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, <[email protected]> 12 * Florian La Roche, <[email protected]> 13 * Alan Cox, <[email protected]> 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 sk_set_prio_allowed(const struct sock *sk, int val) 458 { 459 return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) || 460 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 461 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)); 462 } 463 464 static bool sock_needs_netstamp(const struct sock *sk) 465 { 466 switch (sk->sk_family) { 467 case AF_UNSPEC: 468 case AF_UNIX: 469 return false; 470 default: 471 return true; 472 } 473 } 474 475 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 476 { 477 if (sk->sk_flags & flags) { 478 sk->sk_flags &= ~flags; 479 if (sock_needs_netstamp(sk) && 480 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 481 net_disable_timestamp(); 482 } 483 } 484 485 486 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 487 { 488 unsigned long flags; 489 struct sk_buff_head *list = &sk->sk_receive_queue; 490 491 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) { 492 atomic_inc(&sk->sk_drops); 493 trace_sock_rcvqueue_full(sk, skb); 494 return -ENOMEM; 495 } 496 497 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 498 atomic_inc(&sk->sk_drops); 499 return -ENOBUFS; 500 } 501 502 skb->dev = NULL; 503 skb_set_owner_r(skb, sk); 504 505 /* we escape from rcu protected region, make sure we dont leak 506 * a norefcounted dst 507 */ 508 skb_dst_force(skb); 509 510 spin_lock_irqsave(&list->lock, flags); 511 sock_skb_set_dropcount(sk, skb); 512 __skb_queue_tail(list, skb); 513 spin_unlock_irqrestore(&list->lock, flags); 514 515 if (!sock_flag(sk, SOCK_DEAD)) 516 sk->sk_data_ready(sk); 517 return 0; 518 } 519 EXPORT_SYMBOL(__sock_queue_rcv_skb); 520 521 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 522 enum skb_drop_reason *reason) 523 { 524 enum skb_drop_reason drop_reason; 525 int err; 526 527 err = sk_filter(sk, skb); 528 if (err) { 529 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 530 goto out; 531 } 532 err = __sock_queue_rcv_skb(sk, skb); 533 switch (err) { 534 case -ENOMEM: 535 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 536 break; 537 case -ENOBUFS: 538 drop_reason = SKB_DROP_REASON_PROTO_MEM; 539 break; 540 default: 541 drop_reason = SKB_NOT_DROPPED_YET; 542 break; 543 } 544 out: 545 if (reason) 546 *reason = drop_reason; 547 return err; 548 } 549 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 550 551 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 552 const int nested, unsigned int trim_cap, bool refcounted) 553 { 554 int rc = NET_RX_SUCCESS; 555 556 if (sk_filter_trim_cap(sk, skb, trim_cap)) 557 goto discard_and_relse; 558 559 skb->dev = NULL; 560 561 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) { 562 atomic_inc(&sk->sk_drops); 563 goto discard_and_relse; 564 } 565 if (nested) 566 bh_lock_sock_nested(sk); 567 else 568 bh_lock_sock(sk); 569 if (!sock_owned_by_user(sk)) { 570 /* 571 * trylock + unlock semantics: 572 */ 573 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 574 575 rc = sk_backlog_rcv(sk, skb); 576 577 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 578 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 579 bh_unlock_sock(sk); 580 atomic_inc(&sk->sk_drops); 581 goto discard_and_relse; 582 } 583 584 bh_unlock_sock(sk); 585 out: 586 if (refcounted) 587 sock_put(sk); 588 return rc; 589 discard_and_relse: 590 kfree_skb(skb); 591 goto out; 592 } 593 EXPORT_SYMBOL(__sk_receive_skb); 594 595 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 596 u32)); 597 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 598 u32)); 599 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 600 { 601 struct dst_entry *dst = __sk_dst_get(sk); 602 603 if (dst && dst->obsolete && 604 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 605 dst, cookie) == NULL) { 606 sk_tx_queue_clear(sk); 607 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 608 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 609 dst_release(dst); 610 return NULL; 611 } 612 613 return dst; 614 } 615 EXPORT_SYMBOL(__sk_dst_check); 616 617 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 618 { 619 struct dst_entry *dst = sk_dst_get(sk); 620 621 if (dst && dst->obsolete && 622 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 623 dst, cookie) == NULL) { 624 sk_dst_reset(sk); 625 dst_release(dst); 626 return NULL; 627 } 628 629 return dst; 630 } 631 EXPORT_SYMBOL(sk_dst_check); 632 633 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 634 { 635 int ret = -ENOPROTOOPT; 636 #ifdef CONFIG_NETDEVICES 637 struct net *net = sock_net(sk); 638 639 /* Sorry... */ 640 ret = -EPERM; 641 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 642 goto out; 643 644 ret = -EINVAL; 645 if (ifindex < 0) 646 goto out; 647 648 /* Paired with all READ_ONCE() done locklessly. */ 649 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 650 651 if (sk->sk_prot->rehash) 652 sk->sk_prot->rehash(sk); 653 sk_dst_reset(sk); 654 655 ret = 0; 656 657 out: 658 #endif 659 660 return ret; 661 } 662 663 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 664 { 665 int ret; 666 667 if (lock_sk) 668 lock_sock(sk); 669 ret = sock_bindtoindex_locked(sk, ifindex); 670 if (lock_sk) 671 release_sock(sk); 672 673 return ret; 674 } 675 EXPORT_SYMBOL(sock_bindtoindex); 676 677 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 678 { 679 int ret = -ENOPROTOOPT; 680 #ifdef CONFIG_NETDEVICES 681 struct net *net = sock_net(sk); 682 char devname[IFNAMSIZ]; 683 int index; 684 685 ret = -EINVAL; 686 if (optlen < 0) 687 goto out; 688 689 /* Bind this socket to a particular device like "eth0", 690 * as specified in the passed interface name. If the 691 * name is "" or the option length is zero the socket 692 * is not bound. 693 */ 694 if (optlen > IFNAMSIZ - 1) 695 optlen = IFNAMSIZ - 1; 696 memset(devname, 0, sizeof(devname)); 697 698 ret = -EFAULT; 699 if (copy_from_sockptr(devname, optval, optlen)) 700 goto out; 701 702 index = 0; 703 if (devname[0] != '\0') { 704 struct net_device *dev; 705 706 rcu_read_lock(); 707 dev = dev_get_by_name_rcu(net, devname); 708 if (dev) 709 index = dev->ifindex; 710 rcu_read_unlock(); 711 ret = -ENODEV; 712 if (!dev) 713 goto out; 714 } 715 716 sockopt_lock_sock(sk); 717 ret = sock_bindtoindex_locked(sk, index); 718 sockopt_release_sock(sk); 719 out: 720 #endif 721 722 return ret; 723 } 724 725 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 726 sockptr_t optlen, int len) 727 { 728 int ret = -ENOPROTOOPT; 729 #ifdef CONFIG_NETDEVICES 730 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 731 struct net *net = sock_net(sk); 732 char devname[IFNAMSIZ]; 733 734 if (bound_dev_if == 0) { 735 len = 0; 736 goto zero; 737 } 738 739 ret = -EINVAL; 740 if (len < IFNAMSIZ) 741 goto out; 742 743 ret = netdev_get_name(net, devname, bound_dev_if); 744 if (ret) 745 goto out; 746 747 len = strlen(devname) + 1; 748 749 ret = -EFAULT; 750 if (copy_to_sockptr(optval, devname, len)) 751 goto out; 752 753 zero: 754 ret = -EFAULT; 755 if (copy_to_sockptr(optlen, &len, sizeof(int))) 756 goto out; 757 758 ret = 0; 759 760 out: 761 #endif 762 763 return ret; 764 } 765 766 bool sk_mc_loop(const struct sock *sk) 767 { 768 if (dev_recursion_level()) 769 return false; 770 if (!sk) 771 return true; 772 /* IPV6_ADDRFORM can change sk->sk_family under us. */ 773 switch (READ_ONCE(sk->sk_family)) { 774 case AF_INET: 775 return inet_test_bit(MC_LOOP, sk); 776 #if IS_ENABLED(CONFIG_IPV6) 777 case AF_INET6: 778 return inet6_test_bit(MC6_LOOP, sk); 779 #endif 780 } 781 WARN_ON_ONCE(1); 782 return true; 783 } 784 EXPORT_SYMBOL(sk_mc_loop); 785 786 void sock_set_reuseaddr(struct sock *sk) 787 { 788 lock_sock(sk); 789 sk->sk_reuse = SK_CAN_REUSE; 790 release_sock(sk); 791 } 792 EXPORT_SYMBOL(sock_set_reuseaddr); 793 794 void sock_set_reuseport(struct sock *sk) 795 { 796 lock_sock(sk); 797 sk->sk_reuseport = true; 798 release_sock(sk); 799 } 800 EXPORT_SYMBOL(sock_set_reuseport); 801 802 void sock_no_linger(struct sock *sk) 803 { 804 lock_sock(sk); 805 WRITE_ONCE(sk->sk_lingertime, 0); 806 sock_set_flag(sk, SOCK_LINGER); 807 release_sock(sk); 808 } 809 EXPORT_SYMBOL(sock_no_linger); 810 811 void sock_set_priority(struct sock *sk, u32 priority) 812 { 813 WRITE_ONCE(sk->sk_priority, priority); 814 } 815 EXPORT_SYMBOL(sock_set_priority); 816 817 void sock_set_sndtimeo(struct sock *sk, s64 secs) 818 { 819 lock_sock(sk); 820 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 821 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 822 else 823 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 824 release_sock(sk); 825 } 826 EXPORT_SYMBOL(sock_set_sndtimeo); 827 828 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 829 { 830 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val); 831 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns); 832 if (val) { 833 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 834 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 835 } 836 } 837 838 void sock_enable_timestamps(struct sock *sk) 839 { 840 lock_sock(sk); 841 __sock_set_timestamps(sk, true, false, true); 842 release_sock(sk); 843 } 844 EXPORT_SYMBOL(sock_enable_timestamps); 845 846 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 847 { 848 switch (optname) { 849 case SO_TIMESTAMP_OLD: 850 __sock_set_timestamps(sk, valbool, false, false); 851 break; 852 case SO_TIMESTAMP_NEW: 853 __sock_set_timestamps(sk, valbool, true, false); 854 break; 855 case SO_TIMESTAMPNS_OLD: 856 __sock_set_timestamps(sk, valbool, false, true); 857 break; 858 case SO_TIMESTAMPNS_NEW: 859 __sock_set_timestamps(sk, valbool, true, true); 860 break; 861 } 862 } 863 864 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 865 { 866 struct net *net = sock_net(sk); 867 struct net_device *dev = NULL; 868 bool match = false; 869 int *vclock_index; 870 int i, num; 871 872 if (sk->sk_bound_dev_if) 873 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 874 875 if (!dev) { 876 pr_err("%s: sock not bind to device\n", __func__); 877 return -EOPNOTSUPP; 878 } 879 880 num = ethtool_get_phc_vclocks(dev, &vclock_index); 881 dev_put(dev); 882 883 for (i = 0; i < num; i++) { 884 if (*(vclock_index + i) == phc_index) { 885 match = true; 886 break; 887 } 888 } 889 890 if (num > 0) 891 kfree(vclock_index); 892 893 if (!match) 894 return -EINVAL; 895 896 WRITE_ONCE(sk->sk_bind_phc, phc_index); 897 898 return 0; 899 } 900 901 int sock_set_timestamping(struct sock *sk, int optname, 902 struct so_timestamping timestamping) 903 { 904 int val = timestamping.flags; 905 int ret; 906 907 if (val & ~SOF_TIMESTAMPING_MASK) 908 return -EINVAL; 909 910 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 911 !(val & SOF_TIMESTAMPING_OPT_ID)) 912 return -EINVAL; 913 914 if (val & SOF_TIMESTAMPING_OPT_ID && 915 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 916 if (sk_is_tcp(sk)) { 917 if ((1 << sk->sk_state) & 918 (TCPF_CLOSE | TCPF_LISTEN)) 919 return -EINVAL; 920 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 921 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 922 else 923 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 924 } else { 925 atomic_set(&sk->sk_tskey, 0); 926 } 927 } 928 929 if (val & SOF_TIMESTAMPING_OPT_STATS && 930 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 931 return -EINVAL; 932 933 if (val & SOF_TIMESTAMPING_BIND_PHC) { 934 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 935 if (ret) 936 return ret; 937 } 938 939 WRITE_ONCE(sk->sk_tsflags, val); 940 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 941 sock_valbool_flag(sk, SOCK_TIMESTAMPING_ANY, !!(val & TSFLAGS_ANY)); 942 943 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 944 sock_enable_timestamp(sk, 945 SOCK_TIMESTAMPING_RX_SOFTWARE); 946 else 947 sock_disable_timestamp(sk, 948 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 949 return 0; 950 } 951 952 #if defined(CONFIG_CGROUP_BPF) 953 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 954 { 955 struct bpf_sock_ops_kern sock_ops; 956 957 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 958 sock_ops.op = op; 959 sock_ops.is_fullsock = 1; 960 sock_ops.sk = sk; 961 bpf_skops_init_skb(&sock_ops, skb, 0); 962 __cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS); 963 } 964 #endif 965 966 void sock_set_keepalive(struct sock *sk) 967 { 968 lock_sock(sk); 969 if (sk->sk_prot->keepalive) 970 sk->sk_prot->keepalive(sk, true); 971 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 972 release_sock(sk); 973 } 974 EXPORT_SYMBOL(sock_set_keepalive); 975 976 static void __sock_set_rcvbuf(struct sock *sk, int val) 977 { 978 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 979 * as a negative value. 980 */ 981 val = min_t(int, val, INT_MAX / 2); 982 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 983 984 /* We double it on the way in to account for "struct sk_buff" etc. 985 * overhead. Applications assume that the SO_RCVBUF setting they make 986 * will allow that much actual data to be received on that socket. 987 * 988 * Applications are unaware that "struct sk_buff" and other overheads 989 * allocate from the receive buffer during socket buffer allocation. 990 * 991 * And after considering the possible alternatives, returning the value 992 * we actually used in getsockopt is the most desirable behavior. 993 */ 994 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 995 } 996 997 void sock_set_rcvbuf(struct sock *sk, int val) 998 { 999 lock_sock(sk); 1000 __sock_set_rcvbuf(sk, val); 1001 release_sock(sk); 1002 } 1003 EXPORT_SYMBOL(sock_set_rcvbuf); 1004 1005 static void __sock_set_mark(struct sock *sk, u32 val) 1006 { 1007 if (val != sk->sk_mark) { 1008 WRITE_ONCE(sk->sk_mark, val); 1009 sk_dst_reset(sk); 1010 } 1011 } 1012 1013 void sock_set_mark(struct sock *sk, u32 val) 1014 { 1015 lock_sock(sk); 1016 __sock_set_mark(sk, val); 1017 release_sock(sk); 1018 } 1019 EXPORT_SYMBOL(sock_set_mark); 1020 1021 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1022 { 1023 /* Round down bytes to multiple of pages */ 1024 bytes = round_down(bytes, PAGE_SIZE); 1025 1026 WARN_ON(bytes > sk->sk_reserved_mem); 1027 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1028 sk_mem_reclaim(sk); 1029 } 1030 1031 static int sock_reserve_memory(struct sock *sk, int bytes) 1032 { 1033 long allocated; 1034 bool charged; 1035 int pages; 1036 1037 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1038 return -EOPNOTSUPP; 1039 1040 if (!bytes) 1041 return 0; 1042 1043 pages = sk_mem_pages(bytes); 1044 1045 /* pre-charge to memcg */ 1046 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1047 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1048 if (!charged) 1049 return -ENOMEM; 1050 1051 /* pre-charge to forward_alloc */ 1052 sk_memory_allocated_add(sk, pages); 1053 allocated = sk_memory_allocated(sk); 1054 /* If the system goes into memory pressure with this 1055 * precharge, give up and return error. 1056 */ 1057 if (allocated > sk_prot_mem_limits(sk, 1)) { 1058 sk_memory_allocated_sub(sk, pages); 1059 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1060 return -ENOMEM; 1061 } 1062 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1063 1064 WRITE_ONCE(sk->sk_reserved_mem, 1065 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1066 1067 return 0; 1068 } 1069 1070 #ifdef CONFIG_PAGE_POOL 1071 1072 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED 1073 * in 1 syscall. The limit exists to limit the amount of memory the kernel 1074 * allocates to copy these tokens, and to prevent looping over the frags for 1075 * too long. 1076 */ 1077 #define MAX_DONTNEED_TOKENS 128 1078 #define MAX_DONTNEED_FRAGS 1024 1079 1080 static noinline_for_stack int 1081 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1082 { 1083 unsigned int num_tokens, i, j, k, netmem_num = 0; 1084 struct dmabuf_token *tokens; 1085 int ret = 0, num_frags = 0; 1086 netmem_ref netmems[16]; 1087 1088 if (!sk_is_tcp(sk)) 1089 return -EBADF; 1090 1091 if (optlen % sizeof(*tokens) || 1092 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1093 return -EINVAL; 1094 1095 num_tokens = optlen / sizeof(*tokens); 1096 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL); 1097 if (!tokens) 1098 return -ENOMEM; 1099 1100 if (copy_from_sockptr(tokens, optval, optlen)) { 1101 kvfree(tokens); 1102 return -EFAULT; 1103 } 1104 1105 xa_lock_bh(&sk->sk_user_frags); 1106 for (i = 0; i < num_tokens; i++) { 1107 for (j = 0; j < tokens[i].token_count; j++) { 1108 if (++num_frags > MAX_DONTNEED_FRAGS) 1109 goto frag_limit_reached; 1110 1111 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1112 &sk->sk_user_frags, tokens[i].token_start + j); 1113 1114 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem))) 1115 continue; 1116 1117 netmems[netmem_num++] = netmem; 1118 if (netmem_num == ARRAY_SIZE(netmems)) { 1119 xa_unlock_bh(&sk->sk_user_frags); 1120 for (k = 0; k < netmem_num; k++) 1121 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1122 netmem_num = 0; 1123 xa_lock_bh(&sk->sk_user_frags); 1124 } 1125 ret++; 1126 } 1127 } 1128 1129 frag_limit_reached: 1130 xa_unlock_bh(&sk->sk_user_frags); 1131 for (k = 0; k < netmem_num; k++) 1132 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1133 1134 kvfree(tokens); 1135 return ret; 1136 } 1137 #endif 1138 1139 void sockopt_lock_sock(struct sock *sk) 1140 { 1141 /* When current->bpf_ctx is set, the setsockopt is called from 1142 * a bpf prog. bpf has ensured the sk lock has been 1143 * acquired before calling setsockopt(). 1144 */ 1145 if (has_current_bpf_ctx()) 1146 return; 1147 1148 lock_sock(sk); 1149 } 1150 EXPORT_SYMBOL(sockopt_lock_sock); 1151 1152 void sockopt_release_sock(struct sock *sk) 1153 { 1154 if (has_current_bpf_ctx()) 1155 return; 1156 1157 release_sock(sk); 1158 } 1159 EXPORT_SYMBOL(sockopt_release_sock); 1160 1161 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1162 { 1163 return has_current_bpf_ctx() || ns_capable(ns, cap); 1164 } 1165 EXPORT_SYMBOL(sockopt_ns_capable); 1166 1167 bool sockopt_capable(int cap) 1168 { 1169 return has_current_bpf_ctx() || capable(cap); 1170 } 1171 EXPORT_SYMBOL(sockopt_capable); 1172 1173 static int sockopt_validate_clockid(__kernel_clockid_t value) 1174 { 1175 switch (value) { 1176 case CLOCK_REALTIME: 1177 case CLOCK_MONOTONIC: 1178 case CLOCK_TAI: 1179 return 0; 1180 } 1181 return -EINVAL; 1182 } 1183 1184 /* 1185 * This is meant for all protocols to use and covers goings on 1186 * at the socket level. Everything here is generic. 1187 */ 1188 1189 int sk_setsockopt(struct sock *sk, int level, int optname, 1190 sockptr_t optval, unsigned int optlen) 1191 { 1192 struct so_timestamping timestamping; 1193 struct socket *sock = sk->sk_socket; 1194 struct sock_txtime sk_txtime; 1195 int val; 1196 int valbool; 1197 struct linger ling; 1198 int ret = 0; 1199 1200 /* 1201 * Options without arguments 1202 */ 1203 1204 if (optname == SO_BINDTODEVICE) 1205 return sock_setbindtodevice(sk, optval, optlen); 1206 1207 if (optlen < sizeof(int)) 1208 return -EINVAL; 1209 1210 if (copy_from_sockptr(&val, optval, sizeof(val))) 1211 return -EFAULT; 1212 1213 valbool = val ? 1 : 0; 1214 1215 /* handle options which do not require locking the socket. */ 1216 switch (optname) { 1217 case SO_PRIORITY: 1218 if (sk_set_prio_allowed(sk, val)) { 1219 sock_set_priority(sk, val); 1220 return 0; 1221 } 1222 return -EPERM; 1223 case SO_PASSSEC: 1224 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1225 return 0; 1226 case SO_PASSCRED: 1227 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1228 return 0; 1229 case SO_PASSPIDFD: 1230 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1231 return 0; 1232 case SO_TYPE: 1233 case SO_PROTOCOL: 1234 case SO_DOMAIN: 1235 case SO_ERROR: 1236 return -ENOPROTOOPT; 1237 #ifdef CONFIG_NET_RX_BUSY_POLL 1238 case SO_BUSY_POLL: 1239 if (val < 0) 1240 return -EINVAL; 1241 WRITE_ONCE(sk->sk_ll_usec, val); 1242 return 0; 1243 case SO_PREFER_BUSY_POLL: 1244 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1245 return -EPERM; 1246 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1247 return 0; 1248 case SO_BUSY_POLL_BUDGET: 1249 if (val > READ_ONCE(sk->sk_busy_poll_budget) && 1250 !sockopt_capable(CAP_NET_ADMIN)) 1251 return -EPERM; 1252 if (val < 0 || val > U16_MAX) 1253 return -EINVAL; 1254 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1255 return 0; 1256 #endif 1257 case SO_MAX_PACING_RATE: 1258 { 1259 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1260 unsigned long pacing_rate; 1261 1262 if (sizeof(ulval) != sizeof(val) && 1263 optlen >= sizeof(ulval) && 1264 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1265 return -EFAULT; 1266 } 1267 if (ulval != ~0UL) 1268 cmpxchg(&sk->sk_pacing_status, 1269 SK_PACING_NONE, 1270 SK_PACING_NEEDED); 1271 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1272 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1273 pacing_rate = READ_ONCE(sk->sk_pacing_rate); 1274 if (ulval < pacing_rate) 1275 WRITE_ONCE(sk->sk_pacing_rate, ulval); 1276 return 0; 1277 } 1278 case SO_TXREHASH: 1279 if (val < -1 || val > 1) 1280 return -EINVAL; 1281 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1282 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1283 /* Paired with READ_ONCE() in tcp_rtx_synack() 1284 * and sk_getsockopt(). 1285 */ 1286 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1287 return 0; 1288 case SO_PEEK_OFF: 1289 { 1290 int (*set_peek_off)(struct sock *sk, int val); 1291 1292 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1293 if (set_peek_off) 1294 ret = set_peek_off(sk, val); 1295 else 1296 ret = -EOPNOTSUPP; 1297 return ret; 1298 } 1299 #ifdef CONFIG_PAGE_POOL 1300 case SO_DEVMEM_DONTNEED: 1301 return sock_devmem_dontneed(sk, optval, optlen); 1302 #endif 1303 } 1304 1305 sockopt_lock_sock(sk); 1306 1307 switch (optname) { 1308 case SO_DEBUG: 1309 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1310 ret = -EACCES; 1311 else 1312 sock_valbool_flag(sk, SOCK_DBG, valbool); 1313 break; 1314 case SO_REUSEADDR: 1315 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1316 break; 1317 case SO_REUSEPORT: 1318 if (valbool && !sk_is_inet(sk)) 1319 ret = -EOPNOTSUPP; 1320 else 1321 sk->sk_reuseport = valbool; 1322 break; 1323 case SO_DONTROUTE: 1324 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1325 sk_dst_reset(sk); 1326 break; 1327 case SO_BROADCAST: 1328 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1329 break; 1330 case SO_SNDBUF: 1331 /* Don't error on this BSD doesn't and if you think 1332 * about it this is right. Otherwise apps have to 1333 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1334 * are treated in BSD as hints 1335 */ 1336 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1337 set_sndbuf: 1338 /* Ensure val * 2 fits into an int, to prevent max_t() 1339 * from treating it as a negative value. 1340 */ 1341 val = min_t(int, val, INT_MAX / 2); 1342 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1343 WRITE_ONCE(sk->sk_sndbuf, 1344 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1345 /* Wake up sending tasks if we upped the value. */ 1346 sk->sk_write_space(sk); 1347 break; 1348 1349 case SO_SNDBUFFORCE: 1350 if (!sockopt_capable(CAP_NET_ADMIN)) { 1351 ret = -EPERM; 1352 break; 1353 } 1354 1355 /* No negative values (to prevent underflow, as val will be 1356 * multiplied by 2). 1357 */ 1358 if (val < 0) 1359 val = 0; 1360 goto set_sndbuf; 1361 1362 case SO_RCVBUF: 1363 /* Don't error on this BSD doesn't and if you think 1364 * about it this is right. Otherwise apps have to 1365 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1366 * are treated in BSD as hints 1367 */ 1368 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1369 break; 1370 1371 case SO_RCVBUFFORCE: 1372 if (!sockopt_capable(CAP_NET_ADMIN)) { 1373 ret = -EPERM; 1374 break; 1375 } 1376 1377 /* No negative values (to prevent underflow, as val will be 1378 * multiplied by 2). 1379 */ 1380 __sock_set_rcvbuf(sk, max(val, 0)); 1381 break; 1382 1383 case SO_KEEPALIVE: 1384 if (sk->sk_prot->keepalive) 1385 sk->sk_prot->keepalive(sk, valbool); 1386 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1387 break; 1388 1389 case SO_OOBINLINE: 1390 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1391 break; 1392 1393 case SO_NO_CHECK: 1394 sk->sk_no_check_tx = valbool; 1395 break; 1396 1397 case SO_LINGER: 1398 if (optlen < sizeof(ling)) { 1399 ret = -EINVAL; /* 1003.1g */ 1400 break; 1401 } 1402 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1403 ret = -EFAULT; 1404 break; 1405 } 1406 if (!ling.l_onoff) { 1407 sock_reset_flag(sk, SOCK_LINGER); 1408 } else { 1409 unsigned long t_sec = ling.l_linger; 1410 1411 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1412 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1413 else 1414 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1415 sock_set_flag(sk, SOCK_LINGER); 1416 } 1417 break; 1418 1419 case SO_BSDCOMPAT: 1420 break; 1421 1422 case SO_TIMESTAMP_OLD: 1423 case SO_TIMESTAMP_NEW: 1424 case SO_TIMESTAMPNS_OLD: 1425 case SO_TIMESTAMPNS_NEW: 1426 sock_set_timestamp(sk, optname, valbool); 1427 break; 1428 1429 case SO_TIMESTAMPING_NEW: 1430 case SO_TIMESTAMPING_OLD: 1431 if (optlen == sizeof(timestamping)) { 1432 if (copy_from_sockptr(×tamping, optval, 1433 sizeof(timestamping))) { 1434 ret = -EFAULT; 1435 break; 1436 } 1437 } else { 1438 memset(×tamping, 0, sizeof(timestamping)); 1439 timestamping.flags = val; 1440 } 1441 ret = sock_set_timestamping(sk, optname, timestamping); 1442 break; 1443 1444 case SO_RCVLOWAT: 1445 { 1446 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1447 1448 if (val < 0) 1449 val = INT_MAX; 1450 if (sock) 1451 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1452 if (set_rcvlowat) 1453 ret = set_rcvlowat(sk, val); 1454 else 1455 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1456 break; 1457 } 1458 case SO_RCVTIMEO_OLD: 1459 case SO_RCVTIMEO_NEW: 1460 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1461 optlen, optname == SO_RCVTIMEO_OLD); 1462 break; 1463 1464 case SO_SNDTIMEO_OLD: 1465 case SO_SNDTIMEO_NEW: 1466 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1467 optlen, optname == SO_SNDTIMEO_OLD); 1468 break; 1469 1470 case SO_ATTACH_FILTER: { 1471 struct sock_fprog fprog; 1472 1473 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1474 if (!ret) 1475 ret = sk_attach_filter(&fprog, sk); 1476 break; 1477 } 1478 case SO_ATTACH_BPF: 1479 ret = -EINVAL; 1480 if (optlen == sizeof(u32)) { 1481 u32 ufd; 1482 1483 ret = -EFAULT; 1484 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1485 break; 1486 1487 ret = sk_attach_bpf(ufd, sk); 1488 } 1489 break; 1490 1491 case SO_ATTACH_REUSEPORT_CBPF: { 1492 struct sock_fprog fprog; 1493 1494 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1495 if (!ret) 1496 ret = sk_reuseport_attach_filter(&fprog, sk); 1497 break; 1498 } 1499 case SO_ATTACH_REUSEPORT_EBPF: 1500 ret = -EINVAL; 1501 if (optlen == sizeof(u32)) { 1502 u32 ufd; 1503 1504 ret = -EFAULT; 1505 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1506 break; 1507 1508 ret = sk_reuseport_attach_bpf(ufd, sk); 1509 } 1510 break; 1511 1512 case SO_DETACH_REUSEPORT_BPF: 1513 ret = reuseport_detach_prog(sk); 1514 break; 1515 1516 case SO_DETACH_FILTER: 1517 ret = sk_detach_filter(sk); 1518 break; 1519 1520 case SO_LOCK_FILTER: 1521 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1522 ret = -EPERM; 1523 else 1524 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1525 break; 1526 1527 case SO_MARK: 1528 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1529 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1530 ret = -EPERM; 1531 break; 1532 } 1533 1534 __sock_set_mark(sk, val); 1535 break; 1536 case SO_RCVMARK: 1537 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1538 break; 1539 1540 case SO_RCVPRIORITY: 1541 sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool); 1542 break; 1543 1544 case SO_RXQ_OVFL: 1545 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1546 break; 1547 1548 case SO_WIFI_STATUS: 1549 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1550 break; 1551 1552 case SO_NOFCS: 1553 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1554 break; 1555 1556 case SO_SELECT_ERR_QUEUE: 1557 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1558 break; 1559 1560 1561 case SO_INCOMING_CPU: 1562 reuseport_update_incoming_cpu(sk, val); 1563 break; 1564 1565 case SO_CNX_ADVICE: 1566 if (val == 1) 1567 dst_negative_advice(sk); 1568 break; 1569 1570 case SO_ZEROCOPY: 1571 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1572 if (!(sk_is_tcp(sk) || 1573 (sk->sk_type == SOCK_DGRAM && 1574 sk->sk_protocol == IPPROTO_UDP))) 1575 ret = -EOPNOTSUPP; 1576 } else if (sk->sk_family != PF_RDS) { 1577 ret = -EOPNOTSUPP; 1578 } 1579 if (!ret) { 1580 if (val < 0 || val > 1) 1581 ret = -EINVAL; 1582 else 1583 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1584 } 1585 break; 1586 1587 case SO_TXTIME: 1588 if (optlen != sizeof(struct sock_txtime)) { 1589 ret = -EINVAL; 1590 break; 1591 } else if (copy_from_sockptr(&sk_txtime, optval, 1592 sizeof(struct sock_txtime))) { 1593 ret = -EFAULT; 1594 break; 1595 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1596 ret = -EINVAL; 1597 break; 1598 } 1599 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1600 * scheduler has enough safe guards. 1601 */ 1602 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1603 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1604 ret = -EPERM; 1605 break; 1606 } 1607 1608 ret = sockopt_validate_clockid(sk_txtime.clockid); 1609 if (ret) 1610 break; 1611 1612 sock_valbool_flag(sk, SOCK_TXTIME, true); 1613 sk->sk_clockid = sk_txtime.clockid; 1614 sk->sk_txtime_deadline_mode = 1615 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1616 sk->sk_txtime_report_errors = 1617 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1618 break; 1619 1620 case SO_BINDTOIFINDEX: 1621 ret = sock_bindtoindex_locked(sk, val); 1622 break; 1623 1624 case SO_BUF_LOCK: 1625 if (val & ~SOCK_BUF_LOCK_MASK) { 1626 ret = -EINVAL; 1627 break; 1628 } 1629 sk->sk_userlocks = val | (sk->sk_userlocks & 1630 ~SOCK_BUF_LOCK_MASK); 1631 break; 1632 1633 case SO_RESERVE_MEM: 1634 { 1635 int delta; 1636 1637 if (val < 0) { 1638 ret = -EINVAL; 1639 break; 1640 } 1641 1642 delta = val - sk->sk_reserved_mem; 1643 if (delta < 0) 1644 sock_release_reserved_memory(sk, -delta); 1645 else 1646 ret = sock_reserve_memory(sk, delta); 1647 break; 1648 } 1649 1650 default: 1651 ret = -ENOPROTOOPT; 1652 break; 1653 } 1654 sockopt_release_sock(sk); 1655 return ret; 1656 } 1657 1658 int sock_setsockopt(struct socket *sock, int level, int optname, 1659 sockptr_t optval, unsigned int optlen) 1660 { 1661 return sk_setsockopt(sock->sk, level, optname, 1662 optval, optlen); 1663 } 1664 EXPORT_SYMBOL(sock_setsockopt); 1665 1666 static const struct cred *sk_get_peer_cred(struct sock *sk) 1667 { 1668 const struct cred *cred; 1669 1670 spin_lock(&sk->sk_peer_lock); 1671 cred = get_cred(sk->sk_peer_cred); 1672 spin_unlock(&sk->sk_peer_lock); 1673 1674 return cred; 1675 } 1676 1677 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1678 struct ucred *ucred) 1679 { 1680 ucred->pid = pid_vnr(pid); 1681 ucred->uid = ucred->gid = -1; 1682 if (cred) { 1683 struct user_namespace *current_ns = current_user_ns(); 1684 1685 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1686 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1687 } 1688 } 1689 1690 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1691 { 1692 struct user_namespace *user_ns = current_user_ns(); 1693 int i; 1694 1695 for (i = 0; i < src->ngroups; i++) { 1696 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1697 1698 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1699 return -EFAULT; 1700 } 1701 1702 return 0; 1703 } 1704 1705 int sk_getsockopt(struct sock *sk, int level, int optname, 1706 sockptr_t optval, sockptr_t optlen) 1707 { 1708 struct socket *sock = sk->sk_socket; 1709 1710 union { 1711 int val; 1712 u64 val64; 1713 unsigned long ulval; 1714 struct linger ling; 1715 struct old_timeval32 tm32; 1716 struct __kernel_old_timeval tm; 1717 struct __kernel_sock_timeval stm; 1718 struct sock_txtime txtime; 1719 struct so_timestamping timestamping; 1720 } v; 1721 1722 int lv = sizeof(int); 1723 int len; 1724 1725 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1726 return -EFAULT; 1727 if (len < 0) 1728 return -EINVAL; 1729 1730 memset(&v, 0, sizeof(v)); 1731 1732 switch (optname) { 1733 case SO_DEBUG: 1734 v.val = sock_flag(sk, SOCK_DBG); 1735 break; 1736 1737 case SO_DONTROUTE: 1738 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1739 break; 1740 1741 case SO_BROADCAST: 1742 v.val = sock_flag(sk, SOCK_BROADCAST); 1743 break; 1744 1745 case SO_SNDBUF: 1746 v.val = READ_ONCE(sk->sk_sndbuf); 1747 break; 1748 1749 case SO_RCVBUF: 1750 v.val = READ_ONCE(sk->sk_rcvbuf); 1751 break; 1752 1753 case SO_REUSEADDR: 1754 v.val = sk->sk_reuse; 1755 break; 1756 1757 case SO_REUSEPORT: 1758 v.val = sk->sk_reuseport; 1759 break; 1760 1761 case SO_KEEPALIVE: 1762 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1763 break; 1764 1765 case SO_TYPE: 1766 v.val = sk->sk_type; 1767 break; 1768 1769 case SO_PROTOCOL: 1770 v.val = sk->sk_protocol; 1771 break; 1772 1773 case SO_DOMAIN: 1774 v.val = sk->sk_family; 1775 break; 1776 1777 case SO_ERROR: 1778 v.val = -sock_error(sk); 1779 if (v.val == 0) 1780 v.val = xchg(&sk->sk_err_soft, 0); 1781 break; 1782 1783 case SO_OOBINLINE: 1784 v.val = sock_flag(sk, SOCK_URGINLINE); 1785 break; 1786 1787 case SO_NO_CHECK: 1788 v.val = sk->sk_no_check_tx; 1789 break; 1790 1791 case SO_PRIORITY: 1792 v.val = READ_ONCE(sk->sk_priority); 1793 break; 1794 1795 case SO_LINGER: 1796 lv = sizeof(v.ling); 1797 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1798 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1799 break; 1800 1801 case SO_BSDCOMPAT: 1802 break; 1803 1804 case SO_TIMESTAMP_OLD: 1805 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1806 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1807 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1808 break; 1809 1810 case SO_TIMESTAMPNS_OLD: 1811 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1812 break; 1813 1814 case SO_TIMESTAMP_NEW: 1815 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1816 break; 1817 1818 case SO_TIMESTAMPNS_NEW: 1819 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1820 break; 1821 1822 case SO_TIMESTAMPING_OLD: 1823 case SO_TIMESTAMPING_NEW: 1824 lv = sizeof(v.timestamping); 1825 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1826 * returning the flags when they were set through the same option. 1827 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1828 */ 1829 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1830 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1831 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1832 } 1833 break; 1834 1835 case SO_RCVTIMEO_OLD: 1836 case SO_RCVTIMEO_NEW: 1837 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1838 SO_RCVTIMEO_OLD == optname); 1839 break; 1840 1841 case SO_SNDTIMEO_OLD: 1842 case SO_SNDTIMEO_NEW: 1843 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1844 SO_SNDTIMEO_OLD == optname); 1845 break; 1846 1847 case SO_RCVLOWAT: 1848 v.val = READ_ONCE(sk->sk_rcvlowat); 1849 break; 1850 1851 case SO_SNDLOWAT: 1852 v.val = 1; 1853 break; 1854 1855 case SO_PASSCRED: 1856 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1857 break; 1858 1859 case SO_PASSPIDFD: 1860 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1861 break; 1862 1863 case SO_PEERCRED: 1864 { 1865 struct ucred peercred; 1866 if (len > sizeof(peercred)) 1867 len = sizeof(peercred); 1868 1869 spin_lock(&sk->sk_peer_lock); 1870 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1871 spin_unlock(&sk->sk_peer_lock); 1872 1873 if (copy_to_sockptr(optval, &peercred, len)) 1874 return -EFAULT; 1875 goto lenout; 1876 } 1877 1878 case SO_PEERPIDFD: 1879 { 1880 struct pid *peer_pid; 1881 struct file *pidfd_file = NULL; 1882 int pidfd; 1883 1884 if (len > sizeof(pidfd)) 1885 len = sizeof(pidfd); 1886 1887 spin_lock(&sk->sk_peer_lock); 1888 peer_pid = get_pid(sk->sk_peer_pid); 1889 spin_unlock(&sk->sk_peer_lock); 1890 1891 if (!peer_pid) 1892 return -ENODATA; 1893 1894 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1895 put_pid(peer_pid); 1896 if (pidfd < 0) 1897 return pidfd; 1898 1899 if (copy_to_sockptr(optval, &pidfd, len) || 1900 copy_to_sockptr(optlen, &len, sizeof(int))) { 1901 put_unused_fd(pidfd); 1902 fput(pidfd_file); 1903 1904 return -EFAULT; 1905 } 1906 1907 fd_install(pidfd, pidfd_file); 1908 return 0; 1909 } 1910 1911 case SO_PEERGROUPS: 1912 { 1913 const struct cred *cred; 1914 int ret, n; 1915 1916 cred = sk_get_peer_cred(sk); 1917 if (!cred) 1918 return -ENODATA; 1919 1920 n = cred->group_info->ngroups; 1921 if (len < n * sizeof(gid_t)) { 1922 len = n * sizeof(gid_t); 1923 put_cred(cred); 1924 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1925 } 1926 len = n * sizeof(gid_t); 1927 1928 ret = groups_to_user(optval, cred->group_info); 1929 put_cred(cred); 1930 if (ret) 1931 return ret; 1932 goto lenout; 1933 } 1934 1935 case SO_PEERNAME: 1936 { 1937 struct sockaddr_storage address; 1938 1939 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 1940 if (lv < 0) 1941 return -ENOTCONN; 1942 if (lv < len) 1943 return -EINVAL; 1944 if (copy_to_sockptr(optval, &address, len)) 1945 return -EFAULT; 1946 goto lenout; 1947 } 1948 1949 /* Dubious BSD thing... Probably nobody even uses it, but 1950 * the UNIX standard wants it for whatever reason... -DaveM 1951 */ 1952 case SO_ACCEPTCONN: 1953 v.val = sk->sk_state == TCP_LISTEN; 1954 break; 1955 1956 case SO_PASSSEC: 1957 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1958 break; 1959 1960 case SO_PEERSEC: 1961 return security_socket_getpeersec_stream(sock, 1962 optval, optlen, len); 1963 1964 case SO_MARK: 1965 v.val = READ_ONCE(sk->sk_mark); 1966 break; 1967 1968 case SO_RCVMARK: 1969 v.val = sock_flag(sk, SOCK_RCVMARK); 1970 break; 1971 1972 case SO_RCVPRIORITY: 1973 v.val = sock_flag(sk, SOCK_RCVPRIORITY); 1974 break; 1975 1976 case SO_RXQ_OVFL: 1977 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1978 break; 1979 1980 case SO_WIFI_STATUS: 1981 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1982 break; 1983 1984 case SO_PEEK_OFF: 1985 if (!READ_ONCE(sock->ops)->set_peek_off) 1986 return -EOPNOTSUPP; 1987 1988 v.val = READ_ONCE(sk->sk_peek_off); 1989 break; 1990 case SO_NOFCS: 1991 v.val = sock_flag(sk, SOCK_NOFCS); 1992 break; 1993 1994 case SO_BINDTODEVICE: 1995 return sock_getbindtodevice(sk, optval, optlen, len); 1996 1997 case SO_GET_FILTER: 1998 len = sk_get_filter(sk, optval, len); 1999 if (len < 0) 2000 return len; 2001 2002 goto lenout; 2003 2004 case SO_LOCK_FILTER: 2005 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 2006 break; 2007 2008 case SO_BPF_EXTENSIONS: 2009 v.val = bpf_tell_extensions(); 2010 break; 2011 2012 case SO_SELECT_ERR_QUEUE: 2013 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 2014 break; 2015 2016 #ifdef CONFIG_NET_RX_BUSY_POLL 2017 case SO_BUSY_POLL: 2018 v.val = READ_ONCE(sk->sk_ll_usec); 2019 break; 2020 case SO_PREFER_BUSY_POLL: 2021 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 2022 break; 2023 #endif 2024 2025 case SO_MAX_PACING_RATE: 2026 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 2027 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 2028 lv = sizeof(v.ulval); 2029 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 2030 } else { 2031 /* 32bit version */ 2032 v.val = min_t(unsigned long, ~0U, 2033 READ_ONCE(sk->sk_max_pacing_rate)); 2034 } 2035 break; 2036 2037 case SO_INCOMING_CPU: 2038 v.val = READ_ONCE(sk->sk_incoming_cpu); 2039 break; 2040 2041 case SO_MEMINFO: 2042 { 2043 u32 meminfo[SK_MEMINFO_VARS]; 2044 2045 sk_get_meminfo(sk, meminfo); 2046 2047 len = min_t(unsigned int, len, sizeof(meminfo)); 2048 if (copy_to_sockptr(optval, &meminfo, len)) 2049 return -EFAULT; 2050 2051 goto lenout; 2052 } 2053 2054 #ifdef CONFIG_NET_RX_BUSY_POLL 2055 case SO_INCOMING_NAPI_ID: 2056 v.val = READ_ONCE(sk->sk_napi_id); 2057 2058 /* aggregate non-NAPI IDs down to 0 */ 2059 if (!napi_id_valid(v.val)) 2060 v.val = 0; 2061 2062 break; 2063 #endif 2064 2065 case SO_COOKIE: 2066 lv = sizeof(u64); 2067 if (len < lv) 2068 return -EINVAL; 2069 v.val64 = sock_gen_cookie(sk); 2070 break; 2071 2072 case SO_ZEROCOPY: 2073 v.val = sock_flag(sk, SOCK_ZEROCOPY); 2074 break; 2075 2076 case SO_TXTIME: 2077 lv = sizeof(v.txtime); 2078 v.txtime.clockid = sk->sk_clockid; 2079 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 2080 SOF_TXTIME_DEADLINE_MODE : 0; 2081 v.txtime.flags |= sk->sk_txtime_report_errors ? 2082 SOF_TXTIME_REPORT_ERRORS : 0; 2083 break; 2084 2085 case SO_BINDTOIFINDEX: 2086 v.val = READ_ONCE(sk->sk_bound_dev_if); 2087 break; 2088 2089 case SO_NETNS_COOKIE: 2090 lv = sizeof(u64); 2091 if (len != lv) 2092 return -EINVAL; 2093 v.val64 = sock_net(sk)->net_cookie; 2094 break; 2095 2096 case SO_BUF_LOCK: 2097 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 2098 break; 2099 2100 case SO_RESERVE_MEM: 2101 v.val = READ_ONCE(sk->sk_reserved_mem); 2102 break; 2103 2104 case SO_TXREHASH: 2105 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2106 v.val = READ_ONCE(sk->sk_txrehash); 2107 break; 2108 2109 default: 2110 /* We implement the SO_SNDLOWAT etc to not be settable 2111 * (1003.1g 7). 2112 */ 2113 return -ENOPROTOOPT; 2114 } 2115 2116 if (len > lv) 2117 len = lv; 2118 if (copy_to_sockptr(optval, &v, len)) 2119 return -EFAULT; 2120 lenout: 2121 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2122 return -EFAULT; 2123 return 0; 2124 } 2125 2126 /* 2127 * Initialize an sk_lock. 2128 * 2129 * (We also register the sk_lock with the lock validator.) 2130 */ 2131 static inline void sock_lock_init(struct sock *sk) 2132 { 2133 if (sk->sk_kern_sock) 2134 sock_lock_init_class_and_name( 2135 sk, 2136 af_family_kern_slock_key_strings[sk->sk_family], 2137 af_family_kern_slock_keys + sk->sk_family, 2138 af_family_kern_key_strings[sk->sk_family], 2139 af_family_kern_keys + sk->sk_family); 2140 else 2141 sock_lock_init_class_and_name( 2142 sk, 2143 af_family_slock_key_strings[sk->sk_family], 2144 af_family_slock_keys + sk->sk_family, 2145 af_family_key_strings[sk->sk_family], 2146 af_family_keys + sk->sk_family); 2147 } 2148 2149 /* 2150 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2151 * even temporarily, because of RCU lookups. sk_node should also be left as is. 2152 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2153 */ 2154 static void sock_copy(struct sock *nsk, const struct sock *osk) 2155 { 2156 const struct proto *prot = READ_ONCE(osk->sk_prot); 2157 #ifdef CONFIG_SECURITY_NETWORK 2158 void *sptr = nsk->sk_security; 2159 #endif 2160 2161 /* If we move sk_tx_queue_mapping out of the private section, 2162 * we must check if sk_tx_queue_clear() is called after 2163 * sock_copy() in sk_clone_lock(). 2164 */ 2165 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2166 offsetof(struct sock, sk_dontcopy_begin) || 2167 offsetof(struct sock, sk_tx_queue_mapping) >= 2168 offsetof(struct sock, sk_dontcopy_end)); 2169 2170 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2171 2172 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2173 prot->obj_size - offsetof(struct sock, sk_dontcopy_end), 2174 /* alloc is larger than struct, see sk_prot_alloc() */); 2175 2176 #ifdef CONFIG_SECURITY_NETWORK 2177 nsk->sk_security = sptr; 2178 security_sk_clone(osk, nsk); 2179 #endif 2180 } 2181 2182 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2183 int family) 2184 { 2185 struct sock *sk; 2186 struct kmem_cache *slab; 2187 2188 slab = prot->slab; 2189 if (slab != NULL) { 2190 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2191 if (!sk) 2192 return sk; 2193 if (want_init_on_alloc(priority)) 2194 sk_prot_clear_nulls(sk, prot->obj_size); 2195 } else 2196 sk = kmalloc(prot->obj_size, priority); 2197 2198 if (sk != NULL) { 2199 if (security_sk_alloc(sk, family, priority)) 2200 goto out_free; 2201 2202 if (!try_module_get(prot->owner)) 2203 goto out_free_sec; 2204 } 2205 2206 return sk; 2207 2208 out_free_sec: 2209 security_sk_free(sk); 2210 out_free: 2211 if (slab != NULL) 2212 kmem_cache_free(slab, sk); 2213 else 2214 kfree(sk); 2215 return NULL; 2216 } 2217 2218 static void sk_prot_free(struct proto *prot, struct sock *sk) 2219 { 2220 struct kmem_cache *slab; 2221 struct module *owner; 2222 2223 owner = prot->owner; 2224 slab = prot->slab; 2225 2226 cgroup_sk_free(&sk->sk_cgrp_data); 2227 mem_cgroup_sk_free(sk); 2228 security_sk_free(sk); 2229 if (slab != NULL) 2230 kmem_cache_free(slab, sk); 2231 else 2232 kfree(sk); 2233 module_put(owner); 2234 } 2235 2236 /** 2237 * sk_alloc - All socket objects are allocated here 2238 * @net: the applicable net namespace 2239 * @family: protocol family 2240 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2241 * @prot: struct proto associated with this new sock instance 2242 * @kern: is this to be a kernel socket? 2243 */ 2244 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2245 struct proto *prot, int kern) 2246 { 2247 struct sock *sk; 2248 2249 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2250 if (sk) { 2251 sk->sk_family = family; 2252 /* 2253 * See comment in struct sock definition to understand 2254 * why we need sk_prot_creator -acme 2255 */ 2256 sk->sk_prot = sk->sk_prot_creator = prot; 2257 sk->sk_kern_sock = kern; 2258 sock_lock_init(sk); 2259 sk->sk_net_refcnt = kern ? 0 : 1; 2260 if (likely(sk->sk_net_refcnt)) { 2261 get_net_track(net, &sk->ns_tracker, priority); 2262 sock_inuse_add(net, 1); 2263 } else { 2264 __netns_tracker_alloc(net, &sk->ns_tracker, 2265 false, priority); 2266 } 2267 2268 sock_net_set(sk, net); 2269 refcount_set(&sk->sk_wmem_alloc, 1); 2270 2271 mem_cgroup_sk_alloc(sk); 2272 cgroup_sk_alloc(&sk->sk_cgrp_data); 2273 sock_update_classid(&sk->sk_cgrp_data); 2274 sock_update_netprioidx(&sk->sk_cgrp_data); 2275 sk_tx_queue_clear(sk); 2276 } 2277 2278 return sk; 2279 } 2280 EXPORT_SYMBOL(sk_alloc); 2281 2282 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2283 * grace period. This is the case for UDP sockets and TCP listeners. 2284 */ 2285 static void __sk_destruct(struct rcu_head *head) 2286 { 2287 struct sock *sk = container_of(head, struct sock, sk_rcu); 2288 struct sk_filter *filter; 2289 2290 if (sk->sk_destruct) 2291 sk->sk_destruct(sk); 2292 2293 filter = rcu_dereference_check(sk->sk_filter, 2294 refcount_read(&sk->sk_wmem_alloc) == 0); 2295 if (filter) { 2296 sk_filter_uncharge(sk, filter); 2297 RCU_INIT_POINTER(sk->sk_filter, NULL); 2298 } 2299 2300 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2301 2302 #ifdef CONFIG_BPF_SYSCALL 2303 bpf_sk_storage_free(sk); 2304 #endif 2305 2306 if (atomic_read(&sk->sk_omem_alloc)) 2307 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2308 __func__, atomic_read(&sk->sk_omem_alloc)); 2309 2310 if (sk->sk_frag.page) { 2311 put_page(sk->sk_frag.page); 2312 sk->sk_frag.page = NULL; 2313 } 2314 2315 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2316 put_cred(sk->sk_peer_cred); 2317 put_pid(sk->sk_peer_pid); 2318 2319 if (likely(sk->sk_net_refcnt)) 2320 put_net_track(sock_net(sk), &sk->ns_tracker); 2321 else 2322 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2323 2324 sk_prot_free(sk->sk_prot_creator, sk); 2325 } 2326 2327 void sk_destruct(struct sock *sk) 2328 { 2329 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2330 2331 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2332 reuseport_detach_sock(sk); 2333 use_call_rcu = true; 2334 } 2335 2336 if (use_call_rcu) 2337 call_rcu(&sk->sk_rcu, __sk_destruct); 2338 else 2339 __sk_destruct(&sk->sk_rcu); 2340 } 2341 2342 static void __sk_free(struct sock *sk) 2343 { 2344 if (likely(sk->sk_net_refcnt)) 2345 sock_inuse_add(sock_net(sk), -1); 2346 2347 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2348 sock_diag_broadcast_destroy(sk); 2349 else 2350 sk_destruct(sk); 2351 } 2352 2353 void sk_free(struct sock *sk) 2354 { 2355 /* 2356 * We subtract one from sk_wmem_alloc and can know if 2357 * some packets are still in some tx queue. 2358 * If not null, sock_wfree() will call __sk_free(sk) later 2359 */ 2360 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2361 __sk_free(sk); 2362 } 2363 EXPORT_SYMBOL(sk_free); 2364 2365 static void sk_init_common(struct sock *sk) 2366 { 2367 skb_queue_head_init(&sk->sk_receive_queue); 2368 skb_queue_head_init(&sk->sk_write_queue); 2369 skb_queue_head_init(&sk->sk_error_queue); 2370 2371 rwlock_init(&sk->sk_callback_lock); 2372 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2373 af_rlock_keys + sk->sk_family, 2374 af_family_rlock_key_strings[sk->sk_family]); 2375 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2376 af_wlock_keys + sk->sk_family, 2377 af_family_wlock_key_strings[sk->sk_family]); 2378 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2379 af_elock_keys + sk->sk_family, 2380 af_family_elock_key_strings[sk->sk_family]); 2381 if (sk->sk_kern_sock) 2382 lockdep_set_class_and_name(&sk->sk_callback_lock, 2383 af_kern_callback_keys + sk->sk_family, 2384 af_family_kern_clock_key_strings[sk->sk_family]); 2385 else 2386 lockdep_set_class_and_name(&sk->sk_callback_lock, 2387 af_callback_keys + sk->sk_family, 2388 af_family_clock_key_strings[sk->sk_family]); 2389 } 2390 2391 /** 2392 * sk_clone_lock - clone a socket, and lock its clone 2393 * @sk: the socket to clone 2394 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2395 * 2396 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2397 */ 2398 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2399 { 2400 struct proto *prot = READ_ONCE(sk->sk_prot); 2401 struct sk_filter *filter; 2402 bool is_charged = true; 2403 struct sock *newsk; 2404 2405 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2406 if (!newsk) 2407 goto out; 2408 2409 sock_copy(newsk, sk); 2410 2411 newsk->sk_prot_creator = prot; 2412 2413 /* SANITY */ 2414 if (likely(newsk->sk_net_refcnt)) { 2415 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2416 sock_inuse_add(sock_net(newsk), 1); 2417 } else { 2418 /* Kernel sockets are not elevating the struct net refcount. 2419 * Instead, use a tracker to more easily detect if a layer 2420 * is not properly dismantling its kernel sockets at netns 2421 * destroy time. 2422 */ 2423 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2424 false, priority); 2425 } 2426 sk_node_init(&newsk->sk_node); 2427 sock_lock_init(newsk); 2428 bh_lock_sock(newsk); 2429 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2430 newsk->sk_backlog.len = 0; 2431 2432 atomic_set(&newsk->sk_rmem_alloc, 0); 2433 2434 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2435 refcount_set(&newsk->sk_wmem_alloc, 1); 2436 2437 atomic_set(&newsk->sk_omem_alloc, 0); 2438 sk_init_common(newsk); 2439 2440 newsk->sk_dst_cache = NULL; 2441 newsk->sk_dst_pending_confirm = 0; 2442 newsk->sk_wmem_queued = 0; 2443 newsk->sk_forward_alloc = 0; 2444 newsk->sk_reserved_mem = 0; 2445 atomic_set(&newsk->sk_drops, 0); 2446 newsk->sk_send_head = NULL; 2447 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2448 atomic_set(&newsk->sk_zckey, 0); 2449 2450 sock_reset_flag(newsk, SOCK_DONE); 2451 2452 /* sk->sk_memcg will be populated at accept() time */ 2453 newsk->sk_memcg = NULL; 2454 2455 cgroup_sk_clone(&newsk->sk_cgrp_data); 2456 2457 rcu_read_lock(); 2458 filter = rcu_dereference(sk->sk_filter); 2459 if (filter != NULL) 2460 /* though it's an empty new sock, the charging may fail 2461 * if sysctl_optmem_max was changed between creation of 2462 * original socket and cloning 2463 */ 2464 is_charged = sk_filter_charge(newsk, filter); 2465 RCU_INIT_POINTER(newsk->sk_filter, filter); 2466 rcu_read_unlock(); 2467 2468 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2469 /* We need to make sure that we don't uncharge the new 2470 * socket if we couldn't charge it in the first place 2471 * as otherwise we uncharge the parent's filter. 2472 */ 2473 if (!is_charged) 2474 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2475 sk_free_unlock_clone(newsk); 2476 newsk = NULL; 2477 goto out; 2478 } 2479 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2480 2481 if (bpf_sk_storage_clone(sk, newsk)) { 2482 sk_free_unlock_clone(newsk); 2483 newsk = NULL; 2484 goto out; 2485 } 2486 2487 /* Clear sk_user_data if parent had the pointer tagged 2488 * as not suitable for copying when cloning. 2489 */ 2490 if (sk_user_data_is_nocopy(newsk)) 2491 newsk->sk_user_data = NULL; 2492 2493 newsk->sk_err = 0; 2494 newsk->sk_err_soft = 0; 2495 newsk->sk_priority = 0; 2496 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2497 2498 /* Before updating sk_refcnt, we must commit prior changes to memory 2499 * (Documentation/RCU/rculist_nulls.rst for details) 2500 */ 2501 smp_wmb(); 2502 refcount_set(&newsk->sk_refcnt, 2); 2503 2504 sk_set_socket(newsk, NULL); 2505 sk_tx_queue_clear(newsk); 2506 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2507 2508 if (newsk->sk_prot->sockets_allocated) 2509 sk_sockets_allocated_inc(newsk); 2510 2511 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2512 net_enable_timestamp(); 2513 out: 2514 return newsk; 2515 } 2516 EXPORT_SYMBOL_GPL(sk_clone_lock); 2517 2518 void sk_free_unlock_clone(struct sock *sk) 2519 { 2520 /* It is still raw copy of parent, so invalidate 2521 * destructor and make plain sk_free() */ 2522 sk->sk_destruct = NULL; 2523 bh_unlock_sock(sk); 2524 sk_free(sk); 2525 } 2526 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2527 2528 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2529 { 2530 bool is_ipv6 = false; 2531 u32 max_size; 2532 2533 #if IS_ENABLED(CONFIG_IPV6) 2534 is_ipv6 = (sk->sk_family == AF_INET6 && 2535 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2536 #endif 2537 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2538 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2539 READ_ONCE(dst->dev->gso_ipv4_max_size); 2540 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2541 max_size = GSO_LEGACY_MAX_SIZE; 2542 2543 return max_size - (MAX_TCP_HEADER + 1); 2544 } 2545 2546 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2547 { 2548 u32 max_segs = 1; 2549 2550 sk->sk_route_caps = dst->dev->features; 2551 if (sk_is_tcp(sk)) 2552 sk->sk_route_caps |= NETIF_F_GSO; 2553 if (sk->sk_route_caps & NETIF_F_GSO) 2554 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2555 if (unlikely(sk->sk_gso_disabled)) 2556 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2557 if (sk_can_gso(sk)) { 2558 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2559 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2560 } else { 2561 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2562 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2563 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2564 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2565 } 2566 } 2567 sk->sk_gso_max_segs = max_segs; 2568 sk_dst_set(sk, dst); 2569 } 2570 EXPORT_SYMBOL_GPL(sk_setup_caps); 2571 2572 /* 2573 * Simple resource managers for sockets. 2574 */ 2575 2576 2577 /* 2578 * Write buffer destructor automatically called from kfree_skb. 2579 */ 2580 void sock_wfree(struct sk_buff *skb) 2581 { 2582 struct sock *sk = skb->sk; 2583 unsigned int len = skb->truesize; 2584 bool free; 2585 2586 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2587 if (sock_flag(sk, SOCK_RCU_FREE) && 2588 sk->sk_write_space == sock_def_write_space) { 2589 rcu_read_lock(); 2590 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2591 sock_def_write_space_wfree(sk); 2592 rcu_read_unlock(); 2593 if (unlikely(free)) 2594 __sk_free(sk); 2595 return; 2596 } 2597 2598 /* 2599 * Keep a reference on sk_wmem_alloc, this will be released 2600 * after sk_write_space() call 2601 */ 2602 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2603 sk->sk_write_space(sk); 2604 len = 1; 2605 } 2606 /* 2607 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2608 * could not do because of in-flight packets 2609 */ 2610 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2611 __sk_free(sk); 2612 } 2613 EXPORT_SYMBOL(sock_wfree); 2614 2615 /* This variant of sock_wfree() is used by TCP, 2616 * since it sets SOCK_USE_WRITE_QUEUE. 2617 */ 2618 void __sock_wfree(struct sk_buff *skb) 2619 { 2620 struct sock *sk = skb->sk; 2621 2622 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2623 __sk_free(sk); 2624 } 2625 2626 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2627 { 2628 skb_orphan(skb); 2629 #ifdef CONFIG_INET 2630 if (unlikely(!sk_fullsock(sk))) 2631 return skb_set_owner_edemux(skb, sk); 2632 #endif 2633 skb->sk = sk; 2634 skb->destructor = sock_wfree; 2635 skb_set_hash_from_sk(skb, sk); 2636 /* 2637 * We used to take a refcount on sk, but following operation 2638 * is enough to guarantee sk_free() won't free this sock until 2639 * all in-flight packets are completed 2640 */ 2641 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2642 } 2643 EXPORT_SYMBOL(skb_set_owner_w); 2644 2645 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2646 { 2647 /* Drivers depend on in-order delivery for crypto offload, 2648 * partial orphan breaks out-of-order-OK logic. 2649 */ 2650 if (skb_is_decrypted(skb)) 2651 return false; 2652 2653 return (skb->destructor == sock_wfree || 2654 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2655 } 2656 2657 /* This helper is used by netem, as it can hold packets in its 2658 * delay queue. We want to allow the owner socket to send more 2659 * packets, as if they were already TX completed by a typical driver. 2660 * But we also want to keep skb->sk set because some packet schedulers 2661 * rely on it (sch_fq for example). 2662 */ 2663 void skb_orphan_partial(struct sk_buff *skb) 2664 { 2665 if (skb_is_tcp_pure_ack(skb)) 2666 return; 2667 2668 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2669 return; 2670 2671 skb_orphan(skb); 2672 } 2673 EXPORT_SYMBOL(skb_orphan_partial); 2674 2675 /* 2676 * Read buffer destructor automatically called from kfree_skb. 2677 */ 2678 void sock_rfree(struct sk_buff *skb) 2679 { 2680 struct sock *sk = skb->sk; 2681 unsigned int len = skb->truesize; 2682 2683 atomic_sub(len, &sk->sk_rmem_alloc); 2684 sk_mem_uncharge(sk, len); 2685 } 2686 EXPORT_SYMBOL(sock_rfree); 2687 2688 /* 2689 * Buffer destructor for skbs that are not used directly in read or write 2690 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2691 */ 2692 void sock_efree(struct sk_buff *skb) 2693 { 2694 sock_put(skb->sk); 2695 } 2696 EXPORT_SYMBOL(sock_efree); 2697 2698 /* Buffer destructor for prefetch/receive path where reference count may 2699 * not be held, e.g. for listen sockets. 2700 */ 2701 #ifdef CONFIG_INET 2702 void sock_pfree(struct sk_buff *skb) 2703 { 2704 struct sock *sk = skb->sk; 2705 2706 if (!sk_is_refcounted(sk)) 2707 return; 2708 2709 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2710 inet_reqsk(sk)->rsk_listener = NULL; 2711 reqsk_free(inet_reqsk(sk)); 2712 return; 2713 } 2714 2715 sock_gen_put(sk); 2716 } 2717 EXPORT_SYMBOL(sock_pfree); 2718 #endif /* CONFIG_INET */ 2719 2720 kuid_t sock_i_uid(struct sock *sk) 2721 { 2722 kuid_t uid; 2723 2724 read_lock_bh(&sk->sk_callback_lock); 2725 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2726 read_unlock_bh(&sk->sk_callback_lock); 2727 return uid; 2728 } 2729 EXPORT_SYMBOL(sock_i_uid); 2730 2731 unsigned long __sock_i_ino(struct sock *sk) 2732 { 2733 unsigned long ino; 2734 2735 read_lock(&sk->sk_callback_lock); 2736 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2737 read_unlock(&sk->sk_callback_lock); 2738 return ino; 2739 } 2740 EXPORT_SYMBOL(__sock_i_ino); 2741 2742 unsigned long sock_i_ino(struct sock *sk) 2743 { 2744 unsigned long ino; 2745 2746 local_bh_disable(); 2747 ino = __sock_i_ino(sk); 2748 local_bh_enable(); 2749 return ino; 2750 } 2751 EXPORT_SYMBOL(sock_i_ino); 2752 2753 /* 2754 * Allocate a skb from the socket's send buffer. 2755 */ 2756 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2757 gfp_t priority) 2758 { 2759 if (force || 2760 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2761 struct sk_buff *skb = alloc_skb(size, priority); 2762 2763 if (skb) { 2764 skb_set_owner_w(skb, sk); 2765 return skb; 2766 } 2767 } 2768 return NULL; 2769 } 2770 EXPORT_SYMBOL(sock_wmalloc); 2771 2772 static void sock_ofree(struct sk_buff *skb) 2773 { 2774 struct sock *sk = skb->sk; 2775 2776 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2777 } 2778 2779 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2780 gfp_t priority) 2781 { 2782 struct sk_buff *skb; 2783 2784 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2785 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2786 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2787 return NULL; 2788 2789 skb = alloc_skb(size, priority); 2790 if (!skb) 2791 return NULL; 2792 2793 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2794 skb->sk = sk; 2795 skb->destructor = sock_ofree; 2796 return skb; 2797 } 2798 2799 /* 2800 * Allocate a memory block from the socket's option memory buffer. 2801 */ 2802 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2803 { 2804 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2805 2806 if ((unsigned int)size <= optmem_max && 2807 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2808 void *mem; 2809 /* First do the add, to avoid the race if kmalloc 2810 * might sleep. 2811 */ 2812 atomic_add(size, &sk->sk_omem_alloc); 2813 mem = kmalloc(size, priority); 2814 if (mem) 2815 return mem; 2816 atomic_sub(size, &sk->sk_omem_alloc); 2817 } 2818 return NULL; 2819 } 2820 EXPORT_SYMBOL(sock_kmalloc); 2821 2822 /* Free an option memory block. Note, we actually want the inline 2823 * here as this allows gcc to detect the nullify and fold away the 2824 * condition entirely. 2825 */ 2826 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2827 const bool nullify) 2828 { 2829 if (WARN_ON_ONCE(!mem)) 2830 return; 2831 if (nullify) 2832 kfree_sensitive(mem); 2833 else 2834 kfree(mem); 2835 atomic_sub(size, &sk->sk_omem_alloc); 2836 } 2837 2838 void sock_kfree_s(struct sock *sk, void *mem, int size) 2839 { 2840 __sock_kfree_s(sk, mem, size, false); 2841 } 2842 EXPORT_SYMBOL(sock_kfree_s); 2843 2844 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2845 { 2846 __sock_kfree_s(sk, mem, size, true); 2847 } 2848 EXPORT_SYMBOL(sock_kzfree_s); 2849 2850 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2851 I think, these locks should be removed for datagram sockets. 2852 */ 2853 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2854 { 2855 DEFINE_WAIT(wait); 2856 2857 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2858 for (;;) { 2859 if (!timeo) 2860 break; 2861 if (signal_pending(current)) 2862 break; 2863 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2864 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2865 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2866 break; 2867 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2868 break; 2869 if (READ_ONCE(sk->sk_err)) 2870 break; 2871 timeo = schedule_timeout(timeo); 2872 } 2873 finish_wait(sk_sleep(sk), &wait); 2874 return timeo; 2875 } 2876 2877 2878 /* 2879 * Generic send/receive buffer handlers 2880 */ 2881 2882 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2883 unsigned long data_len, int noblock, 2884 int *errcode, int max_page_order) 2885 { 2886 struct sk_buff *skb; 2887 long timeo; 2888 int err; 2889 2890 timeo = sock_sndtimeo(sk, noblock); 2891 for (;;) { 2892 err = sock_error(sk); 2893 if (err != 0) 2894 goto failure; 2895 2896 err = -EPIPE; 2897 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2898 goto failure; 2899 2900 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2901 break; 2902 2903 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2904 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2905 err = -EAGAIN; 2906 if (!timeo) 2907 goto failure; 2908 if (signal_pending(current)) 2909 goto interrupted; 2910 timeo = sock_wait_for_wmem(sk, timeo); 2911 } 2912 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2913 errcode, sk->sk_allocation); 2914 if (skb) 2915 skb_set_owner_w(skb, sk); 2916 return skb; 2917 2918 interrupted: 2919 err = sock_intr_errno(timeo); 2920 failure: 2921 *errcode = err; 2922 return NULL; 2923 } 2924 EXPORT_SYMBOL(sock_alloc_send_pskb); 2925 2926 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2927 struct sockcm_cookie *sockc) 2928 { 2929 u32 tsflags; 2930 2931 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 2932 2933 switch (cmsg->cmsg_type) { 2934 case SO_MARK: 2935 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2936 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2937 return -EPERM; 2938 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2939 return -EINVAL; 2940 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2941 break; 2942 case SO_TIMESTAMPING_OLD: 2943 case SO_TIMESTAMPING_NEW: 2944 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2945 return -EINVAL; 2946 2947 tsflags = *(u32 *)CMSG_DATA(cmsg); 2948 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2949 return -EINVAL; 2950 2951 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2952 sockc->tsflags |= tsflags; 2953 break; 2954 case SCM_TXTIME: 2955 if (!sock_flag(sk, SOCK_TXTIME)) 2956 return -EINVAL; 2957 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2958 return -EINVAL; 2959 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2960 break; 2961 case SCM_TS_OPT_ID: 2962 if (sk_is_tcp(sk)) 2963 return -EINVAL; 2964 tsflags = READ_ONCE(sk->sk_tsflags); 2965 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 2966 return -EINVAL; 2967 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2968 return -EINVAL; 2969 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 2970 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 2971 break; 2972 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2973 case SCM_RIGHTS: 2974 case SCM_CREDENTIALS: 2975 break; 2976 case SO_PRIORITY: 2977 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2978 return -EINVAL; 2979 if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg))) 2980 return -EPERM; 2981 sockc->priority = *(u32 *)CMSG_DATA(cmsg); 2982 break; 2983 default: 2984 return -EINVAL; 2985 } 2986 return 0; 2987 } 2988 EXPORT_SYMBOL(__sock_cmsg_send); 2989 2990 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2991 struct sockcm_cookie *sockc) 2992 { 2993 struct cmsghdr *cmsg; 2994 int ret; 2995 2996 for_each_cmsghdr(cmsg, msg) { 2997 if (!CMSG_OK(msg, cmsg)) 2998 return -EINVAL; 2999 if (cmsg->cmsg_level != SOL_SOCKET) 3000 continue; 3001 ret = __sock_cmsg_send(sk, cmsg, sockc); 3002 if (ret) 3003 return ret; 3004 } 3005 return 0; 3006 } 3007 EXPORT_SYMBOL(sock_cmsg_send); 3008 3009 static void sk_enter_memory_pressure(struct sock *sk) 3010 { 3011 if (!sk->sk_prot->enter_memory_pressure) 3012 return; 3013 3014 sk->sk_prot->enter_memory_pressure(sk); 3015 } 3016 3017 static void sk_leave_memory_pressure(struct sock *sk) 3018 { 3019 if (sk->sk_prot->leave_memory_pressure) { 3020 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 3021 tcp_leave_memory_pressure, sk); 3022 } else { 3023 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 3024 3025 if (memory_pressure && READ_ONCE(*memory_pressure)) 3026 WRITE_ONCE(*memory_pressure, 0); 3027 } 3028 } 3029 3030 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3031 3032 /** 3033 * skb_page_frag_refill - check that a page_frag contains enough room 3034 * @sz: minimum size of the fragment we want to get 3035 * @pfrag: pointer to page_frag 3036 * @gfp: priority for memory allocation 3037 * 3038 * Note: While this allocator tries to use high order pages, there is 3039 * no guarantee that allocations succeed. Therefore, @sz MUST be 3040 * less or equal than PAGE_SIZE. 3041 */ 3042 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 3043 { 3044 if (pfrag->page) { 3045 if (page_ref_count(pfrag->page) == 1) { 3046 pfrag->offset = 0; 3047 return true; 3048 } 3049 if (pfrag->offset + sz <= pfrag->size) 3050 return true; 3051 put_page(pfrag->page); 3052 } 3053 3054 pfrag->offset = 0; 3055 if (SKB_FRAG_PAGE_ORDER && 3056 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3057 /* Avoid direct reclaim but allow kswapd to wake */ 3058 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3059 __GFP_COMP | __GFP_NOWARN | 3060 __GFP_NORETRY, 3061 SKB_FRAG_PAGE_ORDER); 3062 if (likely(pfrag->page)) { 3063 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3064 return true; 3065 } 3066 } 3067 pfrag->page = alloc_page(gfp); 3068 if (likely(pfrag->page)) { 3069 pfrag->size = PAGE_SIZE; 3070 return true; 3071 } 3072 return false; 3073 } 3074 EXPORT_SYMBOL(skb_page_frag_refill); 3075 3076 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3077 { 3078 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3079 return true; 3080 3081 sk_enter_memory_pressure(sk); 3082 sk_stream_moderate_sndbuf(sk); 3083 return false; 3084 } 3085 EXPORT_SYMBOL(sk_page_frag_refill); 3086 3087 void __lock_sock(struct sock *sk) 3088 __releases(&sk->sk_lock.slock) 3089 __acquires(&sk->sk_lock.slock) 3090 { 3091 DEFINE_WAIT(wait); 3092 3093 for (;;) { 3094 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3095 TASK_UNINTERRUPTIBLE); 3096 spin_unlock_bh(&sk->sk_lock.slock); 3097 schedule(); 3098 spin_lock_bh(&sk->sk_lock.slock); 3099 if (!sock_owned_by_user(sk)) 3100 break; 3101 } 3102 finish_wait(&sk->sk_lock.wq, &wait); 3103 } 3104 3105 void __release_sock(struct sock *sk) 3106 __releases(&sk->sk_lock.slock) 3107 __acquires(&sk->sk_lock.slock) 3108 { 3109 struct sk_buff *skb, *next; 3110 3111 while ((skb = sk->sk_backlog.head) != NULL) { 3112 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3113 3114 spin_unlock_bh(&sk->sk_lock.slock); 3115 3116 do { 3117 next = skb->next; 3118 prefetch(next); 3119 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3120 skb_mark_not_on_list(skb); 3121 sk_backlog_rcv(sk, skb); 3122 3123 cond_resched(); 3124 3125 skb = next; 3126 } while (skb != NULL); 3127 3128 spin_lock_bh(&sk->sk_lock.slock); 3129 } 3130 3131 /* 3132 * Doing the zeroing here guarantee we can not loop forever 3133 * while a wild producer attempts to flood us. 3134 */ 3135 sk->sk_backlog.len = 0; 3136 } 3137 3138 void __sk_flush_backlog(struct sock *sk) 3139 { 3140 spin_lock_bh(&sk->sk_lock.slock); 3141 __release_sock(sk); 3142 3143 if (sk->sk_prot->release_cb) 3144 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3145 tcp_release_cb, sk); 3146 3147 spin_unlock_bh(&sk->sk_lock.slock); 3148 } 3149 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3150 3151 /** 3152 * sk_wait_data - wait for data to arrive at sk_receive_queue 3153 * @sk: sock to wait on 3154 * @timeo: for how long 3155 * @skb: last skb seen on sk_receive_queue 3156 * 3157 * Now socket state including sk->sk_err is changed only under lock, 3158 * hence we may omit checks after joining wait queue. 3159 * We check receive queue before schedule() only as optimization; 3160 * it is very likely that release_sock() added new data. 3161 */ 3162 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3163 { 3164 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3165 int rc; 3166 3167 add_wait_queue(sk_sleep(sk), &wait); 3168 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3169 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3170 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3171 remove_wait_queue(sk_sleep(sk), &wait); 3172 return rc; 3173 } 3174 EXPORT_SYMBOL(sk_wait_data); 3175 3176 /** 3177 * __sk_mem_raise_allocated - increase memory_allocated 3178 * @sk: socket 3179 * @size: memory size to allocate 3180 * @amt: pages to allocate 3181 * @kind: allocation type 3182 * 3183 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3184 * 3185 * Unlike the globally shared limits among the sockets under same protocol, 3186 * consuming the budget of a memcg won't have direct effect on other ones. 3187 * So be optimistic about memcg's tolerance, and leave the callers to decide 3188 * whether or not to raise allocated through sk_under_memory_pressure() or 3189 * its variants. 3190 */ 3191 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3192 { 3193 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL; 3194 struct proto *prot = sk->sk_prot; 3195 bool charged = false; 3196 long allocated; 3197 3198 sk_memory_allocated_add(sk, amt); 3199 allocated = sk_memory_allocated(sk); 3200 3201 if (memcg) { 3202 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge())) 3203 goto suppress_allocation; 3204 charged = true; 3205 } 3206 3207 /* Under limit. */ 3208 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3209 sk_leave_memory_pressure(sk); 3210 return 1; 3211 } 3212 3213 /* Under pressure. */ 3214 if (allocated > sk_prot_mem_limits(sk, 1)) 3215 sk_enter_memory_pressure(sk); 3216 3217 /* Over hard limit. */ 3218 if (allocated > sk_prot_mem_limits(sk, 2)) 3219 goto suppress_allocation; 3220 3221 /* Guarantee minimum buffer size under pressure (either global 3222 * or memcg) to make sure features described in RFC 7323 (TCP 3223 * Extensions for High Performance) work properly. 3224 * 3225 * This rule does NOT stand when exceeds global or memcg's hard 3226 * limit, or else a DoS attack can be taken place by spawning 3227 * lots of sockets whose usage are under minimum buffer size. 3228 */ 3229 if (kind == SK_MEM_RECV) { 3230 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3231 return 1; 3232 3233 } else { /* SK_MEM_SEND */ 3234 int wmem0 = sk_get_wmem0(sk, prot); 3235 3236 if (sk->sk_type == SOCK_STREAM) { 3237 if (sk->sk_wmem_queued < wmem0) 3238 return 1; 3239 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3240 return 1; 3241 } 3242 } 3243 3244 if (sk_has_memory_pressure(sk)) { 3245 u64 alloc; 3246 3247 /* The following 'average' heuristic is within the 3248 * scope of global accounting, so it only makes 3249 * sense for global memory pressure. 3250 */ 3251 if (!sk_under_global_memory_pressure(sk)) 3252 return 1; 3253 3254 /* Try to be fair among all the sockets under global 3255 * pressure by allowing the ones that below average 3256 * usage to raise. 3257 */ 3258 alloc = sk_sockets_allocated_read_positive(sk); 3259 if (sk_prot_mem_limits(sk, 2) > alloc * 3260 sk_mem_pages(sk->sk_wmem_queued + 3261 atomic_read(&sk->sk_rmem_alloc) + 3262 sk->sk_forward_alloc)) 3263 return 1; 3264 } 3265 3266 suppress_allocation: 3267 3268 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3269 sk_stream_moderate_sndbuf(sk); 3270 3271 /* Fail only if socket is _under_ its sndbuf. 3272 * In this case we cannot block, so that we have to fail. 3273 */ 3274 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3275 /* Force charge with __GFP_NOFAIL */ 3276 if (memcg && !charged) { 3277 mem_cgroup_charge_skmem(memcg, amt, 3278 gfp_memcg_charge() | __GFP_NOFAIL); 3279 } 3280 return 1; 3281 } 3282 } 3283 3284 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3285 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3286 3287 sk_memory_allocated_sub(sk, amt); 3288 3289 if (charged) 3290 mem_cgroup_uncharge_skmem(memcg, amt); 3291 3292 return 0; 3293 } 3294 3295 /** 3296 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3297 * @sk: socket 3298 * @size: memory size to allocate 3299 * @kind: allocation type 3300 * 3301 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3302 * rmem allocation. This function assumes that protocols which have 3303 * memory_pressure use sk_wmem_queued as write buffer accounting. 3304 */ 3305 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3306 { 3307 int ret, amt = sk_mem_pages(size); 3308 3309 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3310 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3311 if (!ret) 3312 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3313 return ret; 3314 } 3315 EXPORT_SYMBOL(__sk_mem_schedule); 3316 3317 /** 3318 * __sk_mem_reduce_allocated - reclaim memory_allocated 3319 * @sk: socket 3320 * @amount: number of quanta 3321 * 3322 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3323 */ 3324 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3325 { 3326 sk_memory_allocated_sub(sk, amount); 3327 3328 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3329 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3330 3331 if (sk_under_global_memory_pressure(sk) && 3332 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3333 sk_leave_memory_pressure(sk); 3334 } 3335 3336 /** 3337 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3338 * @sk: socket 3339 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3340 */ 3341 void __sk_mem_reclaim(struct sock *sk, int amount) 3342 { 3343 amount >>= PAGE_SHIFT; 3344 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3345 __sk_mem_reduce_allocated(sk, amount); 3346 } 3347 EXPORT_SYMBOL(__sk_mem_reclaim); 3348 3349 int sk_set_peek_off(struct sock *sk, int val) 3350 { 3351 WRITE_ONCE(sk->sk_peek_off, val); 3352 return 0; 3353 } 3354 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3355 3356 /* 3357 * Set of default routines for initialising struct proto_ops when 3358 * the protocol does not support a particular function. In certain 3359 * cases where it makes no sense for a protocol to have a "do nothing" 3360 * function, some default processing is provided. 3361 */ 3362 3363 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3364 { 3365 return -EOPNOTSUPP; 3366 } 3367 EXPORT_SYMBOL(sock_no_bind); 3368 3369 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3370 int len, int flags) 3371 { 3372 return -EOPNOTSUPP; 3373 } 3374 EXPORT_SYMBOL(sock_no_connect); 3375 3376 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3377 { 3378 return -EOPNOTSUPP; 3379 } 3380 EXPORT_SYMBOL(sock_no_socketpair); 3381 3382 int sock_no_accept(struct socket *sock, struct socket *newsock, 3383 struct proto_accept_arg *arg) 3384 { 3385 return -EOPNOTSUPP; 3386 } 3387 EXPORT_SYMBOL(sock_no_accept); 3388 3389 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3390 int peer) 3391 { 3392 return -EOPNOTSUPP; 3393 } 3394 EXPORT_SYMBOL(sock_no_getname); 3395 3396 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3397 { 3398 return -EOPNOTSUPP; 3399 } 3400 EXPORT_SYMBOL(sock_no_ioctl); 3401 3402 int sock_no_listen(struct socket *sock, int backlog) 3403 { 3404 return -EOPNOTSUPP; 3405 } 3406 EXPORT_SYMBOL(sock_no_listen); 3407 3408 int sock_no_shutdown(struct socket *sock, int how) 3409 { 3410 return -EOPNOTSUPP; 3411 } 3412 EXPORT_SYMBOL(sock_no_shutdown); 3413 3414 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3415 { 3416 return -EOPNOTSUPP; 3417 } 3418 EXPORT_SYMBOL(sock_no_sendmsg); 3419 3420 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3421 { 3422 return -EOPNOTSUPP; 3423 } 3424 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3425 3426 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3427 int flags) 3428 { 3429 return -EOPNOTSUPP; 3430 } 3431 EXPORT_SYMBOL(sock_no_recvmsg); 3432 3433 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3434 { 3435 /* Mirror missing mmap method error code */ 3436 return -ENODEV; 3437 } 3438 EXPORT_SYMBOL(sock_no_mmap); 3439 3440 /* 3441 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3442 * various sock-based usage counts. 3443 */ 3444 void __receive_sock(struct file *file) 3445 { 3446 struct socket *sock; 3447 3448 sock = sock_from_file(file); 3449 if (sock) { 3450 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3451 sock_update_classid(&sock->sk->sk_cgrp_data); 3452 } 3453 } 3454 3455 /* 3456 * Default Socket Callbacks 3457 */ 3458 3459 static void sock_def_wakeup(struct sock *sk) 3460 { 3461 struct socket_wq *wq; 3462 3463 rcu_read_lock(); 3464 wq = rcu_dereference(sk->sk_wq); 3465 if (skwq_has_sleeper(wq)) 3466 wake_up_interruptible_all(&wq->wait); 3467 rcu_read_unlock(); 3468 } 3469 3470 static void sock_def_error_report(struct sock *sk) 3471 { 3472 struct socket_wq *wq; 3473 3474 rcu_read_lock(); 3475 wq = rcu_dereference(sk->sk_wq); 3476 if (skwq_has_sleeper(wq)) 3477 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3478 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3479 rcu_read_unlock(); 3480 } 3481 3482 void sock_def_readable(struct sock *sk) 3483 { 3484 struct socket_wq *wq; 3485 3486 trace_sk_data_ready(sk); 3487 3488 rcu_read_lock(); 3489 wq = rcu_dereference(sk->sk_wq); 3490 if (skwq_has_sleeper(wq)) 3491 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3492 EPOLLRDNORM | EPOLLRDBAND); 3493 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3494 rcu_read_unlock(); 3495 } 3496 3497 static void sock_def_write_space(struct sock *sk) 3498 { 3499 struct socket_wq *wq; 3500 3501 rcu_read_lock(); 3502 3503 /* Do not wake up a writer until he can make "significant" 3504 * progress. --DaveM 3505 */ 3506 if (sock_writeable(sk)) { 3507 wq = rcu_dereference(sk->sk_wq); 3508 if (skwq_has_sleeper(wq)) 3509 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3510 EPOLLWRNORM | EPOLLWRBAND); 3511 3512 /* Should agree with poll, otherwise some programs break */ 3513 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3514 } 3515 3516 rcu_read_unlock(); 3517 } 3518 3519 /* An optimised version of sock_def_write_space(), should only be called 3520 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3521 * ->sk_wmem_alloc. 3522 */ 3523 static void sock_def_write_space_wfree(struct sock *sk) 3524 { 3525 /* Do not wake up a writer until he can make "significant" 3526 * progress. --DaveM 3527 */ 3528 if (sock_writeable(sk)) { 3529 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3530 3531 /* rely on refcount_sub from sock_wfree() */ 3532 smp_mb__after_atomic(); 3533 if (wq && waitqueue_active(&wq->wait)) 3534 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3535 EPOLLWRNORM | EPOLLWRBAND); 3536 3537 /* Should agree with poll, otherwise some programs break */ 3538 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3539 } 3540 } 3541 3542 static void sock_def_destruct(struct sock *sk) 3543 { 3544 } 3545 3546 void sk_send_sigurg(struct sock *sk) 3547 { 3548 if (sk->sk_socket && sk->sk_socket->file) 3549 if (send_sigurg(sk->sk_socket->file)) 3550 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3551 } 3552 EXPORT_SYMBOL(sk_send_sigurg); 3553 3554 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3555 unsigned long expires) 3556 { 3557 if (!mod_timer(timer, expires)) 3558 sock_hold(sk); 3559 } 3560 EXPORT_SYMBOL(sk_reset_timer); 3561 3562 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3563 { 3564 if (del_timer(timer)) 3565 __sock_put(sk); 3566 } 3567 EXPORT_SYMBOL(sk_stop_timer); 3568 3569 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3570 { 3571 if (del_timer_sync(timer)) 3572 __sock_put(sk); 3573 } 3574 EXPORT_SYMBOL(sk_stop_timer_sync); 3575 3576 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3577 { 3578 sk_init_common(sk); 3579 sk->sk_send_head = NULL; 3580 3581 timer_setup(&sk->sk_timer, NULL, 0); 3582 3583 sk->sk_allocation = GFP_KERNEL; 3584 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3585 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3586 sk->sk_state = TCP_CLOSE; 3587 sk->sk_use_task_frag = true; 3588 sk_set_socket(sk, sock); 3589 3590 sock_set_flag(sk, SOCK_ZAPPED); 3591 3592 if (sock) { 3593 sk->sk_type = sock->type; 3594 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3595 sock->sk = sk; 3596 } else { 3597 RCU_INIT_POINTER(sk->sk_wq, NULL); 3598 } 3599 sk->sk_uid = uid; 3600 3601 sk->sk_state_change = sock_def_wakeup; 3602 sk->sk_data_ready = sock_def_readable; 3603 sk->sk_write_space = sock_def_write_space; 3604 sk->sk_error_report = sock_def_error_report; 3605 sk->sk_destruct = sock_def_destruct; 3606 3607 sk->sk_frag.page = NULL; 3608 sk->sk_frag.offset = 0; 3609 sk->sk_peek_off = -1; 3610 3611 sk->sk_peer_pid = NULL; 3612 sk->sk_peer_cred = NULL; 3613 spin_lock_init(&sk->sk_peer_lock); 3614 3615 sk->sk_write_pending = 0; 3616 sk->sk_rcvlowat = 1; 3617 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3618 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3619 3620 sk->sk_stamp = SK_DEFAULT_STAMP; 3621 #if BITS_PER_LONG==32 3622 seqlock_init(&sk->sk_stamp_seq); 3623 #endif 3624 atomic_set(&sk->sk_zckey, 0); 3625 3626 #ifdef CONFIG_NET_RX_BUSY_POLL 3627 sk->sk_napi_id = 0; 3628 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3629 #endif 3630 3631 sk->sk_max_pacing_rate = ~0UL; 3632 sk->sk_pacing_rate = ~0UL; 3633 WRITE_ONCE(sk->sk_pacing_shift, 10); 3634 sk->sk_incoming_cpu = -1; 3635 3636 sk_rx_queue_clear(sk); 3637 /* 3638 * Before updating sk_refcnt, we must commit prior changes to memory 3639 * (Documentation/RCU/rculist_nulls.rst for details) 3640 */ 3641 smp_wmb(); 3642 refcount_set(&sk->sk_refcnt, 1); 3643 atomic_set(&sk->sk_drops, 0); 3644 } 3645 EXPORT_SYMBOL(sock_init_data_uid); 3646 3647 void sock_init_data(struct socket *sock, struct sock *sk) 3648 { 3649 kuid_t uid = sock ? 3650 SOCK_INODE(sock)->i_uid : 3651 make_kuid(sock_net(sk)->user_ns, 0); 3652 3653 sock_init_data_uid(sock, sk, uid); 3654 } 3655 EXPORT_SYMBOL(sock_init_data); 3656 3657 void lock_sock_nested(struct sock *sk, int subclass) 3658 { 3659 /* The sk_lock has mutex_lock() semantics here. */ 3660 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3661 3662 might_sleep(); 3663 spin_lock_bh(&sk->sk_lock.slock); 3664 if (sock_owned_by_user_nocheck(sk)) 3665 __lock_sock(sk); 3666 sk->sk_lock.owned = 1; 3667 spin_unlock_bh(&sk->sk_lock.slock); 3668 } 3669 EXPORT_SYMBOL(lock_sock_nested); 3670 3671 void release_sock(struct sock *sk) 3672 { 3673 spin_lock_bh(&sk->sk_lock.slock); 3674 if (sk->sk_backlog.tail) 3675 __release_sock(sk); 3676 3677 if (sk->sk_prot->release_cb) 3678 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3679 tcp_release_cb, sk); 3680 3681 sock_release_ownership(sk); 3682 if (waitqueue_active(&sk->sk_lock.wq)) 3683 wake_up(&sk->sk_lock.wq); 3684 spin_unlock_bh(&sk->sk_lock.slock); 3685 } 3686 EXPORT_SYMBOL(release_sock); 3687 3688 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3689 { 3690 might_sleep(); 3691 spin_lock_bh(&sk->sk_lock.slock); 3692 3693 if (!sock_owned_by_user_nocheck(sk)) { 3694 /* 3695 * Fast path return with bottom halves disabled and 3696 * sock::sk_lock.slock held. 3697 * 3698 * The 'mutex' is not contended and holding 3699 * sock::sk_lock.slock prevents all other lockers to 3700 * proceed so the corresponding unlock_sock_fast() can 3701 * avoid the slow path of release_sock() completely and 3702 * just release slock. 3703 * 3704 * From a semantical POV this is equivalent to 'acquiring' 3705 * the 'mutex', hence the corresponding lockdep 3706 * mutex_release() has to happen in the fast path of 3707 * unlock_sock_fast(). 3708 */ 3709 return false; 3710 } 3711 3712 __lock_sock(sk); 3713 sk->sk_lock.owned = 1; 3714 __acquire(&sk->sk_lock.slock); 3715 spin_unlock_bh(&sk->sk_lock.slock); 3716 return true; 3717 } 3718 EXPORT_SYMBOL(__lock_sock_fast); 3719 3720 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3721 bool timeval, bool time32) 3722 { 3723 struct sock *sk = sock->sk; 3724 struct timespec64 ts; 3725 3726 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3727 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3728 if (ts.tv_sec == -1) 3729 return -ENOENT; 3730 if (ts.tv_sec == 0) { 3731 ktime_t kt = ktime_get_real(); 3732 sock_write_timestamp(sk, kt); 3733 ts = ktime_to_timespec64(kt); 3734 } 3735 3736 if (timeval) 3737 ts.tv_nsec /= 1000; 3738 3739 #ifdef CONFIG_COMPAT_32BIT_TIME 3740 if (time32) 3741 return put_old_timespec32(&ts, userstamp); 3742 #endif 3743 #ifdef CONFIG_SPARC64 3744 /* beware of padding in sparc64 timeval */ 3745 if (timeval && !in_compat_syscall()) { 3746 struct __kernel_old_timeval __user tv = { 3747 .tv_sec = ts.tv_sec, 3748 .tv_usec = ts.tv_nsec, 3749 }; 3750 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3751 return -EFAULT; 3752 return 0; 3753 } 3754 #endif 3755 return put_timespec64(&ts, userstamp); 3756 } 3757 EXPORT_SYMBOL(sock_gettstamp); 3758 3759 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3760 { 3761 if (!sock_flag(sk, flag)) { 3762 unsigned long previous_flags = sk->sk_flags; 3763 3764 sock_set_flag(sk, flag); 3765 /* 3766 * we just set one of the two flags which require net 3767 * time stamping, but time stamping might have been on 3768 * already because of the other one 3769 */ 3770 if (sock_needs_netstamp(sk) && 3771 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3772 net_enable_timestamp(); 3773 } 3774 } 3775 3776 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3777 int level, int type) 3778 { 3779 struct sock_exterr_skb *serr; 3780 struct sk_buff *skb; 3781 int copied, err; 3782 3783 err = -EAGAIN; 3784 skb = sock_dequeue_err_skb(sk); 3785 if (skb == NULL) 3786 goto out; 3787 3788 copied = skb->len; 3789 if (copied > len) { 3790 msg->msg_flags |= MSG_TRUNC; 3791 copied = len; 3792 } 3793 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3794 if (err) 3795 goto out_free_skb; 3796 3797 sock_recv_timestamp(msg, sk, skb); 3798 3799 serr = SKB_EXT_ERR(skb); 3800 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3801 3802 msg->msg_flags |= MSG_ERRQUEUE; 3803 err = copied; 3804 3805 out_free_skb: 3806 kfree_skb(skb); 3807 out: 3808 return err; 3809 } 3810 EXPORT_SYMBOL(sock_recv_errqueue); 3811 3812 /* 3813 * Get a socket option on an socket. 3814 * 3815 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3816 * asynchronous errors should be reported by getsockopt. We assume 3817 * this means if you specify SO_ERROR (otherwise what is the point of it). 3818 */ 3819 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3820 char __user *optval, int __user *optlen) 3821 { 3822 struct sock *sk = sock->sk; 3823 3824 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3825 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3826 } 3827 EXPORT_SYMBOL(sock_common_getsockopt); 3828 3829 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3830 int flags) 3831 { 3832 struct sock *sk = sock->sk; 3833 int addr_len = 0; 3834 int err; 3835 3836 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3837 if (err >= 0) 3838 msg->msg_namelen = addr_len; 3839 return err; 3840 } 3841 EXPORT_SYMBOL(sock_common_recvmsg); 3842 3843 /* 3844 * Set socket options on an inet socket. 3845 */ 3846 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3847 sockptr_t optval, unsigned int optlen) 3848 { 3849 struct sock *sk = sock->sk; 3850 3851 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3852 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3853 } 3854 EXPORT_SYMBOL(sock_common_setsockopt); 3855 3856 void sk_common_release(struct sock *sk) 3857 { 3858 if (sk->sk_prot->destroy) 3859 sk->sk_prot->destroy(sk); 3860 3861 /* 3862 * Observation: when sk_common_release is called, processes have 3863 * no access to socket. But net still has. 3864 * Step one, detach it from networking: 3865 * 3866 * A. Remove from hash tables. 3867 */ 3868 3869 sk->sk_prot->unhash(sk); 3870 3871 /* 3872 * In this point socket cannot receive new packets, but it is possible 3873 * that some packets are in flight because some CPU runs receiver and 3874 * did hash table lookup before we unhashed socket. They will achieve 3875 * receive queue and will be purged by socket destructor. 3876 * 3877 * Also we still have packets pending on receive queue and probably, 3878 * our own packets waiting in device queues. sock_destroy will drain 3879 * receive queue, but transmitted packets will delay socket destruction 3880 * until the last reference will be released. 3881 */ 3882 3883 sock_orphan(sk); 3884 3885 xfrm_sk_free_policy(sk); 3886 3887 sock_put(sk); 3888 } 3889 EXPORT_SYMBOL(sk_common_release); 3890 3891 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3892 { 3893 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3894 3895 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3896 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3897 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3898 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3899 mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc); 3900 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3901 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3902 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3903 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3904 } 3905 3906 #ifdef CONFIG_PROC_FS 3907 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3908 3909 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3910 { 3911 int cpu, idx = prot->inuse_idx; 3912 int res = 0; 3913 3914 for_each_possible_cpu(cpu) 3915 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3916 3917 return res >= 0 ? res : 0; 3918 } 3919 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3920 3921 int sock_inuse_get(struct net *net) 3922 { 3923 int cpu, res = 0; 3924 3925 for_each_possible_cpu(cpu) 3926 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3927 3928 return res; 3929 } 3930 3931 EXPORT_SYMBOL_GPL(sock_inuse_get); 3932 3933 static int __net_init sock_inuse_init_net(struct net *net) 3934 { 3935 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3936 if (net->core.prot_inuse == NULL) 3937 return -ENOMEM; 3938 return 0; 3939 } 3940 3941 static void __net_exit sock_inuse_exit_net(struct net *net) 3942 { 3943 free_percpu(net->core.prot_inuse); 3944 } 3945 3946 static struct pernet_operations net_inuse_ops = { 3947 .init = sock_inuse_init_net, 3948 .exit = sock_inuse_exit_net, 3949 }; 3950 3951 static __init int net_inuse_init(void) 3952 { 3953 if (register_pernet_subsys(&net_inuse_ops)) 3954 panic("Cannot initialize net inuse counters"); 3955 3956 return 0; 3957 } 3958 3959 core_initcall(net_inuse_init); 3960 3961 static int assign_proto_idx(struct proto *prot) 3962 { 3963 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3964 3965 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3966 pr_err("PROTO_INUSE_NR exhausted\n"); 3967 return -ENOSPC; 3968 } 3969 3970 set_bit(prot->inuse_idx, proto_inuse_idx); 3971 return 0; 3972 } 3973 3974 static void release_proto_idx(struct proto *prot) 3975 { 3976 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3977 clear_bit(prot->inuse_idx, proto_inuse_idx); 3978 } 3979 #else 3980 static inline int assign_proto_idx(struct proto *prot) 3981 { 3982 return 0; 3983 } 3984 3985 static inline void release_proto_idx(struct proto *prot) 3986 { 3987 } 3988 3989 #endif 3990 3991 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3992 { 3993 if (!twsk_prot) 3994 return; 3995 kfree(twsk_prot->twsk_slab_name); 3996 twsk_prot->twsk_slab_name = NULL; 3997 kmem_cache_destroy(twsk_prot->twsk_slab); 3998 twsk_prot->twsk_slab = NULL; 3999 } 4000 4001 static int tw_prot_init(const struct proto *prot) 4002 { 4003 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 4004 4005 if (!twsk_prot) 4006 return 0; 4007 4008 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 4009 prot->name); 4010 if (!twsk_prot->twsk_slab_name) 4011 return -ENOMEM; 4012 4013 twsk_prot->twsk_slab = 4014 kmem_cache_create(twsk_prot->twsk_slab_name, 4015 twsk_prot->twsk_obj_size, 0, 4016 SLAB_ACCOUNT | prot->slab_flags, 4017 NULL); 4018 if (!twsk_prot->twsk_slab) { 4019 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 4020 prot->name); 4021 return -ENOMEM; 4022 } 4023 4024 return 0; 4025 } 4026 4027 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 4028 { 4029 if (!rsk_prot) 4030 return; 4031 kfree(rsk_prot->slab_name); 4032 rsk_prot->slab_name = NULL; 4033 kmem_cache_destroy(rsk_prot->slab); 4034 rsk_prot->slab = NULL; 4035 } 4036 4037 static int req_prot_init(const struct proto *prot) 4038 { 4039 struct request_sock_ops *rsk_prot = prot->rsk_prot; 4040 4041 if (!rsk_prot) 4042 return 0; 4043 4044 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4045 prot->name); 4046 if (!rsk_prot->slab_name) 4047 return -ENOMEM; 4048 4049 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4050 rsk_prot->obj_size, 0, 4051 SLAB_ACCOUNT | prot->slab_flags, 4052 NULL); 4053 4054 if (!rsk_prot->slab) { 4055 pr_crit("%s: Can't create request sock SLAB cache!\n", 4056 prot->name); 4057 return -ENOMEM; 4058 } 4059 return 0; 4060 } 4061 4062 int proto_register(struct proto *prot, int alloc_slab) 4063 { 4064 int ret = -ENOBUFS; 4065 4066 if (prot->memory_allocated && !prot->sysctl_mem) { 4067 pr_err("%s: missing sysctl_mem\n", prot->name); 4068 return -EINVAL; 4069 } 4070 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4071 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4072 return -EINVAL; 4073 } 4074 if (alloc_slab) { 4075 prot->slab = kmem_cache_create_usercopy(prot->name, 4076 prot->obj_size, 0, 4077 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4078 prot->slab_flags, 4079 prot->useroffset, prot->usersize, 4080 NULL); 4081 4082 if (prot->slab == NULL) { 4083 pr_crit("%s: Can't create sock SLAB cache!\n", 4084 prot->name); 4085 goto out; 4086 } 4087 4088 if (req_prot_init(prot)) 4089 goto out_free_request_sock_slab; 4090 4091 if (tw_prot_init(prot)) 4092 goto out_free_timewait_sock_slab; 4093 } 4094 4095 mutex_lock(&proto_list_mutex); 4096 ret = assign_proto_idx(prot); 4097 if (ret) { 4098 mutex_unlock(&proto_list_mutex); 4099 goto out_free_timewait_sock_slab; 4100 } 4101 list_add(&prot->node, &proto_list); 4102 mutex_unlock(&proto_list_mutex); 4103 return ret; 4104 4105 out_free_timewait_sock_slab: 4106 if (alloc_slab) 4107 tw_prot_cleanup(prot->twsk_prot); 4108 out_free_request_sock_slab: 4109 if (alloc_slab) { 4110 req_prot_cleanup(prot->rsk_prot); 4111 4112 kmem_cache_destroy(prot->slab); 4113 prot->slab = NULL; 4114 } 4115 out: 4116 return ret; 4117 } 4118 EXPORT_SYMBOL(proto_register); 4119 4120 void proto_unregister(struct proto *prot) 4121 { 4122 mutex_lock(&proto_list_mutex); 4123 release_proto_idx(prot); 4124 list_del(&prot->node); 4125 mutex_unlock(&proto_list_mutex); 4126 4127 kmem_cache_destroy(prot->slab); 4128 prot->slab = NULL; 4129 4130 req_prot_cleanup(prot->rsk_prot); 4131 tw_prot_cleanup(prot->twsk_prot); 4132 } 4133 EXPORT_SYMBOL(proto_unregister); 4134 4135 int sock_load_diag_module(int family, int protocol) 4136 { 4137 if (!protocol) { 4138 if (!sock_is_registered(family)) 4139 return -ENOENT; 4140 4141 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4142 NETLINK_SOCK_DIAG, family); 4143 } 4144 4145 #ifdef CONFIG_INET 4146 if (family == AF_INET && 4147 protocol != IPPROTO_RAW && 4148 protocol < MAX_INET_PROTOS && 4149 !rcu_access_pointer(inet_protos[protocol])) 4150 return -ENOENT; 4151 #endif 4152 4153 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4154 NETLINK_SOCK_DIAG, family, protocol); 4155 } 4156 EXPORT_SYMBOL(sock_load_diag_module); 4157 4158 #ifdef CONFIG_PROC_FS 4159 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4160 __acquires(proto_list_mutex) 4161 { 4162 mutex_lock(&proto_list_mutex); 4163 return seq_list_start_head(&proto_list, *pos); 4164 } 4165 4166 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4167 { 4168 return seq_list_next(v, &proto_list, pos); 4169 } 4170 4171 static void proto_seq_stop(struct seq_file *seq, void *v) 4172 __releases(proto_list_mutex) 4173 { 4174 mutex_unlock(&proto_list_mutex); 4175 } 4176 4177 static char proto_method_implemented(const void *method) 4178 { 4179 return method == NULL ? 'n' : 'y'; 4180 } 4181 static long sock_prot_memory_allocated(struct proto *proto) 4182 { 4183 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4184 } 4185 4186 static const char *sock_prot_memory_pressure(struct proto *proto) 4187 { 4188 return proto->memory_pressure != NULL ? 4189 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4190 } 4191 4192 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4193 { 4194 4195 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4196 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4197 proto->name, 4198 proto->obj_size, 4199 sock_prot_inuse_get(seq_file_net(seq), proto), 4200 sock_prot_memory_allocated(proto), 4201 sock_prot_memory_pressure(proto), 4202 proto->max_header, 4203 proto->slab == NULL ? "no" : "yes", 4204 module_name(proto->owner), 4205 proto_method_implemented(proto->close), 4206 proto_method_implemented(proto->connect), 4207 proto_method_implemented(proto->disconnect), 4208 proto_method_implemented(proto->accept), 4209 proto_method_implemented(proto->ioctl), 4210 proto_method_implemented(proto->init), 4211 proto_method_implemented(proto->destroy), 4212 proto_method_implemented(proto->shutdown), 4213 proto_method_implemented(proto->setsockopt), 4214 proto_method_implemented(proto->getsockopt), 4215 proto_method_implemented(proto->sendmsg), 4216 proto_method_implemented(proto->recvmsg), 4217 proto_method_implemented(proto->bind), 4218 proto_method_implemented(proto->backlog_rcv), 4219 proto_method_implemented(proto->hash), 4220 proto_method_implemented(proto->unhash), 4221 proto_method_implemented(proto->get_port), 4222 proto_method_implemented(proto->enter_memory_pressure)); 4223 } 4224 4225 static int proto_seq_show(struct seq_file *seq, void *v) 4226 { 4227 if (v == &proto_list) 4228 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4229 "protocol", 4230 "size", 4231 "sockets", 4232 "memory", 4233 "press", 4234 "maxhdr", 4235 "slab", 4236 "module", 4237 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4238 else 4239 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4240 return 0; 4241 } 4242 4243 static const struct seq_operations proto_seq_ops = { 4244 .start = proto_seq_start, 4245 .next = proto_seq_next, 4246 .stop = proto_seq_stop, 4247 .show = proto_seq_show, 4248 }; 4249 4250 static __net_init int proto_init_net(struct net *net) 4251 { 4252 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4253 sizeof(struct seq_net_private))) 4254 return -ENOMEM; 4255 4256 return 0; 4257 } 4258 4259 static __net_exit void proto_exit_net(struct net *net) 4260 { 4261 remove_proc_entry("protocols", net->proc_net); 4262 } 4263 4264 4265 static __net_initdata struct pernet_operations proto_net_ops = { 4266 .init = proto_init_net, 4267 .exit = proto_exit_net, 4268 }; 4269 4270 static int __init proto_init(void) 4271 { 4272 return register_pernet_subsys(&proto_net_ops); 4273 } 4274 4275 subsys_initcall(proto_init); 4276 4277 #endif /* PROC_FS */ 4278 4279 #ifdef CONFIG_NET_RX_BUSY_POLL 4280 bool sk_busy_loop_end(void *p, unsigned long start_time) 4281 { 4282 struct sock *sk = p; 4283 4284 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4285 return true; 4286 4287 if (sk_is_udp(sk) && 4288 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4289 return true; 4290 4291 return sk_busy_loop_timeout(sk, start_time); 4292 } 4293 EXPORT_SYMBOL(sk_busy_loop_end); 4294 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4295 4296 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4297 { 4298 if (!sk->sk_prot->bind_add) 4299 return -EOPNOTSUPP; 4300 return sk->sk_prot->bind_add(sk, addr, addr_len); 4301 } 4302 EXPORT_SYMBOL(sock_bind_add); 4303 4304 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4305 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4306 void __user *arg, void *karg, size_t size) 4307 { 4308 int ret; 4309 4310 if (copy_from_user(karg, arg, size)) 4311 return -EFAULT; 4312 4313 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4314 if (ret) 4315 return ret; 4316 4317 if (copy_to_user(arg, karg, size)) 4318 return -EFAULT; 4319 4320 return 0; 4321 } 4322 EXPORT_SYMBOL(sock_ioctl_inout); 4323 4324 /* This is the most common ioctl prep function, where the result (4 bytes) is 4325 * copied back to userspace if the ioctl() returns successfully. No input is 4326 * copied from userspace as input argument. 4327 */ 4328 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4329 { 4330 int ret, karg = 0; 4331 4332 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4333 if (ret) 4334 return ret; 4335 4336 return put_user(karg, (int __user *)arg); 4337 } 4338 4339 /* A wrapper around sock ioctls, which copies the data from userspace 4340 * (depending on the protocol/ioctl), and copies back the result to userspace. 4341 * The main motivation for this function is to pass kernel memory to the 4342 * protocol ioctl callbacks, instead of userspace memory. 4343 */ 4344 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4345 { 4346 int rc = 1; 4347 4348 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4349 rc = ipmr_sk_ioctl(sk, cmd, arg); 4350 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4351 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4352 else if (sk_is_phonet(sk)) 4353 rc = phonet_sk_ioctl(sk, cmd, arg); 4354 4355 /* If ioctl was processed, returns its value */ 4356 if (rc <= 0) 4357 return rc; 4358 4359 /* Otherwise call the default handler */ 4360 return sock_ioctl_out(sk, cmd, arg); 4361 } 4362 EXPORT_SYMBOL(sk_ioctl); 4363 4364 static int __init sock_struct_check(void) 4365 { 4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4367 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4371 4372 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4381 4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4385 4386 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4387 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4388 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4389 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4390 4391 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4392 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4393 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf); 4394 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4395 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4396 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4397 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4398 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4399 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4400 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm); 4401 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status); 4402 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4403 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4404 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4405 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4406 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4407 4408 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4409 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4410 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4411 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4412 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4413 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4414 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4415 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4416 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4417 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4418 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4419 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4420 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4421 return 0; 4422 } 4423 4424 core_initcall(sock_struct_check); 4425