1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 */ 85 86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <linux/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/udp.h> 111 #include <linux/init.h> 112 #include <linux/highmem.h> 113 #include <linux/user_namespace.h> 114 #include <linux/static_key.h> 115 #include <linux/memcontrol.h> 116 #include <linux/prefetch.h> 117 #include <linux/compat.h> 118 #include <linux/mroute.h> 119 #include <linux/mroute6.h> 120 #include <linux/icmpv6.h> 121 122 #include <linux/uaccess.h> 123 124 #include <linux/netdevice.h> 125 #include <net/protocol.h> 126 #include <linux/skbuff.h> 127 #include <linux/skbuff_ref.h> 128 #include <net/net_namespace.h> 129 #include <net/request_sock.h> 130 #include <net/sock.h> 131 #include <net/proto_memory.h> 132 #include <linux/net_tstamp.h> 133 #include <net/xfrm.h> 134 #include <linux/ipsec.h> 135 #include <net/cls_cgroup.h> 136 #include <net/netprio_cgroup.h> 137 #include <linux/sock_diag.h> 138 139 #include <linux/filter.h> 140 #include <net/sock_reuseport.h> 141 #include <net/bpf_sk_storage.h> 142 143 #include <trace/events/sock.h> 144 145 #include <net/tcp.h> 146 #include <net/busy_poll.h> 147 #include <net/phonet/phonet.h> 148 149 #include <linux/ethtool.h> 150 151 #include "dev.h" 152 153 static DEFINE_MUTEX(proto_list_mutex); 154 static LIST_HEAD(proto_list); 155 156 static void sock_def_write_space_wfree(struct sock *sk); 157 static void sock_def_write_space(struct sock *sk); 158 159 /** 160 * sk_ns_capable - General socket capability test 161 * @sk: Socket to use a capability on or through 162 * @user_ns: The user namespace of the capability to use 163 * @cap: The capability to use 164 * 165 * Test to see if the opener of the socket had when the socket was 166 * created and the current process has the capability @cap in the user 167 * namespace @user_ns. 168 */ 169 bool sk_ns_capable(const struct sock *sk, 170 struct user_namespace *user_ns, int cap) 171 { 172 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 173 ns_capable(user_ns, cap); 174 } 175 EXPORT_SYMBOL(sk_ns_capable); 176 177 /** 178 * sk_capable - Socket global capability test 179 * @sk: Socket to use a capability on or through 180 * @cap: The global capability to use 181 * 182 * Test to see if the opener of the socket had when the socket was 183 * created and the current process has the capability @cap in all user 184 * namespaces. 185 */ 186 bool sk_capable(const struct sock *sk, int cap) 187 { 188 return sk_ns_capable(sk, &init_user_ns, cap); 189 } 190 EXPORT_SYMBOL(sk_capable); 191 192 /** 193 * sk_net_capable - Network namespace socket capability test 194 * @sk: Socket to use a capability on or through 195 * @cap: The capability to use 196 * 197 * Test to see if the opener of the socket had when the socket was created 198 * and the current process has the capability @cap over the network namespace 199 * the socket is a member of. 200 */ 201 bool sk_net_capable(const struct sock *sk, int cap) 202 { 203 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 204 } 205 EXPORT_SYMBOL(sk_net_capable); 206 207 /* 208 * Each address family might have different locking rules, so we have 209 * one slock key per address family and separate keys for internal and 210 * userspace sockets. 211 */ 212 static struct lock_class_key af_family_keys[AF_MAX]; 213 static struct lock_class_key af_family_kern_keys[AF_MAX]; 214 static struct lock_class_key af_family_slock_keys[AF_MAX]; 215 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 216 217 /* 218 * Make lock validator output more readable. (we pre-construct these 219 * strings build-time, so that runtime initialization of socket 220 * locks is fast): 221 */ 222 223 #define _sock_locks(x) \ 224 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 225 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 226 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 227 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 228 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 229 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 230 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 231 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 232 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 233 x "27" , x "28" , x "AF_CAN" , \ 234 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 235 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 236 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 237 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 238 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 239 x "AF_MCTP" , \ 240 x "AF_MAX" 241 242 static const char *const af_family_key_strings[AF_MAX+1] = { 243 _sock_locks("sk_lock-") 244 }; 245 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 246 _sock_locks("slock-") 247 }; 248 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 249 _sock_locks("clock-") 250 }; 251 252 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 253 _sock_locks("k-sk_lock-") 254 }; 255 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 256 _sock_locks("k-slock-") 257 }; 258 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 259 _sock_locks("k-clock-") 260 }; 261 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 262 _sock_locks("rlock-") 263 }; 264 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 265 _sock_locks("wlock-") 266 }; 267 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 268 _sock_locks("elock-") 269 }; 270 271 /* 272 * sk_callback_lock and sk queues locking rules are per-address-family, 273 * so split the lock classes by using a per-AF key: 274 */ 275 static struct lock_class_key af_callback_keys[AF_MAX]; 276 static struct lock_class_key af_rlock_keys[AF_MAX]; 277 static struct lock_class_key af_wlock_keys[AF_MAX]; 278 static struct lock_class_key af_elock_keys[AF_MAX]; 279 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 280 281 /* Run time adjustable parameters. */ 282 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 283 EXPORT_SYMBOL(sysctl_wmem_max); 284 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 285 EXPORT_SYMBOL(sysctl_rmem_max); 286 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 287 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 288 289 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 290 EXPORT_SYMBOL_GPL(memalloc_socks_key); 291 292 /** 293 * sk_set_memalloc - sets %SOCK_MEMALLOC 294 * @sk: socket to set it on 295 * 296 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 297 * It's the responsibility of the admin to adjust min_free_kbytes 298 * to meet the requirements 299 */ 300 void sk_set_memalloc(struct sock *sk) 301 { 302 sock_set_flag(sk, SOCK_MEMALLOC); 303 sk->sk_allocation |= __GFP_MEMALLOC; 304 static_branch_inc(&memalloc_socks_key); 305 } 306 EXPORT_SYMBOL_GPL(sk_set_memalloc); 307 308 void sk_clear_memalloc(struct sock *sk) 309 { 310 sock_reset_flag(sk, SOCK_MEMALLOC); 311 sk->sk_allocation &= ~__GFP_MEMALLOC; 312 static_branch_dec(&memalloc_socks_key); 313 314 /* 315 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 316 * progress of swapping. SOCK_MEMALLOC may be cleared while 317 * it has rmem allocations due to the last swapfile being deactivated 318 * but there is a risk that the socket is unusable due to exceeding 319 * the rmem limits. Reclaim the reserves and obey rmem limits again. 320 */ 321 sk_mem_reclaim(sk); 322 } 323 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 324 325 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 326 { 327 int ret; 328 unsigned int noreclaim_flag; 329 330 /* these should have been dropped before queueing */ 331 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 332 333 noreclaim_flag = memalloc_noreclaim_save(); 334 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 335 tcp_v6_do_rcv, 336 tcp_v4_do_rcv, 337 sk, skb); 338 memalloc_noreclaim_restore(noreclaim_flag); 339 340 return ret; 341 } 342 EXPORT_SYMBOL(__sk_backlog_rcv); 343 344 void sk_error_report(struct sock *sk) 345 { 346 sk->sk_error_report(sk); 347 348 switch (sk->sk_family) { 349 case AF_INET: 350 fallthrough; 351 case AF_INET6: 352 trace_inet_sk_error_report(sk); 353 break; 354 default: 355 break; 356 } 357 } 358 EXPORT_SYMBOL(sk_error_report); 359 360 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 361 { 362 struct __kernel_sock_timeval tv; 363 364 if (timeo == MAX_SCHEDULE_TIMEOUT) { 365 tv.tv_sec = 0; 366 tv.tv_usec = 0; 367 } else { 368 tv.tv_sec = timeo / HZ; 369 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 370 } 371 372 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 373 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 374 *(struct old_timeval32 *)optval = tv32; 375 return sizeof(tv32); 376 } 377 378 if (old_timeval) { 379 struct __kernel_old_timeval old_tv; 380 old_tv.tv_sec = tv.tv_sec; 381 old_tv.tv_usec = tv.tv_usec; 382 *(struct __kernel_old_timeval *)optval = old_tv; 383 return sizeof(old_tv); 384 } 385 386 *(struct __kernel_sock_timeval *)optval = tv; 387 return sizeof(tv); 388 } 389 EXPORT_SYMBOL(sock_get_timeout); 390 391 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 392 sockptr_t optval, int optlen, bool old_timeval) 393 { 394 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 395 struct old_timeval32 tv32; 396 397 if (optlen < sizeof(tv32)) 398 return -EINVAL; 399 400 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 401 return -EFAULT; 402 tv->tv_sec = tv32.tv_sec; 403 tv->tv_usec = tv32.tv_usec; 404 } else if (old_timeval) { 405 struct __kernel_old_timeval old_tv; 406 407 if (optlen < sizeof(old_tv)) 408 return -EINVAL; 409 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 410 return -EFAULT; 411 tv->tv_sec = old_tv.tv_sec; 412 tv->tv_usec = old_tv.tv_usec; 413 } else { 414 if (optlen < sizeof(*tv)) 415 return -EINVAL; 416 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 417 return -EFAULT; 418 } 419 420 return 0; 421 } 422 EXPORT_SYMBOL(sock_copy_user_timeval); 423 424 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 425 bool old_timeval) 426 { 427 struct __kernel_sock_timeval tv; 428 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 429 long val; 430 431 if (err) 432 return err; 433 434 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 435 return -EDOM; 436 437 if (tv.tv_sec < 0) { 438 static int warned __read_mostly; 439 440 WRITE_ONCE(*timeo_p, 0); 441 if (warned < 10 && net_ratelimit()) { 442 warned++; 443 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 444 __func__, current->comm, task_pid_nr(current)); 445 } 446 return 0; 447 } 448 val = MAX_SCHEDULE_TIMEOUT; 449 if ((tv.tv_sec || tv.tv_usec) && 450 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 451 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 452 USEC_PER_SEC / HZ); 453 WRITE_ONCE(*timeo_p, val); 454 return 0; 455 } 456 457 static bool 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); 941 942 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 943 sock_enable_timestamp(sk, 944 SOCK_TIMESTAMPING_RX_SOFTWARE); 945 else 946 sock_disable_timestamp(sk, 947 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 948 return 0; 949 } 950 951 void sock_set_keepalive(struct sock *sk) 952 { 953 lock_sock(sk); 954 if (sk->sk_prot->keepalive) 955 sk->sk_prot->keepalive(sk, true); 956 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 957 release_sock(sk); 958 } 959 EXPORT_SYMBOL(sock_set_keepalive); 960 961 static void __sock_set_rcvbuf(struct sock *sk, int val) 962 { 963 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 964 * as a negative value. 965 */ 966 val = min_t(int, val, INT_MAX / 2); 967 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 968 969 /* We double it on the way in to account for "struct sk_buff" etc. 970 * overhead. Applications assume that the SO_RCVBUF setting they make 971 * will allow that much actual data to be received on that socket. 972 * 973 * Applications are unaware that "struct sk_buff" and other overheads 974 * allocate from the receive buffer during socket buffer allocation. 975 * 976 * And after considering the possible alternatives, returning the value 977 * we actually used in getsockopt is the most desirable behavior. 978 */ 979 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 980 } 981 982 void sock_set_rcvbuf(struct sock *sk, int val) 983 { 984 lock_sock(sk); 985 __sock_set_rcvbuf(sk, val); 986 release_sock(sk); 987 } 988 EXPORT_SYMBOL(sock_set_rcvbuf); 989 990 static void __sock_set_mark(struct sock *sk, u32 val) 991 { 992 if (val != sk->sk_mark) { 993 WRITE_ONCE(sk->sk_mark, val); 994 sk_dst_reset(sk); 995 } 996 } 997 998 void sock_set_mark(struct sock *sk, u32 val) 999 { 1000 lock_sock(sk); 1001 __sock_set_mark(sk, val); 1002 release_sock(sk); 1003 } 1004 EXPORT_SYMBOL(sock_set_mark); 1005 1006 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1007 { 1008 /* Round down bytes to multiple of pages */ 1009 bytes = round_down(bytes, PAGE_SIZE); 1010 1011 WARN_ON(bytes > sk->sk_reserved_mem); 1012 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1013 sk_mem_reclaim(sk); 1014 } 1015 1016 static int sock_reserve_memory(struct sock *sk, int bytes) 1017 { 1018 long allocated; 1019 bool charged; 1020 int pages; 1021 1022 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1023 return -EOPNOTSUPP; 1024 1025 if (!bytes) 1026 return 0; 1027 1028 pages = sk_mem_pages(bytes); 1029 1030 /* pre-charge to memcg */ 1031 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1032 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1033 if (!charged) 1034 return -ENOMEM; 1035 1036 /* pre-charge to forward_alloc */ 1037 sk_memory_allocated_add(sk, pages); 1038 allocated = sk_memory_allocated(sk); 1039 /* If the system goes into memory pressure with this 1040 * precharge, give up and return error. 1041 */ 1042 if (allocated > sk_prot_mem_limits(sk, 1)) { 1043 sk_memory_allocated_sub(sk, pages); 1044 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1045 return -ENOMEM; 1046 } 1047 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1048 1049 WRITE_ONCE(sk->sk_reserved_mem, 1050 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1051 1052 return 0; 1053 } 1054 1055 #ifdef CONFIG_PAGE_POOL 1056 1057 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED 1058 * in 1 syscall. The limit exists to limit the amount of memory the kernel 1059 * allocates to copy these tokens, and to prevent looping over the frags for 1060 * too long. 1061 */ 1062 #define MAX_DONTNEED_TOKENS 128 1063 #define MAX_DONTNEED_FRAGS 1024 1064 1065 static noinline_for_stack int 1066 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1067 { 1068 unsigned int num_tokens, i, j, k, netmem_num = 0; 1069 struct dmabuf_token *tokens; 1070 int ret = 0, num_frags = 0; 1071 netmem_ref netmems[16]; 1072 1073 if (!sk_is_tcp(sk)) 1074 return -EBADF; 1075 1076 if (optlen % sizeof(*tokens) || 1077 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1078 return -EINVAL; 1079 1080 num_tokens = optlen / sizeof(*tokens); 1081 tokens = kvmalloc_array(num_tokens, sizeof(*tokens), GFP_KERNEL); 1082 if (!tokens) 1083 return -ENOMEM; 1084 1085 if (copy_from_sockptr(tokens, optval, optlen)) { 1086 kvfree(tokens); 1087 return -EFAULT; 1088 } 1089 1090 xa_lock_bh(&sk->sk_user_frags); 1091 for (i = 0; i < num_tokens; i++) { 1092 for (j = 0; j < tokens[i].token_count; j++) { 1093 if (++num_frags > MAX_DONTNEED_FRAGS) 1094 goto frag_limit_reached; 1095 1096 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1097 &sk->sk_user_frags, tokens[i].token_start + j); 1098 1099 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem))) 1100 continue; 1101 1102 netmems[netmem_num++] = netmem; 1103 if (netmem_num == ARRAY_SIZE(netmems)) { 1104 xa_unlock_bh(&sk->sk_user_frags); 1105 for (k = 0; k < netmem_num; k++) 1106 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1107 netmem_num = 0; 1108 xa_lock_bh(&sk->sk_user_frags); 1109 } 1110 ret++; 1111 } 1112 } 1113 1114 frag_limit_reached: 1115 xa_unlock_bh(&sk->sk_user_frags); 1116 for (k = 0; k < netmem_num; k++) 1117 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1118 1119 kvfree(tokens); 1120 return ret; 1121 } 1122 #endif 1123 1124 void sockopt_lock_sock(struct sock *sk) 1125 { 1126 /* When current->bpf_ctx is set, the setsockopt is called from 1127 * a bpf prog. bpf has ensured the sk lock has been 1128 * acquired before calling setsockopt(). 1129 */ 1130 if (has_current_bpf_ctx()) 1131 return; 1132 1133 lock_sock(sk); 1134 } 1135 EXPORT_SYMBOL(sockopt_lock_sock); 1136 1137 void sockopt_release_sock(struct sock *sk) 1138 { 1139 if (has_current_bpf_ctx()) 1140 return; 1141 1142 release_sock(sk); 1143 } 1144 EXPORT_SYMBOL(sockopt_release_sock); 1145 1146 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1147 { 1148 return has_current_bpf_ctx() || ns_capable(ns, cap); 1149 } 1150 EXPORT_SYMBOL(sockopt_ns_capable); 1151 1152 bool sockopt_capable(int cap) 1153 { 1154 return has_current_bpf_ctx() || capable(cap); 1155 } 1156 EXPORT_SYMBOL(sockopt_capable); 1157 1158 static int sockopt_validate_clockid(__kernel_clockid_t value) 1159 { 1160 switch (value) { 1161 case CLOCK_REALTIME: 1162 case CLOCK_MONOTONIC: 1163 case CLOCK_TAI: 1164 return 0; 1165 } 1166 return -EINVAL; 1167 } 1168 1169 /* 1170 * This is meant for all protocols to use and covers goings on 1171 * at the socket level. Everything here is generic. 1172 */ 1173 1174 int sk_setsockopt(struct sock *sk, int level, int optname, 1175 sockptr_t optval, unsigned int optlen) 1176 { 1177 struct so_timestamping timestamping; 1178 struct socket *sock = sk->sk_socket; 1179 struct sock_txtime sk_txtime; 1180 int val; 1181 int valbool; 1182 struct linger ling; 1183 int ret = 0; 1184 1185 /* 1186 * Options without arguments 1187 */ 1188 1189 if (optname == SO_BINDTODEVICE) 1190 return sock_setbindtodevice(sk, optval, optlen); 1191 1192 if (optlen < sizeof(int)) 1193 return -EINVAL; 1194 1195 if (copy_from_sockptr(&val, optval, sizeof(val))) 1196 return -EFAULT; 1197 1198 valbool = val ? 1 : 0; 1199 1200 /* handle options which do not require locking the socket. */ 1201 switch (optname) { 1202 case SO_PRIORITY: 1203 if (sk_set_prio_allowed(sk, val)) { 1204 sock_set_priority(sk, val); 1205 return 0; 1206 } 1207 return -EPERM; 1208 case SO_PASSSEC: 1209 assign_bit(SOCK_PASSSEC, &sock->flags, valbool); 1210 return 0; 1211 case SO_PASSCRED: 1212 assign_bit(SOCK_PASSCRED, &sock->flags, valbool); 1213 return 0; 1214 case SO_PASSPIDFD: 1215 assign_bit(SOCK_PASSPIDFD, &sock->flags, valbool); 1216 return 0; 1217 case SO_TYPE: 1218 case SO_PROTOCOL: 1219 case SO_DOMAIN: 1220 case SO_ERROR: 1221 return -ENOPROTOOPT; 1222 #ifdef CONFIG_NET_RX_BUSY_POLL 1223 case SO_BUSY_POLL: 1224 if (val < 0) 1225 return -EINVAL; 1226 WRITE_ONCE(sk->sk_ll_usec, val); 1227 return 0; 1228 case SO_PREFER_BUSY_POLL: 1229 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1230 return -EPERM; 1231 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1232 return 0; 1233 case SO_BUSY_POLL_BUDGET: 1234 if (val > READ_ONCE(sk->sk_busy_poll_budget) && 1235 !sockopt_capable(CAP_NET_ADMIN)) 1236 return -EPERM; 1237 if (val < 0 || val > U16_MAX) 1238 return -EINVAL; 1239 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1240 return 0; 1241 #endif 1242 case SO_MAX_PACING_RATE: 1243 { 1244 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1245 unsigned long pacing_rate; 1246 1247 if (sizeof(ulval) != sizeof(val) && 1248 optlen >= sizeof(ulval) && 1249 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1250 return -EFAULT; 1251 } 1252 if (ulval != ~0UL) 1253 cmpxchg(&sk->sk_pacing_status, 1254 SK_PACING_NONE, 1255 SK_PACING_NEEDED); 1256 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1257 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1258 pacing_rate = READ_ONCE(sk->sk_pacing_rate); 1259 if (ulval < pacing_rate) 1260 WRITE_ONCE(sk->sk_pacing_rate, ulval); 1261 return 0; 1262 } 1263 case SO_TXREHASH: 1264 if (val < -1 || val > 1) 1265 return -EINVAL; 1266 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1267 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1268 /* Paired with READ_ONCE() in tcp_rtx_synack() 1269 * and sk_getsockopt(). 1270 */ 1271 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1272 return 0; 1273 case SO_PEEK_OFF: 1274 { 1275 int (*set_peek_off)(struct sock *sk, int val); 1276 1277 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1278 if (set_peek_off) 1279 ret = set_peek_off(sk, val); 1280 else 1281 ret = -EOPNOTSUPP; 1282 return ret; 1283 } 1284 #ifdef CONFIG_PAGE_POOL 1285 case SO_DEVMEM_DONTNEED: 1286 return sock_devmem_dontneed(sk, optval, optlen); 1287 #endif 1288 } 1289 1290 sockopt_lock_sock(sk); 1291 1292 switch (optname) { 1293 case SO_DEBUG: 1294 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1295 ret = -EACCES; 1296 else 1297 sock_valbool_flag(sk, SOCK_DBG, valbool); 1298 break; 1299 case SO_REUSEADDR: 1300 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1301 break; 1302 case SO_REUSEPORT: 1303 sk->sk_reuseport = valbool; 1304 break; 1305 case SO_DONTROUTE: 1306 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1307 sk_dst_reset(sk); 1308 break; 1309 case SO_BROADCAST: 1310 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1311 break; 1312 case SO_SNDBUF: 1313 /* Don't error on this BSD doesn't and if you think 1314 * about it this is right. Otherwise apps have to 1315 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1316 * are treated in BSD as hints 1317 */ 1318 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1319 set_sndbuf: 1320 /* Ensure val * 2 fits into an int, to prevent max_t() 1321 * from treating it as a negative value. 1322 */ 1323 val = min_t(int, val, INT_MAX / 2); 1324 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1325 WRITE_ONCE(sk->sk_sndbuf, 1326 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1327 /* Wake up sending tasks if we upped the value. */ 1328 sk->sk_write_space(sk); 1329 break; 1330 1331 case SO_SNDBUFFORCE: 1332 if (!sockopt_capable(CAP_NET_ADMIN)) { 1333 ret = -EPERM; 1334 break; 1335 } 1336 1337 /* No negative values (to prevent underflow, as val will be 1338 * multiplied by 2). 1339 */ 1340 if (val < 0) 1341 val = 0; 1342 goto set_sndbuf; 1343 1344 case SO_RCVBUF: 1345 /* Don't error on this BSD doesn't and if you think 1346 * about it this is right. Otherwise apps have to 1347 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1348 * are treated in BSD as hints 1349 */ 1350 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1351 break; 1352 1353 case SO_RCVBUFFORCE: 1354 if (!sockopt_capable(CAP_NET_ADMIN)) { 1355 ret = -EPERM; 1356 break; 1357 } 1358 1359 /* No negative values (to prevent underflow, as val will be 1360 * multiplied by 2). 1361 */ 1362 __sock_set_rcvbuf(sk, max(val, 0)); 1363 break; 1364 1365 case SO_KEEPALIVE: 1366 if (sk->sk_prot->keepalive) 1367 sk->sk_prot->keepalive(sk, valbool); 1368 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1369 break; 1370 1371 case SO_OOBINLINE: 1372 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1373 break; 1374 1375 case SO_NO_CHECK: 1376 sk->sk_no_check_tx = valbool; 1377 break; 1378 1379 case SO_LINGER: 1380 if (optlen < sizeof(ling)) { 1381 ret = -EINVAL; /* 1003.1g */ 1382 break; 1383 } 1384 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1385 ret = -EFAULT; 1386 break; 1387 } 1388 if (!ling.l_onoff) { 1389 sock_reset_flag(sk, SOCK_LINGER); 1390 } else { 1391 unsigned long t_sec = ling.l_linger; 1392 1393 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1394 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1395 else 1396 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1397 sock_set_flag(sk, SOCK_LINGER); 1398 } 1399 break; 1400 1401 case SO_BSDCOMPAT: 1402 break; 1403 1404 case SO_TIMESTAMP_OLD: 1405 case SO_TIMESTAMP_NEW: 1406 case SO_TIMESTAMPNS_OLD: 1407 case SO_TIMESTAMPNS_NEW: 1408 sock_set_timestamp(sk, optname, valbool); 1409 break; 1410 1411 case SO_TIMESTAMPING_NEW: 1412 case SO_TIMESTAMPING_OLD: 1413 if (optlen == sizeof(timestamping)) { 1414 if (copy_from_sockptr(×tamping, optval, 1415 sizeof(timestamping))) { 1416 ret = -EFAULT; 1417 break; 1418 } 1419 } else { 1420 memset(×tamping, 0, sizeof(timestamping)); 1421 timestamping.flags = val; 1422 } 1423 ret = sock_set_timestamping(sk, optname, timestamping); 1424 break; 1425 1426 case SO_RCVLOWAT: 1427 { 1428 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1429 1430 if (val < 0) 1431 val = INT_MAX; 1432 if (sock) 1433 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1434 if (set_rcvlowat) 1435 ret = set_rcvlowat(sk, val); 1436 else 1437 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1438 break; 1439 } 1440 case SO_RCVTIMEO_OLD: 1441 case SO_RCVTIMEO_NEW: 1442 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1443 optlen, optname == SO_RCVTIMEO_OLD); 1444 break; 1445 1446 case SO_SNDTIMEO_OLD: 1447 case SO_SNDTIMEO_NEW: 1448 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1449 optlen, optname == SO_SNDTIMEO_OLD); 1450 break; 1451 1452 case SO_ATTACH_FILTER: { 1453 struct sock_fprog fprog; 1454 1455 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1456 if (!ret) 1457 ret = sk_attach_filter(&fprog, sk); 1458 break; 1459 } 1460 case SO_ATTACH_BPF: 1461 ret = -EINVAL; 1462 if (optlen == sizeof(u32)) { 1463 u32 ufd; 1464 1465 ret = -EFAULT; 1466 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1467 break; 1468 1469 ret = sk_attach_bpf(ufd, sk); 1470 } 1471 break; 1472 1473 case SO_ATTACH_REUSEPORT_CBPF: { 1474 struct sock_fprog fprog; 1475 1476 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1477 if (!ret) 1478 ret = sk_reuseport_attach_filter(&fprog, sk); 1479 break; 1480 } 1481 case SO_ATTACH_REUSEPORT_EBPF: 1482 ret = -EINVAL; 1483 if (optlen == sizeof(u32)) { 1484 u32 ufd; 1485 1486 ret = -EFAULT; 1487 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1488 break; 1489 1490 ret = sk_reuseport_attach_bpf(ufd, sk); 1491 } 1492 break; 1493 1494 case SO_DETACH_REUSEPORT_BPF: 1495 ret = reuseport_detach_prog(sk); 1496 break; 1497 1498 case SO_DETACH_FILTER: 1499 ret = sk_detach_filter(sk); 1500 break; 1501 1502 case SO_LOCK_FILTER: 1503 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1504 ret = -EPERM; 1505 else 1506 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1507 break; 1508 1509 case SO_MARK: 1510 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1511 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1512 ret = -EPERM; 1513 break; 1514 } 1515 1516 __sock_set_mark(sk, val); 1517 break; 1518 case SO_RCVMARK: 1519 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1520 break; 1521 1522 case SO_RCVPRIORITY: 1523 sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool); 1524 break; 1525 1526 case SO_RXQ_OVFL: 1527 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1528 break; 1529 1530 case SO_WIFI_STATUS: 1531 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1532 break; 1533 1534 case SO_NOFCS: 1535 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1536 break; 1537 1538 case SO_SELECT_ERR_QUEUE: 1539 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1540 break; 1541 1542 1543 case SO_INCOMING_CPU: 1544 reuseport_update_incoming_cpu(sk, val); 1545 break; 1546 1547 case SO_CNX_ADVICE: 1548 if (val == 1) 1549 dst_negative_advice(sk); 1550 break; 1551 1552 case SO_ZEROCOPY: 1553 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1554 if (!(sk_is_tcp(sk) || 1555 (sk->sk_type == SOCK_DGRAM && 1556 sk->sk_protocol == IPPROTO_UDP))) 1557 ret = -EOPNOTSUPP; 1558 } else if (sk->sk_family != PF_RDS) { 1559 ret = -EOPNOTSUPP; 1560 } 1561 if (!ret) { 1562 if (val < 0 || val > 1) 1563 ret = -EINVAL; 1564 else 1565 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1566 } 1567 break; 1568 1569 case SO_TXTIME: 1570 if (optlen != sizeof(struct sock_txtime)) { 1571 ret = -EINVAL; 1572 break; 1573 } else if (copy_from_sockptr(&sk_txtime, optval, 1574 sizeof(struct sock_txtime))) { 1575 ret = -EFAULT; 1576 break; 1577 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1578 ret = -EINVAL; 1579 break; 1580 } 1581 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1582 * scheduler has enough safe guards. 1583 */ 1584 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1585 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1586 ret = -EPERM; 1587 break; 1588 } 1589 1590 ret = sockopt_validate_clockid(sk_txtime.clockid); 1591 if (ret) 1592 break; 1593 1594 sock_valbool_flag(sk, SOCK_TXTIME, true); 1595 sk->sk_clockid = sk_txtime.clockid; 1596 sk->sk_txtime_deadline_mode = 1597 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1598 sk->sk_txtime_report_errors = 1599 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1600 break; 1601 1602 case SO_BINDTOIFINDEX: 1603 ret = sock_bindtoindex_locked(sk, val); 1604 break; 1605 1606 case SO_BUF_LOCK: 1607 if (val & ~SOCK_BUF_LOCK_MASK) { 1608 ret = -EINVAL; 1609 break; 1610 } 1611 sk->sk_userlocks = val | (sk->sk_userlocks & 1612 ~SOCK_BUF_LOCK_MASK); 1613 break; 1614 1615 case SO_RESERVE_MEM: 1616 { 1617 int delta; 1618 1619 if (val < 0) { 1620 ret = -EINVAL; 1621 break; 1622 } 1623 1624 delta = val - sk->sk_reserved_mem; 1625 if (delta < 0) 1626 sock_release_reserved_memory(sk, -delta); 1627 else 1628 ret = sock_reserve_memory(sk, delta); 1629 break; 1630 } 1631 1632 default: 1633 ret = -ENOPROTOOPT; 1634 break; 1635 } 1636 sockopt_release_sock(sk); 1637 return ret; 1638 } 1639 1640 int sock_setsockopt(struct socket *sock, int level, int optname, 1641 sockptr_t optval, unsigned int optlen) 1642 { 1643 return sk_setsockopt(sock->sk, level, optname, 1644 optval, optlen); 1645 } 1646 EXPORT_SYMBOL(sock_setsockopt); 1647 1648 static const struct cred *sk_get_peer_cred(struct sock *sk) 1649 { 1650 const struct cred *cred; 1651 1652 spin_lock(&sk->sk_peer_lock); 1653 cred = get_cred(sk->sk_peer_cred); 1654 spin_unlock(&sk->sk_peer_lock); 1655 1656 return cred; 1657 } 1658 1659 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1660 struct ucred *ucred) 1661 { 1662 ucred->pid = pid_vnr(pid); 1663 ucred->uid = ucred->gid = -1; 1664 if (cred) { 1665 struct user_namespace *current_ns = current_user_ns(); 1666 1667 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1668 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1669 } 1670 } 1671 1672 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1673 { 1674 struct user_namespace *user_ns = current_user_ns(); 1675 int i; 1676 1677 for (i = 0; i < src->ngroups; i++) { 1678 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1679 1680 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1681 return -EFAULT; 1682 } 1683 1684 return 0; 1685 } 1686 1687 int sk_getsockopt(struct sock *sk, int level, int optname, 1688 sockptr_t optval, sockptr_t optlen) 1689 { 1690 struct socket *sock = sk->sk_socket; 1691 1692 union { 1693 int val; 1694 u64 val64; 1695 unsigned long ulval; 1696 struct linger ling; 1697 struct old_timeval32 tm32; 1698 struct __kernel_old_timeval tm; 1699 struct __kernel_sock_timeval stm; 1700 struct sock_txtime txtime; 1701 struct so_timestamping timestamping; 1702 } v; 1703 1704 int lv = sizeof(int); 1705 int len; 1706 1707 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1708 return -EFAULT; 1709 if (len < 0) 1710 return -EINVAL; 1711 1712 memset(&v, 0, sizeof(v)); 1713 1714 switch (optname) { 1715 case SO_DEBUG: 1716 v.val = sock_flag(sk, SOCK_DBG); 1717 break; 1718 1719 case SO_DONTROUTE: 1720 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1721 break; 1722 1723 case SO_BROADCAST: 1724 v.val = sock_flag(sk, SOCK_BROADCAST); 1725 break; 1726 1727 case SO_SNDBUF: 1728 v.val = READ_ONCE(sk->sk_sndbuf); 1729 break; 1730 1731 case SO_RCVBUF: 1732 v.val = READ_ONCE(sk->sk_rcvbuf); 1733 break; 1734 1735 case SO_REUSEADDR: 1736 v.val = sk->sk_reuse; 1737 break; 1738 1739 case SO_REUSEPORT: 1740 v.val = sk->sk_reuseport; 1741 break; 1742 1743 case SO_KEEPALIVE: 1744 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1745 break; 1746 1747 case SO_TYPE: 1748 v.val = sk->sk_type; 1749 break; 1750 1751 case SO_PROTOCOL: 1752 v.val = sk->sk_protocol; 1753 break; 1754 1755 case SO_DOMAIN: 1756 v.val = sk->sk_family; 1757 break; 1758 1759 case SO_ERROR: 1760 v.val = -sock_error(sk); 1761 if (v.val == 0) 1762 v.val = xchg(&sk->sk_err_soft, 0); 1763 break; 1764 1765 case SO_OOBINLINE: 1766 v.val = sock_flag(sk, SOCK_URGINLINE); 1767 break; 1768 1769 case SO_NO_CHECK: 1770 v.val = sk->sk_no_check_tx; 1771 break; 1772 1773 case SO_PRIORITY: 1774 v.val = READ_ONCE(sk->sk_priority); 1775 break; 1776 1777 case SO_LINGER: 1778 lv = sizeof(v.ling); 1779 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1780 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1781 break; 1782 1783 case SO_BSDCOMPAT: 1784 break; 1785 1786 case SO_TIMESTAMP_OLD: 1787 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1788 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1789 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1790 break; 1791 1792 case SO_TIMESTAMPNS_OLD: 1793 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1794 break; 1795 1796 case SO_TIMESTAMP_NEW: 1797 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1798 break; 1799 1800 case SO_TIMESTAMPNS_NEW: 1801 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1802 break; 1803 1804 case SO_TIMESTAMPING_OLD: 1805 case SO_TIMESTAMPING_NEW: 1806 lv = sizeof(v.timestamping); 1807 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1808 * returning the flags when they were set through the same option. 1809 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1810 */ 1811 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1812 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1813 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1814 } 1815 break; 1816 1817 case SO_RCVTIMEO_OLD: 1818 case SO_RCVTIMEO_NEW: 1819 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1820 SO_RCVTIMEO_OLD == optname); 1821 break; 1822 1823 case SO_SNDTIMEO_OLD: 1824 case SO_SNDTIMEO_NEW: 1825 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1826 SO_SNDTIMEO_OLD == optname); 1827 break; 1828 1829 case SO_RCVLOWAT: 1830 v.val = READ_ONCE(sk->sk_rcvlowat); 1831 break; 1832 1833 case SO_SNDLOWAT: 1834 v.val = 1; 1835 break; 1836 1837 case SO_PASSCRED: 1838 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1839 break; 1840 1841 case SO_PASSPIDFD: 1842 v.val = !!test_bit(SOCK_PASSPIDFD, &sock->flags); 1843 break; 1844 1845 case SO_PEERCRED: 1846 { 1847 struct ucred peercred; 1848 if (len > sizeof(peercred)) 1849 len = sizeof(peercred); 1850 1851 spin_lock(&sk->sk_peer_lock); 1852 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1853 spin_unlock(&sk->sk_peer_lock); 1854 1855 if (copy_to_sockptr(optval, &peercred, len)) 1856 return -EFAULT; 1857 goto lenout; 1858 } 1859 1860 case SO_PEERPIDFD: 1861 { 1862 struct pid *peer_pid; 1863 struct file *pidfd_file = NULL; 1864 int pidfd; 1865 1866 if (len > sizeof(pidfd)) 1867 len = sizeof(pidfd); 1868 1869 spin_lock(&sk->sk_peer_lock); 1870 peer_pid = get_pid(sk->sk_peer_pid); 1871 spin_unlock(&sk->sk_peer_lock); 1872 1873 if (!peer_pid) 1874 return -ENODATA; 1875 1876 pidfd = pidfd_prepare(peer_pid, 0, &pidfd_file); 1877 put_pid(peer_pid); 1878 if (pidfd < 0) 1879 return pidfd; 1880 1881 if (copy_to_sockptr(optval, &pidfd, len) || 1882 copy_to_sockptr(optlen, &len, sizeof(int))) { 1883 put_unused_fd(pidfd); 1884 fput(pidfd_file); 1885 1886 return -EFAULT; 1887 } 1888 1889 fd_install(pidfd, pidfd_file); 1890 return 0; 1891 } 1892 1893 case SO_PEERGROUPS: 1894 { 1895 const struct cred *cred; 1896 int ret, n; 1897 1898 cred = sk_get_peer_cred(sk); 1899 if (!cred) 1900 return -ENODATA; 1901 1902 n = cred->group_info->ngroups; 1903 if (len < n * sizeof(gid_t)) { 1904 len = n * sizeof(gid_t); 1905 put_cred(cred); 1906 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1907 } 1908 len = n * sizeof(gid_t); 1909 1910 ret = groups_to_user(optval, cred->group_info); 1911 put_cred(cred); 1912 if (ret) 1913 return ret; 1914 goto lenout; 1915 } 1916 1917 case SO_PEERNAME: 1918 { 1919 struct sockaddr_storage address; 1920 1921 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 1922 if (lv < 0) 1923 return -ENOTCONN; 1924 if (lv < len) 1925 return -EINVAL; 1926 if (copy_to_sockptr(optval, &address, len)) 1927 return -EFAULT; 1928 goto lenout; 1929 } 1930 1931 /* Dubious BSD thing... Probably nobody even uses it, but 1932 * the UNIX standard wants it for whatever reason... -DaveM 1933 */ 1934 case SO_ACCEPTCONN: 1935 v.val = sk->sk_state == TCP_LISTEN; 1936 break; 1937 1938 case SO_PASSSEC: 1939 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1940 break; 1941 1942 case SO_PEERSEC: 1943 return security_socket_getpeersec_stream(sock, 1944 optval, optlen, len); 1945 1946 case SO_MARK: 1947 v.val = READ_ONCE(sk->sk_mark); 1948 break; 1949 1950 case SO_RCVMARK: 1951 v.val = sock_flag(sk, SOCK_RCVMARK); 1952 break; 1953 1954 case SO_RCVPRIORITY: 1955 v.val = sock_flag(sk, SOCK_RCVPRIORITY); 1956 break; 1957 1958 case SO_RXQ_OVFL: 1959 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1960 break; 1961 1962 case SO_WIFI_STATUS: 1963 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1964 break; 1965 1966 case SO_PEEK_OFF: 1967 if (!READ_ONCE(sock->ops)->set_peek_off) 1968 return -EOPNOTSUPP; 1969 1970 v.val = READ_ONCE(sk->sk_peek_off); 1971 break; 1972 case SO_NOFCS: 1973 v.val = sock_flag(sk, SOCK_NOFCS); 1974 break; 1975 1976 case SO_BINDTODEVICE: 1977 return sock_getbindtodevice(sk, optval, optlen, len); 1978 1979 case SO_GET_FILTER: 1980 len = sk_get_filter(sk, optval, len); 1981 if (len < 0) 1982 return len; 1983 1984 goto lenout; 1985 1986 case SO_LOCK_FILTER: 1987 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1988 break; 1989 1990 case SO_BPF_EXTENSIONS: 1991 v.val = bpf_tell_extensions(); 1992 break; 1993 1994 case SO_SELECT_ERR_QUEUE: 1995 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1996 break; 1997 1998 #ifdef CONFIG_NET_RX_BUSY_POLL 1999 case SO_BUSY_POLL: 2000 v.val = READ_ONCE(sk->sk_ll_usec); 2001 break; 2002 case SO_PREFER_BUSY_POLL: 2003 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 2004 break; 2005 #endif 2006 2007 case SO_MAX_PACING_RATE: 2008 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 2009 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 2010 lv = sizeof(v.ulval); 2011 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 2012 } else { 2013 /* 32bit version */ 2014 v.val = min_t(unsigned long, ~0U, 2015 READ_ONCE(sk->sk_max_pacing_rate)); 2016 } 2017 break; 2018 2019 case SO_INCOMING_CPU: 2020 v.val = READ_ONCE(sk->sk_incoming_cpu); 2021 break; 2022 2023 case SO_MEMINFO: 2024 { 2025 u32 meminfo[SK_MEMINFO_VARS]; 2026 2027 sk_get_meminfo(sk, meminfo); 2028 2029 len = min_t(unsigned int, len, sizeof(meminfo)); 2030 if (copy_to_sockptr(optval, &meminfo, len)) 2031 return -EFAULT; 2032 2033 goto lenout; 2034 } 2035 2036 #ifdef CONFIG_NET_RX_BUSY_POLL 2037 case SO_INCOMING_NAPI_ID: 2038 v.val = READ_ONCE(sk->sk_napi_id); 2039 2040 /* aggregate non-NAPI IDs down to 0 */ 2041 if (v.val < MIN_NAPI_ID) 2042 v.val = 0; 2043 2044 break; 2045 #endif 2046 2047 case SO_COOKIE: 2048 lv = sizeof(u64); 2049 if (len < lv) 2050 return -EINVAL; 2051 v.val64 = sock_gen_cookie(sk); 2052 break; 2053 2054 case SO_ZEROCOPY: 2055 v.val = sock_flag(sk, SOCK_ZEROCOPY); 2056 break; 2057 2058 case SO_TXTIME: 2059 lv = sizeof(v.txtime); 2060 v.txtime.clockid = sk->sk_clockid; 2061 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 2062 SOF_TXTIME_DEADLINE_MODE : 0; 2063 v.txtime.flags |= sk->sk_txtime_report_errors ? 2064 SOF_TXTIME_REPORT_ERRORS : 0; 2065 break; 2066 2067 case SO_BINDTOIFINDEX: 2068 v.val = READ_ONCE(sk->sk_bound_dev_if); 2069 break; 2070 2071 case SO_NETNS_COOKIE: 2072 lv = sizeof(u64); 2073 if (len != lv) 2074 return -EINVAL; 2075 v.val64 = sock_net(sk)->net_cookie; 2076 break; 2077 2078 case SO_BUF_LOCK: 2079 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 2080 break; 2081 2082 case SO_RESERVE_MEM: 2083 v.val = READ_ONCE(sk->sk_reserved_mem); 2084 break; 2085 2086 case SO_TXREHASH: 2087 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2088 v.val = READ_ONCE(sk->sk_txrehash); 2089 break; 2090 2091 default: 2092 /* We implement the SO_SNDLOWAT etc to not be settable 2093 * (1003.1g 7). 2094 */ 2095 return -ENOPROTOOPT; 2096 } 2097 2098 if (len > lv) 2099 len = lv; 2100 if (copy_to_sockptr(optval, &v, len)) 2101 return -EFAULT; 2102 lenout: 2103 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2104 return -EFAULT; 2105 return 0; 2106 } 2107 2108 /* 2109 * Initialize an sk_lock. 2110 * 2111 * (We also register the sk_lock with the lock validator.) 2112 */ 2113 static inline void sock_lock_init(struct sock *sk) 2114 { 2115 if (sk->sk_kern_sock) 2116 sock_lock_init_class_and_name( 2117 sk, 2118 af_family_kern_slock_key_strings[sk->sk_family], 2119 af_family_kern_slock_keys + sk->sk_family, 2120 af_family_kern_key_strings[sk->sk_family], 2121 af_family_kern_keys + sk->sk_family); 2122 else 2123 sock_lock_init_class_and_name( 2124 sk, 2125 af_family_slock_key_strings[sk->sk_family], 2126 af_family_slock_keys + sk->sk_family, 2127 af_family_key_strings[sk->sk_family], 2128 af_family_keys + sk->sk_family); 2129 } 2130 2131 /* 2132 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2133 * even temporarily, because of RCU lookups. sk_node should also be left as is. 2134 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2135 */ 2136 static void sock_copy(struct sock *nsk, const struct sock *osk) 2137 { 2138 const struct proto *prot = READ_ONCE(osk->sk_prot); 2139 #ifdef CONFIG_SECURITY_NETWORK 2140 void *sptr = nsk->sk_security; 2141 #endif 2142 2143 /* If we move sk_tx_queue_mapping out of the private section, 2144 * we must check if sk_tx_queue_clear() is called after 2145 * sock_copy() in sk_clone_lock(). 2146 */ 2147 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2148 offsetof(struct sock, sk_dontcopy_begin) || 2149 offsetof(struct sock, sk_tx_queue_mapping) >= 2150 offsetof(struct sock, sk_dontcopy_end)); 2151 2152 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2153 2154 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2155 prot->obj_size - offsetof(struct sock, sk_dontcopy_end), 2156 /* alloc is larger than struct, see sk_prot_alloc() */); 2157 2158 #ifdef CONFIG_SECURITY_NETWORK 2159 nsk->sk_security = sptr; 2160 security_sk_clone(osk, nsk); 2161 #endif 2162 } 2163 2164 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2165 int family) 2166 { 2167 struct sock *sk; 2168 struct kmem_cache *slab; 2169 2170 slab = prot->slab; 2171 if (slab != NULL) { 2172 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2173 if (!sk) 2174 return sk; 2175 if (want_init_on_alloc(priority)) 2176 sk_prot_clear_nulls(sk, prot->obj_size); 2177 } else 2178 sk = kmalloc(prot->obj_size, priority); 2179 2180 if (sk != NULL) { 2181 if (security_sk_alloc(sk, family, priority)) 2182 goto out_free; 2183 2184 if (!try_module_get(prot->owner)) 2185 goto out_free_sec; 2186 } 2187 2188 return sk; 2189 2190 out_free_sec: 2191 security_sk_free(sk); 2192 out_free: 2193 if (slab != NULL) 2194 kmem_cache_free(slab, sk); 2195 else 2196 kfree(sk); 2197 return NULL; 2198 } 2199 2200 static void sk_prot_free(struct proto *prot, struct sock *sk) 2201 { 2202 struct kmem_cache *slab; 2203 struct module *owner; 2204 2205 owner = prot->owner; 2206 slab = prot->slab; 2207 2208 cgroup_sk_free(&sk->sk_cgrp_data); 2209 mem_cgroup_sk_free(sk); 2210 security_sk_free(sk); 2211 if (slab != NULL) 2212 kmem_cache_free(slab, sk); 2213 else 2214 kfree(sk); 2215 module_put(owner); 2216 } 2217 2218 /** 2219 * sk_alloc - All socket objects are allocated here 2220 * @net: the applicable net namespace 2221 * @family: protocol family 2222 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2223 * @prot: struct proto associated with this new sock instance 2224 * @kern: is this to be a kernel socket? 2225 */ 2226 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2227 struct proto *prot, int kern) 2228 { 2229 struct sock *sk; 2230 2231 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2232 if (sk) { 2233 sk->sk_family = family; 2234 /* 2235 * See comment in struct sock definition to understand 2236 * why we need sk_prot_creator -acme 2237 */ 2238 sk->sk_prot = sk->sk_prot_creator = prot; 2239 sk->sk_kern_sock = kern; 2240 sock_lock_init(sk); 2241 sk->sk_net_refcnt = kern ? 0 : 1; 2242 if (likely(sk->sk_net_refcnt)) { 2243 get_net_track(net, &sk->ns_tracker, priority); 2244 sock_inuse_add(net, 1); 2245 } else { 2246 __netns_tracker_alloc(net, &sk->ns_tracker, 2247 false, priority); 2248 } 2249 2250 sock_net_set(sk, net); 2251 refcount_set(&sk->sk_wmem_alloc, 1); 2252 2253 mem_cgroup_sk_alloc(sk); 2254 cgroup_sk_alloc(&sk->sk_cgrp_data); 2255 sock_update_classid(&sk->sk_cgrp_data); 2256 sock_update_netprioidx(&sk->sk_cgrp_data); 2257 sk_tx_queue_clear(sk); 2258 } 2259 2260 return sk; 2261 } 2262 EXPORT_SYMBOL(sk_alloc); 2263 2264 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2265 * grace period. This is the case for UDP sockets and TCP listeners. 2266 */ 2267 static void __sk_destruct(struct rcu_head *head) 2268 { 2269 struct sock *sk = container_of(head, struct sock, sk_rcu); 2270 struct sk_filter *filter; 2271 2272 if (sk->sk_destruct) 2273 sk->sk_destruct(sk); 2274 2275 filter = rcu_dereference_check(sk->sk_filter, 2276 refcount_read(&sk->sk_wmem_alloc) == 0); 2277 if (filter) { 2278 sk_filter_uncharge(sk, filter); 2279 RCU_INIT_POINTER(sk->sk_filter, NULL); 2280 } 2281 2282 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2283 2284 #ifdef CONFIG_BPF_SYSCALL 2285 bpf_sk_storage_free(sk); 2286 #endif 2287 2288 if (atomic_read(&sk->sk_omem_alloc)) 2289 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2290 __func__, atomic_read(&sk->sk_omem_alloc)); 2291 2292 if (sk->sk_frag.page) { 2293 put_page(sk->sk_frag.page); 2294 sk->sk_frag.page = NULL; 2295 } 2296 2297 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2298 put_cred(sk->sk_peer_cred); 2299 put_pid(sk->sk_peer_pid); 2300 2301 if (likely(sk->sk_net_refcnt)) 2302 put_net_track(sock_net(sk), &sk->ns_tracker); 2303 else 2304 __netns_tracker_free(sock_net(sk), &sk->ns_tracker, false); 2305 2306 sk_prot_free(sk->sk_prot_creator, sk); 2307 } 2308 2309 void sk_destruct(struct sock *sk) 2310 { 2311 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2312 2313 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2314 reuseport_detach_sock(sk); 2315 use_call_rcu = true; 2316 } 2317 2318 if (use_call_rcu) 2319 call_rcu(&sk->sk_rcu, __sk_destruct); 2320 else 2321 __sk_destruct(&sk->sk_rcu); 2322 } 2323 2324 static void __sk_free(struct sock *sk) 2325 { 2326 if (likely(sk->sk_net_refcnt)) 2327 sock_inuse_add(sock_net(sk), -1); 2328 2329 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2330 sock_diag_broadcast_destroy(sk); 2331 else 2332 sk_destruct(sk); 2333 } 2334 2335 void sk_free(struct sock *sk) 2336 { 2337 /* 2338 * We subtract one from sk_wmem_alloc and can know if 2339 * some packets are still in some tx queue. 2340 * If not null, sock_wfree() will call __sk_free(sk) later 2341 */ 2342 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2343 __sk_free(sk); 2344 } 2345 EXPORT_SYMBOL(sk_free); 2346 2347 static void sk_init_common(struct sock *sk) 2348 { 2349 skb_queue_head_init(&sk->sk_receive_queue); 2350 skb_queue_head_init(&sk->sk_write_queue); 2351 skb_queue_head_init(&sk->sk_error_queue); 2352 2353 rwlock_init(&sk->sk_callback_lock); 2354 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2355 af_rlock_keys + sk->sk_family, 2356 af_family_rlock_key_strings[sk->sk_family]); 2357 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2358 af_wlock_keys + sk->sk_family, 2359 af_family_wlock_key_strings[sk->sk_family]); 2360 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2361 af_elock_keys + sk->sk_family, 2362 af_family_elock_key_strings[sk->sk_family]); 2363 if (sk->sk_kern_sock) 2364 lockdep_set_class_and_name(&sk->sk_callback_lock, 2365 af_kern_callback_keys + sk->sk_family, 2366 af_family_kern_clock_key_strings[sk->sk_family]); 2367 else 2368 lockdep_set_class_and_name(&sk->sk_callback_lock, 2369 af_callback_keys + sk->sk_family, 2370 af_family_clock_key_strings[sk->sk_family]); 2371 } 2372 2373 /** 2374 * sk_clone_lock - clone a socket, and lock its clone 2375 * @sk: the socket to clone 2376 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2377 * 2378 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2379 */ 2380 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2381 { 2382 struct proto *prot = READ_ONCE(sk->sk_prot); 2383 struct sk_filter *filter; 2384 bool is_charged = true; 2385 struct sock *newsk; 2386 2387 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2388 if (!newsk) 2389 goto out; 2390 2391 sock_copy(newsk, sk); 2392 2393 newsk->sk_prot_creator = prot; 2394 2395 /* SANITY */ 2396 if (likely(newsk->sk_net_refcnt)) { 2397 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2398 sock_inuse_add(sock_net(newsk), 1); 2399 } else { 2400 /* Kernel sockets are not elevating the struct net refcount. 2401 * Instead, use a tracker to more easily detect if a layer 2402 * is not properly dismantling its kernel sockets at netns 2403 * destroy time. 2404 */ 2405 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2406 false, priority); 2407 } 2408 sk_node_init(&newsk->sk_node); 2409 sock_lock_init(newsk); 2410 bh_lock_sock(newsk); 2411 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2412 newsk->sk_backlog.len = 0; 2413 2414 atomic_set(&newsk->sk_rmem_alloc, 0); 2415 2416 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2417 refcount_set(&newsk->sk_wmem_alloc, 1); 2418 2419 atomic_set(&newsk->sk_omem_alloc, 0); 2420 sk_init_common(newsk); 2421 2422 newsk->sk_dst_cache = NULL; 2423 newsk->sk_dst_pending_confirm = 0; 2424 newsk->sk_wmem_queued = 0; 2425 newsk->sk_forward_alloc = 0; 2426 newsk->sk_reserved_mem = 0; 2427 atomic_set(&newsk->sk_drops, 0); 2428 newsk->sk_send_head = NULL; 2429 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2430 atomic_set(&newsk->sk_zckey, 0); 2431 2432 sock_reset_flag(newsk, SOCK_DONE); 2433 2434 /* sk->sk_memcg will be populated at accept() time */ 2435 newsk->sk_memcg = NULL; 2436 2437 cgroup_sk_clone(&newsk->sk_cgrp_data); 2438 2439 rcu_read_lock(); 2440 filter = rcu_dereference(sk->sk_filter); 2441 if (filter != NULL) 2442 /* though it's an empty new sock, the charging may fail 2443 * if sysctl_optmem_max was changed between creation of 2444 * original socket and cloning 2445 */ 2446 is_charged = sk_filter_charge(newsk, filter); 2447 RCU_INIT_POINTER(newsk->sk_filter, filter); 2448 rcu_read_unlock(); 2449 2450 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2451 /* We need to make sure that we don't uncharge the new 2452 * socket if we couldn't charge it in the first place 2453 * as otherwise we uncharge the parent's filter. 2454 */ 2455 if (!is_charged) 2456 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2457 sk_free_unlock_clone(newsk); 2458 newsk = NULL; 2459 goto out; 2460 } 2461 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2462 2463 if (bpf_sk_storage_clone(sk, newsk)) { 2464 sk_free_unlock_clone(newsk); 2465 newsk = NULL; 2466 goto out; 2467 } 2468 2469 /* Clear sk_user_data if parent had the pointer tagged 2470 * as not suitable for copying when cloning. 2471 */ 2472 if (sk_user_data_is_nocopy(newsk)) 2473 newsk->sk_user_data = NULL; 2474 2475 newsk->sk_err = 0; 2476 newsk->sk_err_soft = 0; 2477 newsk->sk_priority = 0; 2478 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2479 2480 /* Before updating sk_refcnt, we must commit prior changes to memory 2481 * (Documentation/RCU/rculist_nulls.rst for details) 2482 */ 2483 smp_wmb(); 2484 refcount_set(&newsk->sk_refcnt, 2); 2485 2486 sk_set_socket(newsk, NULL); 2487 sk_tx_queue_clear(newsk); 2488 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2489 2490 if (newsk->sk_prot->sockets_allocated) 2491 sk_sockets_allocated_inc(newsk); 2492 2493 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2494 net_enable_timestamp(); 2495 out: 2496 return newsk; 2497 } 2498 EXPORT_SYMBOL_GPL(sk_clone_lock); 2499 2500 void sk_free_unlock_clone(struct sock *sk) 2501 { 2502 /* It is still raw copy of parent, so invalidate 2503 * destructor and make plain sk_free() */ 2504 sk->sk_destruct = NULL; 2505 bh_unlock_sock(sk); 2506 sk_free(sk); 2507 } 2508 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2509 2510 static u32 sk_dst_gso_max_size(struct sock *sk, struct dst_entry *dst) 2511 { 2512 bool is_ipv6 = false; 2513 u32 max_size; 2514 2515 #if IS_ENABLED(CONFIG_IPV6) 2516 is_ipv6 = (sk->sk_family == AF_INET6 && 2517 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2518 #endif 2519 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2520 max_size = is_ipv6 ? READ_ONCE(dst->dev->gso_max_size) : 2521 READ_ONCE(dst->dev->gso_ipv4_max_size); 2522 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2523 max_size = GSO_LEGACY_MAX_SIZE; 2524 2525 return max_size - (MAX_TCP_HEADER + 1); 2526 } 2527 2528 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2529 { 2530 u32 max_segs = 1; 2531 2532 sk->sk_route_caps = dst->dev->features; 2533 if (sk_is_tcp(sk)) 2534 sk->sk_route_caps |= NETIF_F_GSO; 2535 if (sk->sk_route_caps & NETIF_F_GSO) 2536 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2537 if (unlikely(sk->sk_gso_disabled)) 2538 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2539 if (sk_can_gso(sk)) { 2540 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2541 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2542 } else { 2543 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2544 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dst); 2545 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2546 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2547 } 2548 } 2549 sk->sk_gso_max_segs = max_segs; 2550 sk_dst_set(sk, dst); 2551 } 2552 EXPORT_SYMBOL_GPL(sk_setup_caps); 2553 2554 /* 2555 * Simple resource managers for sockets. 2556 */ 2557 2558 2559 /* 2560 * Write buffer destructor automatically called from kfree_skb. 2561 */ 2562 void sock_wfree(struct sk_buff *skb) 2563 { 2564 struct sock *sk = skb->sk; 2565 unsigned int len = skb->truesize; 2566 bool free; 2567 2568 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2569 if (sock_flag(sk, SOCK_RCU_FREE) && 2570 sk->sk_write_space == sock_def_write_space) { 2571 rcu_read_lock(); 2572 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2573 sock_def_write_space_wfree(sk); 2574 rcu_read_unlock(); 2575 if (unlikely(free)) 2576 __sk_free(sk); 2577 return; 2578 } 2579 2580 /* 2581 * Keep a reference on sk_wmem_alloc, this will be released 2582 * after sk_write_space() call 2583 */ 2584 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2585 sk->sk_write_space(sk); 2586 len = 1; 2587 } 2588 /* 2589 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2590 * could not do because of in-flight packets 2591 */ 2592 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2593 __sk_free(sk); 2594 } 2595 EXPORT_SYMBOL(sock_wfree); 2596 2597 /* This variant of sock_wfree() is used by TCP, 2598 * since it sets SOCK_USE_WRITE_QUEUE. 2599 */ 2600 void __sock_wfree(struct sk_buff *skb) 2601 { 2602 struct sock *sk = skb->sk; 2603 2604 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2605 __sk_free(sk); 2606 } 2607 2608 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2609 { 2610 skb_orphan(skb); 2611 #ifdef CONFIG_INET 2612 if (unlikely(!sk_fullsock(sk))) 2613 return skb_set_owner_edemux(skb, sk); 2614 #endif 2615 skb->sk = sk; 2616 skb->destructor = sock_wfree; 2617 skb_set_hash_from_sk(skb, sk); 2618 /* 2619 * We used to take a refcount on sk, but following operation 2620 * is enough to guarantee sk_free() won't free this sock until 2621 * all in-flight packets are completed 2622 */ 2623 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2624 } 2625 EXPORT_SYMBOL(skb_set_owner_w); 2626 2627 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2628 { 2629 /* Drivers depend on in-order delivery for crypto offload, 2630 * partial orphan breaks out-of-order-OK logic. 2631 */ 2632 if (skb_is_decrypted(skb)) 2633 return false; 2634 2635 return (skb->destructor == sock_wfree || 2636 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2637 } 2638 2639 /* This helper is used by netem, as it can hold packets in its 2640 * delay queue. We want to allow the owner socket to send more 2641 * packets, as if they were already TX completed by a typical driver. 2642 * But we also want to keep skb->sk set because some packet schedulers 2643 * rely on it (sch_fq for example). 2644 */ 2645 void skb_orphan_partial(struct sk_buff *skb) 2646 { 2647 if (skb_is_tcp_pure_ack(skb)) 2648 return; 2649 2650 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2651 return; 2652 2653 skb_orphan(skb); 2654 } 2655 EXPORT_SYMBOL(skb_orphan_partial); 2656 2657 /* 2658 * Read buffer destructor automatically called from kfree_skb. 2659 */ 2660 void sock_rfree(struct sk_buff *skb) 2661 { 2662 struct sock *sk = skb->sk; 2663 unsigned int len = skb->truesize; 2664 2665 atomic_sub(len, &sk->sk_rmem_alloc); 2666 sk_mem_uncharge(sk, len); 2667 } 2668 EXPORT_SYMBOL(sock_rfree); 2669 2670 /* 2671 * Buffer destructor for skbs that are not used directly in read or write 2672 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2673 */ 2674 void sock_efree(struct sk_buff *skb) 2675 { 2676 sock_put(skb->sk); 2677 } 2678 EXPORT_SYMBOL(sock_efree); 2679 2680 /* Buffer destructor for prefetch/receive path where reference count may 2681 * not be held, e.g. for listen sockets. 2682 */ 2683 #ifdef CONFIG_INET 2684 void sock_pfree(struct sk_buff *skb) 2685 { 2686 struct sock *sk = skb->sk; 2687 2688 if (!sk_is_refcounted(sk)) 2689 return; 2690 2691 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2692 inet_reqsk(sk)->rsk_listener = NULL; 2693 reqsk_free(inet_reqsk(sk)); 2694 return; 2695 } 2696 2697 sock_gen_put(sk); 2698 } 2699 EXPORT_SYMBOL(sock_pfree); 2700 #endif /* CONFIG_INET */ 2701 2702 kuid_t sock_i_uid(struct sock *sk) 2703 { 2704 kuid_t uid; 2705 2706 read_lock_bh(&sk->sk_callback_lock); 2707 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2708 read_unlock_bh(&sk->sk_callback_lock); 2709 return uid; 2710 } 2711 EXPORT_SYMBOL(sock_i_uid); 2712 2713 unsigned long __sock_i_ino(struct sock *sk) 2714 { 2715 unsigned long ino; 2716 2717 read_lock(&sk->sk_callback_lock); 2718 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2719 read_unlock(&sk->sk_callback_lock); 2720 return ino; 2721 } 2722 EXPORT_SYMBOL(__sock_i_ino); 2723 2724 unsigned long sock_i_ino(struct sock *sk) 2725 { 2726 unsigned long ino; 2727 2728 local_bh_disable(); 2729 ino = __sock_i_ino(sk); 2730 local_bh_enable(); 2731 return ino; 2732 } 2733 EXPORT_SYMBOL(sock_i_ino); 2734 2735 /* 2736 * Allocate a skb from the socket's send buffer. 2737 */ 2738 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2739 gfp_t priority) 2740 { 2741 if (force || 2742 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2743 struct sk_buff *skb = alloc_skb(size, priority); 2744 2745 if (skb) { 2746 skb_set_owner_w(skb, sk); 2747 return skb; 2748 } 2749 } 2750 return NULL; 2751 } 2752 EXPORT_SYMBOL(sock_wmalloc); 2753 2754 static void sock_ofree(struct sk_buff *skb) 2755 { 2756 struct sock *sk = skb->sk; 2757 2758 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2759 } 2760 2761 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2762 gfp_t priority) 2763 { 2764 struct sk_buff *skb; 2765 2766 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2767 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2768 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2769 return NULL; 2770 2771 skb = alloc_skb(size, priority); 2772 if (!skb) 2773 return NULL; 2774 2775 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2776 skb->sk = sk; 2777 skb->destructor = sock_ofree; 2778 return skb; 2779 } 2780 2781 /* 2782 * Allocate a memory block from the socket's option memory buffer. 2783 */ 2784 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2785 { 2786 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2787 2788 if ((unsigned int)size <= optmem_max && 2789 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2790 void *mem; 2791 /* First do the add, to avoid the race if kmalloc 2792 * might sleep. 2793 */ 2794 atomic_add(size, &sk->sk_omem_alloc); 2795 mem = kmalloc(size, priority); 2796 if (mem) 2797 return mem; 2798 atomic_sub(size, &sk->sk_omem_alloc); 2799 } 2800 return NULL; 2801 } 2802 EXPORT_SYMBOL(sock_kmalloc); 2803 2804 /* Free an option memory block. Note, we actually want the inline 2805 * here as this allows gcc to detect the nullify and fold away the 2806 * condition entirely. 2807 */ 2808 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2809 const bool nullify) 2810 { 2811 if (WARN_ON_ONCE(!mem)) 2812 return; 2813 if (nullify) 2814 kfree_sensitive(mem); 2815 else 2816 kfree(mem); 2817 atomic_sub(size, &sk->sk_omem_alloc); 2818 } 2819 2820 void sock_kfree_s(struct sock *sk, void *mem, int size) 2821 { 2822 __sock_kfree_s(sk, mem, size, false); 2823 } 2824 EXPORT_SYMBOL(sock_kfree_s); 2825 2826 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2827 { 2828 __sock_kfree_s(sk, mem, size, true); 2829 } 2830 EXPORT_SYMBOL(sock_kzfree_s); 2831 2832 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2833 I think, these locks should be removed for datagram sockets. 2834 */ 2835 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2836 { 2837 DEFINE_WAIT(wait); 2838 2839 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2840 for (;;) { 2841 if (!timeo) 2842 break; 2843 if (signal_pending(current)) 2844 break; 2845 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2846 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2847 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2848 break; 2849 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2850 break; 2851 if (READ_ONCE(sk->sk_err)) 2852 break; 2853 timeo = schedule_timeout(timeo); 2854 } 2855 finish_wait(sk_sleep(sk), &wait); 2856 return timeo; 2857 } 2858 2859 2860 /* 2861 * Generic send/receive buffer handlers 2862 */ 2863 2864 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2865 unsigned long data_len, int noblock, 2866 int *errcode, int max_page_order) 2867 { 2868 struct sk_buff *skb; 2869 long timeo; 2870 int err; 2871 2872 timeo = sock_sndtimeo(sk, noblock); 2873 for (;;) { 2874 err = sock_error(sk); 2875 if (err != 0) 2876 goto failure; 2877 2878 err = -EPIPE; 2879 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2880 goto failure; 2881 2882 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2883 break; 2884 2885 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2886 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2887 err = -EAGAIN; 2888 if (!timeo) 2889 goto failure; 2890 if (signal_pending(current)) 2891 goto interrupted; 2892 timeo = sock_wait_for_wmem(sk, timeo); 2893 } 2894 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2895 errcode, sk->sk_allocation); 2896 if (skb) 2897 skb_set_owner_w(skb, sk); 2898 return skb; 2899 2900 interrupted: 2901 err = sock_intr_errno(timeo); 2902 failure: 2903 *errcode = err; 2904 return NULL; 2905 } 2906 EXPORT_SYMBOL(sock_alloc_send_pskb); 2907 2908 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 2909 struct sockcm_cookie *sockc) 2910 { 2911 u32 tsflags; 2912 2913 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 2914 2915 switch (cmsg->cmsg_type) { 2916 case SO_MARK: 2917 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2918 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2919 return -EPERM; 2920 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2921 return -EINVAL; 2922 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2923 break; 2924 case SO_TIMESTAMPING_OLD: 2925 case SO_TIMESTAMPING_NEW: 2926 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2927 return -EINVAL; 2928 2929 tsflags = *(u32 *)CMSG_DATA(cmsg); 2930 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2931 return -EINVAL; 2932 2933 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2934 sockc->tsflags |= tsflags; 2935 break; 2936 case SCM_TXTIME: 2937 if (!sock_flag(sk, SOCK_TXTIME)) 2938 return -EINVAL; 2939 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2940 return -EINVAL; 2941 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2942 break; 2943 case SCM_TS_OPT_ID: 2944 if (sk_is_tcp(sk)) 2945 return -EINVAL; 2946 tsflags = READ_ONCE(sk->sk_tsflags); 2947 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 2948 return -EINVAL; 2949 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2950 return -EINVAL; 2951 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 2952 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 2953 break; 2954 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2955 case SCM_RIGHTS: 2956 case SCM_CREDENTIALS: 2957 break; 2958 case SO_PRIORITY: 2959 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2960 return -EINVAL; 2961 if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg))) 2962 return -EPERM; 2963 sockc->priority = *(u32 *)CMSG_DATA(cmsg); 2964 break; 2965 default: 2966 return -EINVAL; 2967 } 2968 return 0; 2969 } 2970 EXPORT_SYMBOL(__sock_cmsg_send); 2971 2972 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2973 struct sockcm_cookie *sockc) 2974 { 2975 struct cmsghdr *cmsg; 2976 int ret; 2977 2978 for_each_cmsghdr(cmsg, msg) { 2979 if (!CMSG_OK(msg, cmsg)) 2980 return -EINVAL; 2981 if (cmsg->cmsg_level != SOL_SOCKET) 2982 continue; 2983 ret = __sock_cmsg_send(sk, cmsg, sockc); 2984 if (ret) 2985 return ret; 2986 } 2987 return 0; 2988 } 2989 EXPORT_SYMBOL(sock_cmsg_send); 2990 2991 static void sk_enter_memory_pressure(struct sock *sk) 2992 { 2993 if (!sk->sk_prot->enter_memory_pressure) 2994 return; 2995 2996 sk->sk_prot->enter_memory_pressure(sk); 2997 } 2998 2999 static void sk_leave_memory_pressure(struct sock *sk) 3000 { 3001 if (sk->sk_prot->leave_memory_pressure) { 3002 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 3003 tcp_leave_memory_pressure, sk); 3004 } else { 3005 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 3006 3007 if (memory_pressure && READ_ONCE(*memory_pressure)) 3008 WRITE_ONCE(*memory_pressure, 0); 3009 } 3010 } 3011 3012 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3013 3014 /** 3015 * skb_page_frag_refill - check that a page_frag contains enough room 3016 * @sz: minimum size of the fragment we want to get 3017 * @pfrag: pointer to page_frag 3018 * @gfp: priority for memory allocation 3019 * 3020 * Note: While this allocator tries to use high order pages, there is 3021 * no guarantee that allocations succeed. Therefore, @sz MUST be 3022 * less or equal than PAGE_SIZE. 3023 */ 3024 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 3025 { 3026 if (pfrag->page) { 3027 if (page_ref_count(pfrag->page) == 1) { 3028 pfrag->offset = 0; 3029 return true; 3030 } 3031 if (pfrag->offset + sz <= pfrag->size) 3032 return true; 3033 put_page(pfrag->page); 3034 } 3035 3036 pfrag->offset = 0; 3037 if (SKB_FRAG_PAGE_ORDER && 3038 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3039 /* Avoid direct reclaim but allow kswapd to wake */ 3040 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3041 __GFP_COMP | __GFP_NOWARN | 3042 __GFP_NORETRY, 3043 SKB_FRAG_PAGE_ORDER); 3044 if (likely(pfrag->page)) { 3045 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3046 return true; 3047 } 3048 } 3049 pfrag->page = alloc_page(gfp); 3050 if (likely(pfrag->page)) { 3051 pfrag->size = PAGE_SIZE; 3052 return true; 3053 } 3054 return false; 3055 } 3056 EXPORT_SYMBOL(skb_page_frag_refill); 3057 3058 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3059 { 3060 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3061 return true; 3062 3063 sk_enter_memory_pressure(sk); 3064 sk_stream_moderate_sndbuf(sk); 3065 return false; 3066 } 3067 EXPORT_SYMBOL(sk_page_frag_refill); 3068 3069 void __lock_sock(struct sock *sk) 3070 __releases(&sk->sk_lock.slock) 3071 __acquires(&sk->sk_lock.slock) 3072 { 3073 DEFINE_WAIT(wait); 3074 3075 for (;;) { 3076 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3077 TASK_UNINTERRUPTIBLE); 3078 spin_unlock_bh(&sk->sk_lock.slock); 3079 schedule(); 3080 spin_lock_bh(&sk->sk_lock.slock); 3081 if (!sock_owned_by_user(sk)) 3082 break; 3083 } 3084 finish_wait(&sk->sk_lock.wq, &wait); 3085 } 3086 3087 void __release_sock(struct sock *sk) 3088 __releases(&sk->sk_lock.slock) 3089 __acquires(&sk->sk_lock.slock) 3090 { 3091 struct sk_buff *skb, *next; 3092 3093 while ((skb = sk->sk_backlog.head) != NULL) { 3094 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3095 3096 spin_unlock_bh(&sk->sk_lock.slock); 3097 3098 do { 3099 next = skb->next; 3100 prefetch(next); 3101 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3102 skb_mark_not_on_list(skb); 3103 sk_backlog_rcv(sk, skb); 3104 3105 cond_resched(); 3106 3107 skb = next; 3108 } while (skb != NULL); 3109 3110 spin_lock_bh(&sk->sk_lock.slock); 3111 } 3112 3113 /* 3114 * Doing the zeroing here guarantee we can not loop forever 3115 * while a wild producer attempts to flood us. 3116 */ 3117 sk->sk_backlog.len = 0; 3118 } 3119 3120 void __sk_flush_backlog(struct sock *sk) 3121 { 3122 spin_lock_bh(&sk->sk_lock.slock); 3123 __release_sock(sk); 3124 3125 if (sk->sk_prot->release_cb) 3126 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3127 tcp_release_cb, sk); 3128 3129 spin_unlock_bh(&sk->sk_lock.slock); 3130 } 3131 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3132 3133 /** 3134 * sk_wait_data - wait for data to arrive at sk_receive_queue 3135 * @sk: sock to wait on 3136 * @timeo: for how long 3137 * @skb: last skb seen on sk_receive_queue 3138 * 3139 * Now socket state including sk->sk_err is changed only under lock, 3140 * hence we may omit checks after joining wait queue. 3141 * We check receive queue before schedule() only as optimization; 3142 * it is very likely that release_sock() added new data. 3143 */ 3144 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3145 { 3146 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3147 int rc; 3148 3149 add_wait_queue(sk_sleep(sk), &wait); 3150 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3151 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3152 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3153 remove_wait_queue(sk_sleep(sk), &wait); 3154 return rc; 3155 } 3156 EXPORT_SYMBOL(sk_wait_data); 3157 3158 /** 3159 * __sk_mem_raise_allocated - increase memory_allocated 3160 * @sk: socket 3161 * @size: memory size to allocate 3162 * @amt: pages to allocate 3163 * @kind: allocation type 3164 * 3165 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3166 * 3167 * Unlike the globally shared limits among the sockets under same protocol, 3168 * consuming the budget of a memcg won't have direct effect on other ones. 3169 * So be optimistic about memcg's tolerance, and leave the callers to decide 3170 * whether or not to raise allocated through sk_under_memory_pressure() or 3171 * its variants. 3172 */ 3173 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3174 { 3175 struct mem_cgroup *memcg = mem_cgroup_sockets_enabled ? sk->sk_memcg : NULL; 3176 struct proto *prot = sk->sk_prot; 3177 bool charged = false; 3178 long allocated; 3179 3180 sk_memory_allocated_add(sk, amt); 3181 allocated = sk_memory_allocated(sk); 3182 3183 if (memcg) { 3184 if (!mem_cgroup_charge_skmem(memcg, amt, gfp_memcg_charge())) 3185 goto suppress_allocation; 3186 charged = true; 3187 } 3188 3189 /* Under limit. */ 3190 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3191 sk_leave_memory_pressure(sk); 3192 return 1; 3193 } 3194 3195 /* Under pressure. */ 3196 if (allocated > sk_prot_mem_limits(sk, 1)) 3197 sk_enter_memory_pressure(sk); 3198 3199 /* Over hard limit. */ 3200 if (allocated > sk_prot_mem_limits(sk, 2)) 3201 goto suppress_allocation; 3202 3203 /* Guarantee minimum buffer size under pressure (either global 3204 * or memcg) to make sure features described in RFC 7323 (TCP 3205 * Extensions for High Performance) work properly. 3206 * 3207 * This rule does NOT stand when exceeds global or memcg's hard 3208 * limit, or else a DoS attack can be taken place by spawning 3209 * lots of sockets whose usage are under minimum buffer size. 3210 */ 3211 if (kind == SK_MEM_RECV) { 3212 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3213 return 1; 3214 3215 } else { /* SK_MEM_SEND */ 3216 int wmem0 = sk_get_wmem0(sk, prot); 3217 3218 if (sk->sk_type == SOCK_STREAM) { 3219 if (sk->sk_wmem_queued < wmem0) 3220 return 1; 3221 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3222 return 1; 3223 } 3224 } 3225 3226 if (sk_has_memory_pressure(sk)) { 3227 u64 alloc; 3228 3229 /* The following 'average' heuristic is within the 3230 * scope of global accounting, so it only makes 3231 * sense for global memory pressure. 3232 */ 3233 if (!sk_under_global_memory_pressure(sk)) 3234 return 1; 3235 3236 /* Try to be fair among all the sockets under global 3237 * pressure by allowing the ones that below average 3238 * usage to raise. 3239 */ 3240 alloc = sk_sockets_allocated_read_positive(sk); 3241 if (sk_prot_mem_limits(sk, 2) > alloc * 3242 sk_mem_pages(sk->sk_wmem_queued + 3243 atomic_read(&sk->sk_rmem_alloc) + 3244 sk->sk_forward_alloc)) 3245 return 1; 3246 } 3247 3248 suppress_allocation: 3249 3250 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3251 sk_stream_moderate_sndbuf(sk); 3252 3253 /* Fail only if socket is _under_ its sndbuf. 3254 * In this case we cannot block, so that we have to fail. 3255 */ 3256 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3257 /* Force charge with __GFP_NOFAIL */ 3258 if (memcg && !charged) { 3259 mem_cgroup_charge_skmem(memcg, amt, 3260 gfp_memcg_charge() | __GFP_NOFAIL); 3261 } 3262 return 1; 3263 } 3264 } 3265 3266 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3267 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3268 3269 sk_memory_allocated_sub(sk, amt); 3270 3271 if (charged) 3272 mem_cgroup_uncharge_skmem(memcg, amt); 3273 3274 return 0; 3275 } 3276 3277 /** 3278 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3279 * @sk: socket 3280 * @size: memory size to allocate 3281 * @kind: allocation type 3282 * 3283 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3284 * rmem allocation. This function assumes that protocols which have 3285 * memory_pressure use sk_wmem_queued as write buffer accounting. 3286 */ 3287 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3288 { 3289 int ret, amt = sk_mem_pages(size); 3290 3291 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3292 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3293 if (!ret) 3294 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3295 return ret; 3296 } 3297 EXPORT_SYMBOL(__sk_mem_schedule); 3298 3299 /** 3300 * __sk_mem_reduce_allocated - reclaim memory_allocated 3301 * @sk: socket 3302 * @amount: number of quanta 3303 * 3304 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3305 */ 3306 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3307 { 3308 sk_memory_allocated_sub(sk, amount); 3309 3310 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3311 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3312 3313 if (sk_under_global_memory_pressure(sk) && 3314 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3315 sk_leave_memory_pressure(sk); 3316 } 3317 3318 /** 3319 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3320 * @sk: socket 3321 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3322 */ 3323 void __sk_mem_reclaim(struct sock *sk, int amount) 3324 { 3325 amount >>= PAGE_SHIFT; 3326 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3327 __sk_mem_reduce_allocated(sk, amount); 3328 } 3329 EXPORT_SYMBOL(__sk_mem_reclaim); 3330 3331 int sk_set_peek_off(struct sock *sk, int val) 3332 { 3333 WRITE_ONCE(sk->sk_peek_off, val); 3334 return 0; 3335 } 3336 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3337 3338 /* 3339 * Set of default routines for initialising struct proto_ops when 3340 * the protocol does not support a particular function. In certain 3341 * cases where it makes no sense for a protocol to have a "do nothing" 3342 * function, some default processing is provided. 3343 */ 3344 3345 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3346 { 3347 return -EOPNOTSUPP; 3348 } 3349 EXPORT_SYMBOL(sock_no_bind); 3350 3351 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3352 int len, int flags) 3353 { 3354 return -EOPNOTSUPP; 3355 } 3356 EXPORT_SYMBOL(sock_no_connect); 3357 3358 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3359 { 3360 return -EOPNOTSUPP; 3361 } 3362 EXPORT_SYMBOL(sock_no_socketpair); 3363 3364 int sock_no_accept(struct socket *sock, struct socket *newsock, 3365 struct proto_accept_arg *arg) 3366 { 3367 return -EOPNOTSUPP; 3368 } 3369 EXPORT_SYMBOL(sock_no_accept); 3370 3371 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3372 int peer) 3373 { 3374 return -EOPNOTSUPP; 3375 } 3376 EXPORT_SYMBOL(sock_no_getname); 3377 3378 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3379 { 3380 return -EOPNOTSUPP; 3381 } 3382 EXPORT_SYMBOL(sock_no_ioctl); 3383 3384 int sock_no_listen(struct socket *sock, int backlog) 3385 { 3386 return -EOPNOTSUPP; 3387 } 3388 EXPORT_SYMBOL(sock_no_listen); 3389 3390 int sock_no_shutdown(struct socket *sock, int how) 3391 { 3392 return -EOPNOTSUPP; 3393 } 3394 EXPORT_SYMBOL(sock_no_shutdown); 3395 3396 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3397 { 3398 return -EOPNOTSUPP; 3399 } 3400 EXPORT_SYMBOL(sock_no_sendmsg); 3401 3402 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3403 { 3404 return -EOPNOTSUPP; 3405 } 3406 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3407 3408 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3409 int flags) 3410 { 3411 return -EOPNOTSUPP; 3412 } 3413 EXPORT_SYMBOL(sock_no_recvmsg); 3414 3415 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3416 { 3417 /* Mirror missing mmap method error code */ 3418 return -ENODEV; 3419 } 3420 EXPORT_SYMBOL(sock_no_mmap); 3421 3422 /* 3423 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3424 * various sock-based usage counts. 3425 */ 3426 void __receive_sock(struct file *file) 3427 { 3428 struct socket *sock; 3429 3430 sock = sock_from_file(file); 3431 if (sock) { 3432 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3433 sock_update_classid(&sock->sk->sk_cgrp_data); 3434 } 3435 } 3436 3437 /* 3438 * Default Socket Callbacks 3439 */ 3440 3441 static void sock_def_wakeup(struct sock *sk) 3442 { 3443 struct socket_wq *wq; 3444 3445 rcu_read_lock(); 3446 wq = rcu_dereference(sk->sk_wq); 3447 if (skwq_has_sleeper(wq)) 3448 wake_up_interruptible_all(&wq->wait); 3449 rcu_read_unlock(); 3450 } 3451 3452 static void sock_def_error_report(struct sock *sk) 3453 { 3454 struct socket_wq *wq; 3455 3456 rcu_read_lock(); 3457 wq = rcu_dereference(sk->sk_wq); 3458 if (skwq_has_sleeper(wq)) 3459 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3460 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3461 rcu_read_unlock(); 3462 } 3463 3464 void sock_def_readable(struct sock *sk) 3465 { 3466 struct socket_wq *wq; 3467 3468 trace_sk_data_ready(sk); 3469 3470 rcu_read_lock(); 3471 wq = rcu_dereference(sk->sk_wq); 3472 if (skwq_has_sleeper(wq)) 3473 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3474 EPOLLRDNORM | EPOLLRDBAND); 3475 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3476 rcu_read_unlock(); 3477 } 3478 3479 static void sock_def_write_space(struct sock *sk) 3480 { 3481 struct socket_wq *wq; 3482 3483 rcu_read_lock(); 3484 3485 /* Do not wake up a writer until he can make "significant" 3486 * progress. --DaveM 3487 */ 3488 if (sock_writeable(sk)) { 3489 wq = rcu_dereference(sk->sk_wq); 3490 if (skwq_has_sleeper(wq)) 3491 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3492 EPOLLWRNORM | EPOLLWRBAND); 3493 3494 /* Should agree with poll, otherwise some programs break */ 3495 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3496 } 3497 3498 rcu_read_unlock(); 3499 } 3500 3501 /* An optimised version of sock_def_write_space(), should only be called 3502 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3503 * ->sk_wmem_alloc. 3504 */ 3505 static void sock_def_write_space_wfree(struct sock *sk) 3506 { 3507 /* Do not wake up a writer until he can make "significant" 3508 * progress. --DaveM 3509 */ 3510 if (sock_writeable(sk)) { 3511 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3512 3513 /* rely on refcount_sub from sock_wfree() */ 3514 smp_mb__after_atomic(); 3515 if (wq && waitqueue_active(&wq->wait)) 3516 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3517 EPOLLWRNORM | EPOLLWRBAND); 3518 3519 /* Should agree with poll, otherwise some programs break */ 3520 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3521 } 3522 } 3523 3524 static void sock_def_destruct(struct sock *sk) 3525 { 3526 } 3527 3528 void sk_send_sigurg(struct sock *sk) 3529 { 3530 if (sk->sk_socket && sk->sk_socket->file) 3531 if (send_sigurg(sk->sk_socket->file)) 3532 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3533 } 3534 EXPORT_SYMBOL(sk_send_sigurg); 3535 3536 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3537 unsigned long expires) 3538 { 3539 if (!mod_timer(timer, expires)) 3540 sock_hold(sk); 3541 } 3542 EXPORT_SYMBOL(sk_reset_timer); 3543 3544 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3545 { 3546 if (del_timer(timer)) 3547 __sock_put(sk); 3548 } 3549 EXPORT_SYMBOL(sk_stop_timer); 3550 3551 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3552 { 3553 if (del_timer_sync(timer)) 3554 __sock_put(sk); 3555 } 3556 EXPORT_SYMBOL(sk_stop_timer_sync); 3557 3558 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3559 { 3560 sk_init_common(sk); 3561 sk->sk_send_head = NULL; 3562 3563 timer_setup(&sk->sk_timer, NULL, 0); 3564 3565 sk->sk_allocation = GFP_KERNEL; 3566 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3567 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3568 sk->sk_state = TCP_CLOSE; 3569 sk->sk_use_task_frag = true; 3570 sk_set_socket(sk, sock); 3571 3572 sock_set_flag(sk, SOCK_ZAPPED); 3573 3574 if (sock) { 3575 sk->sk_type = sock->type; 3576 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3577 sock->sk = sk; 3578 } else { 3579 RCU_INIT_POINTER(sk->sk_wq, NULL); 3580 } 3581 sk->sk_uid = uid; 3582 3583 sk->sk_state_change = sock_def_wakeup; 3584 sk->sk_data_ready = sock_def_readable; 3585 sk->sk_write_space = sock_def_write_space; 3586 sk->sk_error_report = sock_def_error_report; 3587 sk->sk_destruct = sock_def_destruct; 3588 3589 sk->sk_frag.page = NULL; 3590 sk->sk_frag.offset = 0; 3591 sk->sk_peek_off = -1; 3592 3593 sk->sk_peer_pid = NULL; 3594 sk->sk_peer_cred = NULL; 3595 spin_lock_init(&sk->sk_peer_lock); 3596 3597 sk->sk_write_pending = 0; 3598 sk->sk_rcvlowat = 1; 3599 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3600 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3601 3602 sk->sk_stamp = SK_DEFAULT_STAMP; 3603 #if BITS_PER_LONG==32 3604 seqlock_init(&sk->sk_stamp_seq); 3605 #endif 3606 atomic_set(&sk->sk_zckey, 0); 3607 3608 #ifdef CONFIG_NET_RX_BUSY_POLL 3609 sk->sk_napi_id = 0; 3610 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3611 #endif 3612 3613 sk->sk_max_pacing_rate = ~0UL; 3614 sk->sk_pacing_rate = ~0UL; 3615 WRITE_ONCE(sk->sk_pacing_shift, 10); 3616 sk->sk_incoming_cpu = -1; 3617 3618 sk_rx_queue_clear(sk); 3619 /* 3620 * Before updating sk_refcnt, we must commit prior changes to memory 3621 * (Documentation/RCU/rculist_nulls.rst for details) 3622 */ 3623 smp_wmb(); 3624 refcount_set(&sk->sk_refcnt, 1); 3625 atomic_set(&sk->sk_drops, 0); 3626 } 3627 EXPORT_SYMBOL(sock_init_data_uid); 3628 3629 void sock_init_data(struct socket *sock, struct sock *sk) 3630 { 3631 kuid_t uid = sock ? 3632 SOCK_INODE(sock)->i_uid : 3633 make_kuid(sock_net(sk)->user_ns, 0); 3634 3635 sock_init_data_uid(sock, sk, uid); 3636 } 3637 EXPORT_SYMBOL(sock_init_data); 3638 3639 void lock_sock_nested(struct sock *sk, int subclass) 3640 { 3641 /* The sk_lock has mutex_lock() semantics here. */ 3642 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3643 3644 might_sleep(); 3645 spin_lock_bh(&sk->sk_lock.slock); 3646 if (sock_owned_by_user_nocheck(sk)) 3647 __lock_sock(sk); 3648 sk->sk_lock.owned = 1; 3649 spin_unlock_bh(&sk->sk_lock.slock); 3650 } 3651 EXPORT_SYMBOL(lock_sock_nested); 3652 3653 void release_sock(struct sock *sk) 3654 { 3655 spin_lock_bh(&sk->sk_lock.slock); 3656 if (sk->sk_backlog.tail) 3657 __release_sock(sk); 3658 3659 if (sk->sk_prot->release_cb) 3660 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3661 tcp_release_cb, sk); 3662 3663 sock_release_ownership(sk); 3664 if (waitqueue_active(&sk->sk_lock.wq)) 3665 wake_up(&sk->sk_lock.wq); 3666 spin_unlock_bh(&sk->sk_lock.slock); 3667 } 3668 EXPORT_SYMBOL(release_sock); 3669 3670 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3671 { 3672 might_sleep(); 3673 spin_lock_bh(&sk->sk_lock.slock); 3674 3675 if (!sock_owned_by_user_nocheck(sk)) { 3676 /* 3677 * Fast path return with bottom halves disabled and 3678 * sock::sk_lock.slock held. 3679 * 3680 * The 'mutex' is not contended and holding 3681 * sock::sk_lock.slock prevents all other lockers to 3682 * proceed so the corresponding unlock_sock_fast() can 3683 * avoid the slow path of release_sock() completely and 3684 * just release slock. 3685 * 3686 * From a semantical POV this is equivalent to 'acquiring' 3687 * the 'mutex', hence the corresponding lockdep 3688 * mutex_release() has to happen in the fast path of 3689 * unlock_sock_fast(). 3690 */ 3691 return false; 3692 } 3693 3694 __lock_sock(sk); 3695 sk->sk_lock.owned = 1; 3696 __acquire(&sk->sk_lock.slock); 3697 spin_unlock_bh(&sk->sk_lock.slock); 3698 return true; 3699 } 3700 EXPORT_SYMBOL(__lock_sock_fast); 3701 3702 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3703 bool timeval, bool time32) 3704 { 3705 struct sock *sk = sock->sk; 3706 struct timespec64 ts; 3707 3708 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3709 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3710 if (ts.tv_sec == -1) 3711 return -ENOENT; 3712 if (ts.tv_sec == 0) { 3713 ktime_t kt = ktime_get_real(); 3714 sock_write_timestamp(sk, kt); 3715 ts = ktime_to_timespec64(kt); 3716 } 3717 3718 if (timeval) 3719 ts.tv_nsec /= 1000; 3720 3721 #ifdef CONFIG_COMPAT_32BIT_TIME 3722 if (time32) 3723 return put_old_timespec32(&ts, userstamp); 3724 #endif 3725 #ifdef CONFIG_SPARC64 3726 /* beware of padding in sparc64 timeval */ 3727 if (timeval && !in_compat_syscall()) { 3728 struct __kernel_old_timeval __user tv = { 3729 .tv_sec = ts.tv_sec, 3730 .tv_usec = ts.tv_nsec, 3731 }; 3732 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3733 return -EFAULT; 3734 return 0; 3735 } 3736 #endif 3737 return put_timespec64(&ts, userstamp); 3738 } 3739 EXPORT_SYMBOL(sock_gettstamp); 3740 3741 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3742 { 3743 if (!sock_flag(sk, flag)) { 3744 unsigned long previous_flags = sk->sk_flags; 3745 3746 sock_set_flag(sk, flag); 3747 /* 3748 * we just set one of the two flags which require net 3749 * time stamping, but time stamping might have been on 3750 * already because of the other one 3751 */ 3752 if (sock_needs_netstamp(sk) && 3753 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3754 net_enable_timestamp(); 3755 } 3756 } 3757 3758 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3759 int level, int type) 3760 { 3761 struct sock_exterr_skb *serr; 3762 struct sk_buff *skb; 3763 int copied, err; 3764 3765 err = -EAGAIN; 3766 skb = sock_dequeue_err_skb(sk); 3767 if (skb == NULL) 3768 goto out; 3769 3770 copied = skb->len; 3771 if (copied > len) { 3772 msg->msg_flags |= MSG_TRUNC; 3773 copied = len; 3774 } 3775 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3776 if (err) 3777 goto out_free_skb; 3778 3779 sock_recv_timestamp(msg, sk, skb); 3780 3781 serr = SKB_EXT_ERR(skb); 3782 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3783 3784 msg->msg_flags |= MSG_ERRQUEUE; 3785 err = copied; 3786 3787 out_free_skb: 3788 kfree_skb(skb); 3789 out: 3790 return err; 3791 } 3792 EXPORT_SYMBOL(sock_recv_errqueue); 3793 3794 /* 3795 * Get a socket option on an socket. 3796 * 3797 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3798 * asynchronous errors should be reported by getsockopt. We assume 3799 * this means if you specify SO_ERROR (otherwise what is the point of it). 3800 */ 3801 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3802 char __user *optval, int __user *optlen) 3803 { 3804 struct sock *sk = sock->sk; 3805 3806 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3807 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3808 } 3809 EXPORT_SYMBOL(sock_common_getsockopt); 3810 3811 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3812 int flags) 3813 { 3814 struct sock *sk = sock->sk; 3815 int addr_len = 0; 3816 int err; 3817 3818 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3819 if (err >= 0) 3820 msg->msg_namelen = addr_len; 3821 return err; 3822 } 3823 EXPORT_SYMBOL(sock_common_recvmsg); 3824 3825 /* 3826 * Set socket options on an inet socket. 3827 */ 3828 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3829 sockptr_t optval, unsigned int optlen) 3830 { 3831 struct sock *sk = sock->sk; 3832 3833 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3834 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3835 } 3836 EXPORT_SYMBOL(sock_common_setsockopt); 3837 3838 void sk_common_release(struct sock *sk) 3839 { 3840 if (sk->sk_prot->destroy) 3841 sk->sk_prot->destroy(sk); 3842 3843 /* 3844 * Observation: when sk_common_release is called, processes have 3845 * no access to socket. But net still has. 3846 * Step one, detach it from networking: 3847 * 3848 * A. Remove from hash tables. 3849 */ 3850 3851 sk->sk_prot->unhash(sk); 3852 3853 /* 3854 * In this point socket cannot receive new packets, but it is possible 3855 * that some packets are in flight because some CPU runs receiver and 3856 * did hash table lookup before we unhashed socket. They will achieve 3857 * receive queue and will be purged by socket destructor. 3858 * 3859 * Also we still have packets pending on receive queue and probably, 3860 * our own packets waiting in device queues. sock_destroy will drain 3861 * receive queue, but transmitted packets will delay socket destruction 3862 * until the last reference will be released. 3863 */ 3864 3865 sock_orphan(sk); 3866 3867 xfrm_sk_free_policy(sk); 3868 3869 sock_put(sk); 3870 } 3871 EXPORT_SYMBOL(sk_common_release); 3872 3873 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3874 { 3875 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3876 3877 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3878 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3879 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3880 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3881 mem[SK_MEMINFO_FWD_ALLOC] = sk_forward_alloc_get(sk); 3882 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3883 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3884 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3885 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3886 } 3887 3888 #ifdef CONFIG_PROC_FS 3889 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3890 3891 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3892 { 3893 int cpu, idx = prot->inuse_idx; 3894 int res = 0; 3895 3896 for_each_possible_cpu(cpu) 3897 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3898 3899 return res >= 0 ? res : 0; 3900 } 3901 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3902 3903 int sock_inuse_get(struct net *net) 3904 { 3905 int cpu, res = 0; 3906 3907 for_each_possible_cpu(cpu) 3908 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3909 3910 return res; 3911 } 3912 3913 EXPORT_SYMBOL_GPL(sock_inuse_get); 3914 3915 static int __net_init sock_inuse_init_net(struct net *net) 3916 { 3917 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3918 if (net->core.prot_inuse == NULL) 3919 return -ENOMEM; 3920 return 0; 3921 } 3922 3923 static void __net_exit sock_inuse_exit_net(struct net *net) 3924 { 3925 free_percpu(net->core.prot_inuse); 3926 } 3927 3928 static struct pernet_operations net_inuse_ops = { 3929 .init = sock_inuse_init_net, 3930 .exit = sock_inuse_exit_net, 3931 }; 3932 3933 static __init int net_inuse_init(void) 3934 { 3935 if (register_pernet_subsys(&net_inuse_ops)) 3936 panic("Cannot initialize net inuse counters"); 3937 3938 return 0; 3939 } 3940 3941 core_initcall(net_inuse_init); 3942 3943 static int assign_proto_idx(struct proto *prot) 3944 { 3945 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3946 3947 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3948 pr_err("PROTO_INUSE_NR exhausted\n"); 3949 return -ENOSPC; 3950 } 3951 3952 set_bit(prot->inuse_idx, proto_inuse_idx); 3953 return 0; 3954 } 3955 3956 static void release_proto_idx(struct proto *prot) 3957 { 3958 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3959 clear_bit(prot->inuse_idx, proto_inuse_idx); 3960 } 3961 #else 3962 static inline int assign_proto_idx(struct proto *prot) 3963 { 3964 return 0; 3965 } 3966 3967 static inline void release_proto_idx(struct proto *prot) 3968 { 3969 } 3970 3971 #endif 3972 3973 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3974 { 3975 if (!twsk_prot) 3976 return; 3977 kfree(twsk_prot->twsk_slab_name); 3978 twsk_prot->twsk_slab_name = NULL; 3979 kmem_cache_destroy(twsk_prot->twsk_slab); 3980 twsk_prot->twsk_slab = NULL; 3981 } 3982 3983 static int tw_prot_init(const struct proto *prot) 3984 { 3985 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3986 3987 if (!twsk_prot) 3988 return 0; 3989 3990 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3991 prot->name); 3992 if (!twsk_prot->twsk_slab_name) 3993 return -ENOMEM; 3994 3995 twsk_prot->twsk_slab = 3996 kmem_cache_create(twsk_prot->twsk_slab_name, 3997 twsk_prot->twsk_obj_size, 0, 3998 SLAB_ACCOUNT | prot->slab_flags, 3999 NULL); 4000 if (!twsk_prot->twsk_slab) { 4001 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 4002 prot->name); 4003 return -ENOMEM; 4004 } 4005 4006 return 0; 4007 } 4008 4009 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 4010 { 4011 if (!rsk_prot) 4012 return; 4013 kfree(rsk_prot->slab_name); 4014 rsk_prot->slab_name = NULL; 4015 kmem_cache_destroy(rsk_prot->slab); 4016 rsk_prot->slab = NULL; 4017 } 4018 4019 static int req_prot_init(const struct proto *prot) 4020 { 4021 struct request_sock_ops *rsk_prot = prot->rsk_prot; 4022 4023 if (!rsk_prot) 4024 return 0; 4025 4026 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4027 prot->name); 4028 if (!rsk_prot->slab_name) 4029 return -ENOMEM; 4030 4031 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4032 rsk_prot->obj_size, 0, 4033 SLAB_ACCOUNT | prot->slab_flags, 4034 NULL); 4035 4036 if (!rsk_prot->slab) { 4037 pr_crit("%s: Can't create request sock SLAB cache!\n", 4038 prot->name); 4039 return -ENOMEM; 4040 } 4041 return 0; 4042 } 4043 4044 int proto_register(struct proto *prot, int alloc_slab) 4045 { 4046 int ret = -ENOBUFS; 4047 4048 if (prot->memory_allocated && !prot->sysctl_mem) { 4049 pr_err("%s: missing sysctl_mem\n", prot->name); 4050 return -EINVAL; 4051 } 4052 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4053 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4054 return -EINVAL; 4055 } 4056 if (alloc_slab) { 4057 prot->slab = kmem_cache_create_usercopy(prot->name, 4058 prot->obj_size, 0, 4059 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4060 prot->slab_flags, 4061 prot->useroffset, prot->usersize, 4062 NULL); 4063 4064 if (prot->slab == NULL) { 4065 pr_crit("%s: Can't create sock SLAB cache!\n", 4066 prot->name); 4067 goto out; 4068 } 4069 4070 if (req_prot_init(prot)) 4071 goto out_free_request_sock_slab; 4072 4073 if (tw_prot_init(prot)) 4074 goto out_free_timewait_sock_slab; 4075 } 4076 4077 mutex_lock(&proto_list_mutex); 4078 ret = assign_proto_idx(prot); 4079 if (ret) { 4080 mutex_unlock(&proto_list_mutex); 4081 goto out_free_timewait_sock_slab; 4082 } 4083 list_add(&prot->node, &proto_list); 4084 mutex_unlock(&proto_list_mutex); 4085 return ret; 4086 4087 out_free_timewait_sock_slab: 4088 if (alloc_slab) 4089 tw_prot_cleanup(prot->twsk_prot); 4090 out_free_request_sock_slab: 4091 if (alloc_slab) { 4092 req_prot_cleanup(prot->rsk_prot); 4093 4094 kmem_cache_destroy(prot->slab); 4095 prot->slab = NULL; 4096 } 4097 out: 4098 return ret; 4099 } 4100 EXPORT_SYMBOL(proto_register); 4101 4102 void proto_unregister(struct proto *prot) 4103 { 4104 mutex_lock(&proto_list_mutex); 4105 release_proto_idx(prot); 4106 list_del(&prot->node); 4107 mutex_unlock(&proto_list_mutex); 4108 4109 kmem_cache_destroy(prot->slab); 4110 prot->slab = NULL; 4111 4112 req_prot_cleanup(prot->rsk_prot); 4113 tw_prot_cleanup(prot->twsk_prot); 4114 } 4115 EXPORT_SYMBOL(proto_unregister); 4116 4117 int sock_load_diag_module(int family, int protocol) 4118 { 4119 if (!protocol) { 4120 if (!sock_is_registered(family)) 4121 return -ENOENT; 4122 4123 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4124 NETLINK_SOCK_DIAG, family); 4125 } 4126 4127 #ifdef CONFIG_INET 4128 if (family == AF_INET && 4129 protocol != IPPROTO_RAW && 4130 protocol < MAX_INET_PROTOS && 4131 !rcu_access_pointer(inet_protos[protocol])) 4132 return -ENOENT; 4133 #endif 4134 4135 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4136 NETLINK_SOCK_DIAG, family, protocol); 4137 } 4138 EXPORT_SYMBOL(sock_load_diag_module); 4139 4140 #ifdef CONFIG_PROC_FS 4141 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4142 __acquires(proto_list_mutex) 4143 { 4144 mutex_lock(&proto_list_mutex); 4145 return seq_list_start_head(&proto_list, *pos); 4146 } 4147 4148 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4149 { 4150 return seq_list_next(v, &proto_list, pos); 4151 } 4152 4153 static void proto_seq_stop(struct seq_file *seq, void *v) 4154 __releases(proto_list_mutex) 4155 { 4156 mutex_unlock(&proto_list_mutex); 4157 } 4158 4159 static char proto_method_implemented(const void *method) 4160 { 4161 return method == NULL ? 'n' : 'y'; 4162 } 4163 static long sock_prot_memory_allocated(struct proto *proto) 4164 { 4165 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4166 } 4167 4168 static const char *sock_prot_memory_pressure(struct proto *proto) 4169 { 4170 return proto->memory_pressure != NULL ? 4171 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4172 } 4173 4174 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4175 { 4176 4177 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4178 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4179 proto->name, 4180 proto->obj_size, 4181 sock_prot_inuse_get(seq_file_net(seq), proto), 4182 sock_prot_memory_allocated(proto), 4183 sock_prot_memory_pressure(proto), 4184 proto->max_header, 4185 proto->slab == NULL ? "no" : "yes", 4186 module_name(proto->owner), 4187 proto_method_implemented(proto->close), 4188 proto_method_implemented(proto->connect), 4189 proto_method_implemented(proto->disconnect), 4190 proto_method_implemented(proto->accept), 4191 proto_method_implemented(proto->ioctl), 4192 proto_method_implemented(proto->init), 4193 proto_method_implemented(proto->destroy), 4194 proto_method_implemented(proto->shutdown), 4195 proto_method_implemented(proto->setsockopt), 4196 proto_method_implemented(proto->getsockopt), 4197 proto_method_implemented(proto->sendmsg), 4198 proto_method_implemented(proto->recvmsg), 4199 proto_method_implemented(proto->bind), 4200 proto_method_implemented(proto->backlog_rcv), 4201 proto_method_implemented(proto->hash), 4202 proto_method_implemented(proto->unhash), 4203 proto_method_implemented(proto->get_port), 4204 proto_method_implemented(proto->enter_memory_pressure)); 4205 } 4206 4207 static int proto_seq_show(struct seq_file *seq, void *v) 4208 { 4209 if (v == &proto_list) 4210 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4211 "protocol", 4212 "size", 4213 "sockets", 4214 "memory", 4215 "press", 4216 "maxhdr", 4217 "slab", 4218 "module", 4219 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4220 else 4221 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4222 return 0; 4223 } 4224 4225 static const struct seq_operations proto_seq_ops = { 4226 .start = proto_seq_start, 4227 .next = proto_seq_next, 4228 .stop = proto_seq_stop, 4229 .show = proto_seq_show, 4230 }; 4231 4232 static __net_init int proto_init_net(struct net *net) 4233 { 4234 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4235 sizeof(struct seq_net_private))) 4236 return -ENOMEM; 4237 4238 return 0; 4239 } 4240 4241 static __net_exit void proto_exit_net(struct net *net) 4242 { 4243 remove_proc_entry("protocols", net->proc_net); 4244 } 4245 4246 4247 static __net_initdata struct pernet_operations proto_net_ops = { 4248 .init = proto_init_net, 4249 .exit = proto_exit_net, 4250 }; 4251 4252 static int __init proto_init(void) 4253 { 4254 return register_pernet_subsys(&proto_net_ops); 4255 } 4256 4257 subsys_initcall(proto_init); 4258 4259 #endif /* PROC_FS */ 4260 4261 #ifdef CONFIG_NET_RX_BUSY_POLL 4262 bool sk_busy_loop_end(void *p, unsigned long start_time) 4263 { 4264 struct sock *sk = p; 4265 4266 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4267 return true; 4268 4269 if (sk_is_udp(sk) && 4270 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4271 return true; 4272 4273 return sk_busy_loop_timeout(sk, start_time); 4274 } 4275 EXPORT_SYMBOL(sk_busy_loop_end); 4276 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4277 4278 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4279 { 4280 if (!sk->sk_prot->bind_add) 4281 return -EOPNOTSUPP; 4282 return sk->sk_prot->bind_add(sk, addr, addr_len); 4283 } 4284 EXPORT_SYMBOL(sock_bind_add); 4285 4286 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4287 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4288 void __user *arg, void *karg, size_t size) 4289 { 4290 int ret; 4291 4292 if (copy_from_user(karg, arg, size)) 4293 return -EFAULT; 4294 4295 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4296 if (ret) 4297 return ret; 4298 4299 if (copy_to_user(arg, karg, size)) 4300 return -EFAULT; 4301 4302 return 0; 4303 } 4304 EXPORT_SYMBOL(sock_ioctl_inout); 4305 4306 /* This is the most common ioctl prep function, where the result (4 bytes) is 4307 * copied back to userspace if the ioctl() returns successfully. No input is 4308 * copied from userspace as input argument. 4309 */ 4310 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4311 { 4312 int ret, karg = 0; 4313 4314 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4315 if (ret) 4316 return ret; 4317 4318 return put_user(karg, (int __user *)arg); 4319 } 4320 4321 /* A wrapper around sock ioctls, which copies the data from userspace 4322 * (depending on the protocol/ioctl), and copies back the result to userspace. 4323 * The main motivation for this function is to pass kernel memory to the 4324 * protocol ioctl callbacks, instead of userspace memory. 4325 */ 4326 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4327 { 4328 int rc = 1; 4329 4330 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4331 rc = ipmr_sk_ioctl(sk, cmd, arg); 4332 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4333 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4334 else if (sk_is_phonet(sk)) 4335 rc = phonet_sk_ioctl(sk, cmd, arg); 4336 4337 /* If ioctl was processed, returns its value */ 4338 if (rc <= 0) 4339 return rc; 4340 4341 /* Otherwise call the default handler */ 4342 return sock_ioctl_out(sk, cmd, arg); 4343 } 4344 EXPORT_SYMBOL(sk_ioctl); 4345 4346 static int __init sock_struct_check(void) 4347 { 4348 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4349 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4350 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4351 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4352 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4353 4354 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4355 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4356 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4357 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4358 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4359 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4360 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4361 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4362 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4363 4364 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4365 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4366 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4367 4368 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4369 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4370 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4371 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4372 4373 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4374 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4375 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_sndbuf); 4376 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4377 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4378 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4379 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4380 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4381 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4382 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_dst_pending_confirm); 4383 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_status); 4384 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4385 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4386 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4387 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4388 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4389 4390 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4391 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4392 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4393 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4394 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4395 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4396 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4397 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4398 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4399 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4400 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4401 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4402 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4403 return 0; 4404 } 4405 4406 core_initcall(sock_struct_check); 4407