236 sock_update_classid(&sk->sk_cgrp_data); 2237 sock_update_netprioidx(&sk->sk_cgrp_data); 2238 sk_tx_queue_clear(sk); 2239 } 2240 2241 return sk; 2242 } 2243 EXPORT_SYMBOL(sk_alloc); 2244 2245 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2246 * grace period. This is the case for UDP sockets and TCP listeners. 2247 */ 2248 static void __sk_destruct(struct rcu_head *head) 2249 { 2250 struct sock *sk = container_of(head, struct sock, sk_rcu); 2251 struct sk_filter *filter; 2252 2253 if (sk->sk_destruct) 2254 sk->sk_destruct(sk); 2255 2256 filter = rcu_dereference_check(sk->sk_filter, 2257 refcount_read(&sk->sk_wmem_alloc) == 0); 2258 if (filter) { 2259 sk_filter_uncharge(sk, filter); 2260 RCU_INIT_POINTER(sk->sk_filter, NULL); 2261 } 2262 2263 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2264 2265 #ifdef CONFIG_BPF_SYSCALL 2266 bpf_sk_storage_free(sk); 2267 #endif 2268 2269 if (atomic_read(&sk->sk_omem_alloc)) 2270 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2271 __func__, atomic_read(&sk->sk_omem_alloc)); 2272 2273 if (sk->sk_frag.page) { 2274 put_page(sk->sk_frag.page); 2275 sk->sk_frag.page = NULL; 2276 } 2277 2278 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2279 put_cred(sk->sk_peer_cred); 2280 put_pid(sk->sk_peer_pid); 2281 2282 if (likely(sk->sk_net_refcnt)) 2283 put_net_track(sock_net(sk), &sk->ns_tracker); 2284 else 2285 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2286 2287 sk_prot_free(sk->sk_prot_creator, sk); 2288 } 2289 2290 void sk_destruct(struct sock *sk) 2291 { 2292 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2293 2294 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2295 reuseport_detach_sock(sk); 2296 use_call_rcu = true; 2297 } 2298 2299 if (use_call_rcu) 2300 call_rcu(&sk->sk_rcu, __sk_destruct); 2301 else 2302 __sk_destruct(&sk->sk_rcu); 2303 } 2304 2305 static void __sk_free(struct sock *sk) 2306 { 2307 if (likely(sk->sk_net_refcnt)) 2308 sock_inuse_add(sock_net(sk), -1); 2309 2310 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2311 sock_diag_broadcast_destroy(sk); 2312 else 2313 sk_destruct(sk); 2314 } 2315 2316 void sk_free(struct sock *sk) 2317 { 2318 /* 2319 * We subtract one from sk_wmem_alloc and can know if 2320 * some packets are still in some tx queue. 2321 * If not null, sock_wfree() will call __sk_free(sk) later 2322 */ 2323 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2324 __sk_free(sk); 2325 } 2326 EXPORT_SYMBOL(sk_free); 2327 2328 static void sk_init_common(struct sock *sk) 2329 { 2330 skb_queue_head_init(&sk->sk_receive_queue); 2331 skb_queue_head_init(&sk->sk_write_queue); 2332 skb_queue_head_init(&sk->sk_error_queue); 2333 2334 rwlock_init(&sk->sk_callback_lock); 2335 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2336 af_rlock_keys + sk->sk_family, 2337 af_family_rlock_key_strings[sk->sk_family]); 2338 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2339 af_wlock_keys + sk->sk_family, 2340 af_family_wlock_key_strings[sk->sk_family]); 2341 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2342 af_elock_keys + sk->sk_family, 2343 af_family_elock_key_strings[sk->sk_family]); 2344 if (sk->sk_kern_sock) 2345 lockdep_set_class_and_name(&sk->sk_callback_lock, 2346 af_kern_callback_keys + sk->sk_family, 2347 af_family_kern_clock_key_strings[sk->sk_family]); 2348 else 2349 lockdep_set_class_and_name(&sk->sk_callback_lock, 2350 af_callback_keys + sk->sk_family, 2351 af_family_clock_key_strings[sk->sk_family]); 2352 } 2353 2354 /** 2355 * sk_clone_lock - clone a socket, and lock its clone 2356 * @sk: the socket to clone 2357 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2358 * 2359 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2360 */ 2361 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2362 { 2363 struct proto *prot = READ_ONCE(sk->sk_prot); 2364 struct sk_filter *filter; 2365 bool is_charged = true; 2366 struct sock *newsk; 2367 2368 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2369 if (!newsk) 2370 goto out; 2371 2372 sock_copy(newsk, sk); 2373 2374 newsk->sk_prot_creator = prot; 2375 2376 /* SANITY */ 2377 if (likely(newsk->sk_net_refcnt)) { 2378 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2379 sock_inuse_add(sock_net(newsk), 1); 2380 } else { 2381 /* Kernel sockets are not elevating the struct net refcount. 2382 * Instead, use a tracker to more easily detect if a layer 2383 * is not properly dismantling its kernel sockets at netns 2384 * destroy time. 2385 */ 2386 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2387 false, priority); 2388 } 2389 sk_node_init(&newsk->sk_node); 2390 sock_lock_init(newsk); 2391 bh_lock_sock(newsk); 2392 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2393 newsk->sk_backlog.len = 0; 2394 2395 atomic_set(&newsk->sk_rmem_alloc, 0); 2396 2397 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2398 refcount_set(&newsk->sk_wmem_alloc, 1); 2399 2400 atomic_set(&newsk->sk_omem_alloc, 0); 2401 sk_init_common(newsk); 2402 2403 newsk->sk_dst_cache = NULL; 2404 newsk->sk_dst_pending_confirm = 0; 2405 newsk->sk_wmem_queued = 0; 2406 newsk->sk_forward_alloc = 0; 2407 newsk->sk_reserved_mem = 0; 2408 atomic_set(&newsk->sk_drops, 0); 2409 newsk->sk_send_head = NULL; 2410 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2411 atomic_set(&newsk->sk_zckey, 0); 2412 2413 sock_reset_flag(newsk, SOCK_DONE); 2414 2415 /* sk->sk_memcg will be populated at accept() time */ 2416 newsk->sk_memcg = NULL; 2417 2418 cgroup_sk_clone(&newsk->sk_cgrp_data); 2419 2420 rcu_read_lock(); 2421 filter = rcu_dereference(sk->sk_filter); 2422 if (filter != NULL) 2423 /* though it's an empty new sock, the charging may fail 2424 * if sysctl_optmem_max was changed between creation of 2425 * original socket and cloning 2426 */ 2427 is_charged = sk_filter_charge(newsk, filter); 2428 RCU_INIT_POINTER(newsk->sk_filter, filter); 2429 rcu_read_unlock(); 2430 2431 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2432 /* We need to make sure that we don't uncharge the new 2433 * socket if we couldn't charge it in the first place 2434 * as otherwise we uncharge the parent's filter. 2435 */ 2436 if (!is_charged) 2437 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2438 sk_free_unlock_clone(newsk); 2439 newsk = NULL; 2440 goto out; 2441 } 2442 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2443 2444 if (bpf_sk_storage_clone(sk, newsk)) { 2445 sk_free_unlock_clone(newsk); 2446 newsk = NULL; 2447 goto out; 2448 } 2449 2450 /* Clear sk_user_data if parent had the pointer tagged 2451 * as not suitable for copying when cloning. 2452 */ 2453 if (sk_user_data_is_nocopy(newsk)) 2454 newsk->sk_user_data = NULL; 2455 2456 newsk->sk_err = 0; 2457 newsk->sk_err_soft = 0; 2458 newsk->sk_priority = 0; 2459 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2460 2461 /* Before updating sk_refcnt, we must commit prior changes to memory 2462 * (Documentation/RCU/rculist_nulls.rst for details) 2463 */ 2464 smp_wmb(); 2465 refcount_set(&newsk->sk_refcnt, 2); 2466 2467 sk_set_socket(newsk, NULL); 2468 sk_tx_queue_clear(newsk); 2469 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2470 2471 if (newsk->sk_prot->sockets_allocated) 2472 sk_sockets_allocated_inc(newsk); 2473 2474 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2475 net_enable_timestamp(); 2476 out: 2477 return newsk; 2478 } 2479 EXPORT_SYMBOL_GPL(sk_clone_lock); 2480 2481 void sk_free_unlock_clone(struct sock *sk) 2482 { 2483 /* It is still raw copy of parent, so invalidate 2484 * destructor and make plain sk_free() */ 2485 sk->sk_destruct = NULL; 2486 bh_unlock_sock(sk); 2487 sk_free(sk); 2488 } 2489 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2490 2491 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2492 { 2493 bool is_ipv6 = false; 2494 u32 max_size; 2495 2496 #if IS_ENABLED(CONFIG_IPV6) 2497 is_ipv6 = (sk->sk_family == AF_INET6 && 2498 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2499 #endif 2500 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2501 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2502 READ_ONCE(dst->dev->gso_ipv4_max_size); 2503 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2504 max_size = GSO_LEGACY_MAX_SIZE; 2505 2506 return max_size - (MAX_TCP_HEADER + 1); 2507 } 2508 2509 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2510 { 2511 u32 max_segs = 1; 2512 2513 sk->sk_route_caps = dst->dev->features; 2514 if (sk_is_tcp(sk)) 2515 sk->sk_route_caps |= NETIF_F_GSO; 2516 if (sk->sk_route_caps & NETIF_F_GSO) 2517 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2518 if (unlikely(sk->sk_gso_disabled)) 2519 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2520 if (sk_can_gso(sk)) { 2521 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2522 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2523 } else { 2524 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2525 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2526 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2527 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2528 } 2529 } 2530 sk->sk_gso_max_segs = max_segs; 2531 sk_dst_set(sk, dst); 2532 } 2533 EXPORT_SYMBOL_GPL(sk_setup_caps); 2534 2535 /* 2536 * Simple resource managers for sockets. 2537 */ 2538 2539 2540 /* 2541 * Write buffer destructor automatically called from kfree_skb. 2542 */ 2543 void sock_wfree(struct sk_buff *skb) 2544 { 2545 struct sock *sk = skb->sk; 2546 unsigned int len = skb->truesize; 2547 bool free; 2548 2549 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2550 if (sock_flag(sk, SOCK_RCU_FREE) && 2551 sk->sk_write_space == sock_def_write_space) { 2552 rcu_read_lock(); 2553 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2554 sock_def_write_space_wfree(sk); 2555 rcu_read_unlock(); 2556 if (unlikely(free)) 2557 __sk_free(sk); 2558 return; 2559 } 2560 2561 /* 2562 * Keep a reference on sk_wmem_alloc, this will be released 2563 * after sk_write_space() call 2564 */ 2565 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2566 sk->sk_write_space(sk); 2567 len = 1; 2568 } 2569 /* 2570 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2571 * could not do because of in-flight packets 2572 */ 2573 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2574 __sk_free(sk); 2575 } 2576 EXPORT_SYMBOL(sock_wfree); 2577 2578 /* This variant of sock_wfree() is used by TCP, 2579 * since it sets SOCK_USE_WRITE_QUEUE. 2580 */ 2581 void __sock_wfree(struct sk_buff *skb) 2582 { 2583 struct sock *sk = skb->sk; 2584 2585 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2586 __sk_free(sk); 2587 } 2588 2589 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2590 { 2591 skb_orphan(skb); 2592 #ifdef CONFIG_INET 2593 if (unlikely(!sk_fullsock(sk))) 2594 return skb_set_owner_edemux(skb, sk); 2595 #endif 2596 skb->sk = sk; 2597 skb->destructor = sock_wfree; 2598 skb_set_hash_from_sk(skb, sk); 2599 /* 2600 * We used to take a refcount on sk, but following operation 2601 * is enough to guarantee sk_free() won't free this sock until 2602 * all in-flight packets are completed 2603 */ 2604 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2605 } 2606 EXPORT_SYMBOL(skb_set_owner_w); 2607 2608 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2609 { 2610 /* Drivers depend on in-order delivery for crypto offload, 2611 * partial orphan breaks out-of-order-OK logic. 2612 */ 2613 if (skb_is_decrypted(skb)) 2614 return false; 2615 2616 return (skb->destructor == sock_wfree || 2617 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2618 } 2619 2620 /* This helper is used by netem, as it can hold packets in its 2621 * delay queue. We want to allow the owner socket to send more 2622 * packets, as if they were already TX completed by a typical driver. 2623 * But we also want to keep skb->sk set because some packet schedulers 2624 * rely on it (sch_fq for example). 2625 */ 2626 void skb_orphan_partial(struct sk_buff *skb) 2627 { 2628 if (skb_is_tcp_pure_ack(skb)) 2629 return; 2630 2631 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2632 return; 2633 2634 skb_orphan(skb); 2635 } 2636 EXPORT_SYMBOL(skb_orphan_partial); 2637 2638 /* 2639 * Read buffer destructor automatically called from kfree_skb. 2640 */ 2641 void sock_rfree(struct sk_buff *skb) 2642 { 2643 struct sock *sk = skb->sk; 2644 unsigned int len = skb->truesize; 2645 2646 atomic_sub(len, &sk->sk_rmem_alloc); 2647 sk_mem_uncharge(sk, len); 2648 } 2649 EXPORT_SYMBOL(sock_rfree); 2650 2651 /* 2652 * Buffer destructor for skbs that are not used directly in read or write 2653 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2654 */ 2655 void sock_efree(struct sk_buff *skb) 2656 { 2657 sock_put(skb->sk); 2658 } 2659 EXPORT_SYMBOL(sock_efree); 2660 2661 /* Buffer destructor for prefetch/receive path where reference count may 2662 * not be held, e.g. for listen sockets. 2663 */ 2664 #ifdef CONFIG_INET 2665 void sock_pfree(struct sk_buff *skb) 2666 { 2667 struct sock *sk = skb->sk; 2668 2669 if (!sk_is_refcounted(sk)) 2670 return; 2671 2672 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2673 inet_reqsk(sk)->rsk_listener = NULL; 2674 reqsk_free(inet_reqsk(sk)); 2675 return; 2676 } 2677 2678 sock_gen_put(sk); 2679 } 2680 EXPORT_SYMBOL(sock_pfree); 2681 #endif /* CONFIG_INET */ 2682 2683 kuid_t sock_i_uid(struct sock *sk) 2684 { 2685 kuid_t uid; 2686 2687 read_lock_bh(&sk->sk_callback_lock); 2688 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2689 read_unlock_bh(&sk->sk_callback_lock); 2690 return uid; 2691 } 2692 EXPORT_SYMBOL(sock_i_uid); 2693 2694 unsigned long __sock_i_ino(struct sock *sk) 2695 { 2696 unsigned long ino; 2697 2698 read_lock(&sk->sk_callback_lock); 2699 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2700 read_unlock(&sk->sk_callback_lock); 2701 return ino; 2702 } 2703 EXPORT_SYMBOL(__sock_i_ino); 2704 2705 unsigned long sock_i_ino(struct sock *sk) 2706 { 2707 unsigned long ino; 2708 2709 local_bh_disable(); 2710 ino = __sock_i_ino(sk); 2711 local_bh_enable(); 2712 return ino; 2713 } 2714 EXPORT_SYMBOL(sock_i_ino); 2715 2716 /* 2717 * Allocate a skb from the socket's send buffer. 2718 */ 2719 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2720 gfp_t priority) 2721 { 2722 if (force || 2723 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2724 struct sk_buff *skb = alloc_skb(size, priority); 2725 2726 if (skb) { 2727 skb_set_owner_w(skb, sk); 2728 return skb; 2729 } 2730 } 2731 return NULL; 2732 } 2733 EXPORT_SYMBOL(sock_wmalloc); 2734 2735 static void sock_ofree(struct sk_buff *skb) 2736 { 2737 struct sock *sk = skb->sk; 2738 2739 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2740 } 2741 2742 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2743 gfp_t priority) 2744 { 2745 struct sk_buff *skb; 2746 2747 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2748 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2749 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2750 return NULL; 2751 2752 skb = alloc_skb(size, priority); 2753 if (!skb) 2754 return NULL; 2755 2756 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2757 skb->sk = sk; 2758 skb->destructor = sock_ofree; 2759 return skb; 2760 } 2761 2762 /* 2763 * Allocate a memory block from the socket's option memory buffer. 2764 */ 2765 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2766 { 2767 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2768 2769 if ((unsigned int)size <= optmem_max && 2770 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2771 void *mem; 2772 /* First do the add, to avoid the race if kmalloc 2773 * might sleep. 2774 */ 2775 atomic_add(size, &sk->sk_omem_alloc); 2776 mem = kmalloc(size, priority); 2777 if (mem) 2778 return mem; 2779 atomic_sub(size, &sk->sk_omem_alloc); 2780 } 2781 return NULL; 2782 } 2783 EXPORT_SYMBOL(sock_kmalloc); 2784 2785 /* Free an option memory block. Note, we actually want the inline 2786 * here as this allows gcc to detect the nullify and fold away the 2787 * condition entirely. 2788 */ 2789 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2790 const bool nullify) 2791 { 2792 if (WARN_ON_ONCE(!mem)) 2793 return; 2794 if (nullify) 2795 kfree_sensitive(mem); 2796 else 2797 kfree(mem); 2798 atomic_sub(size, &sk->sk_omem_alloc); 2799 } 2800 2801 void sock_kfree_s(struct sock *sk, void *mem, int size) 2802 { 2803 __sock_kfree_s(sk, mem, size, false); 2804 } 2805 EXPORT_SYMBOL(sock_kfree_s); 2806 2807 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2808 { 2809 __sock_kfree_s(sk, mem, size, true); 2810 } 2811 EXPORT_SYMBOL(sock_kzfree_s); 2812 2813 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2814 I think, these locks should be removed for datagram sockets. 2815 */ 2816 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2817 { 2818 DEFINE_WAIT(wait); 2819 2820 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2821 for (;;) { 2822 if (!timeo) 2823 break; 2824 if (signal_pending(current)) 2825 break; 2826 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2827 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2828 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2829 break; 2830 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2831 break; 2832 if (READ_ONCE(sk->sk_err)) 2833 break; 2834 timeo = schedule_timeout(timeo); 2835 } 2836 finish_wait(sk_sleep(sk), &wait); 2837 return timeo; 2838 } 2839 2840 2841 /* 2842 * Generic send/receive buffer handlers 2843 */ 2844 2845 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2846 unsigned long data_len, int noblock, 2847 int *errcode, int max_page_order) 2848 { 2849 struct sk_buff *skb; 2850 long timeo; 2851 int err; 2852 2853 timeo = sock_sndtimeo(sk, noblock); 2854 for (;;) { 2855 err = sock_error(sk); 2856 if (err != 0) 2857 goto failure; 2858 2859 err = -EPIPE; 2860 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2861 goto failure; 2862 2863 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2864 break; 2865 2866 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2867 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2868 err = -EAGAIN; 2869 if (!timeo) 2870 goto failure; 2871 if (signal_pending(current)) 2872 goto interrupted; 2873 timeo = sock_wait_for_wmem(sk, timeo); 2874 } 2875 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2876 errcode, sk->sk_allocation); 2877 if (skb) 2878 skb_set_owner_w(skb, sk); 2879 return skb; 2880 2881 interrupted: 2882 err = sock_intr_errno(timeo); 2883 failure: 2884 *errcode = err; 2885 return NULL; 2886 } 2887 EXPORT_SYMBOL(sock_alloc_send_pskb); 2888 2889 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2890 struct sockcm_cookie *sockc) 2891 { 2892 u32 tsflags; 2893 2894 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 2895 2896 switch (cmsg->cmsg_type) { 2897 case SO_MARK: 2898 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2899 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2900 return -EPERM; 2901 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2902 return -EINVAL; 2903 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2904 break; 2905 case SO_TIMESTAMPING_OLD: 2906 case SO_TIMESTAMPING_NEW: 2907 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2908 return -EINVAL; 2909 2910 tsflags = *(u32 *)CMSG_DATA(cmsg); 2911 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2912 return -EINVAL; 2913 2914 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2915 sockc->tsflags |= tsflags; 2916 break; 2917 case SCM_TXTIME: 2918 if (!sock_flag(sk, SOCK_TXTIME)) 2919 return -EINVAL; 2920 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2921 return -EINVAL; 2922 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2923 break; 2924 case SCM_TS_OPT_ID: 2925 if (sk_is_tcp(sk)) 2926 return -EINVAL; 2927 tsflags = READ_ONCE(sk->sk_tsflags); 2928 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 2929 return -EINVAL; 2930 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2931 return -EINVAL; 2932 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 2933 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 2934 break; 2935 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2936 case SCM_RIGHTS: 2937 case SCM_CREDENTIALS: 2938 break; 2939 default: 2940 return -EINVAL; 2941 } 2942 return 0; 2943 } 2944 EXPORT_SYMBOL(__sock_cmsg_send); 2945 2946 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2947 struct sockcm_cookie *sockc) 2948 { 2949 struct cmsghdr *cmsg; 2950 int ret; 2951 2952 for_each_cmsghdr(cmsg, msg) { 2953 if (!CMSG_OK(msg, cmsg)) 2954 return -EINVAL; 2955 if (cmsg->cmsg_level != SOL_SOCKET) 2956 continue; 2957 ret = __sock_cmsg_send(sk, cmsg, sockc); 2958 if (ret) 2959 return ret; 2960 } 2961 return 0; 2962 } 2963 EXPORT_SYMBOL(sock_cmsg_send); 2964 2965 static void sk_enter_memory_pressure(struct sock *sk) 2966 { 2967 if (!sk->sk_prot->enter_memory_pressure) 2968 return; 2969 2970 sk->sk_prot->enter_memory_pressure(sk); 2971 } 2972 2973 static void sk_leave_memory_pressure(struct sock *sk) 2974 { 2975 if (sk->sk_prot->leave_memory_pressure) { 2976 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2977 tcp_leave_memory_pressure, sk); 2978 } else { 2979 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2980 2981 if (memory_pressure && READ_ONCE(*memory_pressure)) 2982 WRITE_ONCE(*memory_pressure, 0); 2983 } 2984 } 2985 2986 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2987 2988 /** 2989 * skb_page_frag_refill - check that a page_frag contains enough room 2990 * @sz: minimum size of the fragment we want to get 2991 * @pfrag: pointer to page_frag 2992 * @gfp: priority for memory allocation 2993 * 2994 * Note: While this allocator tries to use high order pages, there is 2995 * no guarantee that allocations succeed. Therefore, @sz MUST be 2996 * less or equal than PAGE_SIZE. 2997 */ 2998 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2999 { 3000 if (pfrag->page) { 3001 if (page_ref_count(pfrag->page) == 1) { 3002 pfrag->offset = 0; 3003 return true; 3004 } 3005 if (pfrag->offset + sz <= pfrag->size) 3006 return true; 3007 put_page(pfrag->page); 3008 } 3009 3010 pfrag->offset = 0; 3011 if (SKB_FRAG_PAGE_ORDER && 3012 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3013 /* Avoid direct reclaim but allow kswapd to wake */ 3014 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3015 __GFP_COMP | __GFP_NOWARN | 3016 __GFP_NORETRY, 3017 SKB_FRAG_PAGE_ORDER); 3018 if (likely(pfrag->page)) { 3019 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3020 return true; 3021 } 3022 } 3023 pfrag->page = alloc_page(gfp); 3024 if (likely(pfrag->page)) { 3025 pfrag->size = PAGE_SIZE; 3026 return true; 3027 } 3028 return false; 3029 } 3030 EXPORT_SYMBOL(skb_page_frag_refill); 3031 3032 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3033 { 3034 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3035 return true; 3036 3037 sk_enter_memory_pressure(sk); 3038 sk_stream_moderate_sndbuf(sk); 3039 return false; 3040 } 3041 EXPORT_SYMBOL(sk_page_frag_refill); 3042 3043 void __lock_sock(struct sock *sk) 3044 __releases(&sk->sk_lock.slock) 3045 __acquires(&sk->sk_lock.slock) 3046 { 3047 DEFINE_WAIT(wait); 3048 3049 for (;;) { 3050 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3051 TASK_UNINTERRUPTIBLE); 3052 spin_unlock_bh(&sk->sk_lock.slock); 3053 schedule(); 3054 spin_lock_bh(&sk->sk_lock.slock); 3055 if (!sock_owned_by_user(sk)) 3056 break; 3057 } 3058 finish_wait(&sk->sk_lock.wq, &wait); 3059 } 3060 3061 void __release_sock(struct sock *sk) 3062 __releases(&sk->sk_lock.slock) 3063 __acquires(&sk->sk_lock.slock) 3064 { 3065 struct sk_buff *skb, *next; 3066 3067 while ((skb = sk->sk_backlog.head) != NULL) { 3068 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3069 3070 spin_unlock_bh(&sk->sk_lock.slock); 3071 3072 do { 3073 next = skb->next; 3074 prefetch(next); 3075 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3076 skb_mark_not_on_list(skb); 3077 sk_backlog_rcv(sk, skb); 3078 3079 cond_resched(); 3080 3081 skb = next; 3082 } while (skb != NULL); 3083 3084 spin_lock_bh(&sk->sk_lock.slock); 3085 } 3086 3087 /* 3088 * Doing the zeroing here guarantee we can not loop forever 3089 * while a wild producer attempts to flood us. 3090 */ 3091 sk->sk_backlog.len = 0; 3092 } 3093 3094 void __sk_flush_backlog(struct sock *sk) 3095 { 3096 spin_lock_bh(&sk->sk_lock.slock); 3097 __release_sock(sk); 3098 3099 if (sk->sk_prot->release_cb) 3100 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3101 tcp_release_cb, sk); 3102 3103 spin_unlock_bh(&sk->sk_lock.slock); 3104 } 3105 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3106 3107 /** 3108 * sk_wait_data - wait for data to arrive at sk_receive_queue 3109 * @sk: sock to wait on 3110 * @timeo: for how long 3111 * @skb: last skb seen on sk_receive_queue 3112 * 3113 * Now socket state including sk->sk_err is changed only under lock, 3114 * hence we may omit checks after joining wait queue. 3115 * We check receive queue before schedule() only as optimization; 3116 * it is very likely that release_sock() added new data. 3117 */ 3118 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3119 { 3120 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3121 int rc; 3122 3123 add_wait_queue(sk_sleep(sk), &wait); 3124 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3125 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3126 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3127 remove_wait_queue(sk_sleep(sk), &wait); 3128 return rc; 3129 } 3130 EXPORT_SYMBOL(sk_wait_data); 3131 3132 /** 3133 * __sk_mem_raise_allocated - increase memory_allocated 3134 * @sk: socket 3135 * @size: memory size to allocate 3136 * @amt: pages to allocate 3137 * @kind: allocation type 3138 * 3139 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3140 * 3141 * Unlike the globally shared limits among the sockets under same protocol, 3142 * consuming the budget of a memcg won't have direct effect on other ones. 3143 * So be optimistic about memcg's tolerance, and leave the callers to decide 3144 * whether or not to raise allocated through sk_under_memory_pressure() or 3145 * its variants. 3146 */ 3147 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3148 { 3149 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL; 3150 struct proto *prot = sk->sk_prot; 3151 bool charged = false; 3152 long allocated; 3153 3154 sk_memory_allocated_add(sk, amt); 3155 allocated = sk_memory_allocated(sk); 3156 3157 if (memcg) { 3158 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge())) 3159 goto suppress_allocation; 3160 charged = true; 3161 } 3162 3163 /* Under limit. */ 3164 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3165 sk_leave_memory_pressure(sk); 3166 return 1; 3167 } 3168 3169 /* Under pressure. */ 3170 if (allocated > sk_prot_mem_limits(sk, 1)) 3171 sk_enter_memory_pressure(sk); 3172 3173 /* Over hard limit. */ 3174 if (allocated > sk_prot_mem_limits(sk, 2)) 3175 goto suppress_allocation; 3176 3177 /* Guarantee minimum buffer size under pressure (either global 3178 * or memcg) to make sure features described in RFC 7323 (TCP 3179 * Extensions for High Performance) work properly. 3180 * 3181 * This rule does NOT stand when exceeds global or memcg's hard 3182 * limit, or else a DoS attack can be taken place by spawning 3183 * lots of sockets whose usage are under minimum buffer size. 3184 */ 3185 if (kind == SK_MEM_RECV) { 3186 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3187 return 1; 3188 3189 } else { /* SK_MEM_SEND */ 3190 int wmem0 = sk_get_wmem0(sk, prot); 3191 3192 if (sk->sk_type == SOCK_STREAM) { 3193 if (sk->sk_wmem_queued < wmem0) 3194 return 1; 3195 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3196 return 1; 3197 } 3198 } 3199 3200 if (sk_has_memory_pressure(sk)) { 3201 u64 alloc; 3202 3203 /* The following 'average' heuristic is within the 3204 * scope of global accounting, so it only makes 3205 * sense for global memory pressure. 3206 */ 3207 if (!sk_under_global_memory_pressure(sk)) 3208 return 1; 3209 3210 /* Try to be fair among all the sockets under global 3211 * pressure by allowing the ones that below average 3212 * usage to raise. 3213 */ 3214 alloc = sk_sockets_allocated_read_positive(sk); 3215 if (sk_prot_mem_limits(sk, 2) > alloc * 3216 sk_mem_pages(sk->sk_wmem_queued + 3217 atomic_read(&sk->sk_rmem_alloc) + 3218 sk->sk_forward_alloc)) 3219 return 1; 3220 } 3221 3222 suppress_allocation: 3223 3224 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3225 sk_stream_moderate_sndbuf(sk); 3226 3227 /* Fail only if socket is _under_ its sndbuf. 3228 * In this case we cannot block, so that we have to fail. 3229 */ 3230 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3231 /* Force charge with __GFP_NOFAIL */ 3232 if (memcg && !charged) { 3233 mem_cgroup_charge_skmem(memcg, amt, 3234 gfp_memcg_charge() | __GFP_NOFAIL); 3235 } 3236 return 1; 3237 } 3238 } 3239 3240 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3241 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3242 3243 sk_memory_allocated_sub(sk, amt); 3244 3245 if (charged) 3246 mem_cgroup_uncharge_skmem(memcg, amt); 3247 3248 return 0; 3249 } 3250 3251 /** 3252 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3253 * @sk: socket 3254 * @size: memory size to allocate 3255 * @kind: allocation type 3256 * 3257 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3258 * rmem allocation. This function assumes that protocols which have 3259 * memory_pressure use sk_wmem_queued as write buffer accounting. 3260 */ 3261 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3262 { 3263 int ret, amt = sk_mem_pages(size); 3264 3265 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3266 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3267 if (!ret) 3268 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3269 return ret; 3270 } 3271 EXPORT_SYMBOL(__sk_mem_schedule); 3272 3273 /** 3274 * __sk_mem_reduce_allocated - reclaim memory_allocated 3275 * @sk: socket 3276 * @amount: number of quanta 3277 * 3278 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3279 */ 3280 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3281 { 3282 sk_memory_allocated_sub(sk, amount); 3283 3284 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3285 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3286 3287 if (sk_under_global_memory_pressure(sk) && 3288 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3289 sk_leave_memory_pressure(sk); 3290 } 3291 3292 /** 3293 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3294 * @sk: socket 3295 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3296 */ 3297 void __sk_mem_reclaim(struct sock *sk, int amount) 3298 { 3299 amount >>= PAGE_SHIFT; 3300 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3301 __sk_mem_reduce_allocated(sk, amount); 3302 } 3303 EXPORT_SYMBOL(__sk_mem_reclaim); 3304 3305 int sk_set_peek_off(struct sock *sk, int val) 3306 { 3307 WRITE_ONCE(sk->sk_peek_off, val); 3308 return 0; 3309 } 3310 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3311 3312 /* 3313 * Set of default routines for initialising struct proto_ops when 3314 * the protocol does not support a particular function. In certain 3315 * cases where it makes no sense for a protocol to have a "do nothing" 3316 * function, some default processing is provided. 3317 */ 3318 3319 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3320 { 3321 return -EOPNOTSUPP; 3322 } 3323 EXPORT_SYMBOL(sock_no_bind); 3324 3325 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3326 int len, int flags) 3327 { 3328 return -EOPNOTSUPP; 3329 } 3330 EXPORT_SYMBOL(sock_no_connect); 3331 3332 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3333 { 3334 return -EOPNOTSUPP; 3335 } 3336 EXPORT_SYMBOL(sock_no_socketpair); 3337 3338 int sock_no_accept(struct socket *sock, struct socket *newsock, 3339 struct proto_accept_arg *arg) 3340 { 3341 return -EOPNOTSUPP; 3342 } 3343 EXPORT_SYMBOL(sock_no_accept); 3344 3345 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3346 int peer) 3347 { 3348 return -EOPNOTSUPP; 3349 } 3350 EXPORT_SYMBOL(sock_no_getname); 3351 3352 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3353 { 3354 return -EOPNOTSUPP; 3355 } 3356 EXPORT_SYMBOL(sock_no_ioctl); 3357 3358 int sock_no_listen(struct socket *sock, int backlog) 3359 { 3360 return -EOPNOTSUPP; 3361 } 3362 EXPORT_SYMBOL(sock_no_listen); 3363 3364 int sock_no_shutdown(struct socket *sock, int how) 3365 { 3366 return -EOPNOTSUPP; 3367 } 3368 EXPORT_SYMBOL(sock_no_shutdown); 3369 3370 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3371 { 3372 return -EOPNOTSUPP; 3373 } 3374 EXPORT_SYMBOL(sock_no_sendmsg); 3375 3376 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3377 { 3378 return -EOPNOTSUPP; 3379 } 3380 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3381 3382 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3383 int flags) 3384 { 3385 return -EOPNOTSUPP; 3386 } 3387 EXPORT_SYMBOL(sock_no_recvmsg); 3388 3389 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3390 { 3391 /* Mirror missing mmap method error code */ 3392 return -ENODEV; 3393 } 3394 EXPORT_SYMBOL(sock_no_mmap); 3395 3396 /* 3397 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3398 * various sock-based usage counts. 3399 */ 3400 void __receive_sock(struct file *file) 3401 { 3402 struct socket *sock; 3403 3404 sock = sock_from_file(file); 3405 if (sock) { 3406 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3407 sock_update_classid(&sock->sk->sk_cgrp_data); 3408 } 3409 } 3410 3411 /* 3412 * Default Socket Callbacks 3413 */ 3414 3415 static void sock_def_wakeup(struct sock *sk) 3416 { 3417 struct socket_wq *wq; 3418 3419 rcu_read_lock(); 3420 wq = rcu_dereference(sk->sk_wq); 3421 if (skwq_has_sleeper(wq)) 3422 wake_up_interruptible_all(&wq->wait); 3423 rcu_read_unlock(); 3424 } 3425 3426 static void sock_def_error_report(struct sock *sk) 3427 { 3428 struct socket_wq *wq; 3429 3430 rcu_read_lock(); 3431 wq = rcu_dereference(sk->sk_wq); 3432 if (skwq_has_sleeper(wq)) 3433 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3434 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3435 rcu_read_unlock(); 3436 } 3437 3438 void sock_def_readable(struct sock *sk) 3439 { 3440 struct socket_wq *wq; 3441 3442 trace_sk_data_ready(sk); 3443 3444 rcu_read_lock(); 3445 wq = rcu_dereference(sk->sk_wq); 3446 if (skwq_has_sleeper(wq)) 3447 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3448 EPOLLRDNORM | EPOLLRDBAND); 3449 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3450 rcu_read_unlock(); 3451 } 3452 3453 static void sock_def_write_space(struct sock *sk) 3454 { 3455 struct socket_wq *wq; 3456 3457 rcu_read_lock(); 3458 3459 /* Do not wake up a writer until he can make "significant" 3460 * progress. --DaveM 3461 */ 3462 if (sock_writeable(sk)) { 3463 wq = rcu_dereference(sk->sk_wq); 3464 if (skwq_has_sleeper(wq)) 3465 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3466 EPOLLWRNORM | EPOLLWRBAND); 3467 3468 /* Should agree with poll, otherwise some programs break */ 3469 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3470 } 3471 3472 rcu_read_unlock(); 3473 } 3474 3475 /* An optimised version of sock_def_write_space(), should only be called 3476 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3477 * ->sk_wmem_alloc. 3478 */ 3479 static void sock_def_write_space_wfree(struct sock *sk) 3480 { 3481 /* Do not wake up a writer until he can make "significant" 3482 * progress. --DaveM 3483 */ 3484 if (sock_writeable(sk)) { 3485 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3486 3487 /* rely on refcount_sub from sock_wfree() */ 3488 smp_mb__after_atomic(); 3489 if (wq && waitqueue_active(&wq->wait)) 3490 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3491 EPOLLWRNORM | EPOLLWRBAND); 3492 3493 /* Should agree with poll, otherwise some programs break */ 3494 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3495 } 3496 } 3497 3498 static void sock_def_destruct(struct sock *sk) 3499 { 3500 } 3501 3502 void sk_send_sigurg(struct sock *sk) 3503 { 3504 if (sk->sk_socket && sk->sk_socket->file) 3505 if (send_sigurg(sk->sk_socket->file)) 3506 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3507 } 3508 EXPORT_SYMBOL(sk_send_sigurg); 3509 3510 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3511 unsigned long expires) 3512 { 3513 if (!mod_timer(timer, expires)) 3514 sock_hold(sk); 3515 } 3516 EXPORT_SYMBOL(sk_reset_timer); 3517 3518 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3519 { 3520 if (del_timer(timer)) 3521 __sock_put(sk); 3522 } 3523 EXPORT_SYMBOL(sk_stop_timer); 3524 3525 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3526 { 3527 if (del_timer_sync(timer)) 3528 __sock_put(sk); 3529 } 3530 EXPORT_SYMBOL(sk_stop_timer_sync); 3531 3532 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3533 { 3534 sk_init_common(sk); 3535 sk->sk_send_head = NULL; 3536 3537 timer_setup(&sk->sk_timer, NULL, 0); 3538 3539 sk->sk_allocation = GFP_KERNEL; 3540 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3541 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3542 sk->sk_state = TCP_CLOSE; 3543 sk->sk_use_task_frag = true; 3544 sk_set_socket(sk, sock); 3545 3546 sock_set_flag(sk, SOCK_ZAPPED); 3547 3548 if (sock) { 3549 sk->sk_type = sock->type; 3550 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3551 sock->sk = sk; 3552 } else { 3553 RCU_INIT_POINTER(sk->sk_wq, NULL); 3554 } 3555 sk->sk_uid = uid; 3556 3557 sk->sk_state_change = sock_def_wakeup; 3558 sk->sk_data_ready = sock_def_readable; 3559 sk->sk_write_space = sock_def_write_space; 3560 sk->sk_error_report = sock_def_error_report; 3561 sk->sk_destruct = sock_def_destruct; 3562 3563 sk->sk_frag.page = NULL; 3564 sk->sk_frag.offset = 0; 3565 sk->sk_peek_off = -1; 3566 3567 sk->sk_peer_pid = NULL; 3568 sk->sk_peer_cred = NULL; 3569 spin_lock_init(&sk->sk_peer_lock); 3570 3571 sk->sk_write_pending = 0; 3572 sk->sk_rcvlowat = 1; 3573 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3574 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3575 3576 sk->sk_stamp = SK_DEFAULT_STAMP; 3577 #if BITS_PER_LONG==32 3578 seqlock_init(&sk->sk_stamp_seq); 3579 #endif 3580 atomic_set(&sk->sk_zckey, 0); 3581 3582 #ifdef CONFIG_NET_RX_BUSY_POLL 3583 sk->sk_napi_id = 0; 3584 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3585 #endif 3586 3587 sk->sk_max_pacing_rate = ~0UL; 3588 sk->sk_pacing_rate = ~0UL; 3589 WRITE_ONCE(sk->sk_pacing_shift, 10); 3590 sk->sk_incoming_cpu = -1; 3591 3592 sk_rx_queue_clear(sk); 3593 /* 3594 * Before updating sk_refcnt, we must commit prior changes to memory 3595 * (Documentation/RCU/rculist_nulls.rst for details) 3596 */ 3597 smp_wmb(); 3598 refcount_set(&sk->sk_refcnt, 1); 3599 atomic_set(&sk->sk_drops, 0); 3600 } 3601 EXPORT_SYMBOL(sock_init_data_uid); 3602 3603 void sock_init_data(struct socket *sock, struct sock *sk) 3604 { 3605 kuid_t uid = sock ? 3606 SOCK_INODE(sock)->i_uid : 3607 make_kuid(sock_net(sk)->user_ns, 0); 3608 3609 sock_init_data_uid(sock, sk, uid); 3610 } 3611 EXPORT_SYMBOL(sock_init_data); 3612 3613 void lock_sock_nested(struct sock *sk, int subclass) 3614 { 3615 /* The sk_lock has mutex_lock() semantics here. */ 3616 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3617 3618 might_sleep(); 3619 spin_lock_bh(&sk->sk_lock.slock); 3620 if (sock_owned_by_user_nocheck(sk)) 3621 __lock_sock(sk); 3622 sk->sk_lock.owned = 1; 3623 spin_unlock_bh(&sk->sk_lock.slock); 3624 } 3625 EXPORT_SYMBOL(lock_sock_nested); 3626 3627 void release_sock(struct sock *sk) 3628 { 3629 spin_lock_bh(&sk->sk_lock.slock); 3630 if (sk->sk_backlog.tail) 3631 __release_sock(sk); 3632 3633 if (sk->sk_prot->release_cb) 3634 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3635 tcp_release_cb, sk); 3636 3637 sock_release_ownership(sk); 3638 if (waitqueue_active(&sk->sk_lock.wq)) 3639 wake_up(&sk->sk_lock.wq); 3640 spin_unlock_bh(&sk->sk_lock.slock); 3641 } 3642 EXPORT_SYMBOL(release_sock); 3643 3644 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3645 { 3646 might_sleep(); 3647 spin_lock_bh(&sk->sk_lock.slock); 3648 3649 if (!sock_owned_by_user_nocheck(sk)) { 3650 /* 3651 * Fast path return with bottom halves disabled and 3652 * sock::sk_lock.slock held. 3653 * 3654 * The 'mutex' is not contended and holding 3655 * sock::sk_lock.slock prevents all other lockers to 3656 * proceed so the corresponding unlock_sock_fast() can 3657 * avoid the slow path of release_sock() completely and 3658 * just release slock. 3659 * 3660 * From a semantical POV this is equivalent to 'acquiring' 3661 * the 'mutex', hence the corresponding lockdep 3662 * mutex_release() has to happen in the fast path of 3663 * unlock_sock_fast(). 3664 */ 3665 return false; 3666 } 3667 3668 __lock_sock(sk); 3669 sk->sk_lock.owned = 1; 3670 __acquire(&sk->sk_lock.slock); 3671 spin_unlock_bh(&sk->sk_lock.slock); 3672 return true; 3673 } 3674 EXPORT_SYMBOL(__lock_sock_fast); 3675 3676 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3677 bool timeval, bool time32) 3678 { 3679 struct sock *sk = sock->sk; 3680 struct timespec64 ts; 3681 3682 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3683 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3684 if (ts.tv_sec == -1) 3685 return -ENOENT; 3686 if (ts.tv_sec == 0) { 3687 ktime_t kt = ktime_get_real(); 3688 sock_write_timestamp(sk, kt); 3689 ts = ktime_to_timespec64(kt); 3690 } 3691 3692 if (timeval) 3693 ts.tv_nsec /= 1000; 3694 3695 #ifdef CONFIG_COMPAT_32BIT_TIME 3696 if (time32) 3697 return put_old_timespec32(&ts, userstamp); 3698 #endif 3699 #ifdef CONFIG_SPARC64 3700 /* beware of padding in sparc64 timeval */ 3701 if (timeval && !in_compat_syscall()) { 3702 struct __kernel_old_timeval __user tv = { 3703 .tv_sec = ts.tv_sec, 3704 .tv_usec = ts.tv_nsec, 3705 }; 3706 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3707 return -EFAULT; 3708 return 0; 3709 } 3710 #endif 3711 return put_timespec64(&ts, userstamp); 3712 } 3713 EXPORT_SYMBOL(sock_gettstamp); 3714 3715 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3716 { 3717 if (!sock_flag(sk, flag)) { 3718 unsigned long previous_flags = sk->sk_flags; 3719 3720 sock_set_flag(sk, flag); 3721 /* 3722 * we just set one of the two flags which require net 3723 * time stamping, but time stamping might have been on 3724 * already because of the other one 3725 */ 3726 if (sock_needs_netstamp(sk) && 3727 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3728 net_enable_timestamp(); 3729 } 3730 } 3731 3732 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3733 int level, int type) 3734 { 3735 struct sock_exterr_skb *serr; 3736 struct sk_buff *skb; 3737 int copied, err; 3738 3739 err = -EAGAIN; 3740 skb = sock_dequeue_err_skb(sk); 3741 if (skb == NULL) 3742 goto out; 3743 3744 copied = skb->len; 3745 if (copied > len) { 3746 msg->msg_flags |= MSG_TRUNC; 3747 copied = len; 3748 } 3749 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3750 if (err) 3751 goto out_free_skb; 3752 3753 sock_recv_timestamp(msg, sk, skb); 3754 3755 serr = SKB_EXT_ERR(skb); 3756 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3757 3758 msg->msg_flags |= MSG_ERRQUEUE; 3759 err = copied; 3760 3761 out_free_skb: 3762 kfree_skb(skb); 3763 out: 3764 return err; 3765 } 3766 EXPORT_SYMBOL(sock_recv_errqueue); 3767 3768 /* 3769 * Get a socket option on an socket. 3770 * 3771 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3772 * asynchronous errors should be reported by getsockopt. We assume 3773 * this means if you specify SO_ERROR (otherwise what is the point of it). 3774 */ 3775 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3776 char __user *optval, int __user *optlen) 3777 { 3778 struct sock *sk = sock->sk; 3779 3780 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3781 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3782 } 3783 EXPORT_SYMBOL(sock_common_getsockopt); 3784 3785 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3786 int flags) 3787 { 3788 struct sock *sk = sock->sk; 3789 int addr_len = 0; 3790 int err; 3791 3792 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3793 if (err >= 0) 3794 msg->msg_namelen = addr_len; 3795 return err; 3796 } 3797 EXPORT_SYMBOL(sock_common_recvmsg); 3798 3799 /* 3800 * Set socket options on an inet socket. 3801 */ 3802 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3803 sockptr_t optval, unsigned int optlen) 3804 { 3805 struct sock *sk = sock->sk; 3806 3807 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3808 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3809 } 3810 EXPORT_SYMBOL(sock_common_setsockopt); 3811 3812 void sk_common_release(struct sock *sk) 3813 { 3814 if (sk->sk_prot->destroy) 3815 sk->sk_prot->destroy(sk); 3816 3817 /* 3818 * Observation: when sk_common_release is called, processes have 3819 * no access to socket. But net still has. 3820 * Step one, detach it from networking: 3821 * 3822 * A. Remove from hash tables. 3823 */ 3824 3825 sk->sk_prot->unhash(sk); 3826 3827 /* 3828 * In this point socket cannot receive new packets, but it is possible 3829 * that some packets are in flight because some CPU runs receiver and 3830 * did hash table lookup before we unhashed socket. They will achieve 3831 * receive queue and will be purged by socket destructor. 3832 * 3833 * Also we still have packets pending on receive queue and probably, 3834 * our own packets waiting in device queues. sock_destroy will drain 3835 * receive queue, but transmitted packets will delay socket destruction 3836 * until the last reference will be released. 3837 */ 3838 3839 sock_orphan(sk); 3840 3841 xfrm_sk_free_policy(sk); 3842 3843 sock_put(sk); 3844 } 3845 EXPORT_SYMBOL(sk_common_release); 3846 3847 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3848 { 3849 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3850 3851 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3852 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3853 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3854 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3855 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk); 3856 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3857 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3858 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3859 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3860 } 3861 3862 #ifdef CONFIG_PROC_FS 3863 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3864 3865 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3866 { 3867 int cpu, idx = prot->inuse_idx; 3868 int res = 0; 3869 3870 for_each_possible_cpu(cpu) 3871 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3872 3873 return res >= 0 ? res : 0; 3874 } 3875 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3876 3877 int sock_inuse_get(struct net *net) 3878 { 3879 int cpu, res = 0; 3880 3881 for_each_possible_cpu(cpu) 3882 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3883 3884 return res; 3885 } 3886 3887 EXPORT_SYMBOL_GPL(sock_inuse_get); 3888 3889 static int __net_init sock_inuse_init_net(struct net *net) 3890 { 3891 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3892 if (net->core.prot_inuse == NULL) 3893 return -ENOMEM; 3894 return 0; 3895 } 3896 3897 static void __net_exit sock_inuse_exit_net(struct net *net) 3898 { 3899 free_percpu(net->core.prot_inuse); 3900 } 3901 3902 static struct pernet_operations net_inuse_ops = { 3903 .init = sock_inuse_init_net, 3904 .exit = sock_inuse_exit_net, 3905 }; 3906 3907 static __init int net_inuse_init(void) 3908 { 3909 if (register_pernet_subsys(&net_inuse_ops)) 3910 panic("Cannot initialize net inuse counters"); 3911 3912 return 0; 3913 } 3914 3915 core_initcall(net_inuse_init); 3916 3917 static int assign_proto_idx(struct proto *prot) 3918 { 3919 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3920 3921 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3922 pr_err("PROTO_INUSE_NR exhausted\n"); 3923 return -ENOSPC; 3924 } 3925 3926 set_bit(prot->inuse_idx, proto_inuse_idx); 3927 return 0; 3928 } 3929 3930 static void release_proto_idx(struct proto *prot) 3931 { 3932 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3933 clear_bit(prot->inuse_idx, proto_inuse_idx); 3934 } 3935 #else 3936 static inline int assign_proto_idx(struct proto *prot) 3937 { 3938 return 0; 3939 } 3940 3941 static inline void release_proto_idx(struct proto *prot) 3942 { 3943 } 3944 3945 #endif 3946 3947 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3948 { 3949 if (!twsk_prot) 3950 return; 3951 kfree(twsk_prot->twsk_slab_name); 3952 twsk_prot->twsk_slab_name = NULL; 3953 kmem_cache_destroy(twsk_prot->twsk_slab); 3954 twsk_prot->twsk_slab = NULL; 3955 } 3956 3957 static int tw_prot_init(const struct proto *prot) 3958 { 3959 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3960 3961 if (!twsk_prot) 3962 return 0; 3963 3964 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3965 prot->name); 3966 if (!twsk_prot->twsk_slab_name) 3967 return -ENOMEM; 3968 3969 twsk_prot->twsk_slab = 3970 kmem_cache_create(twsk_prot->twsk_slab_name, 3971 twsk_prot->twsk_obj_size, 0, 3972 SLAB_ACCOUNT | prot->slab_flags, 3973 NULL); 3974 if (!twsk_prot->twsk_slab) { 3975 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3976 prot->name); 3977 return -ENOMEM; 3978 } 3979 3980 return 0; 3981 } 3982 3983 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3984 { 3985 if (!rsk_prot) 3986 return; 3987 kfree(rsk_prot->slab_name); 3988 rsk_prot->slab_name = NULL; 3989 kmem_cache_destroy(rsk_prot->slab); 3990 rsk_prot->slab = NULL; 3991 } 3992 3993 static int req_prot_init(const struct proto *prot) 3994 { 3995 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3996 3997 if (!rsk_prot) 3998 return 0; 3999 4000 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4001 prot->name); 4002 if (!rsk_prot->slab_name) 4003 return -ENOMEM; 4004 4005 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4006 rsk_prot->obj_size, 0, 4007 SLAB_ACCOUNT | prot->slab_flags, 4008 NULL); 4009 4010 if (!rsk_prot->slab) { 4011 pr_crit("%s: Can't create request sock SLAB cache!\n", 4012 prot->name); 4013 return -ENOMEM; 4014 } 4015 return 0; 4016 } 4017 4018 int proto_register(struct proto *prot, int alloc_slab) 4019 { 4020 int ret = -ENOBUFS; 4021 4022 if (prot->memory_allocated && !prot->sysctl_mem) { 4023 pr_err("%s: missing sysctl_mem\n", prot->name); 4024 return -EINVAL; 4025 } 4026 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4027 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4028 return -EINVAL; 4029 } 4030 if (alloc_slab) { 4031 prot->slab = kmem_cache_create_usercopy(prot->name, 4032 prot->obj_size, 0, 4033 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4034 prot->slab_flags, 4035 prot->useroffset, prot->usersize, 4036 NULL); 4037 4038 if (prot->slab == NULL) { 4039 pr_crit("%s: Can't create sock SLAB cache!\n", 4040 prot->name); 4041 goto out; 4042 } 4043 4044 if (req_prot_init(prot)) 4045 goto out_free_request_sock_slab; 4046 4047 if (tw_prot_init(prot)) 4048 goto out_free_timewait_sock_slab; 4049 } 4050 4051 mutex_lock(&proto_list_mutex); 4052 ret = assign_proto_idx(prot); 4053 if (ret) { 4054 mutex_unlock(&proto_list_mutex); 4055 goto out_free_timewait_sock_slab; 4056 } 4057 list_add(&prot->node, &proto_list); 4058 mutex_unlock(&proto_list_mutex); 4059 return ret; 4060 4061 out_free_timewait_sock_slab: 4062 if (alloc_slab) 4063 tw_prot_cleanup(prot->twsk_prot); 4064 out_free_request_sock_slab: 4065 if (alloc_slab) { 4066 req_prot_cleanup(prot->rsk_prot); 4067 4068 kmem_cache_destroy(prot->slab); 4069 prot->slab = NULL; 4070 } 4071 out: 4072 return ret; 4073 } 4074 EXPORT_SYMBOL(proto_register); 4075 4076 void proto_unregister(struct proto *prot) 4077 { 4078 mutex_lock(&proto_list_mutex); 4079 release_proto_idx(prot); 4080 list_del(&prot->node); 4081 mutex_unlock(&proto_list_mutex); 4082 4083 kmem_cache_destroy(prot->slab); 4084 prot->slab = NULL; 4085 4086 req_prot_cleanup(prot->rsk_prot); 4087 tw_prot_cleanup(prot->twsk_prot); 4088 } 4089 EXPORT_SYMBOL(proto_unregister); 4090 4091 int sock_load_diag_module(int family, int protocol) 4092 { 4093 if (!protocol) { 4094 if (!sock_is_registered(family)) 4095 return -ENOENT; 4096 4097 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4098 NETLINK_SOCK_DIAG, family); 4099 } 4100 4101 #ifdef CONFIG_INET 4102 if (family == AF_INET && 4103 protocol != IPPROTO_RAW && 4104 protocol < MAX_INET_PROTOS && 4105 !rcu_access_pointer(inet_protos[protocol])) 4106 return -ENOENT; 4107 #endif 4108 4109 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4110 NETLINK_SOCK_DIAG, family, protocol); 4111 } 4112 EXPORT_SYMBOL(sock_load_diag_module); 4113 4114 #ifdef CONFIG_PROC_FS 4115 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4116 __acquires(proto_list_mutex) 4117 { 4118 mutex_lock(&proto_list_mutex); 4119 return seq_list_start_head(&proto_list, *pos); 4120 } 4121 4122 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4123 { 4124 return seq_list_next(v, &proto_list, pos); 4125 } 4126 4127 static void proto_seq_stop(struct seq_file *seq, void *v) 4128 __releases(proto_list_mutex) 4129 { 4130 mutex_unlock(&proto_list_mutex); 4131 } 4132 4133 static char proto_method_implemented(const void *method) 4134 { 4135 return method == NULL ? 'n' : 'y'; 4136 } 4137 static long sock_prot_memory_allocated(struct proto *proto) 4138 { 4139 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4140 } 4141 4142 static const char *sock_prot_memory_pressure(struct proto *proto) 4143 { 4144 return proto->memory_pressure != NULL ? 4145 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4146 } 4147 4148 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4149 { 4150 4151 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4152 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4153 proto->name, 4154 proto->obj_size, 4155 sock_prot_inuse_get(seq_file_net(seq), proto), 4156 sock_prot_memory_allocated(proto), 4157 sock_prot_memory_pressure(proto), 4158 proto->max_header, 4159 proto->slab == NULL ? "no" : "yes", 4160 module_name(proto->owner), 4161 proto_method_implemented(proto->close), 4162 proto_method_implemented(proto->connect), 4163 proto_method_implemented(proto->disconnect), 4164 proto_method_implemented(proto->accept), 4165 proto_method_implemented(proto->ioctl), 4166 proto_method_implemented(proto->init), 4167 proto_method_implemented(proto->destroy), 4168 proto_method_implemented(proto->shutdown), 4169 proto_method_implemented(proto->setsockopt), 4170 proto_method_implemented(proto->getsockopt), 4171 proto_method_implemented(proto->sendmsg), 4172 proto_method_implemented(proto->recvmsg), 4173 proto_method_implemented(proto->bind), 4174 proto_method_implemented(proto->backlog_rcv), 4175 proto_method_implemented(proto->hash), 4176 proto_method_implemented(proto->unhash), 4177 proto_method_implemented(proto->get_port), 4178 proto_method_implemented(proto->enter_memory_pressure)); 4179 } 4180 4181 static int proto_seq_show(struct seq_file *seq, void *v) 4182 { 4183 if (v == &proto_list) 4184 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4185 "protocol", 4186 "size", 4187 "sockets", 4188 "memory", 4189 "press", 4190 "maxhdr", 4191 "slab", 4192 "module", 4193 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4194 else 4195 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4196 return 0; 4197 } 4198 4199 static const struct seq_operations proto_seq_ops = { 4200 .start = proto_seq_start, 4201 .next = proto_seq_next, 4202 .stop = proto_seq_stop, 4203 .show = proto_seq_show, 4204 }; 4205 4206 static __net_init int proto_init_net(struct net *net) 4207 { 4208 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4209 sizeof(struct seq_net_private))) 4210 return -ENOMEM; 4211 4212 return 0; 4213 } 4214 4215 static __net_exit void proto_exit_net(struct net *net) 4216 { 4217 remove_proc_entry("protocols", net->proc_net); 4218 } 4219 4220 4221 static __net_initdata struct pernet_operations proto_net_ops = { 4222 .init = proto_init_net, 4223 .exit = proto_exit_net, 4224 }; 4225 4226 static int __init proto_init(void) 4227 { 4228 return register_pernet_subsys(&proto_net_ops); 4229 } 4230 4231 subsys_initcall(proto_init); 4232 4233 #endif /* PROC_FS */ 4234 4235 #ifdef CONFIG_NET_RX_BUSY_POLL 4236 bool sk_busy_loop_end(void *p, unsigned long start_time) 4237 { 4238 struct sock *sk = p; 4239 4240 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4241 return true; 4242 4243 if (sk_is_udp(sk) && 4244 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4245 return true; 4246 4247 return sk_busy_loop_timeout(sk, start_time); 4248 } 4249 EXPORT_SYMBOL(sk_busy_loop_end); 4250 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4251 4252 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4253 { 4254 if (!sk->sk_prot->bind_add) 4255 return -EOPNOTSUPP; 4256 return sk->sk_prot->bind_add(sk, addr, addr_len); 4257 } 4258 EXPORT_SYMBOL(sock_bind_add); 4259 4260 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4261 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4262 void __user *arg, void *karg, size_t size) 4263 { 4264 int ret; 4265 4266 if (copy_from_user(karg, arg, size)) 4267 return -EFAULT; 4268 4269 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4270 if (ret) 4271 return ret; 4272 4273 if (copy_to_user(arg, karg, size)) 4274 return -EFAULT; 4275 4276 return 0; 4277 } 4278 EXPORT_SYMBOL(sock_ioctl_inout); 4279 4280 /* This is the most common ioctl prep function, where the result (4 bytes) is 4281 * copied back to userspace if the ioctl() returns successfully. No input is 4282 * copied from userspace as input argument. 4283 */ 4284 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4285 { 4286 int ret, karg = 0; 4287 4288 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4289 if (ret) 4290 return ret; 4291 4292 return put_user(karg, (int __user *)arg); 4293 } 4294 4295 /* A wrapper around sock ioctls, which copies the data from userspace 4296 * (depending on the protocol/ioctl), and copies back the result to userspace. 4297 * The main motivation for this function is to pass kernel memory to the 4298 * protocol ioctl callbacks, instead of userspace memory. 4299 */ 4300 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4301 { 4302 int rc = 1; 4303 4304 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4305 rc = ipmr_sk_ioctl(sk, cmd, arg); 4306 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4307 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4308 else if (sk_is_phonet(sk)) 4309 rc = phonet_sk_ioctl(sk, cmd, arg); 4310 4311 /* If ioctl was processed, returns its value */ 4312 if (rc <= 0) 4313 return rc; 4314 4315 /* Otherwise call the default handler */ 4316 return sock_ioctl_out(sk, cmd, arg); 4317 } 4318 EXPORT_SYMBOL(sk_ioctl); 4319 4320 static int __init sock_struct_check(void) 4321 { 4322 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4323 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4324 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4325 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4326 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4327 4328 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4329 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4330 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4331 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4332 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4333 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4334 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4335 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4336 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4337 4338 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4339 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4340 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4341 4342 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4343 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4344 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4345 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4346 4347 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf); 4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4353 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm); 4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status); 4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4363 4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4367 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4372 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4377 return 0; 4378 } 4379 4380 core_initcall(sock_struct_check); 4381