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