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 and frags that the user can SO_DEVMEM_DONTNEED 1051 * in 1 syscall. The limit exists to limit the amount of memory the kernel 1052 * allocates to copy these tokens, and to prevent looping over the frags for 1053 * too long. 1054 */ 1055 #define MAX_DONTNEED_TOKENS 128 1056 #define MAX_DONTNEED_FRAGS 1024 1057 1058 static noinline_for_stack int 1059 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1060 { 1061 unsigned int num_tokens, i, j, k, netmem_num = 0; 1062 struct dmabuf_token *tokens; 1063 int ret = 0, num_frags = 0; 1064 netmem_ref netmems[16]; 1065 1066 if (!sk_is_tcp(sk)) 1067 return -EBADF; 1068 1069 if (optlen % sizeof(*tokens) || 1070 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1071 return -EINVAL; 1072 1073 num_tokens = optlen / sizeof(*tokens); 1074 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL); 1075 if (!tokens) 1076 return -ENOMEM; 1077 1078 if (copy_from_sockptr(tokens, optval, optlen)) { 1079 kvfree(tokens); 1080 return -EFAULT; 1081 } 1082 1083 xa_lock_bh(&sk->sk_user_frags); 1084 for (i = 0; i < num_tokens; i++) { 1085 for (j = 0; j < tokens[i].token_count; j++) { 1086 if (++num_frags > MAX_DONTNEED_FRAGS) 1087 goto frag_limit_reached; 1088 1089 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1090 &sk->sk_user_frags, tokens[i].token_start + j); 1091 1092 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem))) 1093 continue; 1094 1095 netmems[netmem_num++] = netmem; 1096 if (netmem_num == ARRAY_SIZE(netmems)) { 1097 xa_unlock_bh(&sk->sk_user_frags); 1098 for (k = 0; k < netmem_num; k++) 1099 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1100 netmem_num = 0; 1101 xa_lock_bh(&sk->sk_user_frags); 1102 } 1103 ret++; 1104 } 1105 } 1106 1107 frag_limit_reached: 1108 xa_unlock_bh(&sk->sk_user_frags); 1109 for (k = 0; k < netmem_num; k++) 1110 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1111 1112 kvfree(tokens); 1113 return ret; 1114 } 1115 #endif 1116 1117 void sockopt_lock_sock(struct sock *sk) 1118 { 1119 /* When current->bpf_ctx is set, the setsockopt is called from 1120 * a bpf prog. bpf has ensured the sk lock has been 1121 * acquired before calling setsockopt(). 1122 */ 1123 if (has_current_bpf_ctx()) 1124 return; 1125 1126 lock_sock(sk); 1127 } 1128 EXPORT_SYMBOL(sockopt_lock_sock); 1129 1130 void sockopt_release_sock(struct sock *sk) 1131 { 1132 if (has_current_bpf_ctx()) 1133 return; 1134 1135 release_sock(sk); 1136 } 1137 EXPORT_SYMBOL(sockopt_release_sock); 1138 1139 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1140 { 1141 return has_current_bpf_ctx() || ns_capable(ns, cap); 1142 } 1143 EXPORT_SYMBOL(sockopt_ns_capable); 1144 1145 bool sockopt_capable(int cap) 1146 { 1147 return has_current_bpf_ctx() || capable(cap); 1148 } 1149 EXPORT_SYMBOL(sockopt_capable); 1150 1151 static int sockopt_validate_clockid(__kernel_clockid_t value) 1152 { 1153 switch (value) { 1154 case CLOCK_REALTIME: 1155 case CLOCK_MONOTONIC: 1156 case CLOCK_TAI: 1157 return 0; 1158 } 1159 return -EINVAL; 1160 } 1161 1162 /* 1163 * This is meant for all protocols to use and covers goings on 1164 * at the socket level. Everything here is generic. 1165 */ 1166 1167 int sk_setsockopt(struct sock *sk, int level, int optname, 1168 sockptr_t optval, unsigned int optlen) 1169 { 1170 struct so_timestamping timestamping; 1171 struct socket *sock = sk->sk_socket; 1172 struct sock_txtime sk_txtime; 1173 int val; 1174 int valbool; 1175 struct linger ling; 1176 int ret = 0; 1177 1178 /* 1179 * Options without arguments 1180 */ 1181 1182 if (optname == SO_BINDTODEVICE) 1183 return sock_setbindtodevice(sk, optval, optlen); 1184 1185 if (optlen < sizeof(int)) 1186 return -EINVAL; 1187 1188 if (copy_from_sockptr(&val, optval, sizeof(val))) 1189 return -EFAULT; 1190 1191 valbool = val ? 1 : 0; 1192 1193 /* handle options which do not require locking the socket. */ 1194 switch (optname) { 1195 case SO_PRIORITY: 1196 if ((val >= 0 && val <= 6) || 1197 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1198 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1199 sock_set_priority(sk, val); 1200 return 0; 1201 } 1202 return -EPERM; 1203 case SO_PASSSEC: 1204 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1205 return 0; 1206 case SO_PASSCRED: 1207 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1208 return 0; 1209 case SO_PASSPIDFD: 1210 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1211 return 0; 1212 case SO_TYPE: 1213 case SO_PROTOCOL: 1214 case SO_DOMAIN: 1215 case SO_ERROR: 1216 return -ENOPROTOOPT; 1217 #ifdef CONFIG_NET_RX_BUSY_POLL 1218 case SO_BUSY_POLL: 1219 if (val < 0) 1220 return -EINVAL; 1221 WRITE_ONCE(sk->sk_ll_usec, val); 1222 return 0; 1223 case SO_PREFER_BUSY_POLL: 1224 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1225 return -EPERM; 1226 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1227 return 0; 1228 case SO_BUSY_POLL_BUDGET: 1229 if (val > READ_ONCE(sk->sk_busy_poll_budget) && 1230 !sockopt_capable(CAP_NET_ADMIN)) 1231 return -EPERM; 1232 if (val < 0 || val > U16_MAX) 1233 return -EINVAL; 1234 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1235 return 0; 1236 #endif 1237 case SO_MAX_PACING_RATE: 1238 { 1239 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1240 unsigned long pacing_rate; 1241 1242 if (sizeof(ulval) != sizeof(val) && 1243 optlen >= sizeof(ulval) && 1244 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1245 return -EFAULT; 1246 } 1247 if (ulval != ~0UL) 1248 cmpxchg(&sk->sk_pacing_status, 1249 SK_PACING_NONE, 1250 SK_PACING_NEEDED); 1251 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1252 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1253 pacing_rate = READ_ONCE(sk->sk_pacing_rate); 1254 if (ulval < pacing_rate) 1255 WRITE_ONCE(sk->sk_pacing_rate, ulval); 1256 return 0; 1257 } 1258 case SO_TXREHASH: 1259 if (val < -1 || val > 1) 1260 return -EINVAL; 1261 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1262 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1263 /* Paired with READ_ONCE() in tcp_rtx_synack() 1264 * and sk_getsockopt(). 1265 */ 1266 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1267 return 0; 1268 case SO_PEEK_OFF: 1269 { 1270 int (*set_peek_off)(struct sock *sk, int val); 1271 1272 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1273 if (set_peek_off) 1274 ret = set_peek_off(sk, val); 1275 else 1276 ret = -EOPNOTSUPP; 1277 return ret; 1278 } 1279 #ifdef CONFIG_PAGE_POOL 1280 case SO_DEVMEM_DONTNEED: 1281 return sock_devmem_dontneed(sk, optval, optlen); 1282 #endif 1283 } 1284 1285 sockopt_lock_sock(sk); 1286 1287 switch (optname) { 1288 case SO_DEBUG: 1289 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1290 ret = -EACCES; 1291 else 1292 sock_valbool_flag(sk, SOCK_DBG, valbool); 1293 break; 1294 case SO_REUSEADDR: 1295 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1296 break; 1297 case SO_REUSEPORT: 1298 sk->sk_reuseport = valbool; 1299 break; 1300 case SO_DONTROUTE: 1301 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1302 sk_dst_reset(sk); 1303 break; 1304 case SO_BROADCAST: 1305 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1306 break; 1307 case SO_SNDBUF: 1308 /* Don't error on this BSD doesn't and if you think 1309 * about it this is right. Otherwise apps have to 1310 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1311 * are treated in BSD as hints 1312 */ 1313 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1314 set_sndbuf: 1315 /* Ensure val * 2 fits into an int, to prevent max_t() 1316 * from treating it as a negative value. 1317 */ 1318 val = min_t(int, val, INT_MAX / 2); 1319 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1320 WRITE_ONCE(sk->sk_sndbuf, 1321 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1322 /* Wake up sending tasks if we upped the value. */ 1323 sk->sk_write_space(sk); 1324 break; 1325 1326 case SO_SNDBUFFORCE: 1327 if (!sockopt_capable(CAP_NET_ADMIN)) { 1328 ret = -EPERM; 1329 break; 1330 } 1331 1332 /* No negative values (to prevent underflow, as val will be 1333 * multiplied by 2). 1334 */ 1335 if (val < 0) 1336 val = 0; 1337 goto set_sndbuf; 1338 1339 case SO_RCVBUF: 1340 /* Don't error on this BSD doesn't and if you think 1341 * about it this is right. Otherwise apps have to 1342 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1343 * are treated in BSD as hints 1344 */ 1345 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1346 break; 1347 1348 case SO_RCVBUFFORCE: 1349 if (!sockopt_capable(CAP_NET_ADMIN)) { 1350 ret = -EPERM; 1351 break; 1352 } 1353 1354 /* No negative values (to prevent underflow, as val will be 1355 * multiplied by 2). 1356 */ 1357 __sock_set_rcvbuf(sk, max(val, 0)); 1358 break; 1359 1360 case SO_KEEPALIVE: 1361 if (sk->sk_prot->keepalive) 1362 sk->sk_prot->keepalive(sk, valbool); 1363 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1364 break; 1365 1366 case SO_OOBINLINE: 1367 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1368 break; 1369 1370 case SO_NO_CHECK: 1371 sk->sk_no_check_tx = valbool; 1372 break; 1373 1374 case SO_LINGER: 1375 if (optlen < sizeof(ling)) { 1376 ret = -EINVAL; /* 1003.1g */ 1377 break; 1378 } 1379 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1380 ret = -EFAULT; 1381 break; 1382 } 1383 if (!ling.l_onoff) { 1384 sock_reset_flag(sk, SOCK_LINGER); 1385 } else { 1386 unsigned long t_sec = ling.l_linger; 1387 1388 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1389 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1390 else 1391 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1392 sock_set_flag(sk, SOCK_LINGER); 1393 } 1394 break; 1395 1396 case SO_BSDCOMPAT: 1397 break; 1398 1399 case SO_TIMESTAMP_OLD: 1400 case SO_TIMESTAMP_NEW: 1401 case SO_TIMESTAMPNS_OLD: 1402 case SO_TIMESTAMPNS_NEW: 1403 sock_set_timestamp(sk, optname, valbool); 1404 break; 1405 1406 case SO_TIMESTAMPING_NEW: 1407 case SO_TIMESTAMPING_OLD: 1408 if (optlen == sizeof(timestamping)) { 1409 if (copy_from_sockptr(×tamping, optval, 1410 sizeof(timestamping))) { 1411 ret = -EFAULT; 1412 break; 1413 } 1414 } else { 1415 memset(×tamping, 0, sizeof(timestamping)); 1416 timestamping.flags = val; 1417 } 1418 ret = sock_set_timestamping(sk, optname, timestamping); 1419 break; 1420 1421 case SO_RCVLOWAT: 1422 { 1423 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1424 1425 if (val < 0) 1426 val = INT_MAX; 1427 if (sock) 1428 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1429 if (set_rcvlowat) 1430 ret = set_rcvlowat(sk, val); 1431 else 1432 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1433 break; 1434 } 1435 case SO_RCVTIMEO_OLD: 1436 case SO_RCVTIMEO_NEW: 1437 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1438 optlen, optname == SO_RCVTIMEO_OLD); 1439 break; 1440 1441 case SO_SNDTIMEO_OLD: 1442 case SO_SNDTIMEO_NEW: 1443 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1444 optlen, optname == SO_SNDTIMEO_OLD); 1445 break; 1446 1447 case SO_ATTACH_FILTER: { 1448 struct sock_fprog fprog; 1449 1450 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1451 if (!ret) 1452 ret = sk_attach_filter(&fprog, sk); 1453 break; 1454 } 1455 case SO_ATTACH_BPF: 1456 ret = -EINVAL; 1457 if (optlen == sizeof(u32)) { 1458 u32 ufd; 1459 1460 ret = -EFAULT; 1461 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1462 break; 1463 1464 ret = sk_attach_bpf(ufd, sk); 1465 } 1466 break; 1467 1468 case SO_ATTACH_REUSEPORT_CBPF: { 1469 struct sock_fprog fprog; 1470 1471 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1472 if (!ret) 1473 ret = sk_reuseport_attach_filter(&fprog, sk); 1474 break; 1475 } 1476 case SO_ATTACH_REUSEPORT_EBPF: 1477 ret = -EINVAL; 1478 if (optlen == sizeof(u32)) { 1479 u32 ufd; 1480 1481 ret = -EFAULT; 1482 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1483 break; 1484 1485 ret = sk_reuseport_attach_bpf(ufd, sk); 1486 } 1487 break; 1488 1489 case SO_DETACH_REUSEPORT_BPF: 1490 ret = reuseport_detach_prog(sk); 1491 break; 1492 1493 case SO_DETACH_FILTER: 1494 ret = sk_detach_filter(sk); 1495 break; 1496 1497 case SO_LOCK_FILTER: 1498 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1499 ret = -EPERM; 1500 else 1501 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1502 break; 1503 1504 case SO_MARK: 1505 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1506 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1507 ret = -EPERM; 1508 break; 1509 } 1510 1511 __sock_set_mark(sk, val); 1512 break; 1513 case SO_RCVMARK: 1514 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1515 break; 1516 1517 case SO_RXQ_OVFL: 1518 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1519 break; 1520 1521 case SO_WIFI_STATUS: 1522 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1523 break; 1524 1525 case SO_NOFCS: 1526 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1527 break; 1528 1529 case SO_SELECT_ERR_QUEUE: 1530 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1531 break; 1532 1533 1534 case SO_INCOMING_CPU: 1535 reuseport_update_incoming_cpu(sk, val); 1536 break; 1537 1538 case SO_CNX_ADVICE: 1539 if (val == 1) 1540 dst_negative_advice(sk); 1541 break; 1542 1543 case SO_ZEROCOPY: 1544 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1545 if (!(sk_is_tcp(sk) || 1546 (sk->sk_type == SOCK_DGRAM && 1547 sk->sk_protocol == IPPROTO_UDP))) 1548 ret = -EOPNOTSUPP; 1549 } else if (sk->sk_family != PF_RDS) { 1550 ret = -EOPNOTSUPP; 1551 } 1552 if (!ret) { 1553 if (val < 0 || val > 1) 1554 ret = -EINVAL; 1555 else 1556 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1557 } 1558 break; 1559 1560 case SO_TXTIME: 1561 if (optlen != sizeof(struct sock_txtime)) { 1562 ret = -EINVAL; 1563 break; 1564 } else if (copy_from_sockptr(&sk_txtime, optval, 1565 sizeof(struct sock_txtime))) { 1566 ret = -EFAULT; 1567 break; 1568 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1569 ret = -EINVAL; 1570 break; 1571 } 1572 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1573 * scheduler has enough safe guards. 1574 */ 1575 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1576 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1577 ret = -EPERM; 1578 break; 1579 } 1580 1581 ret = sockopt_validate_clockid(sk_txtime.clockid); 1582 if (ret) 1583 break; 1584 1585 sock_valbool_flag(sk, SOCK_TXTIME, true); 1586 sk->sk_clockid = sk_txtime.clockid; 1587 sk->sk_txtime_deadline_mode = 1588 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1589 sk->sk_txtime_report_errors = 1590 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1591 break; 1592 1593 case SO_BINDTOIFINDEX: 1594 ret = sock_bindtoindex_locked(sk, val); 1595 break; 1596 1597 case SO_BUF_LOCK: 1598 if (val & ~SOCK_BUF_LOCK_MASK) { 1599 ret = -EINVAL; 1600 break; 1601 } 1602 sk->sk_userlocks = val | (sk->sk_userlocks & 1603 ~SOCK_BUF_LOCK_MASK); 1604 break; 1605 1606 case SO_RESERVE_MEM: 1607 { 1608 int delta; 1609 1610 if (val < 0) { 1611 ret = -EINVAL; 1612 break; 1613 } 1614 1615 delta = val - sk->sk_reserved_mem; 1616 if (delta < 0) 1617 sock_release_reserved_memory(sk, -delta); 1618 else 1619 ret = sock_reserve_memory(sk, delta); 1620 break; 1621 } 1622 1623 default: 1624 ret = -ENOPROTOOPT; 1625 break; 1626 } 1627 sockopt_release_sock(sk); 1628 return ret; 1629 } 1630 1631 int sock_setsockopt(struct socket *sock, int level, int optname, 1632 sockptr_t optval, unsigned int optlen) 1633 { 1634 return sk_setsockopt(sock->sk, level, optname, 1635 optval, optlen); 1636 } 1637 EXPORT_SYMBOL(sock_setsockopt); 1638 1639 static const struct cred *sk_get_peer_cred(struct sock *sk) 1640 { 1641 const struct cred *cred; 1642 1643 spin_lock(&sk->sk_peer_lock); 1644 cred = get_cred(sk->sk_peer_cred); 1645 spin_unlock(&sk->sk_peer_lock); 1646 1647 return cred; 1648 } 1649 1650 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1651 struct ucred *ucred) 1652 { 1653 ucred->pid = pid_vnr(pid); 1654 ucred->uid = ucred->gid = -1; 1655 if (cred) { 1656 struct user_namespace *current_ns = current_user_ns(); 1657 1658 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1659 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1660 } 1661 } 1662 1663 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1664 { 1665 struct user_namespace *user_ns = current_user_ns(); 1666 int i; 1667 1668 for (i = 0; i < src->ngroups; i++) { 1669 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1670 1671 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1672 return -EFAULT; 1673 } 1674 1675 return 0; 1676 } 1677 1678 int sk_getsockopt(struct sock *sk, int level, int optname, 1679 sockptr_t optval, sockptr_t optlen) 1680 { 1681 struct socket *sock = sk->sk_socket; 1682 1683 union { 1684 int val; 1685 u64 val64; 1686 unsigned long ulval; 1687 struct linger ling; 1688 struct old_timeval32 tm32; 1689 struct __kernel_old_timeval tm; 1690 struct __kernel_sock_timeval stm; 1691 struct sock_txtime txtime; 1692 struct so_timestamping timestamping; 1693 } v; 1694 1695 int lv = sizeof(int); 1696 int len; 1697 1698 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1699 return -EFAULT; 1700 if (len < 0) 1701 return -EINVAL; 1702 1703 memset(&v, 0, sizeof(v)); 1704 1705 switch (optname) { 1706 case SO_DEBUG: 1707 v.val = sock_flag(sk, SOCK_DBG); 1708 break; 1709 1710 case SO_DONTROUTE: 1711 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1712 break; 1713 1714 case SO_BROADCAST: 1715 v.val = sock_flag(sk, SOCK_BROADCAST); 1716 break; 1717 1718 case SO_SNDBUF: 1719 v.val = READ_ONCE(sk->sk_sndbuf); 1720 break; 1721 1722 case SO_RCVBUF: 1723 v.val = READ_ONCE(sk->sk_rcvbuf); 1724 break; 1725 1726 case SO_REUSEADDR: 1727 v.val = sk->sk_reuse; 1728 break; 1729 1730 case SO_REUSEPORT: 1731 v.val = sk->sk_reuseport; 1732 break; 1733 1734 case SO_KEEPALIVE: 1735 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1736 break; 1737 1738 case SO_TYPE: 1739 v.val = sk->sk_type; 1740 break; 1741 1742 case SO_PROTOCOL: 1743 v.val = sk->sk_protocol; 1744 break; 1745 1746 case SO_DOMAIN: 1747 v.val = sk->sk_family; 1748 break; 1749 1750 case SO_ERROR: 1751 v.val = -sock_error(sk); 1752 if (v.val == 0) 1753 v.val = xchg(&sk->sk_err_soft, 0); 1754 break; 1755 1756 case SO_OOBINLINE: 1757 v.val = sock_flag(sk, SOCK_URGINLINE); 1758 break; 1759 1760 case SO_NO_CHECK: 1761 v.val = sk->sk_no_check_tx; 1762 break; 1763 1764 case SO_PRIORITY: 1765 v.val = READ_ONCE(sk->sk_priority); 1766 break; 1767 1768 case SO_LINGER: 1769 lv = sizeof(v.ling); 1770 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1771 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1772 break; 1773 1774 case SO_BSDCOMPAT: 1775 break; 1776 1777 case SO_TIMESTAMP_OLD: 1778 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1779 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1780 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1781 break; 1782 1783 case SO_TIMESTAMPNS_OLD: 1784 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1785 break; 1786 1787 case SO_TIMESTAMP_NEW: 1788 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1789 break; 1790 1791 case SO_TIMESTAMPNS_NEW: 1792 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1793 break; 1794 1795 case SO_TIMESTAMPING_OLD: 1796 case SO_TIMESTAMPING_NEW: 1797 lv = sizeof(v.timestamping); 1798 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1799 * returning the flags when they were set through the same option. 1800 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1801 */ 1802 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1803 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1804 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1805 } 1806 break; 1807 1808 case SO_RCVTIMEO_OLD: 1809 case SO_RCVTIMEO_NEW: 1810 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1811 SO_RCVTIMEO_OLD == optname); 1812 break; 1813 1814 case SO_SNDTIMEO_OLD: 1815 case SO_SNDTIMEO_NEW: 1816 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1817 SO_SNDTIMEO_OLD == optname); 1818 break; 1819 1820 case SO_RCVLOWAT: 1821 v.val = READ_ONCE(sk->sk_rcvlowat); 1822 break; 1823 1824 case SO_SNDLOWAT: 1825 v.val = 1; 1826 break; 1827 1828 case SO_PASSCRED: 1829 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1830 break; 1831 1832 case SO_PASSPIDFD: 1833 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1834 break; 1835 1836 case SO_PEERCRED: 1837 { 1838 struct ucred peercred; 1839 if (len > sizeof(peercred)) 1840 len = sizeof(peercred); 1841 1842 spin_lock(&sk->sk_peer_lock); 1843 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1844 spin_unlock(&sk->sk_peer_lock); 1845 1846 if (copy_to_sockptr(optval, &peercred, len)) 1847 return -EFAULT; 1848 goto lenout; 1849 } 1850 1851 case SO_PEERPIDFD: 1852 { 1853 struct pid *peer_pid; 1854 struct file *pidfd_file = NULL; 1855 int pidfd; 1856 1857 if (len > sizeof(pidfd)) 1858 len = sizeof(pidfd); 1859 1860 spin_lock(&sk->sk_peer_lock); 1861 peer_pid = get_pid(sk->sk_peer_pid); 1862 spin_unlock(&sk->sk_peer_lock); 1863 1864 if (!peer_pid) 1865 return -ENODATA; 1866 1867 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1868 put_pid(peer_pid); 1869 if (pidfd < 0) 1870 return pidfd; 1871 1872 if (copy_to_sockptr(optval, &pidfd, len) || 1873 copy_to_sockptr(optlen, &len, sizeof(int))) { 1874 put_unused_fd(pidfd); 1875 fput(pidfd_file); 1876 1877 return -EFAULT; 1878 } 1879 1880 fd_install(pidfd, pidfd_file); 1881 return 0; 1882 } 1883 1884 case SO_PEERGROUPS: 1885 { 1886 const struct cred *cred; 1887 int ret, n; 1888 1889 cred = sk_get_peer_cred(sk); 1890 if (!cred) 1891 return -ENODATA; 1892 1893 n = cred->group_info->ngroups; 1894 if (len < n * sizeof(gid_t)) { 1895 len = n * sizeof(gid_t); 1896 put_cred(cred); 1897 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1898 } 1899 len = n * sizeof(gid_t); 1900 1901 ret = groups_to_user(optval, cred->group_info); 1902 put_cred(cred); 1903 if (ret) 1904 return ret; 1905 goto lenout; 1906 } 1907 1908 case SO_PEERNAME: 1909 { 1910 struct sockaddr_storage address; 1911 1912 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 1913 if (lv < 0) 1914 return -ENOTCONN; 1915 if (lv < len) 1916 return -EINVAL; 1917 if (copy_to_sockptr(optval, &address, len)) 1918 return -EFAULT; 1919 goto lenout; 1920 } 1921 1922 /* Dubious BSD thing... Probably nobody even uses it, but 1923 * the UNIX standard wants it for whatever reason... -DaveM 1924 */ 1925 case SO_ACCEPTCONN: 1926 v.val = sk->sk_state == TCP_LISTEN; 1927 break; 1928 1929 case SO_PASSSEC: 1930 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1931 break; 1932 1933 case SO_PEERSEC: 1934 return security_socket_getpeersec_stream(sock, 1935 optval, optlen, len); 1936 1937 case SO_MARK: 1938 v.val = READ_ONCE(sk->sk_mark); 1939 break; 1940 1941 case SO_RCVMARK: 1942 v.val = sock_flag(sk, SOCK_RCVMARK); 1943 break; 1944 1945 case SO_RXQ_OVFL: 1946 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1947 break; 1948 1949 case SO_WIFI_STATUS: 1950 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1951 break; 1952 1953 case SO_PEEK_OFF: 1954 if (!READ_ONCE(sock->ops)->set_peek_off) 1955 return -EOPNOTSUPP; 1956 1957 v.val = READ_ONCE(sk->sk_peek_off); 1958 break; 1959 case SO_NOFCS: 1960 v.val = sock_flag(sk, SOCK_NOFCS); 1961 break; 1962 1963 case SO_BINDTODEVICE: 1964 return sock_getbindtodevice(sk, optval, optlen, len); 1965 1966 case SO_GET_FILTER: 1967 len = sk_get_filter(sk, optval, len); 1968 if (len < 0) 1969 return len; 1970 1971 goto lenout; 1972 1973 case SO_LOCK_FILTER: 1974 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1975 break; 1976 1977 case SO_BPF_EXTENSIONS: 1978 v.val = bpf_tell_extensions(); 1979 break; 1980 1981 case SO_SELECT_ERR_QUEUE: 1982 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1983 break; 1984 1985 #ifdef CONFIG_NET_RX_BUSY_POLL 1986 case SO_BUSY_POLL: 1987 v.val = READ_ONCE(sk->sk_ll_usec); 1988 break; 1989 case SO_PREFER_BUSY_POLL: 1990 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1991 break; 1992 #endif 1993 1994 case SO_MAX_PACING_RATE: 1995 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 1996 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1997 lv = sizeof(v.ulval); 1998 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 1999 } else { 2000 /* 32bit version */ 2001 v.val = min_t(unsigned long, ~0U, 2002 READ_ONCE(sk->sk_max_pacing_rate)); 2003 } 2004 break; 2005 2006 case SO_INCOMING_CPU: 2007 v.val = READ_ONCE(sk->sk_incoming_cpu); 2008 break; 2009 2010 case SO_MEMINFO: 2011 { 2012 u32 meminfo[SK_MEMINFO_VARS]; 2013 2014 sk_get_meminfo(sk, meminfo); 2015 2016 len = min_t(unsigned int, len, sizeof(meminfo)); 2017 if (copy_to_sockptr(optval, &meminfo, len)) 2018 return -EFAULT; 2019 2020 goto lenout; 2021 } 2022 2023 #ifdef CONFIG_NET_RX_BUSY_POLL 2024 case SO_INCOMING_NAPI_ID: 2025 v.val = READ_ONCE(sk->sk_napi_id); 2026 2027 /* aggregate non-NAPI IDs down to 0 */ 2028 if (v.val < MIN_NAPI_ID) 2029 v.val = 0; 2030 2031 break; 2032 #endif 2033 2034 case SO_COOKIE: 2035 lv = sizeof(u64); 2036 if (len < lv) 2037 return -EINVAL; 2038 v.val64 = sock_gen_cookie(sk); 2039 break; 2040 2041 case SO_ZEROCOPY: 2042 v.val = sock_flag(sk, SOCK_ZEROCOPY); 2043 break; 2044 2045 case SO_TXTIME: 2046 lv = sizeof(v.txtime); 2047 v.txtime.clockid = sk->sk_clockid; 2048 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 2049 SOF_TXTIME_DEADLINE_MODE : 0; 2050 v.txtime.flags |= sk->sk_txtime_report_errors ? 2051 SOF_TXTIME_REPORT_ERRORS : 0; 2052 break; 2053 2054 case SO_BINDTOIFINDEX: 2055 v.val = READ_ONCE(sk->sk_bound_dev_if); 2056 break; 2057 2058 case SO_NETNS_COOKIE: 2059 lv = sizeof(u64); 2060 if (len != lv) 2061 return -EINVAL; 2062 v.val64 = sock_net(sk)->net_cookie; 2063 break; 2064 2065 case SO_BUF_LOCK: 2066 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 2067 break; 2068 2069 case SO_RESERVE_MEM: 2070 v.val = READ_ONCE(sk->sk_reserved_mem); 2071 break; 2072 2073 case SO_TXREHASH: 2074 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2075 v.val = READ_ONCE(sk->sk_txrehash); 2076 break; 2077 2078 default: 2079 /* We implement the SO_SNDLOWAT etc to not be settable 2080 * (1003.1g 7). 2081 */ 2082 return -ENOPROTOOPT; 2083 } 2084 2085 if (len > lv) 2086 len = lv; 2087 if (copy_to_sockptr(optval, &v, len)) 2088 return -EFAULT; 2089 lenout: 2090 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2091 return -EFAULT; 2092 return 0; 2093 } 2094 2095 /* 2096 * Initialize an sk_lock. 2097 * 2098 * (We also register the sk_lock with the lock validator.) 2099 */ 2100 static inline void sock_lock_init(struct sock *sk) 2101 { 2102 if (sk->sk_kern_sock) 2103 sock_lock_init_class_and_name( 2104 sk, 2105 af_family_kern_slock_key_strings[sk->sk_family], 2106 af_family_kern_slock_keys + sk->sk_family, 2107 af_family_kern_key_strings[sk->sk_family], 2108 af_family_kern_keys + sk->sk_family); 2109 else 2110 sock_lock_init_class_and_name( 2111 sk, 2112 af_family_slock_key_strings[sk->sk_family], 2113 af_family_slock_keys + sk->sk_family, 2114 af_family_key_strings[sk->sk_family], 2115 af_family_keys + sk->sk_family); 2116 } 2117 2118 /* 2119 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2120 * even temporarily, because of RCU lookups. sk_node should also be left as is. 2121 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2122 */ 2123 static void sock_copy(struct sock *nsk, const struct sock *osk) 2124 { 2125 const struct proto *prot = READ_ONCE(osk->sk_prot); 2126 #ifdef CONFIG_SECURITY_NETWORK 2127 void *sptr = nsk->sk_security; 2128 #endif 2129 2130 /* If we move sk_tx_queue_mapping out of the private section, 2131 * we must check if sk_tx_queue_clear() is called after 2132 * sock_copy() in sk_clone_lock(). 2133 */ 2134 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2135 offsetof(struct sock, sk_dontcopy_begin) || 2136 offsetof(struct sock, sk_tx_queue_mapping) >= 2137 offsetof(struct sock, sk_dontcopy_end)); 2138 2139 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2140 2141 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2142 prot->obj_size - offsetof(struct sock, sk_dontcopy_end), 2143 /* alloc is larger than struct, see sk_prot_alloc() */); 2144 2145 #ifdef CONFIG_SECURITY_NETWORK 2146 nsk->sk_security = sptr; 2147 security_sk_clone(osk, nsk); 2148 #endif 2149 } 2150 2151 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2152 int family) 2153 { 2154 struct sock *sk; 2155 struct kmem_cache *slab; 2156 2157 slab = prot->slab; 2158 if (slab != NULL) { 2159 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2160 if (!sk) 2161 return sk; 2162 if (want_init_on_alloc(priority)) 2163 sk_prot_clear_nulls(sk, prot->obj_size); 2164 } else 2165 sk = kmalloc(prot->obj_size, priority); 2166 2167 if (sk != NULL) { 2168 if (security_sk_alloc(sk, family, priority)) 2169 goto out_free; 2170 2171 if (!try_module_get(prot->owner)) 2172 goto out_free_sec; 2173 } 2174 2175 return sk; 2176 2177 out_free_sec: 2178 security_sk_free(sk); 2179 out_free: 2180 if (slab != NULL) 2181 kmem_cache_free(slab, sk); 2182 else 2183 kfree(sk); 2184 return NULL; 2185 } 2186 2187 static void sk_prot_free(struct proto *prot, struct sock *sk) 2188 { 2189 struct kmem_cache *slab; 2190 struct module *owner; 2191 2192 owner = prot->owner; 2193 slab = prot->slab; 2194 2195 cgroup_sk_free(&sk->sk_cgrp_data); 2196 mem_cgroup_sk_free(sk); 2197 security_sk_free(sk); 2198 if (slab != NULL) 2199 kmem_cache_free(slab, sk); 2200 else 2201 kfree(sk); 2202 module_put(owner); 2203 } 2204 2205 /** 2206 * sk_alloc - All socket objects are allocated here 2207 * @net: the applicable net namespace 2208 * @family: protocol family 2209 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2210 * @prot: struct proto associated with this new sock instance 2211 * @kern: is this to be a kernel socket? 2212 */ 2213 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2214 struct proto *prot, int kern) 2215 { 2216 struct sock *sk; 2217 2218 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2219 if (sk) { 2220 sk->sk_family = family; 2221 /* 2222 * See comment in struct sock definition to understand 2223 * why we need sk_prot_creator -acme 2224 */ 2225 sk->sk_prot = sk->sk_prot_creator = prot; 2226 sk->sk_kern_sock = kern; 2227 sock_lock_init(sk); 2228 sk->sk_net_refcnt = kern ? 0 : 1; 2229 if (likely(sk->sk_net_refcnt)) { 2230 get_net_track(net, &sk->ns_tracker, priority); 2231 sock_inuse_add(net, 1); 2232 } else { 2233 __netns_tracker_alloc(net, &sk->ns_tracker, 2234 false, priority); 2235 } 2236 2237 sock_net_set(sk, net); 2238 refcount_set(&sk->sk_wmem_alloc, 1); 2239 2240 mem_cgroup_sk_alloc(sk); 2241 cgroup_sk_alloc(&sk->sk_cgrp_data); 2242 sock_update_classid(&sk->sk_cgrp_data); 2243 sock_update_netprioidx(&sk->sk_cgrp_data); 2244 sk_tx_queue_clear(sk); 2245 } 2246 2247 return sk; 2248 } 2249 EXPORT_SYMBOL(sk_alloc); 2250 2251 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2252 * grace period. This is the case for UDP sockets and TCP listeners. 2253 */ 2254 static void __sk_destruct(struct rcu_head *head) 2255 { 2256 struct sock *sk = container_of(head, struct sock, sk_rcu); 2257 struct sk_filter *filter; 2258 2259 if (sk->sk_destruct) 2260 sk->sk_destruct(sk); 2261 2262 filter = rcu_dereference_check(sk->sk_filter, 2263 refcount_read(&sk->sk_wmem_alloc) == 0); 2264 if (filter) { 2265 sk_filter_uncharge(sk, filter); 2266 RCU_INIT_POINTER(sk->sk_filter, NULL); 2267 } 2268 2269 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2270 2271 #ifdef CONFIG_BPF_SYSCALL 2272 bpf_sk_storage_free(sk); 2273 #endif 2274 2275 if (atomic_read(&sk->sk_omem_alloc)) 2276 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2277 __func__, atomic_read(&sk->sk_omem_alloc)); 2278 2279 if (sk->sk_frag.page) { 2280 put_page(sk->sk_frag.page); 2281 sk->sk_frag.page = NULL; 2282 } 2283 2284 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2285 put_cred(sk->sk_peer_cred); 2286 put_pid(sk->sk_peer_pid); 2287 2288 if (likely(sk->sk_net_refcnt)) 2289 put_net_track(sock_net(sk), &sk->ns_tracker); 2290 else 2291 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2292 2293 sk_prot_free(sk->sk_prot_creator, sk); 2294 } 2295 2296 void sk_destruct(struct sock *sk) 2297 { 2298 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2299 2300 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2301 reuseport_detach_sock(sk); 2302 use_call_rcu = true; 2303 } 2304 2305 if (use_call_rcu) 2306 call_rcu(&sk->sk_rcu, __sk_destruct); 2307 else 2308 __sk_destruct(&sk->sk_rcu); 2309 } 2310 2311 static void __sk_free(struct sock *sk) 2312 { 2313 if (likely(sk->sk_net_refcnt)) 2314 sock_inuse_add(sock_net(sk), -1); 2315 2316 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2317 sock_diag_broadcast_destroy(sk); 2318 else 2319 sk_destruct(sk); 2320 } 2321 2322 void sk_free(struct sock *sk) 2323 { 2324 /* 2325 * We subtract one from sk_wmem_alloc and can know if 2326 * some packets are still in some tx queue. 2327 * If not null, sock_wfree() will call __sk_free(sk) later 2328 */ 2329 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2330 __sk_free(sk); 2331 } 2332 EXPORT_SYMBOL(sk_free); 2333 2334 static void sk_init_common(struct sock *sk) 2335 { 2336 skb_queue_head_init(&sk->sk_receive_queue); 2337 skb_queue_head_init(&sk->sk_write_queue); 2338 skb_queue_head_init(&sk->sk_error_queue); 2339 2340 rwlock_init(&sk->sk_callback_lock); 2341 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2342 af_rlock_keys + sk->sk_family, 2343 af_family_rlock_key_strings[sk->sk_family]); 2344 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2345 af_wlock_keys + sk->sk_family, 2346 af_family_wlock_key_strings[sk->sk_family]); 2347 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2348 af_elock_keys + sk->sk_family, 2349 af_family_elock_key_strings[sk->sk_family]); 2350 if (sk->sk_kern_sock) 2351 lockdep_set_class_and_name(&sk->sk_callback_lock, 2352 af_kern_callback_keys + sk->sk_family, 2353 af_family_kern_clock_key_strings[sk->sk_family]); 2354 else 2355 lockdep_set_class_and_name(&sk->sk_callback_lock, 2356 af_callback_keys + sk->sk_family, 2357 af_family_clock_key_strings[sk->sk_family]); 2358 } 2359 2360 /** 2361 * sk_clone_lock - clone a socket, and lock its clone 2362 * @sk: the socket to clone 2363 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2364 * 2365 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2366 */ 2367 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2368 { 2369 struct proto *prot = READ_ONCE(sk->sk_prot); 2370 struct sk_filter *filter; 2371 bool is_charged = true; 2372 struct sock *newsk; 2373 2374 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2375 if (!newsk) 2376 goto out; 2377 2378 sock_copy(newsk, sk); 2379 2380 newsk->sk_prot_creator = prot; 2381 2382 /* SANITY */ 2383 if (likely(newsk->sk_net_refcnt)) { 2384 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2385 sock_inuse_add(sock_net(newsk), 1); 2386 } else { 2387 /* Kernel sockets are not elevating the struct net refcount. 2388 * Instead, use a tracker to more easily detect if a layer 2389 * is not properly dismantling its kernel sockets at netns 2390 * destroy time. 2391 */ 2392 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2393 false, priority); 2394 } 2395 sk_node_init(&newsk->sk_node); 2396 sock_lock_init(newsk); 2397 bh_lock_sock(newsk); 2398 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2399 newsk->sk_backlog.len = 0; 2400 2401 atomic_set(&newsk->sk_rmem_alloc, 0); 2402 2403 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2404 refcount_set(&newsk->sk_wmem_alloc, 1); 2405 2406 atomic_set(&newsk->sk_omem_alloc, 0); 2407 sk_init_common(newsk); 2408 2409 newsk->sk_dst_cache = NULL; 2410 newsk->sk_dst_pending_confirm = 0; 2411 newsk->sk_wmem_queued = 0; 2412 newsk->sk_forward_alloc = 0; 2413 newsk->sk_reserved_mem = 0; 2414 atomic_set(&newsk->sk_drops, 0); 2415 newsk->sk_send_head = NULL; 2416 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2417 atomic_set(&newsk->sk_zckey, 0); 2418 2419 sock_reset_flag(newsk, SOCK_DONE); 2420 2421 /* sk->sk_memcg will be populated at accept() time */ 2422 newsk->sk_memcg = NULL; 2423 2424 cgroup_sk_clone(&newsk->sk_cgrp_data); 2425 2426 rcu_read_lock(); 2427 filter = rcu_dereference(sk->sk_filter); 2428 if (filter != NULL) 2429 /* though it's an empty new sock, the charging may fail 2430 * if sysctl_optmem_max was changed between creation of 2431 * original socket and cloning 2432 */ 2433 is_charged = sk_filter_charge(newsk, filter); 2434 RCU_INIT_POINTER(newsk->sk_filter, filter); 2435 rcu_read_unlock(); 2436 2437 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2438 /* We need to make sure that we don't uncharge the new 2439 * socket if we couldn't charge it in the first place 2440 * as otherwise we uncharge the parent's filter. 2441 */ 2442 if (!is_charged) 2443 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2444 sk_free_unlock_clone(newsk); 2445 newsk = NULL; 2446 goto out; 2447 } 2448 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2449 2450 if (bpf_sk_storage_clone(sk, newsk)) { 2451 sk_free_unlock_clone(newsk); 2452 newsk = NULL; 2453 goto out; 2454 } 2455 2456 /* Clear sk_user_data if parent had the pointer tagged 2457 * as not suitable for copying when cloning. 2458 */ 2459 if (sk_user_data_is_nocopy(newsk)) 2460 newsk->sk_user_data = NULL; 2461 2462 newsk->sk_err = 0; 2463 newsk->sk_err_soft = 0; 2464 newsk->sk_priority = 0; 2465 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2466 2467 /* Before updating sk_refcnt, we must commit prior changes to memory 2468 * (Documentation/RCU/rculist_nulls.rst for details) 2469 */ 2470 smp_wmb(); 2471 refcount_set(&newsk->sk_refcnt, 2); 2472 2473 sk_set_socket(newsk, NULL); 2474 sk_tx_queue_clear(newsk); 2475 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2476 2477 if (newsk->sk_prot->sockets_allocated) 2478 sk_sockets_allocated_inc(newsk); 2479 2480 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2481 net_enable_timestamp(); 2482 out: 2483 return newsk; 2484 } 2485 EXPORT_SYMBOL_GPL(sk_clone_lock); 2486 2487 void sk_free_unlock_clone(struct sock *sk) 2488 { 2489 /* It is still raw copy of parent, so invalidate 2490 * destructor and make plain sk_free() */ 2491 sk->sk_destruct = NULL; 2492 bh_unlock_sock(sk); 2493 sk_free(sk); 2494 } 2495 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2496 2497 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2498 { 2499 bool is_ipv6 = false; 2500 u32 max_size; 2501 2502 #if IS_ENABLED(CONFIG_IPV6) 2503 is_ipv6 = (sk->sk_family == AF_INET6 && 2504 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2505 #endif 2506 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2507 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2508 READ_ONCE(dst->dev->gso_ipv4_max_size); 2509 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2510 max_size = GSO_LEGACY_MAX_SIZE; 2511 2512 return max_size - (MAX_TCP_HEADER + 1); 2513 } 2514 2515 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2516 { 2517 u32 max_segs = 1; 2518 2519 sk->sk_route_caps = dst->dev->features; 2520 if (sk_is_tcp(sk)) 2521 sk->sk_route_caps |= NETIF_F_GSO; 2522 if (sk->sk_route_caps & NETIF_F_GSO) 2523 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2524 if (unlikely(sk->sk_gso_disabled)) 2525 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2526 if (sk_can_gso(sk)) { 2527 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2528 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2529 } else { 2530 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2531 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2532 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2533 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2534 } 2535 } 2536 sk->sk_gso_max_segs = max_segs; 2537 sk_dst_set(sk, dst); 2538 } 2539 EXPORT_SYMBOL_GPL(sk_setup_caps); 2540 2541 /* 2542 * Simple resource managers for sockets. 2543 */ 2544 2545 2546 /* 2547 * Write buffer destructor automatically called from kfree_skb. 2548 */ 2549 void sock_wfree(struct sk_buff *skb) 2550 { 2551 struct sock *sk = skb->sk; 2552 unsigned int len = skb->truesize; 2553 bool free; 2554 2555 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2556 if (sock_flag(sk, SOCK_RCU_FREE) && 2557 sk->sk_write_space == sock_def_write_space) { 2558 rcu_read_lock(); 2559 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2560 sock_def_write_space_wfree(sk); 2561 rcu_read_unlock(); 2562 if (unlikely(free)) 2563 __sk_free(sk); 2564 return; 2565 } 2566 2567 /* 2568 * Keep a reference on sk_wmem_alloc, this will be released 2569 * after sk_write_space() call 2570 */ 2571 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2572 sk->sk_write_space(sk); 2573 len = 1; 2574 } 2575 /* 2576 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2577 * could not do because of in-flight packets 2578 */ 2579 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2580 __sk_free(sk); 2581 } 2582 EXPORT_SYMBOL(sock_wfree); 2583 2584 /* This variant of sock_wfree() is used by TCP, 2585 * since it sets SOCK_USE_WRITE_QUEUE. 2586 */ 2587 void __sock_wfree(struct sk_buff *skb) 2588 { 2589 struct sock *sk = skb->sk; 2590 2591 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2592 __sk_free(sk); 2593 } 2594 2595 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2596 { 2597 skb_orphan(skb); 2598 #ifdef CONFIG_INET 2599 if (unlikely(!sk_fullsock(sk))) 2600 return skb_set_owner_edemux(skb, sk); 2601 #endif 2602 skb->sk = sk; 2603 skb->destructor = sock_wfree; 2604 skb_set_hash_from_sk(skb, sk); 2605 /* 2606 * We used to take a refcount on sk, but following operation 2607 * is enough to guarantee sk_free() won't free this sock until 2608 * all in-flight packets are completed 2609 */ 2610 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2611 } 2612 EXPORT_SYMBOL(skb_set_owner_w); 2613 2614 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2615 { 2616 /* Drivers depend on in-order delivery for crypto offload, 2617 * partial orphan breaks out-of-order-OK logic. 2618 */ 2619 if (skb_is_decrypted(skb)) 2620 return false; 2621 2622 return (skb->destructor == sock_wfree || 2623 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2624 } 2625 2626 /* This helper is used by netem, as it can hold packets in its 2627 * delay queue. We want to allow the owner socket to send more 2628 * packets, as if they were already TX completed by a typical driver. 2629 * But we also want to keep skb->sk set because some packet schedulers 2630 * rely on it (sch_fq for example). 2631 */ 2632 void skb_orphan_partial(struct sk_buff *skb) 2633 { 2634 if (skb_is_tcp_pure_ack(skb)) 2635 return; 2636 2637 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2638 return; 2639 2640 skb_orphan(skb); 2641 } 2642 EXPORT_SYMBOL(skb_orphan_partial); 2643 2644 /* 2645 * Read buffer destructor automatically called from kfree_skb. 2646 */ 2647 void sock_rfree(struct sk_buff *skb) 2648 { 2649 struct sock *sk = skb->sk; 2650 unsigned int len = skb->truesize; 2651 2652 atomic_sub(len, &sk->sk_rmem_alloc); 2653 sk_mem_uncharge(sk, len); 2654 } 2655 EXPORT_SYMBOL(sock_rfree); 2656 2657 /* 2658 * Buffer destructor for skbs that are not used directly in read or write 2659 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2660 */ 2661 void sock_efree(struct sk_buff *skb) 2662 { 2663 sock_put(skb->sk); 2664 } 2665 EXPORT_SYMBOL(sock_efree); 2666 2667 /* Buffer destructor for prefetch/receive path where reference count may 2668 * not be held, e.g. for listen sockets. 2669 */ 2670 #ifdef CONFIG_INET 2671 void sock_pfree(struct sk_buff *skb) 2672 { 2673 struct sock *sk = skb->sk; 2674 2675 if (!sk_is_refcounted(sk)) 2676 return; 2677 2678 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2679 inet_reqsk(sk)->rsk_listener = NULL; 2680 reqsk_free(inet_reqsk(sk)); 2681 return; 2682 } 2683 2684 sock_gen_put(sk); 2685 } 2686 EXPORT_SYMBOL(sock_pfree); 2687 #endif /* CONFIG_INET */ 2688 2689 kuid_t sock_i_uid(struct sock *sk) 2690 { 2691 kuid_t uid; 2692 2693 read_lock_bh(&sk->sk_callback_lock); 2694 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2695 read_unlock_bh(&sk->sk_callback_lock); 2696 return uid; 2697 } 2698 EXPORT_SYMBOL(sock_i_uid); 2699 2700 unsigned long __sock_i_ino(struct sock *sk) 2701 { 2702 unsigned long ino; 2703 2704 read_lock(&sk->sk_callback_lock); 2705 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2706 read_unlock(&sk->sk_callback_lock); 2707 return ino; 2708 } 2709 EXPORT_SYMBOL(__sock_i_ino); 2710 2711 unsigned long sock_i_ino(struct sock *sk) 2712 { 2713 unsigned long ino; 2714 2715 local_bh_disable(); 2716 ino = __sock_i_ino(sk); 2717 local_bh_enable(); 2718 return ino; 2719 } 2720 EXPORT_SYMBOL(sock_i_ino); 2721 2722 /* 2723 * Allocate a skb from the socket's send buffer. 2724 */ 2725 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2726 gfp_t priority) 2727 { 2728 if (force || 2729 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2730 struct sk_buff *skb = alloc_skb(size, priority); 2731 2732 if (skb) { 2733 skb_set_owner_w(skb, sk); 2734 return skb; 2735 } 2736 } 2737 return NULL; 2738 } 2739 EXPORT_SYMBOL(sock_wmalloc); 2740 2741 static void sock_ofree(struct sk_buff *skb) 2742 { 2743 struct sock *sk = skb->sk; 2744 2745 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2746 } 2747 2748 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2749 gfp_t priority) 2750 { 2751 struct sk_buff *skb; 2752 2753 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2754 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2755 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2756 return NULL; 2757 2758 skb = alloc_skb(size, priority); 2759 if (!skb) 2760 return NULL; 2761 2762 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2763 skb->sk = sk; 2764 skb->destructor = sock_ofree; 2765 return skb; 2766 } 2767 2768 /* 2769 * Allocate a memory block from the socket's option memory buffer. 2770 */ 2771 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2772 { 2773 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2774 2775 if ((unsigned int)size <= optmem_max && 2776 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2777 void *mem; 2778 /* First do the add, to avoid the race if kmalloc 2779 * might sleep. 2780 */ 2781 atomic_add(size, &sk->sk_omem_alloc); 2782 mem = kmalloc(size, priority); 2783 if (mem) 2784 return mem; 2785 atomic_sub(size, &sk->sk_omem_alloc); 2786 } 2787 return NULL; 2788 } 2789 EXPORT_SYMBOL(sock_kmalloc); 2790 2791 /* Free an option memory block. Note, we actually want the inline 2792 * here as this allows gcc to detect the nullify and fold away the 2793 * condition entirely. 2794 */ 2795 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2796 const bool nullify) 2797 { 2798 if (WARN_ON_ONCE(!mem)) 2799 return; 2800 if (nullify) 2801 kfree_sensitive(mem); 2802 else 2803 kfree(mem); 2804 atomic_sub(size, &sk->sk_omem_alloc); 2805 } 2806 2807 void sock_kfree_s(struct sock *sk, void *mem, int size) 2808 { 2809 __sock_kfree_s(sk, mem, size, false); 2810 } 2811 EXPORT_SYMBOL(sock_kfree_s); 2812 2813 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2814 { 2815 __sock_kfree_s(sk, mem, size, true); 2816 } 2817 EXPORT_SYMBOL(sock_kzfree_s); 2818 2819 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2820 I think, these locks should be removed for datagram sockets. 2821 */ 2822 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2823 { 2824 DEFINE_WAIT(wait); 2825 2826 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2827 for (;;) { 2828 if (!timeo) 2829 break; 2830 if (signal_pending(current)) 2831 break; 2832 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2833 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2834 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2835 break; 2836 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2837 break; 2838 if (READ_ONCE(sk->sk_err)) 2839 break; 2840 timeo = schedule_timeout(timeo); 2841 } 2842 finish_wait(sk_sleep(sk), &wait); 2843 return timeo; 2844 } 2845 2846 2847 /* 2848 * Generic send/receive buffer handlers 2849 */ 2850 2851 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2852 unsigned long data_len, int noblock, 2853 int *errcode, int max_page_order) 2854 { 2855 struct sk_buff *skb; 2856 long timeo; 2857 int err; 2858 2859 timeo = sock_sndtimeo(sk, noblock); 2860 for (;;) { 2861 err = sock_error(sk); 2862 if (err != 0) 2863 goto failure; 2864 2865 err = -EPIPE; 2866 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2867 goto failure; 2868 2869 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2870 break; 2871 2872 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2873 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2874 err = -EAGAIN; 2875 if (!timeo) 2876 goto failure; 2877 if (signal_pending(current)) 2878 goto interrupted; 2879 timeo = sock_wait_for_wmem(sk, timeo); 2880 } 2881 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2882 errcode, sk->sk_allocation); 2883 if (skb) 2884 skb_set_owner_w(skb, sk); 2885 return skb; 2886 2887 interrupted: 2888 err = sock_intr_errno(timeo); 2889 failure: 2890 *errcode = err; 2891 return NULL; 2892 } 2893 EXPORT_SYMBOL(sock_alloc_send_pskb); 2894 2895 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2896 struct sockcm_cookie *sockc) 2897 { 2898 u32 tsflags; 2899 2900 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 2901 2902 switch (cmsg->cmsg_type) { 2903 case SO_MARK: 2904 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2905 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2906 return -EPERM; 2907 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2908 return -EINVAL; 2909 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2910 break; 2911 case SO_TIMESTAMPING_OLD: 2912 case SO_TIMESTAMPING_NEW: 2913 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2914 return -EINVAL; 2915 2916 tsflags = *(u32 *)CMSG_DATA(cmsg); 2917 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2918 return -EINVAL; 2919 2920 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2921 sockc->tsflags |= tsflags; 2922 break; 2923 case SCM_TXTIME: 2924 if (!sock_flag(sk, SOCK_TXTIME)) 2925 return -EINVAL; 2926 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2927 return -EINVAL; 2928 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2929 break; 2930 case SCM_TS_OPT_ID: 2931 if (sk_is_tcp(sk)) 2932 return -EINVAL; 2933 tsflags = READ_ONCE(sk->sk_tsflags); 2934 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 2935 return -EINVAL; 2936 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2937 return -EINVAL; 2938 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 2939 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 2940 break; 2941 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2942 case SCM_RIGHTS: 2943 case SCM_CREDENTIALS: 2944 break; 2945 default: 2946 return -EINVAL; 2947 } 2948 return 0; 2949 } 2950 EXPORT_SYMBOL(__sock_cmsg_send); 2951 2952 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2953 struct sockcm_cookie *sockc) 2954 { 2955 struct cmsghdr *cmsg; 2956 int ret; 2957 2958 for_each_cmsghdr(cmsg, msg) { 2959 if (!CMSG_OK(msg, cmsg)) 2960 return -EINVAL; 2961 if (cmsg->cmsg_level != SOL_SOCKET) 2962 continue; 2963 ret = __sock_cmsg_send(sk, cmsg, sockc); 2964 if (ret) 2965 return ret; 2966 } 2967 return 0; 2968 } 2969 EXPORT_SYMBOL(sock_cmsg_send); 2970 2971 static void sk_enter_memory_pressure(struct sock *sk) 2972 { 2973 if (!sk->sk_prot->enter_memory_pressure) 2974 return; 2975 2976 sk->sk_prot->enter_memory_pressure(sk); 2977 } 2978 2979 static void sk_leave_memory_pressure(struct sock *sk) 2980 { 2981 if (sk->sk_prot->leave_memory_pressure) { 2982 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 2983 tcp_leave_memory_pressure, sk); 2984 } else { 2985 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2986 2987 if (memory_pressure && READ_ONCE(*memory_pressure)) 2988 WRITE_ONCE(*memory_pressure, 0); 2989 } 2990 } 2991 2992 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2993 2994 /** 2995 * skb_page_frag_refill - check that a page_frag contains enough room 2996 * @sz: minimum size of the fragment we want to get 2997 * @pfrag: pointer to page_frag 2998 * @gfp: priority for memory allocation 2999 * 3000 * Note: While this allocator tries to use high order pages, there is 3001 * no guarantee that allocations succeed. Therefore, @sz MUST be 3002 * less or equal than PAGE_SIZE. 3003 */ 3004 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 3005 { 3006 if (pfrag->page) { 3007 if (page_ref_count(pfrag->page) == 1) { 3008 pfrag->offset = 0; 3009 return true; 3010 } 3011 if (pfrag->offset + sz <= pfrag->size) 3012 return true; 3013 put_page(pfrag->page); 3014 } 3015 3016 pfrag->offset = 0; 3017 if (SKB_FRAG_PAGE_ORDER && 3018 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3019 /* Avoid direct reclaim but allow kswapd to wake */ 3020 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3021 __GFP_COMP | __GFP_NOWARN | 3022 __GFP_NORETRY, 3023 SKB_FRAG_PAGE_ORDER); 3024 if (likely(pfrag->page)) { 3025 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3026 return true; 3027 } 3028 } 3029 pfrag->page = alloc_page(gfp); 3030 if (likely(pfrag->page)) { 3031 pfrag->size = PAGE_SIZE; 3032 return true; 3033 } 3034 return false; 3035 } 3036 EXPORT_SYMBOL(skb_page_frag_refill); 3037 3038 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3039 { 3040 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3041 return true; 3042 3043 sk_enter_memory_pressure(sk); 3044 sk_stream_moderate_sndbuf(sk); 3045 return false; 3046 } 3047 EXPORT_SYMBOL(sk_page_frag_refill); 3048 3049 void __lock_sock(struct sock *sk) 3050 __releases(&sk->sk_lock.slock) 3051 __acquires(&sk->sk_lock.slock) 3052 { 3053 DEFINE_WAIT(wait); 3054 3055 for (;;) { 3056 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3057 TASK_UNINTERRUPTIBLE); 3058 spin_unlock_bh(&sk->sk_lock.slock); 3059 schedule(); 3060 spin_lock_bh(&sk->sk_lock.slock); 3061 if (!sock_owned_by_user(sk)) 3062 break; 3063 } 3064 finish_wait(&sk->sk_lock.wq, &wait); 3065 } 3066 3067 void __release_sock(struct sock *sk) 3068 __releases(&sk->sk_lock.slock) 3069 __acquires(&sk->sk_lock.slock) 3070 { 3071 struct sk_buff *skb, *next; 3072 3073 while ((skb = sk->sk_backlog.head) != NULL) { 3074 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3075 3076 spin_unlock_bh(&sk->sk_lock.slock); 3077 3078 do { 3079 next = skb->next; 3080 prefetch(next); 3081 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3082 skb_mark_not_on_list(skb); 3083 sk_backlog_rcv(sk, skb); 3084 3085 cond_resched(); 3086 3087 skb = next; 3088 } while (skb != NULL); 3089 3090 spin_lock_bh(&sk->sk_lock.slock); 3091 } 3092 3093 /* 3094 * Doing the zeroing here guarantee we can not loop forever 3095 * while a wild producer attempts to flood us. 3096 */ 3097 sk->sk_backlog.len = 0; 3098 } 3099 3100 void __sk_flush_backlog(struct sock *sk) 3101 { 3102 spin_lock_bh(&sk->sk_lock.slock); 3103 __release_sock(sk); 3104 3105 if (sk->sk_prot->release_cb) 3106 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3107 tcp_release_cb, sk); 3108 3109 spin_unlock_bh(&sk->sk_lock.slock); 3110 } 3111 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3112 3113 /** 3114 * sk_wait_data - wait for data to arrive at sk_receive_queue 3115 * @sk: sock to wait on 3116 * @timeo: for how long 3117 * @skb: last skb seen on sk_receive_queue 3118 * 3119 * Now socket state including sk->sk_err is changed only under lock, 3120 * hence we may omit checks after joining wait queue. 3121 * We check receive queue before schedule() only as optimization; 3122 * it is very likely that release_sock() added new data. 3123 */ 3124 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3125 { 3126 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3127 int rc; 3128 3129 add_wait_queue(sk_sleep(sk), &wait); 3130 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3131 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3132 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3133 remove_wait_queue(sk_sleep(sk), &wait); 3134 return rc; 3135 } 3136 EXPORT_SYMBOL(sk_wait_data); 3137 3138 /** 3139 * __sk_mem_raise_allocated - increase memory_allocated 3140 * @sk: socket 3141 * @size: memory size to allocate 3142 * @amt: pages to allocate 3143 * @kind: allocation type 3144 * 3145 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3146 * 3147 * Unlike the globally shared limits among the sockets under same protocol, 3148 * consuming the budget of a memcg won't have direct effect on other ones. 3149 * So be optimistic about memcg's tolerance, and leave the callers to decide 3150 * whether or not to raise allocated through sk_under_memory_pressure() or 3151 * its variants. 3152 */ 3153 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3154 { 3155 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL; 3156 struct proto *prot = sk->sk_prot; 3157 bool charged = false; 3158 long allocated; 3159 3160 sk_memory_allocated_add(sk, amt); 3161 allocated = sk_memory_allocated(sk); 3162 3163 if (memcg) { 3164 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge())) 3165 goto suppress_allocation; 3166 charged = true; 3167 } 3168 3169 /* Under limit. */ 3170 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3171 sk_leave_memory_pressure(sk); 3172 return 1; 3173 } 3174 3175 /* Under pressure. */ 3176 if (allocated > sk_prot_mem_limits(sk, 1)) 3177 sk_enter_memory_pressure(sk); 3178 3179 /* Over hard limit. */ 3180 if (allocated > sk_prot_mem_limits(sk, 2)) 3181 goto suppress_allocation; 3182 3183 /* Guarantee minimum buffer size under pressure (either global 3184 * or memcg) to make sure features described in RFC 7323 (TCP 3185 * Extensions for High Performance) work properly. 3186 * 3187 * This rule does NOT stand when exceeds global or memcg's hard 3188 * limit, or else a DoS attack can be taken place by spawning 3189 * lots of sockets whose usage are under minimum buffer size. 3190 */ 3191 if (kind == SK_MEM_RECV) { 3192 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3193 return 1; 3194 3195 } else { /* SK_MEM_SEND */ 3196 int wmem0 = sk_get_wmem0(sk, prot); 3197 3198 if (sk->sk_type == SOCK_STREAM) { 3199 if (sk->sk_wmem_queued < wmem0) 3200 return 1; 3201 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3202 return 1; 3203 } 3204 } 3205 3206 if (sk_has_memory_pressure(sk)) { 3207 u64 alloc; 3208 3209 /* The following 'average' heuristic is within the 3210 * scope of global accounting, so it only makes 3211 * sense for global memory pressure. 3212 */ 3213 if (!sk_under_global_memory_pressure(sk)) 3214 return 1; 3215 3216 /* Try to be fair among all the sockets under global 3217 * pressure by allowing the ones that below average 3218 * usage to raise. 3219 */ 3220 alloc = sk_sockets_allocated_read_positive(sk); 3221 if (sk_prot_mem_limits(sk, 2) > alloc * 3222 sk_mem_pages(sk->sk_wmem_queued + 3223 atomic_read(&sk->sk_rmem_alloc) + 3224 sk->sk_forward_alloc)) 3225 return 1; 3226 } 3227 3228 suppress_allocation: 3229 3230 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3231 sk_stream_moderate_sndbuf(sk); 3232 3233 /* Fail only if socket is _under_ its sndbuf. 3234 * In this case we cannot block, so that we have to fail. 3235 */ 3236 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3237 /* Force charge with __GFP_NOFAIL */ 3238 if (memcg && !charged) { 3239 mem_cgroup_charge_skmem(memcg, amt, 3240 gfp_memcg_charge() | __GFP_NOFAIL); 3241 } 3242 return 1; 3243 } 3244 } 3245 3246 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3247 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3248 3249 sk_memory_allocated_sub(sk, amt); 3250 3251 if (charged) 3252 mem_cgroup_uncharge_skmem(memcg, amt); 3253 3254 return 0; 3255 } 3256 3257 /** 3258 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3259 * @sk: socket 3260 * @size: memory size to allocate 3261 * @kind: allocation type 3262 * 3263 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3264 * rmem allocation. This function assumes that protocols which have 3265 * memory_pressure use sk_wmem_queued as write buffer accounting. 3266 */ 3267 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3268 { 3269 int ret, amt = sk_mem_pages(size); 3270 3271 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3272 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3273 if (!ret) 3274 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3275 return ret; 3276 } 3277 EXPORT_SYMBOL(__sk_mem_schedule); 3278 3279 /** 3280 * __sk_mem_reduce_allocated - reclaim memory_allocated 3281 * @sk: socket 3282 * @amount: number of quanta 3283 * 3284 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3285 */ 3286 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3287 { 3288 sk_memory_allocated_sub(sk, amount); 3289 3290 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3291 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3292 3293 if (sk_under_global_memory_pressure(sk) && 3294 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3295 sk_leave_memory_pressure(sk); 3296 } 3297 3298 /** 3299 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3300 * @sk: socket 3301 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3302 */ 3303 void __sk_mem_reclaim(struct sock *sk, int amount) 3304 { 3305 amount >>= PAGE_SHIFT; 3306 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3307 __sk_mem_reduce_allocated(sk, amount); 3308 } 3309 EXPORT_SYMBOL(__sk_mem_reclaim); 3310 3311 int sk_set_peek_off(struct sock *sk, int val) 3312 { 3313 WRITE_ONCE(sk->sk_peek_off, val); 3314 return 0; 3315 } 3316 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3317 3318 /* 3319 * Set of default routines for initialising struct proto_ops when 3320 * the protocol does not support a particular function. In certain 3321 * cases where it makes no sense for a protocol to have a "do nothing" 3322 * function, some default processing is provided. 3323 */ 3324 3325 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3326 { 3327 return -EOPNOTSUPP; 3328 } 3329 EXPORT_SYMBOL(sock_no_bind); 3330 3331 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3332 int len, int flags) 3333 { 3334 return -EOPNOTSUPP; 3335 } 3336 EXPORT_SYMBOL(sock_no_connect); 3337 3338 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3339 { 3340 return -EOPNOTSUPP; 3341 } 3342 EXPORT_SYMBOL(sock_no_socketpair); 3343 3344 int sock_no_accept(struct socket *sock, struct socket *newsock, 3345 struct proto_accept_arg *arg) 3346 { 3347 return -EOPNOTSUPP; 3348 } 3349 EXPORT_SYMBOL(sock_no_accept); 3350 3351 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3352 int peer) 3353 { 3354 return -EOPNOTSUPP; 3355 } 3356 EXPORT_SYMBOL(sock_no_getname); 3357 3358 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3359 { 3360 return -EOPNOTSUPP; 3361 } 3362 EXPORT_SYMBOL(sock_no_ioctl); 3363 3364 int sock_no_listen(struct socket *sock, int backlog) 3365 { 3366 return -EOPNOTSUPP; 3367 } 3368 EXPORT_SYMBOL(sock_no_listen); 3369 3370 int sock_no_shutdown(struct socket *sock, int how) 3371 { 3372 return -EOPNOTSUPP; 3373 } 3374 EXPORT_SYMBOL(sock_no_shutdown); 3375 3376 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3377 { 3378 return -EOPNOTSUPP; 3379 } 3380 EXPORT_SYMBOL(sock_no_sendmsg); 3381 3382 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3383 { 3384 return -EOPNOTSUPP; 3385 } 3386 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3387 3388 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3389 int flags) 3390 { 3391 return -EOPNOTSUPP; 3392 } 3393 EXPORT_SYMBOL(sock_no_recvmsg); 3394 3395 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3396 { 3397 /* Mirror missing mmap method error code */ 3398 return -ENODEV; 3399 } 3400 EXPORT_SYMBOL(sock_no_mmap); 3401 3402 /* 3403 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3404 * various sock-based usage counts. 3405 */ 3406 void __receive_sock(struct file *file) 3407 { 3408 struct socket *sock; 3409 3410 sock = sock_from_file(file); 3411 if (sock) { 3412 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3413 sock_update_classid(&sock->sk->sk_cgrp_data); 3414 } 3415 } 3416 3417 /* 3418 * Default Socket Callbacks 3419 */ 3420 3421 static void sock_def_wakeup(struct sock *sk) 3422 { 3423 struct socket_wq *wq; 3424 3425 rcu_read_lock(); 3426 wq = rcu_dereference(sk->sk_wq); 3427 if (skwq_has_sleeper(wq)) 3428 wake_up_interruptible_all(&wq->wait); 3429 rcu_read_unlock(); 3430 } 3431 3432 static void sock_def_error_report(struct sock *sk) 3433 { 3434 struct socket_wq *wq; 3435 3436 rcu_read_lock(); 3437 wq = rcu_dereference(sk->sk_wq); 3438 if (skwq_has_sleeper(wq)) 3439 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3440 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3441 rcu_read_unlock(); 3442 } 3443 3444 void sock_def_readable(struct sock *sk) 3445 { 3446 struct socket_wq *wq; 3447 3448 trace_sk_data_ready(sk); 3449 3450 rcu_read_lock(); 3451 wq = rcu_dereference(sk->sk_wq); 3452 if (skwq_has_sleeper(wq)) 3453 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3454 EPOLLRDNORM | EPOLLRDBAND); 3455 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3456 rcu_read_unlock(); 3457 } 3458 3459 static void sock_def_write_space(struct sock *sk) 3460 { 3461 struct socket_wq *wq; 3462 3463 rcu_read_lock(); 3464 3465 /* Do not wake up a writer until he can make "significant" 3466 * progress. --DaveM 3467 */ 3468 if (sock_writeable(sk)) { 3469 wq = rcu_dereference(sk->sk_wq); 3470 if (skwq_has_sleeper(wq)) 3471 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3472 EPOLLWRNORM | EPOLLWRBAND); 3473 3474 /* Should agree with poll, otherwise some programs break */ 3475 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3476 } 3477 3478 rcu_read_unlock(); 3479 } 3480 3481 /* An optimised version of sock_def_write_space(), should only be called 3482 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3483 * ->sk_wmem_alloc. 3484 */ 3485 static void sock_def_write_space_wfree(struct sock *sk) 3486 { 3487 /* Do not wake up a writer until he can make "significant" 3488 * progress. --DaveM 3489 */ 3490 if (sock_writeable(sk)) { 3491 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3492 3493 /* rely on refcount_sub from sock_wfree() */ 3494 smp_mb__after_atomic(); 3495 if (wq && waitqueue_active(&wq->wait)) 3496 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3497 EPOLLWRNORM | EPOLLWRBAND); 3498 3499 /* Should agree with poll, otherwise some programs break */ 3500 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3501 } 3502 } 3503 3504 static void sock_def_destruct(struct sock *sk) 3505 { 3506 } 3507 3508 void sk_send_sigurg(struct sock *sk) 3509 { 3510 if (sk->sk_socket && sk->sk_socket->file) 3511 if (send_sigurg(sk->sk_socket->file)) 3512 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3513 } 3514 EXPORT_SYMBOL(sk_send_sigurg); 3515 3516 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3517 unsigned long expires) 3518 { 3519 if (!mod_timer(timer, expires)) 3520 sock_hold(sk); 3521 } 3522 EXPORT_SYMBOL(sk_reset_timer); 3523 3524 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3525 { 3526 if (del_timer(timer)) 3527 __sock_put(sk); 3528 } 3529 EXPORT_SYMBOL(sk_stop_timer); 3530 3531 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3532 { 3533 if (del_timer_sync(timer)) 3534 __sock_put(sk); 3535 } 3536 EXPORT_SYMBOL(sk_stop_timer_sync); 3537 3538 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3539 { 3540 sk_init_common(sk); 3541 sk->sk_send_head = NULL; 3542 3543 timer_setup(&sk->sk_timer, NULL, 0); 3544 3545 sk->sk_allocation = GFP_KERNEL; 3546 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3547 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3548 sk->sk_state = TCP_CLOSE; 3549 sk->sk_use_task_frag = true; 3550 sk_set_socket(sk, sock); 3551 3552 sock_set_flag(sk, SOCK_ZAPPED); 3553 3554 if (sock) { 3555 sk->sk_type = sock->type; 3556 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3557 sock->sk = sk; 3558 } else { 3559 RCU_INIT_POINTER(sk->sk_wq, NULL); 3560 } 3561 sk->sk_uid = uid; 3562 3563 sk->sk_state_change = sock_def_wakeup; 3564 sk->sk_data_ready = sock_def_readable; 3565 sk->sk_write_space = sock_def_write_space; 3566 sk->sk_error_report = sock_def_error_report; 3567 sk->sk_destruct = sock_def_destruct; 3568 3569 sk->sk_frag.page = NULL; 3570 sk->sk_frag.offset = 0; 3571 sk->sk_peek_off = -1; 3572 3573 sk->sk_peer_pid = NULL; 3574 sk->sk_peer_cred = NULL; 3575 spin_lock_init(&sk->sk_peer_lock); 3576 3577 sk->sk_write_pending = 0; 3578 sk->sk_rcvlowat = 1; 3579 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3580 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3581 3582 sk->sk_stamp = SK_DEFAULT_STAMP; 3583 #if BITS_PER_LONG==32 3584 seqlock_init(&sk->sk_stamp_seq); 3585 #endif 3586 atomic_set(&sk->sk_zckey, 0); 3587 3588 #ifdef CONFIG_NET_RX_BUSY_POLL 3589 sk->sk_napi_id = 0; 3590 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3591 #endif 3592 3593 sk->sk_max_pacing_rate = ~0UL; 3594 sk->sk_pacing_rate = ~0UL; 3595 WRITE_ONCE(sk->sk_pacing_shift, 10); 3596 sk->sk_incoming_cpu = -1; 3597 3598 sk_rx_queue_clear(sk); 3599 /* 3600 * Before updating sk_refcnt, we must commit prior changes to memory 3601 * (Documentation/RCU/rculist_nulls.rst for details) 3602 */ 3603 smp_wmb(); 3604 refcount_set(&sk->sk_refcnt, 1); 3605 atomic_set(&sk->sk_drops, 0); 3606 } 3607 EXPORT_SYMBOL(sock_init_data_uid); 3608 3609 void sock_init_data(struct socket *sock, struct sock *sk) 3610 { 3611 kuid_t uid = sock ? 3612 SOCK_INODE(sock)->i_uid : 3613 make_kuid(sock_net(sk)->user_ns, 0); 3614 3615 sock_init_data_uid(sock, sk, uid); 3616 } 3617 EXPORT_SYMBOL(sock_init_data); 3618 3619 void lock_sock_nested(struct sock *sk, int subclass) 3620 { 3621 /* The sk_lock has mutex_lock() semantics here. */ 3622 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3623 3624 might_sleep(); 3625 spin_lock_bh(&sk->sk_lock.slock); 3626 if (sock_owned_by_user_nocheck(sk)) 3627 __lock_sock(sk); 3628 sk->sk_lock.owned = 1; 3629 spin_unlock_bh(&sk->sk_lock.slock); 3630 } 3631 EXPORT_SYMBOL(lock_sock_nested); 3632 3633 void release_sock(struct sock *sk) 3634 { 3635 spin_lock_bh(&sk->sk_lock.slock); 3636 if (sk->sk_backlog.tail) 3637 __release_sock(sk); 3638 3639 if (sk->sk_prot->release_cb) 3640 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3641 tcp_release_cb, sk); 3642 3643 sock_release_ownership(sk); 3644 if (waitqueue_active(&sk->sk_lock.wq)) 3645 wake_up(&sk->sk_lock.wq); 3646 spin_unlock_bh(&sk->sk_lock.slock); 3647 } 3648 EXPORT_SYMBOL(release_sock); 3649 3650 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3651 { 3652 might_sleep(); 3653 spin_lock_bh(&sk->sk_lock.slock); 3654 3655 if (!sock_owned_by_user_nocheck(sk)) { 3656 /* 3657 * Fast path return with bottom halves disabled and 3658 * sock::sk_lock.slock held. 3659 * 3660 * The 'mutex' is not contended and holding 3661 * sock::sk_lock.slock prevents all other lockers to 3662 * proceed so the corresponding unlock_sock_fast() can 3663 * avoid the slow path of release_sock() completely and 3664 * just release slock. 3665 * 3666 * From a semantical POV this is equivalent to 'acquiring' 3667 * the 'mutex', hence the corresponding lockdep 3668 * mutex_release() has to happen in the fast path of 3669 * unlock_sock_fast(). 3670 */ 3671 return false; 3672 } 3673 3674 __lock_sock(sk); 3675 sk->sk_lock.owned = 1; 3676 __acquire(&sk->sk_lock.slock); 3677 spin_unlock_bh(&sk->sk_lock.slock); 3678 return true; 3679 } 3680 EXPORT_SYMBOL(__lock_sock_fast); 3681 3682 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3683 bool timeval, bool time32) 3684 { 3685 struct sock *sk = sock->sk; 3686 struct timespec64 ts; 3687 3688 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3689 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3690 if (ts.tv_sec == -1) 3691 return -ENOENT; 3692 if (ts.tv_sec == 0) { 3693 ktime_t kt = ktime_get_real(); 3694 sock_write_timestamp(sk, kt); 3695 ts = ktime_to_timespec64(kt); 3696 } 3697 3698 if (timeval) 3699 ts.tv_nsec /= 1000; 3700 3701 #ifdef CONFIG_COMPAT_32BIT_TIME 3702 if (time32) 3703 return put_old_timespec32(&ts, userstamp); 3704 #endif 3705 #ifdef CONFIG_SPARC64 3706 /* beware of padding in sparc64 timeval */ 3707 if (timeval && !in_compat_syscall()) { 3708 struct __kernel_old_timeval __user tv = { 3709 .tv_sec = ts.tv_sec, 3710 .tv_usec = ts.tv_nsec, 3711 }; 3712 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3713 return -EFAULT; 3714 return 0; 3715 } 3716 #endif 3717 return put_timespec64(&ts, userstamp); 3718 } 3719 EXPORT_SYMBOL(sock_gettstamp); 3720 3721 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3722 { 3723 if (!sock_flag(sk, flag)) { 3724 unsigned long previous_flags = sk->sk_flags; 3725 3726 sock_set_flag(sk, flag); 3727 /* 3728 * we just set one of the two flags which require net 3729 * time stamping, but time stamping might have been on 3730 * already because of the other one 3731 */ 3732 if (sock_needs_netstamp(sk) && 3733 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3734 net_enable_timestamp(); 3735 } 3736 } 3737 3738 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3739 int level, int type) 3740 { 3741 struct sock_exterr_skb *serr; 3742 struct sk_buff *skb; 3743 int copied, err; 3744 3745 err = -EAGAIN; 3746 skb = sock_dequeue_err_skb(sk); 3747 if (skb == NULL) 3748 goto out; 3749 3750 copied = skb->len; 3751 if (copied > len) { 3752 msg->msg_flags |= MSG_TRUNC; 3753 copied = len; 3754 } 3755 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3756 if (err) 3757 goto out_free_skb; 3758 3759 sock_recv_timestamp(msg, sk, skb); 3760 3761 serr = SKB_EXT_ERR(skb); 3762 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3763 3764 msg->msg_flags |= MSG_ERRQUEUE; 3765 err = copied; 3766 3767 out_free_skb: 3768 kfree_skb(skb); 3769 out: 3770 return err; 3771 } 3772 EXPORT_SYMBOL(sock_recv_errqueue); 3773 3774 /* 3775 * Get a socket option on an socket. 3776 * 3777 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3778 * asynchronous errors should be reported by getsockopt. We assume 3779 * this means if you specify SO_ERROR (otherwise what is the point of it). 3780 */ 3781 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3782 char __user *optval, int __user *optlen) 3783 { 3784 struct sock *sk = sock->sk; 3785 3786 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3787 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3788 } 3789 EXPORT_SYMBOL(sock_common_getsockopt); 3790 3791 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3792 int flags) 3793 { 3794 struct sock *sk = sock->sk; 3795 int addr_len = 0; 3796 int err; 3797 3798 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3799 if (err >= 0) 3800 msg->msg_namelen = addr_len; 3801 return err; 3802 } 3803 EXPORT_SYMBOL(sock_common_recvmsg); 3804 3805 /* 3806 * Set socket options on an inet socket. 3807 */ 3808 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3809 sockptr_t optval, unsigned int optlen) 3810 { 3811 struct sock *sk = sock->sk; 3812 3813 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3814 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3815 } 3816 EXPORT_SYMBOL(sock_common_setsockopt); 3817 3818 void sk_common_release(struct sock *sk) 3819 { 3820 if (sk->sk_prot->destroy) 3821 sk->sk_prot->destroy(sk); 3822 3823 /* 3824 * Observation: when sk_common_release is called, processes have 3825 * no access to socket. But net still has. 3826 * Step one, detach it from networking: 3827 * 3828 * A. Remove from hash tables. 3829 */ 3830 3831 sk->sk_prot->unhash(sk); 3832 3833 /* 3834 * In this point socket cannot receive new packets, but it is possible 3835 * that some packets are in flight because some CPU runs receiver and 3836 * did hash table lookup before we unhashed socket. They will achieve 3837 * receive queue and will be purged by socket destructor. 3838 * 3839 * Also we still have packets pending on receive queue and probably, 3840 * our own packets waiting in device queues. sock_destroy will drain 3841 * receive queue, but transmitted packets will delay socket destruction 3842 * until the last reference will be released. 3843 */ 3844 3845 sock_orphan(sk); 3846 3847 xfrm_sk_free_policy(sk); 3848 3849 sock_put(sk); 3850 } 3851 EXPORT_SYMBOL(sk_common_release); 3852 3853 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3854 { 3855 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3856 3857 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3858 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3859 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3860 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3861 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk); 3862 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3863 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3864 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3865 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3866 } 3867 3868 #ifdef CONFIG_PROC_FS 3869 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3870 3871 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3872 { 3873 int cpu, idx = prot->inuse_idx; 3874 int res = 0; 3875 3876 for_each_possible_cpu(cpu) 3877 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3878 3879 return res >= 0 ? res : 0; 3880 } 3881 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3882 3883 int sock_inuse_get(struct net *net) 3884 { 3885 int cpu, res = 0; 3886 3887 for_each_possible_cpu(cpu) 3888 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3889 3890 return res; 3891 } 3892 3893 EXPORT_SYMBOL_GPL(sock_inuse_get); 3894 3895 static int __net_init sock_inuse_init_net(struct net *net) 3896 { 3897 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3898 if (net->core.prot_inuse == NULL) 3899 return -ENOMEM; 3900 return 0; 3901 } 3902 3903 static void __net_exit sock_inuse_exit_net(struct net *net) 3904 { 3905 free_percpu(net->core.prot_inuse); 3906 } 3907 3908 static struct pernet_operations net_inuse_ops = { 3909 .init = sock_inuse_init_net, 3910 .exit = sock_inuse_exit_net, 3911 }; 3912 3913 static __init int net_inuse_init(void) 3914 { 3915 if (register_pernet_subsys(&net_inuse_ops)) 3916 panic("Cannot initialize net inuse counters"); 3917 3918 return 0; 3919 } 3920 3921 core_initcall(net_inuse_init); 3922 3923 static int assign_proto_idx(struct proto *prot) 3924 { 3925 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3926 3927 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3928 pr_err("PROTO_INUSE_NR exhausted\n"); 3929 return -ENOSPC; 3930 } 3931 3932 set_bit(prot->inuse_idx, proto_inuse_idx); 3933 return 0; 3934 } 3935 3936 static void release_proto_idx(struct proto *prot) 3937 { 3938 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3939 clear_bit(prot->inuse_idx, proto_inuse_idx); 3940 } 3941 #else 3942 static inline int assign_proto_idx(struct proto *prot) 3943 { 3944 return 0; 3945 } 3946 3947 static inline void release_proto_idx(struct proto *prot) 3948 { 3949 } 3950 3951 #endif 3952 3953 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3954 { 3955 if (!twsk_prot) 3956 return; 3957 kfree(twsk_prot->twsk_slab_name); 3958 twsk_prot->twsk_slab_name = NULL; 3959 kmem_cache_destroy(twsk_prot->twsk_slab); 3960 twsk_prot->twsk_slab = NULL; 3961 } 3962 3963 static int tw_prot_init(const struct proto *prot) 3964 { 3965 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3966 3967 if (!twsk_prot) 3968 return 0; 3969 3970 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3971 prot->name); 3972 if (!twsk_prot->twsk_slab_name) 3973 return -ENOMEM; 3974 3975 twsk_prot->twsk_slab = 3976 kmem_cache_create(twsk_prot->twsk_slab_name, 3977 twsk_prot->twsk_obj_size, 0, 3978 SLAB_ACCOUNT | prot->slab_flags, 3979 NULL); 3980 if (!twsk_prot->twsk_slab) { 3981 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3982 prot->name); 3983 return -ENOMEM; 3984 } 3985 3986 return 0; 3987 } 3988 3989 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3990 { 3991 if (!rsk_prot) 3992 return; 3993 kfree(rsk_prot->slab_name); 3994 rsk_prot->slab_name = NULL; 3995 kmem_cache_destroy(rsk_prot->slab); 3996 rsk_prot->slab = NULL; 3997 } 3998 3999 static int req_prot_init(const struct proto *prot) 4000 { 4001 struct request_sock_ops *rsk_prot = prot->rsk_prot; 4002 4003 if (!rsk_prot) 4004 return 0; 4005 4006 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4007 prot->name); 4008 if (!rsk_prot->slab_name) 4009 return -ENOMEM; 4010 4011 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4012 rsk_prot->obj_size, 0, 4013 SLAB_ACCOUNT | prot->slab_flags, 4014 NULL); 4015 4016 if (!rsk_prot->slab) { 4017 pr_crit("%s: Can't create request sock SLAB cache!\n", 4018 prot->name); 4019 return -ENOMEM; 4020 } 4021 return 0; 4022 } 4023 4024 int proto_register(struct proto *prot, int alloc_slab) 4025 { 4026 int ret = -ENOBUFS; 4027 4028 if (prot->memory_allocated && !prot->sysctl_mem) { 4029 pr_err("%s: missing sysctl_mem\n", prot->name); 4030 return -EINVAL; 4031 } 4032 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4033 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4034 return -EINVAL; 4035 } 4036 if (alloc_slab) { 4037 prot->slab = kmem_cache_create_usercopy(prot->name, 4038 prot->obj_size, 0, 4039 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4040 prot->slab_flags, 4041 prot->useroffset, prot->usersize, 4042 NULL); 4043 4044 if (prot->slab == NULL) { 4045 pr_crit("%s: Can't create sock SLAB cache!\n", 4046 prot->name); 4047 goto out; 4048 } 4049 4050 if (req_prot_init(prot)) 4051 goto out_free_request_sock_slab; 4052 4053 if (tw_prot_init(prot)) 4054 goto out_free_timewait_sock_slab; 4055 } 4056 4057 mutex_lock(&proto_list_mutex); 4058 ret = assign_proto_idx(prot); 4059 if (ret) { 4060 mutex_unlock(&proto_list_mutex); 4061 goto out_free_timewait_sock_slab; 4062 } 4063 list_add(&prot->node, &proto_list); 4064 mutex_unlock(&proto_list_mutex); 4065 return ret; 4066 4067 out_free_timewait_sock_slab: 4068 if (alloc_slab) 4069 tw_prot_cleanup(prot->twsk_prot); 4070 out_free_request_sock_slab: 4071 if (alloc_slab) { 4072 req_prot_cleanup(prot->rsk_prot); 4073 4074 kmem_cache_destroy(prot->slab); 4075 prot->slab = NULL; 4076 } 4077 out: 4078 return ret; 4079 } 4080 EXPORT_SYMBOL(proto_register); 4081 4082 void proto_unregister(struct proto *prot) 4083 { 4084 mutex_lock(&proto_list_mutex); 4085 release_proto_idx(prot); 4086 list_del(&prot->node); 4087 mutex_unlock(&proto_list_mutex); 4088 4089 kmem_cache_destroy(prot->slab); 4090 prot->slab = NULL; 4091 4092 req_prot_cleanup(prot->rsk_prot); 4093 tw_prot_cleanup(prot->twsk_prot); 4094 } 4095 EXPORT_SYMBOL(proto_unregister); 4096 4097 int sock_load_diag_module(int family, int protocol) 4098 { 4099 if (!protocol) { 4100 if (!sock_is_registered(family)) 4101 return -ENOENT; 4102 4103 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4104 NETLINK_SOCK_DIAG, family); 4105 } 4106 4107 #ifdef CONFIG_INET 4108 if (family == AF_INET && 4109 protocol != IPPROTO_RAW && 4110 protocol < MAX_INET_PROTOS && 4111 !rcu_access_pointer(inet_protos[protocol])) 4112 return -ENOENT; 4113 #endif 4114 4115 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4116 NETLINK_SOCK_DIAG, family, protocol); 4117 } 4118 EXPORT_SYMBOL(sock_load_diag_module); 4119 4120 #ifdef CONFIG_PROC_FS 4121 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4122 __acquires(proto_list_mutex) 4123 { 4124 mutex_lock(&proto_list_mutex); 4125 return seq_list_start_head(&proto_list, *pos); 4126 } 4127 4128 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4129 { 4130 return seq_list_next(v, &proto_list, pos); 4131 } 4132 4133 static void proto_seq_stop(struct seq_file *seq, void *v) 4134 __releases(proto_list_mutex) 4135 { 4136 mutex_unlock(&proto_list_mutex); 4137 } 4138 4139 static char proto_method_implemented(const void *method) 4140 { 4141 return method == NULL ? 'n' : 'y'; 4142 } 4143 static long sock_prot_memory_allocated(struct proto *proto) 4144 { 4145 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4146 } 4147 4148 static const char *sock_prot_memory_pressure(struct proto *proto) 4149 { 4150 return proto->memory_pressure != NULL ? 4151 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4152 } 4153 4154 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4155 { 4156 4157 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4158 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4159 proto->name, 4160 proto->obj_size, 4161 sock_prot_inuse_get(seq_file_net(seq), proto), 4162 sock_prot_memory_allocated(proto), 4163 sock_prot_memory_pressure(proto), 4164 proto->max_header, 4165 proto->slab == NULL ? "no" : "yes", 4166 module_name(proto->owner), 4167 proto_method_implemented(proto->close), 4168 proto_method_implemented(proto->connect), 4169 proto_method_implemented(proto->disconnect), 4170 proto_method_implemented(proto->accept), 4171 proto_method_implemented(proto->ioctl), 4172 proto_method_implemented(proto->init), 4173 proto_method_implemented(proto->destroy), 4174 proto_method_implemented(proto->shutdown), 4175 proto_method_implemented(proto->setsockopt), 4176 proto_method_implemented(proto->getsockopt), 4177 proto_method_implemented(proto->sendmsg), 4178 proto_method_implemented(proto->recvmsg), 4179 proto_method_implemented(proto->bind), 4180 proto_method_implemented(proto->backlog_rcv), 4181 proto_method_implemented(proto->hash), 4182 proto_method_implemented(proto->unhash), 4183 proto_method_implemented(proto->get_port), 4184 proto_method_implemented(proto->enter_memory_pressure)); 4185 } 4186 4187 static int proto_seq_show(struct seq_file *seq, void *v) 4188 { 4189 if (v == &proto_list) 4190 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4191 "protocol", 4192 "size", 4193 "sockets", 4194 "memory", 4195 "press", 4196 "maxhdr", 4197 "slab", 4198 "module", 4199 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4200 else 4201 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4202 return 0; 4203 } 4204 4205 static const struct seq_operations proto_seq_ops = { 4206 .start = proto_seq_start, 4207 .next = proto_seq_next, 4208 .stop = proto_seq_stop, 4209 .show = proto_seq_show, 4210 }; 4211 4212 static __net_init int proto_init_net(struct net *net) 4213 { 4214 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4215 sizeof(struct seq_net_private))) 4216 return -ENOMEM; 4217 4218 return 0; 4219 } 4220 4221 static __net_exit void proto_exit_net(struct net *net) 4222 { 4223 remove_proc_entry("protocols", net->proc_net); 4224 } 4225 4226 4227 static __net_initdata struct pernet_operations proto_net_ops = { 4228 .init = proto_init_net, 4229 .exit = proto_exit_net, 4230 }; 4231 4232 static int __init proto_init(void) 4233 { 4234 return register_pernet_subsys(&proto_net_ops); 4235 } 4236 4237 subsys_initcall(proto_init); 4238 4239 #endif /* PROC_FS */ 4240 4241 #ifdef CONFIG_NET_RX_BUSY_POLL 4242 bool sk_busy_loop_end(void *p, unsigned long start_time) 4243 { 4244 struct sock *sk = p; 4245 4246 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4247 return true; 4248 4249 if (sk_is_udp(sk) && 4250 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4251 return true; 4252 4253 return sk_busy_loop_timeout(sk, start_time); 4254 } 4255 EXPORT_SYMBOL(sk_busy_loop_end); 4256 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4257 4258 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4259 { 4260 if (!sk->sk_prot->bind_add) 4261 return -EOPNOTSUPP; 4262 return sk->sk_prot->bind_add(sk, addr, addr_len); 4263 } 4264 EXPORT_SYMBOL(sock_bind_add); 4265 4266 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4267 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4268 void __user *arg, void *karg, size_t size) 4269 { 4270 int ret; 4271 4272 if (copy_from_user(karg, arg, size)) 4273 return -EFAULT; 4274 4275 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4276 if (ret) 4277 return ret; 4278 4279 if (copy_to_user(arg, karg, size)) 4280 return -EFAULT; 4281 4282 return 0; 4283 } 4284 EXPORT_SYMBOL(sock_ioctl_inout); 4285 4286 /* This is the most common ioctl prep function, where the result (4 bytes) is 4287 * copied back to userspace if the ioctl() returns successfully. No input is 4288 * copied from userspace as input argument. 4289 */ 4290 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4291 { 4292 int ret, karg = 0; 4293 4294 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4295 if (ret) 4296 return ret; 4297 4298 return put_user(karg, (int __user *)arg); 4299 } 4300 4301 /* A wrapper around sock ioctls, which copies the data from userspace 4302 * (depending on the protocol/ioctl), and copies back the result to userspace. 4303 * The main motivation for this function is to pass kernel memory to the 4304 * protocol ioctl callbacks, instead of userspace memory. 4305 */ 4306 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4307 { 4308 int rc = 1; 4309 4310 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4311 rc = ipmr_sk_ioctl(sk, cmd, arg); 4312 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4313 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4314 else if (sk_is_phonet(sk)) 4315 rc = phonet_sk_ioctl(sk, cmd, arg); 4316 4317 /* If ioctl was processed, returns its value */ 4318 if (rc <= 0) 4319 return rc; 4320 4321 /* Otherwise call the default handler */ 4322 return sock_ioctl_out(sk, cmd, arg); 4323 } 4324 EXPORT_SYMBOL(sk_ioctl); 4325 4326 static int __init sock_struct_check(void) 4327 { 4328 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4329 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4330 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4331 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4332 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4333 4334 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4335 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4336 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4337 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4338 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4339 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4340 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4341 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4342 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4343 4344 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4345 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4346 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4347 4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4352 4353 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf); 4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm); 4363 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status); 4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4367 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4369 4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4372 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4383 return 0; 4384 } 4385 4386 core_initcall(sock_struct_check); 4387