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 <uapi/linux/pidfd.h> 152 153 #include "dev.h" 154 155 static DEFINE_MUTEX(proto_list_mutex); 156 static LIST_HEAD(proto_list); 157 158 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc); 159 static void sock_def_write_space(struct sock *sk); 160 161 /** 162 * sk_ns_capable - General socket capability test 163 * @sk: Socket to use a capability on or through 164 * @user_ns: The user namespace of the capability to use 165 * @cap: The capability to use 166 * 167 * Test to see if the opener of the socket had when the socket was 168 * created and the current process has the capability @cap in the user 169 * namespace @user_ns. 170 */ 171 bool sk_ns_capable(const struct sock *sk, 172 struct user_namespace *user_ns, int cap) 173 { 174 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 175 ns_capable(user_ns, cap); 176 } 177 EXPORT_SYMBOL(sk_ns_capable); 178 179 /** 180 * sk_capable - Socket global capability test 181 * @sk: Socket to use a capability on or through 182 * @cap: The global capability to use 183 * 184 * Test to see if the opener of the socket had when the socket was 185 * created and the current process has the capability @cap in all user 186 * namespaces. 187 */ 188 bool sk_capable(const struct sock *sk, int cap) 189 { 190 return sk_ns_capable(sk, &init_user_ns, cap); 191 } 192 EXPORT_SYMBOL(sk_capable); 193 194 /** 195 * sk_net_capable - Network namespace socket capability test 196 * @sk: Socket to use a capability on or through 197 * @cap: The capability to use 198 * 199 * Test to see if the opener of the socket had when the socket was created 200 * and the current process has the capability @cap over the network namespace 201 * the socket is a member of. 202 */ 203 bool sk_net_capable(const struct sock *sk, int cap) 204 { 205 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 206 } 207 EXPORT_SYMBOL(sk_net_capable); 208 209 /* 210 * Each address family might have different locking rules, so we have 211 * one slock key per address family and separate keys for internal and 212 * userspace sockets. 213 */ 214 static struct lock_class_key af_family_keys[AF_MAX]; 215 static struct lock_class_key af_family_kern_keys[AF_MAX]; 216 static struct lock_class_key af_family_slock_keys[AF_MAX]; 217 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 218 219 /* 220 * Make lock validator output more readable. (we pre-construct these 221 * strings build-time, so that runtime initialization of socket 222 * locks is fast): 223 */ 224 225 #define _sock_locks(x) \ 226 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 227 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 228 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 229 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 230 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 231 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 232 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 233 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 234 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 235 x "27" , x "28" , x "AF_CAN" , \ 236 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 237 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 238 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 239 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 240 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 241 x "AF_MCTP" , \ 242 x "AF_MAX" 243 244 static const char *const af_family_key_strings[AF_MAX+1] = { 245 _sock_locks("sk_lock-") 246 }; 247 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 248 _sock_locks("slock-") 249 }; 250 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 251 _sock_locks("clock-") 252 }; 253 254 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 255 _sock_locks("k-sk_lock-") 256 }; 257 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 258 _sock_locks("k-slock-") 259 }; 260 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 261 _sock_locks("k-clock-") 262 }; 263 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 264 _sock_locks("rlock-") 265 }; 266 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 267 _sock_locks("wlock-") 268 }; 269 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 270 _sock_locks("elock-") 271 }; 272 273 /* 274 * sk_callback_lock and sk queues locking rules are per-address-family, 275 * so split the lock classes by using a per-AF key: 276 */ 277 static struct lock_class_key af_callback_keys[AF_MAX]; 278 static struct lock_class_key af_rlock_keys[AF_MAX]; 279 static struct lock_class_key af_wlock_keys[AF_MAX]; 280 static struct lock_class_key af_elock_keys[AF_MAX]; 281 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 282 283 /* Run time adjustable parameters. */ 284 __u32 sysctl_wmem_max __read_mostly = 4 << 20; 285 EXPORT_SYMBOL(sysctl_wmem_max); 286 __u32 sysctl_rmem_max __read_mostly = 4 << 20; 287 EXPORT_SYMBOL(sysctl_rmem_max); 288 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_DEFAULT; 289 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_DEFAULT; 290 291 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 292 EXPORT_SYMBOL_GPL(memalloc_socks_key); 293 294 /** 295 * sk_set_memalloc - sets %SOCK_MEMALLOC 296 * @sk: socket to set it on 297 * 298 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 299 * It's the responsibility of the admin to adjust min_free_kbytes 300 * to meet the requirements 301 */ 302 void sk_set_memalloc(struct sock *sk) 303 { 304 sock_set_flag(sk, SOCK_MEMALLOC); 305 sk->sk_allocation |= __GFP_MEMALLOC; 306 static_branch_inc(&memalloc_socks_key); 307 } 308 EXPORT_SYMBOL_GPL(sk_set_memalloc); 309 310 void sk_clear_memalloc(struct sock *sk) 311 { 312 sock_reset_flag(sk, SOCK_MEMALLOC); 313 sk->sk_allocation &= ~__GFP_MEMALLOC; 314 static_branch_dec(&memalloc_socks_key); 315 316 /* 317 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 318 * progress of swapping. SOCK_MEMALLOC may be cleared while 319 * it has rmem allocations due to the last swapfile being deactivated 320 * but there is a risk that the socket is unusable due to exceeding 321 * the rmem limits. Reclaim the reserves and obey rmem limits again. 322 */ 323 sk_mem_reclaim(sk); 324 } 325 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 326 327 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 328 { 329 int ret; 330 unsigned int noreclaim_flag; 331 332 /* these should have been dropped before queueing */ 333 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 334 335 noreclaim_flag = memalloc_noreclaim_save(); 336 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 337 tcp_v6_do_rcv, 338 tcp_v4_do_rcv, 339 sk, skb); 340 memalloc_noreclaim_restore(noreclaim_flag); 341 342 return ret; 343 } 344 EXPORT_SYMBOL(__sk_backlog_rcv); 345 346 void sk_error_report(struct sock *sk) 347 { 348 sk->sk_error_report(sk); 349 350 switch (sk->sk_family) { 351 case AF_INET: 352 fallthrough; 353 case AF_INET6: 354 trace_inet_sk_error_report(sk); 355 break; 356 default: 357 break; 358 } 359 } 360 EXPORT_SYMBOL(sk_error_report); 361 362 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 363 { 364 struct __kernel_sock_timeval tv; 365 366 if (timeo == MAX_SCHEDULE_TIMEOUT) { 367 tv.tv_sec = 0; 368 tv.tv_usec = 0; 369 } else { 370 tv.tv_sec = timeo / HZ; 371 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 372 } 373 374 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 375 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 376 *(struct old_timeval32 *)optval = tv32; 377 return sizeof(tv32); 378 } 379 380 if (old_timeval) { 381 struct __kernel_old_timeval old_tv; 382 old_tv.tv_sec = tv.tv_sec; 383 old_tv.tv_usec = tv.tv_usec; 384 *(struct __kernel_old_timeval *)optval = old_tv; 385 return sizeof(old_tv); 386 } 387 388 *(struct __kernel_sock_timeval *)optval = tv; 389 return sizeof(tv); 390 } 391 EXPORT_SYMBOL(sock_get_timeout); 392 393 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 394 sockptr_t optval, int optlen, bool old_timeval) 395 { 396 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 397 struct old_timeval32 tv32; 398 399 if (optlen < sizeof(tv32)) 400 return -EINVAL; 401 402 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 403 return -EFAULT; 404 tv->tv_sec = tv32.tv_sec; 405 tv->tv_usec = tv32.tv_usec; 406 } else if (old_timeval) { 407 struct __kernel_old_timeval old_tv; 408 409 if (optlen < sizeof(old_tv)) 410 return -EINVAL; 411 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 412 return -EFAULT; 413 tv->tv_sec = old_tv.tv_sec; 414 tv->tv_usec = old_tv.tv_usec; 415 } else { 416 if (optlen < sizeof(*tv)) 417 return -EINVAL; 418 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 419 return -EFAULT; 420 } 421 422 return 0; 423 } 424 EXPORT_SYMBOL(sock_copy_user_timeval); 425 426 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 427 bool old_timeval) 428 { 429 struct __kernel_sock_timeval tv; 430 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 431 long val; 432 433 if (err) 434 return err; 435 436 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 437 return -EDOM; 438 439 if (tv.tv_sec < 0) { 440 static int warned __read_mostly; 441 442 WRITE_ONCE(*timeo_p, 0); 443 if (warned < 10 && net_ratelimit()) { 444 warned++; 445 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 446 __func__, current->comm, task_pid_nr(current)); 447 } 448 return 0; 449 } 450 val = MAX_SCHEDULE_TIMEOUT; 451 if ((tv.tv_sec || tv.tv_usec) && 452 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 453 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 454 USEC_PER_SEC / HZ); 455 WRITE_ONCE(*timeo_p, val); 456 return 0; 457 } 458 459 static bool sk_set_prio_allowed(const struct sock *sk, int val) 460 { 461 return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) || 462 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 463 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)); 464 } 465 466 static bool sock_needs_netstamp(const struct sock *sk) 467 { 468 switch (sk->sk_family) { 469 case AF_UNSPEC: 470 case AF_UNIX: 471 return false; 472 default: 473 return true; 474 } 475 } 476 477 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 478 { 479 if (sk->sk_flags & flags) { 480 sk->sk_flags &= ~flags; 481 if (sock_needs_netstamp(sk) && 482 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 483 net_disable_timestamp(); 484 } 485 } 486 487 488 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 489 { 490 unsigned long flags; 491 struct sk_buff_head *list = &sk->sk_receive_queue; 492 493 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) { 494 sk_drops_inc(sk); 495 trace_sock_rcvqueue_full(sk, skb); 496 return -ENOMEM; 497 } 498 499 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 500 sk_drops_inc(sk); 501 return -ENOBUFS; 502 } 503 504 skb->dev = NULL; 505 skb_set_owner_r(skb, sk); 506 507 /* we escape from rcu protected region, make sure we dont leak 508 * a norefcounted dst 509 */ 510 skb_dst_force(skb); 511 512 spin_lock_irqsave(&list->lock, flags); 513 sock_skb_set_dropcount(sk, skb); 514 __skb_queue_tail(list, skb); 515 spin_unlock_irqrestore(&list->lock, flags); 516 517 if (!sock_flag(sk, SOCK_DEAD)) 518 sk->sk_data_ready(sk); 519 return 0; 520 } 521 EXPORT_SYMBOL(__sock_queue_rcv_skb); 522 523 enum skb_drop_reason 524 sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb) 525 { 526 enum skb_drop_reason drop_reason; 527 int err; 528 529 drop_reason = sk_filter_reason(sk, skb); 530 if (drop_reason) 531 return drop_reason; 532 533 err = __sock_queue_rcv_skb(sk, skb); 534 switch (err) { 535 case -ENOMEM: 536 return SKB_DROP_REASON_SOCKET_RCVBUFF; 537 case -ENOBUFS: 538 return SKB_DROP_REASON_PROTO_MEM; 539 } 540 return SKB_NOT_DROPPED_YET; 541 } 542 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 543 544 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 545 const int nested, unsigned int trim_cap, bool refcounted) 546 { 547 enum skb_drop_reason reason; 548 int rc = NET_RX_SUCCESS; 549 int err; 550 551 reason = sk_filter_trim_cap(sk, skb, trim_cap); 552 if (reason) 553 goto discard_and_relse; 554 555 skb->dev = NULL; 556 557 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) { 558 sk_drops_inc(sk); 559 reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 560 goto discard_and_relse; 561 } 562 if (nested) 563 bh_lock_sock_nested(sk); 564 else 565 bh_lock_sock(sk); 566 if (!sock_owned_by_user(sk)) { 567 /* 568 * trylock + unlock semantics: 569 */ 570 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 571 572 rc = sk_backlog_rcv(sk, skb); 573 574 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 575 } else if ((err = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))) { 576 bh_unlock_sock(sk); 577 if (err == -ENOMEM) 578 reason = SKB_DROP_REASON_PFMEMALLOC; 579 if (err == -ENOBUFS) 580 reason = SKB_DROP_REASON_SOCKET_BACKLOG; 581 sk_drops_inc(sk); 582 goto discard_and_relse; 583 } 584 585 bh_unlock_sock(sk); 586 out: 587 if (refcounted) 588 sock_put(sk); 589 return rc; 590 discard_and_relse: 591 sk_skb_reason_drop(sk, skb, reason); 592 goto out; 593 } 594 EXPORT_SYMBOL(__sk_receive_skb); 595 596 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 597 u32)); 598 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 599 u32)); 600 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 601 { 602 struct dst_entry *dst = __sk_dst_get(sk); 603 604 if (dst && READ_ONCE(dst->obsolete) && 605 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 606 dst, cookie) == NULL) { 607 sk_tx_queue_clear(sk); 608 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 609 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 610 dst_release(dst); 611 return NULL; 612 } 613 614 return dst; 615 } 616 EXPORT_SYMBOL(__sk_dst_check); 617 618 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 619 { 620 struct dst_entry *dst = sk_dst_get(sk); 621 622 if (dst && READ_ONCE(dst->obsolete) && 623 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 624 dst, cookie) == NULL) { 625 sk_dst_reset(sk); 626 dst_release(dst); 627 return NULL; 628 } 629 630 return dst; 631 } 632 EXPORT_SYMBOL(sk_dst_check); 633 634 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 635 { 636 int ret = -ENOPROTOOPT; 637 #ifdef CONFIG_NETDEVICES 638 struct net *net = sock_net(sk); 639 640 /* Sorry... */ 641 ret = -EPERM; 642 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 643 goto out; 644 645 ret = -EINVAL; 646 if (ifindex < 0) 647 goto out; 648 649 /* Paired with all READ_ONCE() done locklessly. */ 650 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 651 652 if (sk->sk_prot->rehash) 653 sk->sk_prot->rehash(sk); 654 sk_dst_reset(sk); 655 656 ret = 0; 657 658 out: 659 #endif 660 661 return ret; 662 } 663 664 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 665 { 666 int ret; 667 668 if (lock_sk) 669 lock_sock(sk); 670 ret = sock_bindtoindex_locked(sk, ifindex); 671 if (lock_sk) 672 release_sock(sk); 673 674 return ret; 675 } 676 EXPORT_SYMBOL(sock_bindtoindex); 677 678 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 679 { 680 int ret = -ENOPROTOOPT; 681 #ifdef CONFIG_NETDEVICES 682 struct net *net = sock_net(sk); 683 char devname[IFNAMSIZ]; 684 int index; 685 686 ret = -EINVAL; 687 if (optlen < 0) 688 goto out; 689 690 /* Bind this socket to a particular device like "eth0", 691 * as specified in the passed interface name. If the 692 * name is "" or the option length is zero the socket 693 * is not bound. 694 */ 695 if (optlen > IFNAMSIZ - 1) 696 optlen = IFNAMSIZ - 1; 697 memset(devname, 0, sizeof(devname)); 698 699 ret = -EFAULT; 700 if (copy_from_sockptr(devname, optval, optlen)) 701 goto out; 702 703 index = 0; 704 if (devname[0] != '\0') { 705 struct net_device *dev; 706 707 rcu_read_lock(); 708 dev = dev_get_by_name_rcu(net, devname); 709 if (dev) 710 index = dev->ifindex; 711 rcu_read_unlock(); 712 ret = -ENODEV; 713 if (!dev) 714 goto out; 715 } 716 717 sockopt_lock_sock(sk); 718 ret = sock_bindtoindex_locked(sk, index); 719 sockopt_release_sock(sk); 720 out: 721 #endif 722 723 return ret; 724 } 725 726 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 727 sockptr_t optlen, int len) 728 { 729 int ret = -ENOPROTOOPT; 730 #ifdef CONFIG_NETDEVICES 731 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 732 struct net *net = sock_net(sk); 733 char devname[IFNAMSIZ]; 734 735 if (bound_dev_if == 0) { 736 len = 0; 737 goto zero; 738 } 739 740 ret = -EINVAL; 741 if (len < IFNAMSIZ) 742 goto out; 743 744 ret = netdev_get_name(net, devname, bound_dev_if); 745 if (ret) 746 goto out; 747 748 len = strlen(devname) + 1; 749 750 ret = -EFAULT; 751 if (copy_to_sockptr(optval, devname, len)) 752 goto out; 753 754 zero: 755 ret = -EFAULT; 756 if (copy_to_sockptr(optlen, &len, sizeof(int))) 757 goto out; 758 759 ret = 0; 760 761 out: 762 #endif 763 764 return ret; 765 } 766 767 bool sk_mc_loop(const struct sock *sk) 768 { 769 if (dev_recursion_level()) 770 return false; 771 if (!sk) 772 return true; 773 /* IPV6_ADDRFORM can change sk->sk_family under us. */ 774 switch (READ_ONCE(sk->sk_family)) { 775 case AF_INET: 776 return inet_test_bit(MC_LOOP, sk); 777 #if IS_ENABLED(CONFIG_IPV6) 778 case AF_INET6: 779 return inet6_test_bit(MC6_LOOP, sk); 780 #endif 781 } 782 WARN_ON_ONCE(1); 783 return true; 784 } 785 EXPORT_SYMBOL(sk_mc_loop); 786 787 void sock_set_reuseaddr(struct sock *sk) 788 { 789 lock_sock(sk); 790 sk->sk_reuse = SK_CAN_REUSE; 791 release_sock(sk); 792 } 793 EXPORT_SYMBOL(sock_set_reuseaddr); 794 795 void sock_set_reuseport(struct sock *sk) 796 { 797 lock_sock(sk); 798 sk->sk_reuseport = true; 799 release_sock(sk); 800 } 801 EXPORT_SYMBOL(sock_set_reuseport); 802 803 void sock_no_linger(struct sock *sk) 804 { 805 lock_sock(sk); 806 WRITE_ONCE(sk->sk_lingertime, 0); 807 sock_set_flag(sk, SOCK_LINGER); 808 release_sock(sk); 809 } 810 EXPORT_SYMBOL(sock_no_linger); 811 812 void sock_set_priority(struct sock *sk, u32 priority) 813 { 814 WRITE_ONCE(sk->sk_priority, priority); 815 } 816 EXPORT_SYMBOL(sock_set_priority); 817 818 void sock_set_sndtimeo(struct sock *sk, s64 secs) 819 { 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 } 825 EXPORT_SYMBOL(sock_set_sndtimeo); 826 827 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 828 { 829 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val); 830 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns); 831 if (val) { 832 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 833 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 834 } 835 } 836 837 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 838 { 839 switch (optname) { 840 case SO_TIMESTAMP_OLD: 841 __sock_set_timestamps(sk, valbool, false, false); 842 break; 843 case SO_TIMESTAMP_NEW: 844 __sock_set_timestamps(sk, valbool, true, false); 845 break; 846 case SO_TIMESTAMPNS_OLD: 847 __sock_set_timestamps(sk, valbool, false, true); 848 break; 849 case SO_TIMESTAMPNS_NEW: 850 __sock_set_timestamps(sk, valbool, true, true); 851 break; 852 } 853 } 854 855 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 856 { 857 struct net *net = sock_net(sk); 858 struct net_device *dev = NULL; 859 bool match = false; 860 int *vclock_index; 861 int i, num; 862 863 if (sk->sk_bound_dev_if) 864 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 865 866 if (!dev) { 867 pr_err("%s: sock not bind to device\n", __func__); 868 return -EOPNOTSUPP; 869 } 870 871 num = ethtool_get_phc_vclocks(dev, &vclock_index); 872 dev_put(dev); 873 874 for (i = 0; i < num; i++) { 875 if (*(vclock_index + i) == phc_index) { 876 match = true; 877 break; 878 } 879 } 880 881 if (num > 0) 882 kfree(vclock_index); 883 884 if (!match) 885 return -EINVAL; 886 887 WRITE_ONCE(sk->sk_bind_phc, phc_index); 888 889 return 0; 890 } 891 892 int sock_set_timestamping(struct sock *sk, int optname, 893 struct so_timestamping timestamping) 894 { 895 int val = timestamping.flags; 896 int ret; 897 898 if (val & ~SOF_TIMESTAMPING_MASK) 899 return -EINVAL; 900 901 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 902 !(val & SOF_TIMESTAMPING_OPT_ID)) 903 return -EINVAL; 904 905 if (val & SOF_TIMESTAMPING_OPT_ID && 906 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 907 if (sk_is_tcp(sk)) { 908 if ((1 << sk->sk_state) & 909 (TCPF_CLOSE | TCPF_LISTEN)) 910 return -EINVAL; 911 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 912 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 913 else 914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 915 } else { 916 atomic_set(&sk->sk_tskey, 0); 917 } 918 } 919 920 if (val & SOF_TIMESTAMPING_OPT_STATS && 921 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 922 return -EINVAL; 923 924 if (val & SOF_TIMESTAMPING_BIND_PHC) { 925 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 926 if (ret) 927 return ret; 928 } 929 930 WRITE_ONCE(sk->sk_tsflags, val); 931 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 932 sock_valbool_flag(sk, SOCK_TIMESTAMPING_ANY, !!(val & TSFLAGS_ANY)); 933 934 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 935 sock_enable_timestamp(sk, 936 SOCK_TIMESTAMPING_RX_SOFTWARE); 937 else 938 sock_disable_timestamp(sk, 939 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 940 return 0; 941 } 942 943 #if defined(CONFIG_CGROUP_BPF) 944 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 945 { 946 struct bpf_sock_ops_kern sock_ops; 947 948 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 949 sock_ops.op = op; 950 sock_ops.is_fullsock = 1; 951 sock_ops.sk = sk; 952 bpf_skops_init_skb(&sock_ops, skb, 0); 953 __cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS); 954 } 955 #endif 956 957 void sock_set_keepalive(struct sock *sk) 958 { 959 lock_sock(sk); 960 if (sk->sk_prot->keepalive) 961 sk->sk_prot->keepalive(sk, true); 962 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 963 release_sock(sk); 964 } 965 EXPORT_SYMBOL(sock_set_keepalive); 966 967 static void __sock_set_rcvbuf(struct sock *sk, int val) 968 { 969 struct socket *sock = sk->sk_socket; 970 971 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 972 * as a negative value. 973 */ 974 val = min_t(int, val, INT_MAX / 2); 975 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 976 977 /* We double it on the way in to account for "struct sk_buff" etc. 978 * overhead. Applications assume that the SO_RCVBUF setting they make 979 * will allow that much actual data to be received on that socket. 980 * 981 * Applications are unaware that "struct sk_buff" and other overheads 982 * allocate from the receive buffer during socket buffer allocation. 983 * 984 * And after considering the possible alternatives, returning the value 985 * we actually used in getsockopt is the most desirable behavior. 986 */ 987 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 988 989 if (sock) { 990 const struct proto_ops *ops = READ_ONCE(sock->ops); 991 992 if (ops->set_rcvbuf) 993 ops->set_rcvbuf(sk, sk->sk_rcvbuf); 994 } 995 } 996 997 void sock_set_rcvbuf(struct sock *sk, int val) 998 { 999 lock_sock(sk); 1000 __sock_set_rcvbuf(sk, val); 1001 release_sock(sk); 1002 } 1003 EXPORT_SYMBOL(sock_set_rcvbuf); 1004 1005 static void __sock_set_mark(struct sock *sk, u32 val) 1006 { 1007 if (val != sk->sk_mark) { 1008 WRITE_ONCE(sk->sk_mark, val); 1009 sk_dst_reset(sk); 1010 } 1011 } 1012 1013 void sock_set_mark(struct sock *sk, u32 val) 1014 { 1015 lock_sock(sk); 1016 __sock_set_mark(sk, val); 1017 release_sock(sk); 1018 } 1019 EXPORT_SYMBOL(sock_set_mark); 1020 1021 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1022 { 1023 /* Round down bytes to multiple of pages */ 1024 bytes = round_down(bytes, PAGE_SIZE); 1025 1026 WARN_ON(bytes > sk->sk_reserved_mem); 1027 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1028 sk_mem_reclaim(sk); 1029 } 1030 1031 static int sock_reserve_memory(struct sock *sk, int bytes) 1032 { 1033 long allocated; 1034 bool charged; 1035 int pages; 1036 1037 if (!mem_cgroup_sk_enabled(sk) || !sk_has_account(sk)) 1038 return -EOPNOTSUPP; 1039 1040 if (!bytes) 1041 return 0; 1042 1043 pages = sk_mem_pages(bytes); 1044 1045 /* pre-charge to memcg */ 1046 charged = mem_cgroup_sk_charge(sk, pages, 1047 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1048 if (!charged) 1049 return -ENOMEM; 1050 1051 if (sk->sk_bypass_prot_mem) 1052 goto success; 1053 1054 /* pre-charge to forward_alloc */ 1055 sk_memory_allocated_add(sk, pages); 1056 allocated = sk_memory_allocated(sk); 1057 1058 /* If the system goes into memory pressure with this 1059 * precharge, give up and return error. 1060 */ 1061 if (allocated > sk_prot_mem_limits(sk, 1)) { 1062 sk_memory_allocated_sub(sk, pages); 1063 mem_cgroup_sk_uncharge(sk, pages); 1064 return -ENOMEM; 1065 } 1066 1067 success: 1068 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1069 1070 WRITE_ONCE(sk->sk_reserved_mem, 1071 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1072 1073 return 0; 1074 } 1075 1076 #ifdef CONFIG_PAGE_POOL 1077 1078 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED 1079 * in 1 syscall. The limit exists to limit the amount of memory the kernel 1080 * allocates to copy these tokens, and to prevent looping over the frags for 1081 * too long. 1082 */ 1083 #define MAX_DONTNEED_TOKENS 128 1084 #define MAX_DONTNEED_FRAGS 1024 1085 1086 static noinline_for_stack int 1087 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1088 { 1089 unsigned int num_tokens, i, j, k, netmem_num = 0; 1090 struct dmabuf_token *tokens; 1091 int ret = 0, num_frags = 0; 1092 netmem_ref netmems[16]; 1093 1094 if (!sk_is_tcp(sk)) 1095 return -EBADF; 1096 1097 if (optlen % sizeof(*tokens) || 1098 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1099 return -EINVAL; 1100 1101 num_tokens = optlen / sizeof(*tokens); 1102 tokens = kvmalloc_objs(*tokens, num_tokens); 1103 if (!tokens) 1104 return -ENOMEM; 1105 1106 if (copy_from_sockptr(tokens, optval, optlen)) { 1107 kvfree(tokens); 1108 return -EFAULT; 1109 } 1110 1111 xa_lock_bh(&sk->sk_user_frags); 1112 for (i = 0; i < num_tokens; i++) { 1113 for (j = 0; j < tokens[i].token_count; j++) { 1114 if (++num_frags > MAX_DONTNEED_FRAGS) 1115 goto frag_limit_reached; 1116 1117 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1118 &sk->sk_user_frags, tokens[i].token_start + j); 1119 1120 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem))) 1121 continue; 1122 1123 netmems[netmem_num++] = netmem; 1124 if (netmem_num == ARRAY_SIZE(netmems)) { 1125 xa_unlock_bh(&sk->sk_user_frags); 1126 for (k = 0; k < netmem_num; k++) 1127 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1128 netmem_num = 0; 1129 xa_lock_bh(&sk->sk_user_frags); 1130 } 1131 ret++; 1132 } 1133 } 1134 1135 frag_limit_reached: 1136 xa_unlock_bh(&sk->sk_user_frags); 1137 for (k = 0; k < netmem_num; k++) 1138 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1139 1140 kvfree(tokens); 1141 return ret; 1142 } 1143 #endif 1144 1145 void sockopt_lock_sock(struct sock *sk) 1146 { 1147 /* When current->bpf_ctx is set, the setsockopt is called from 1148 * a bpf prog. bpf has ensured the sk lock has been 1149 * acquired before calling setsockopt(). 1150 */ 1151 if (has_current_bpf_ctx()) 1152 return; 1153 1154 lock_sock(sk); 1155 } 1156 EXPORT_SYMBOL(sockopt_lock_sock); 1157 1158 void sockopt_release_sock(struct sock *sk) 1159 { 1160 if (has_current_bpf_ctx()) 1161 return; 1162 1163 release_sock(sk); 1164 } 1165 EXPORT_SYMBOL(sockopt_release_sock); 1166 1167 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1168 { 1169 return has_current_bpf_ctx() || ns_capable(ns, cap); 1170 } 1171 EXPORT_SYMBOL(sockopt_ns_capable); 1172 1173 bool sockopt_capable(int cap) 1174 { 1175 return has_current_bpf_ctx() || capable(cap); 1176 } 1177 EXPORT_SYMBOL(sockopt_capable); 1178 1179 static int sockopt_validate_clockid(__kernel_clockid_t value) 1180 { 1181 switch (value) { 1182 case CLOCK_REALTIME: 1183 case CLOCK_MONOTONIC: 1184 case CLOCK_TAI: 1185 return 0; 1186 } 1187 return -EINVAL; 1188 } 1189 1190 /* 1191 * This is meant for all protocols to use and covers goings on 1192 * at the socket level. Everything here is generic. 1193 */ 1194 1195 int sk_setsockopt(struct sock *sk, int level, int optname, 1196 sockptr_t optval, unsigned int optlen) 1197 { 1198 struct so_timestamping timestamping; 1199 struct socket *sock = sk->sk_socket; 1200 struct sock_txtime sk_txtime; 1201 int val; 1202 int valbool; 1203 struct linger ling; 1204 int ret = 0; 1205 1206 /* 1207 * Options without arguments 1208 */ 1209 1210 if (optname == SO_BINDTODEVICE) 1211 return sock_setbindtodevice(sk, optval, optlen); 1212 1213 if (optlen < sizeof(int)) 1214 return -EINVAL; 1215 1216 if (copy_from_sockptr(&val, optval, sizeof(val))) 1217 return -EFAULT; 1218 1219 valbool = val ? 1 : 0; 1220 1221 /* handle options which do not require locking the socket. */ 1222 switch (optname) { 1223 case SO_PRIORITY: 1224 if (sk_set_prio_allowed(sk, val)) { 1225 sock_set_priority(sk, val); 1226 return 0; 1227 } 1228 return -EPERM; 1229 case SO_TYPE: 1230 case SO_PROTOCOL: 1231 case SO_DOMAIN: 1232 case SO_ERROR: 1233 return -ENOPROTOOPT; 1234 #ifdef CONFIG_NET_RX_BUSY_POLL 1235 case SO_BUSY_POLL: 1236 if (val < 0) 1237 return -EINVAL; 1238 WRITE_ONCE(sk->sk_ll_usec, val); 1239 return 0; 1240 case SO_PREFER_BUSY_POLL: 1241 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1242 return -EPERM; 1243 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1244 return 0; 1245 case SO_BUSY_POLL_BUDGET: 1246 if (val > READ_ONCE(sk->sk_busy_poll_budget) && 1247 !sockopt_capable(CAP_NET_ADMIN)) 1248 return -EPERM; 1249 if (val < 0 || val > U16_MAX) 1250 return -EINVAL; 1251 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1252 return 0; 1253 #endif 1254 case SO_MAX_PACING_RATE: 1255 { 1256 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1257 unsigned long pacing_rate; 1258 1259 if (sizeof(ulval) != sizeof(val) && 1260 optlen >= sizeof(ulval) && 1261 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1262 return -EFAULT; 1263 } 1264 if (ulval != ~0UL) 1265 cmpxchg(&sk->sk_pacing_status, 1266 SK_PACING_NONE, 1267 SK_PACING_NEEDED); 1268 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1269 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1270 pacing_rate = READ_ONCE(sk->sk_pacing_rate); 1271 if (ulval < pacing_rate) 1272 WRITE_ONCE(sk->sk_pacing_rate, ulval); 1273 return 0; 1274 } 1275 case SO_TXREHASH: 1276 if (!sk_is_tcp(sk)) 1277 return -EOPNOTSUPP; 1278 if (val < -1 || val > 1) 1279 return -EINVAL; 1280 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1281 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1282 /* Paired with READ_ONCE() in tcp_rtx_synack() 1283 * and sk_getsockopt(). 1284 */ 1285 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1286 return 0; 1287 case SO_PEEK_OFF: 1288 { 1289 int (*set_peek_off)(struct sock *sk, int val); 1290 1291 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1292 if (set_peek_off) 1293 ret = set_peek_off(sk, val); 1294 else 1295 ret = -EOPNOTSUPP; 1296 return ret; 1297 } 1298 #ifdef CONFIG_PAGE_POOL 1299 case SO_DEVMEM_DONTNEED: 1300 return sock_devmem_dontneed(sk, optval, optlen); 1301 #endif 1302 case SO_SNDTIMEO_OLD: 1303 case SO_SNDTIMEO_NEW: 1304 return sock_set_timeout(&sk->sk_sndtimeo, optval, 1305 optlen, optname == SO_SNDTIMEO_OLD); 1306 case SO_RCVTIMEO_OLD: 1307 case SO_RCVTIMEO_NEW: 1308 return sock_set_timeout(&sk->sk_rcvtimeo, optval, 1309 optlen, optname == SO_RCVTIMEO_OLD); 1310 } 1311 1312 sockopt_lock_sock(sk); 1313 1314 switch (optname) { 1315 case SO_DEBUG: 1316 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1317 ret = -EACCES; 1318 else 1319 sock_valbool_flag(sk, SOCK_DBG, valbool); 1320 break; 1321 case SO_REUSEADDR: 1322 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1323 break; 1324 case SO_REUSEPORT: 1325 if (valbool && !sk_is_inet(sk)) 1326 ret = -EOPNOTSUPP; 1327 else 1328 sk->sk_reuseport = valbool; 1329 break; 1330 case SO_DONTROUTE: 1331 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1332 sk_dst_reset(sk); 1333 break; 1334 case SO_BROADCAST: 1335 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1336 break; 1337 case SO_SNDBUF: 1338 /* Don't error on this BSD doesn't and if you think 1339 * about it this is right. Otherwise apps have to 1340 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1341 * are treated in BSD as hints 1342 */ 1343 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1344 set_sndbuf: 1345 /* Ensure val * 2 fits into an int, to prevent max_t() 1346 * from treating it as a negative value. 1347 */ 1348 val = min_t(int, val, INT_MAX / 2); 1349 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1350 WRITE_ONCE(sk->sk_sndbuf, 1351 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1352 /* Wake up sending tasks if we upped the value. */ 1353 sk->sk_write_space(sk); 1354 break; 1355 1356 case SO_SNDBUFFORCE: 1357 if (!sockopt_capable(CAP_NET_ADMIN)) { 1358 ret = -EPERM; 1359 break; 1360 } 1361 1362 /* No negative values (to prevent underflow, as val will be 1363 * multiplied by 2). 1364 */ 1365 if (val < 0) 1366 val = 0; 1367 goto set_sndbuf; 1368 1369 case SO_RCVBUF: 1370 /* Don't error on this BSD doesn't and if you think 1371 * about it this is right. Otherwise apps have to 1372 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1373 * are treated in BSD as hints 1374 */ 1375 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1376 break; 1377 1378 case SO_RCVBUFFORCE: 1379 if (!sockopt_capable(CAP_NET_ADMIN)) { 1380 ret = -EPERM; 1381 break; 1382 } 1383 1384 /* No negative values (to prevent underflow, as val will be 1385 * multiplied by 2). 1386 */ 1387 __sock_set_rcvbuf(sk, max(val, 0)); 1388 break; 1389 1390 case SO_KEEPALIVE: 1391 if (sk->sk_prot->keepalive) 1392 sk->sk_prot->keepalive(sk, valbool); 1393 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1394 break; 1395 1396 case SO_OOBINLINE: 1397 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1398 break; 1399 1400 case SO_NO_CHECK: 1401 sk->sk_no_check_tx = valbool; 1402 break; 1403 1404 case SO_LINGER: 1405 if (optlen < sizeof(ling)) { 1406 ret = -EINVAL; /* 1003.1g */ 1407 break; 1408 } 1409 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1410 ret = -EFAULT; 1411 break; 1412 } 1413 if (!ling.l_onoff) { 1414 sock_reset_flag(sk, SOCK_LINGER); 1415 } else { 1416 unsigned long t_sec = ling.l_linger; 1417 1418 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1419 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1420 else 1421 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1422 sock_set_flag(sk, SOCK_LINGER); 1423 } 1424 break; 1425 1426 case SO_BSDCOMPAT: 1427 break; 1428 1429 case SO_TIMESTAMP_OLD: 1430 case SO_TIMESTAMP_NEW: 1431 case SO_TIMESTAMPNS_OLD: 1432 case SO_TIMESTAMPNS_NEW: 1433 sock_set_timestamp(sk, optname, valbool); 1434 break; 1435 1436 case SO_TIMESTAMPING_NEW: 1437 case SO_TIMESTAMPING_OLD: 1438 if (optlen == sizeof(timestamping)) { 1439 if (copy_from_sockptr(×tamping, optval, 1440 sizeof(timestamping))) { 1441 ret = -EFAULT; 1442 break; 1443 } 1444 } else { 1445 memset(×tamping, 0, sizeof(timestamping)); 1446 timestamping.flags = val; 1447 } 1448 ret = sock_set_timestamping(sk, optname, timestamping); 1449 break; 1450 1451 case SO_RCVLOWAT: 1452 { 1453 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1454 1455 if (val < 0) 1456 val = INT_MAX; 1457 if (sock) 1458 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1459 if (set_rcvlowat) 1460 ret = set_rcvlowat(sk, val); 1461 else 1462 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1463 break; 1464 } 1465 case SO_ATTACH_FILTER: { 1466 struct sock_fprog fprog; 1467 1468 if (sk_is_tcp(sk) && 1469 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1470 ret = -EPERM; 1471 break; 1472 } 1473 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1474 if (!ret) 1475 ret = sk_attach_filter(&fprog, sk); 1476 break; 1477 } 1478 case SO_ATTACH_BPF: 1479 ret = -EINVAL; 1480 if (optlen == sizeof(u32)) { 1481 u32 ufd; 1482 1483 ret = -EFAULT; 1484 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1485 break; 1486 1487 ret = sk_attach_bpf(ufd, sk); 1488 } 1489 break; 1490 1491 case SO_ATTACH_REUSEPORT_CBPF: { 1492 struct sock_fprog fprog; 1493 1494 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1495 if (!ret) 1496 ret = sk_reuseport_attach_filter(&fprog, sk); 1497 break; 1498 } 1499 case SO_ATTACH_REUSEPORT_EBPF: 1500 ret = -EINVAL; 1501 if (optlen == sizeof(u32)) { 1502 u32 ufd; 1503 1504 ret = -EFAULT; 1505 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1506 break; 1507 1508 ret = sk_reuseport_attach_bpf(ufd, sk); 1509 } 1510 break; 1511 1512 case SO_DETACH_REUSEPORT_BPF: 1513 ret = reuseport_detach_prog(sk); 1514 break; 1515 1516 case SO_DETACH_FILTER: 1517 ret = sk_detach_filter(sk); 1518 break; 1519 1520 case SO_LOCK_FILTER: 1521 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1522 ret = -EPERM; 1523 else 1524 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1525 break; 1526 1527 case SO_MARK: 1528 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1529 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1530 ret = -EPERM; 1531 break; 1532 } 1533 1534 __sock_set_mark(sk, val); 1535 break; 1536 case SO_RCVMARK: 1537 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1538 break; 1539 1540 case SO_RCVPRIORITY: 1541 sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool); 1542 break; 1543 1544 case SO_RXQ_OVFL: 1545 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1546 break; 1547 1548 case SO_WIFI_STATUS: 1549 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1550 break; 1551 1552 case SO_NOFCS: 1553 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1554 break; 1555 1556 case SO_SELECT_ERR_QUEUE: 1557 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1558 break; 1559 1560 case SO_PASSCRED: 1561 if (sk_may_scm_recv(sk)) 1562 sk->sk_scm_credentials = valbool; 1563 else 1564 ret = -EOPNOTSUPP; 1565 break; 1566 1567 case SO_PASSSEC: 1568 if (IS_ENABLED(CONFIG_SECURITY_NETWORK) && sk_may_scm_recv(sk)) 1569 sk->sk_scm_security = valbool; 1570 else 1571 ret = -EOPNOTSUPP; 1572 break; 1573 1574 case SO_PASSPIDFD: 1575 if (sk_is_unix(sk)) 1576 sk->sk_scm_pidfd = valbool; 1577 else 1578 ret = -EOPNOTSUPP; 1579 break; 1580 1581 case SO_PASSRIGHTS: 1582 if (sk_is_unix(sk)) 1583 sk->sk_scm_rights = valbool; 1584 else 1585 ret = -EOPNOTSUPP; 1586 break; 1587 1588 case SO_INCOMING_CPU: 1589 reuseport_update_incoming_cpu(sk, val); 1590 break; 1591 1592 case SO_CNX_ADVICE: 1593 if (val == 1) 1594 dst_negative_advice(sk); 1595 break; 1596 1597 case SO_ZEROCOPY: 1598 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1599 if (!(sk_is_tcp(sk) || 1600 (sk->sk_type == SOCK_DGRAM && 1601 sk->sk_protocol == IPPROTO_UDP))) 1602 ret = -EOPNOTSUPP; 1603 } else if (sk->sk_family != PF_RDS) { 1604 ret = -EOPNOTSUPP; 1605 } 1606 if (!ret) { 1607 if (val < 0 || val > 1) 1608 ret = -EINVAL; 1609 else 1610 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1611 } 1612 break; 1613 1614 case SO_TXTIME: 1615 if (optlen != sizeof(struct sock_txtime)) { 1616 ret = -EINVAL; 1617 break; 1618 } else if (copy_from_sockptr(&sk_txtime, optval, 1619 sizeof(struct sock_txtime))) { 1620 ret = -EFAULT; 1621 break; 1622 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1623 ret = -EINVAL; 1624 break; 1625 } 1626 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1627 * scheduler has enough safe guards. 1628 */ 1629 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1630 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1631 ret = -EPERM; 1632 break; 1633 } 1634 1635 ret = sockopt_validate_clockid(sk_txtime.clockid); 1636 if (ret) 1637 break; 1638 1639 sock_valbool_flag(sk, SOCK_TXTIME, true); 1640 sk->sk_clockid = sk_txtime.clockid; 1641 sk->sk_txtime_deadline_mode = 1642 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1643 sk->sk_txtime_report_errors = 1644 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1645 break; 1646 1647 case SO_BINDTOIFINDEX: 1648 ret = sock_bindtoindex_locked(sk, val); 1649 break; 1650 1651 case SO_BUF_LOCK: 1652 if (val & ~SOCK_BUF_LOCK_MASK) { 1653 ret = -EINVAL; 1654 break; 1655 } 1656 sk->sk_userlocks = val | (sk->sk_userlocks & 1657 ~SOCK_BUF_LOCK_MASK); 1658 break; 1659 1660 case SO_RESERVE_MEM: 1661 { 1662 int delta; 1663 1664 if (val < 0) { 1665 ret = -EINVAL; 1666 break; 1667 } 1668 1669 delta = val - sk->sk_reserved_mem; 1670 if (delta < 0) 1671 sock_release_reserved_memory(sk, -delta); 1672 else 1673 ret = sock_reserve_memory(sk, delta); 1674 break; 1675 } 1676 1677 default: 1678 ret = -ENOPROTOOPT; 1679 break; 1680 } 1681 sockopt_release_sock(sk); 1682 return ret; 1683 } 1684 1685 int sock_setsockopt(struct socket *sock, int level, int optname, 1686 sockptr_t optval, unsigned int optlen) 1687 { 1688 return sk_setsockopt(sock->sk, level, optname, 1689 optval, optlen); 1690 } 1691 EXPORT_SYMBOL(sock_setsockopt); 1692 1693 static const struct cred *sk_get_peer_cred(struct sock *sk) 1694 { 1695 const struct cred *cred; 1696 1697 spin_lock(&sk->sk_peer_lock); 1698 cred = get_cred(sk->sk_peer_cred); 1699 spin_unlock(&sk->sk_peer_lock); 1700 1701 return cred; 1702 } 1703 1704 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1705 struct ucred *ucred) 1706 { 1707 ucred->pid = pid_vnr(pid); 1708 ucred->uid = ucred->gid = -1; 1709 if (cred) { 1710 struct user_namespace *current_ns = current_user_ns(); 1711 1712 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1713 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1714 } 1715 } 1716 1717 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1718 { 1719 struct user_namespace *user_ns = current_user_ns(); 1720 int i; 1721 1722 for (i = 0; i < src->ngroups; i++) { 1723 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1724 1725 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1726 return -EFAULT; 1727 } 1728 1729 return 0; 1730 } 1731 1732 int sk_getsockopt(struct sock *sk, int level, int optname, 1733 sockptr_t optval, sockptr_t optlen) 1734 { 1735 struct socket *sock = sk->sk_socket; 1736 1737 union { 1738 int val; 1739 u64 val64; 1740 unsigned long ulval; 1741 struct linger ling; 1742 struct old_timeval32 tm32; 1743 struct __kernel_old_timeval tm; 1744 struct __kernel_sock_timeval stm; 1745 struct sock_txtime txtime; 1746 struct so_timestamping timestamping; 1747 } v; 1748 1749 int lv = sizeof(int); 1750 int len; 1751 1752 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1753 return -EFAULT; 1754 if (len < 0) 1755 return -EINVAL; 1756 1757 memset(&v, 0, sizeof(v)); 1758 1759 switch (optname) { 1760 case SO_DEBUG: 1761 v.val = sock_flag(sk, SOCK_DBG); 1762 break; 1763 1764 case SO_DONTROUTE: 1765 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1766 break; 1767 1768 case SO_BROADCAST: 1769 v.val = sock_flag(sk, SOCK_BROADCAST); 1770 break; 1771 1772 case SO_SNDBUF: 1773 v.val = READ_ONCE(sk->sk_sndbuf); 1774 break; 1775 1776 case SO_RCVBUF: 1777 v.val = READ_ONCE(sk->sk_rcvbuf); 1778 break; 1779 1780 case SO_REUSEADDR: 1781 v.val = sk->sk_reuse; 1782 break; 1783 1784 case SO_REUSEPORT: 1785 v.val = sk->sk_reuseport; 1786 break; 1787 1788 case SO_KEEPALIVE: 1789 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1790 break; 1791 1792 case SO_TYPE: 1793 v.val = sk->sk_type; 1794 break; 1795 1796 case SO_PROTOCOL: 1797 v.val = sk->sk_protocol; 1798 break; 1799 1800 case SO_DOMAIN: 1801 v.val = sk->sk_family; 1802 break; 1803 1804 case SO_ERROR: 1805 v.val = -sock_error(sk); 1806 if (v.val == 0) 1807 v.val = xchg(&sk->sk_err_soft, 0); 1808 break; 1809 1810 case SO_OOBINLINE: 1811 v.val = sock_flag(sk, SOCK_URGINLINE); 1812 break; 1813 1814 case SO_NO_CHECK: 1815 v.val = sk->sk_no_check_tx; 1816 break; 1817 1818 case SO_PRIORITY: 1819 v.val = READ_ONCE(sk->sk_priority); 1820 break; 1821 1822 case SO_LINGER: 1823 lv = sizeof(v.ling); 1824 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1825 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1826 break; 1827 1828 case SO_BSDCOMPAT: 1829 break; 1830 1831 case SO_TIMESTAMP_OLD: 1832 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1833 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1834 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1835 break; 1836 1837 case SO_TIMESTAMPNS_OLD: 1838 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1839 break; 1840 1841 case SO_TIMESTAMP_NEW: 1842 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1843 break; 1844 1845 case SO_TIMESTAMPNS_NEW: 1846 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1847 break; 1848 1849 case SO_TIMESTAMPING_OLD: 1850 case SO_TIMESTAMPING_NEW: 1851 lv = sizeof(v.timestamping); 1852 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1853 * returning the flags when they were set through the same option. 1854 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1855 */ 1856 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1857 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1858 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1859 } 1860 break; 1861 1862 case SO_RCVTIMEO_OLD: 1863 case SO_RCVTIMEO_NEW: 1864 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1865 SO_RCVTIMEO_OLD == optname); 1866 break; 1867 1868 case SO_SNDTIMEO_OLD: 1869 case SO_SNDTIMEO_NEW: 1870 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1871 SO_SNDTIMEO_OLD == optname); 1872 break; 1873 1874 case SO_RCVLOWAT: 1875 v.val = READ_ONCE(sk->sk_rcvlowat); 1876 break; 1877 1878 case SO_SNDLOWAT: 1879 v.val = 1; 1880 break; 1881 1882 case SO_PASSCRED: 1883 if (!sk_may_scm_recv(sk)) 1884 return -EOPNOTSUPP; 1885 1886 v.val = sk->sk_scm_credentials; 1887 break; 1888 1889 case SO_PASSPIDFD: 1890 if (!sk_is_unix(sk)) 1891 return -EOPNOTSUPP; 1892 1893 v.val = sk->sk_scm_pidfd; 1894 break; 1895 1896 case SO_PASSRIGHTS: 1897 if (!sk_is_unix(sk)) 1898 return -EOPNOTSUPP; 1899 1900 v.val = sk->sk_scm_rights; 1901 break; 1902 1903 case SO_PEERCRED: 1904 { 1905 struct ucred peercred; 1906 if (len > sizeof(peercred)) 1907 len = sizeof(peercred); 1908 1909 spin_lock(&sk->sk_peer_lock); 1910 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1911 spin_unlock(&sk->sk_peer_lock); 1912 1913 if (copy_to_sockptr(optval, &peercred, len)) 1914 return -EFAULT; 1915 goto lenout; 1916 } 1917 1918 case SO_PEERPIDFD: 1919 { 1920 struct pid *peer_pid; 1921 struct file *pidfd_file = NULL; 1922 unsigned int flags = 0; 1923 int pidfd; 1924 1925 if (len > sizeof(pidfd)) 1926 len = sizeof(pidfd); 1927 1928 spin_lock(&sk->sk_peer_lock); 1929 peer_pid = get_pid(sk->sk_peer_pid); 1930 spin_unlock(&sk->sk_peer_lock); 1931 1932 if (!peer_pid) 1933 return -ENODATA; 1934 1935 /* The use of PIDFD_STALE requires stashing of struct pid 1936 * on pidfs with pidfs_register_pid() and only AF_UNIX 1937 * were prepared for this. 1938 */ 1939 if (sk->sk_family == AF_UNIX) 1940 flags = PIDFD_STALE; 1941 1942 pidfd = pidfd_prepare(peer_pid, flags, &pidfd_file); 1943 put_pid(peer_pid); 1944 if (pidfd < 0) 1945 return pidfd; 1946 1947 if (copy_to_sockptr(optval, &pidfd, len) || 1948 copy_to_sockptr(optlen, &len, sizeof(int))) { 1949 put_unused_fd(pidfd); 1950 fput(pidfd_file); 1951 1952 return -EFAULT; 1953 } 1954 1955 fd_install(pidfd, pidfd_file); 1956 return 0; 1957 } 1958 1959 case SO_PEERGROUPS: 1960 { 1961 const struct cred *cred; 1962 int ret, n; 1963 1964 cred = sk_get_peer_cred(sk); 1965 if (!cred) 1966 return -ENODATA; 1967 1968 n = cred->group_info->ngroups; 1969 if (len < n * sizeof(gid_t)) { 1970 len = n * sizeof(gid_t); 1971 put_cred(cred); 1972 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 1973 } 1974 len = n * sizeof(gid_t); 1975 1976 ret = groups_to_user(optval, cred->group_info); 1977 put_cred(cred); 1978 if (ret) 1979 return ret; 1980 goto lenout; 1981 } 1982 1983 case SO_PEERNAME: 1984 { 1985 struct sockaddr_storage address; 1986 1987 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 1988 if (lv < 0) 1989 return -ENOTCONN; 1990 if (lv < len) 1991 return -EINVAL; 1992 if (copy_to_sockptr(optval, &address, len)) 1993 return -EFAULT; 1994 goto lenout; 1995 } 1996 1997 /* Dubious BSD thing... Probably nobody even uses it, but 1998 * the UNIX standard wants it for whatever reason... -DaveM 1999 */ 2000 case SO_ACCEPTCONN: 2001 v.val = sk->sk_state == TCP_LISTEN; 2002 break; 2003 2004 case SO_PASSSEC: 2005 if (!IS_ENABLED(CONFIG_SECURITY_NETWORK) || !sk_may_scm_recv(sk)) 2006 return -EOPNOTSUPP; 2007 2008 v.val = sk->sk_scm_security; 2009 break; 2010 2011 case SO_PEERSEC: 2012 return security_socket_getpeersec_stream(sock, 2013 optval, optlen, len); 2014 2015 case SO_MARK: 2016 v.val = READ_ONCE(sk->sk_mark); 2017 break; 2018 2019 case SO_RCVMARK: 2020 v.val = sock_flag(sk, SOCK_RCVMARK); 2021 break; 2022 2023 case SO_RCVPRIORITY: 2024 v.val = sock_flag(sk, SOCK_RCVPRIORITY); 2025 break; 2026 2027 case SO_RXQ_OVFL: 2028 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 2029 break; 2030 2031 case SO_WIFI_STATUS: 2032 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 2033 break; 2034 2035 case SO_PEEK_OFF: 2036 if (!READ_ONCE(sock->ops)->set_peek_off) 2037 return -EOPNOTSUPP; 2038 2039 v.val = READ_ONCE(sk->sk_peek_off); 2040 break; 2041 case SO_NOFCS: 2042 v.val = sock_flag(sk, SOCK_NOFCS); 2043 break; 2044 2045 case SO_BINDTODEVICE: 2046 return sock_getbindtodevice(sk, optval, optlen, len); 2047 2048 case SO_GET_FILTER: 2049 len = sk_get_filter(sk, optval, len); 2050 if (len < 0) 2051 return len; 2052 2053 goto lenout; 2054 2055 case SO_LOCK_FILTER: 2056 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 2057 break; 2058 2059 case SO_BPF_EXTENSIONS: 2060 v.val = bpf_tell_extensions(); 2061 break; 2062 2063 case SO_SELECT_ERR_QUEUE: 2064 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 2065 break; 2066 2067 #ifdef CONFIG_NET_RX_BUSY_POLL 2068 case SO_BUSY_POLL: 2069 v.val = READ_ONCE(sk->sk_ll_usec); 2070 break; 2071 case SO_PREFER_BUSY_POLL: 2072 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 2073 break; 2074 #endif 2075 2076 case SO_MAX_PACING_RATE: 2077 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 2078 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 2079 lv = sizeof(v.ulval); 2080 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 2081 } else { 2082 /* 32bit version */ 2083 v.val = min_t(unsigned long, ~0U, 2084 READ_ONCE(sk->sk_max_pacing_rate)); 2085 } 2086 break; 2087 2088 case SO_INCOMING_CPU: 2089 v.val = READ_ONCE(sk->sk_incoming_cpu); 2090 break; 2091 2092 case SO_MEMINFO: 2093 { 2094 u32 meminfo[SK_MEMINFO_VARS]; 2095 2096 sk_get_meminfo(sk, meminfo); 2097 2098 len = min_t(unsigned int, len, sizeof(meminfo)); 2099 if (copy_to_sockptr(optval, &meminfo, len)) 2100 return -EFAULT; 2101 2102 goto lenout; 2103 } 2104 2105 #ifdef CONFIG_NET_RX_BUSY_POLL 2106 case SO_INCOMING_NAPI_ID: 2107 v.val = READ_ONCE(sk->sk_napi_id); 2108 2109 /* aggregate non-NAPI IDs down to 0 */ 2110 if (!napi_id_valid(v.val)) 2111 v.val = 0; 2112 2113 break; 2114 #endif 2115 2116 case SO_COOKIE: 2117 lv = sizeof(u64); 2118 if (len < lv) 2119 return -EINVAL; 2120 v.val64 = sock_gen_cookie(sk); 2121 break; 2122 2123 case SO_ZEROCOPY: 2124 v.val = sock_flag(sk, SOCK_ZEROCOPY); 2125 break; 2126 2127 case SO_TXTIME: 2128 lv = sizeof(v.txtime); 2129 v.txtime.clockid = sk->sk_clockid; 2130 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 2131 SOF_TXTIME_DEADLINE_MODE : 0; 2132 v.txtime.flags |= sk->sk_txtime_report_errors ? 2133 SOF_TXTIME_REPORT_ERRORS : 0; 2134 break; 2135 2136 case SO_BINDTOIFINDEX: 2137 v.val = READ_ONCE(sk->sk_bound_dev_if); 2138 break; 2139 2140 case SO_NETNS_COOKIE: 2141 lv = sizeof(u64); 2142 if (len != lv) 2143 return -EINVAL; 2144 v.val64 = sock_net(sk)->net_cookie; 2145 break; 2146 2147 case SO_BUF_LOCK: 2148 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 2149 break; 2150 2151 case SO_RESERVE_MEM: 2152 v.val = READ_ONCE(sk->sk_reserved_mem); 2153 break; 2154 2155 case SO_TXREHASH: 2156 if (!sk_is_tcp(sk)) 2157 return -EOPNOTSUPP; 2158 2159 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2160 v.val = READ_ONCE(sk->sk_txrehash); 2161 break; 2162 2163 default: 2164 /* We implement the SO_SNDLOWAT etc to not be settable 2165 * (1003.1g 7). 2166 */ 2167 return -ENOPROTOOPT; 2168 } 2169 2170 if (len > lv) 2171 len = lv; 2172 if (copy_to_sockptr(optval, &v, len)) 2173 return -EFAULT; 2174 lenout: 2175 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2176 return -EFAULT; 2177 return 0; 2178 } 2179 2180 /* 2181 * Initialize an sk_lock. 2182 * 2183 * (We also register the sk_lock with the lock validator.) 2184 */ 2185 static inline void sock_lock_init(struct sock *sk) 2186 { 2187 sk_owner_clear(sk); 2188 2189 if (sk->sk_kern_sock) 2190 sock_lock_init_class_and_name( 2191 sk, 2192 af_family_kern_slock_key_strings[sk->sk_family], 2193 af_family_kern_slock_keys + sk->sk_family, 2194 af_family_kern_key_strings[sk->sk_family], 2195 af_family_kern_keys + sk->sk_family); 2196 else 2197 sock_lock_init_class_and_name( 2198 sk, 2199 af_family_slock_key_strings[sk->sk_family], 2200 af_family_slock_keys + sk->sk_family, 2201 af_family_key_strings[sk->sk_family], 2202 af_family_keys + sk->sk_family); 2203 } 2204 2205 /* 2206 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2207 * even temporarily, because of RCU lookups. sk_node should also be left as is. 2208 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2209 */ 2210 static void sock_copy(struct sock *nsk, const struct sock *osk) 2211 { 2212 const struct proto *prot = READ_ONCE(osk->sk_prot); 2213 #ifdef CONFIG_SECURITY_NETWORK 2214 void *sptr = nsk->sk_security; 2215 #endif 2216 2217 /* If we move sk_tx_queue_mapping out of the private section, 2218 * we must check if sk_tx_queue_clear() is called after 2219 * sock_copy() in sk_clone_lock(). 2220 */ 2221 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2222 offsetof(struct sock, sk_dontcopy_begin) || 2223 offsetof(struct sock, sk_tx_queue_mapping) >= 2224 offsetof(struct sock, sk_dontcopy_end)); 2225 2226 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2227 2228 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2229 prot->obj_size - offsetof(struct sock, sk_dontcopy_end), 2230 /* alloc is larger than struct, see sk_prot_alloc() */); 2231 2232 #ifdef CONFIG_SECURITY_NETWORK 2233 nsk->sk_security = sptr; 2234 security_sk_clone(osk, nsk); 2235 #endif 2236 } 2237 2238 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2239 int family) 2240 { 2241 struct sock *sk; 2242 struct kmem_cache *slab; 2243 2244 slab = prot->slab; 2245 if (slab != NULL) { 2246 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2247 if (!sk) 2248 return sk; 2249 if (want_init_on_alloc(priority)) 2250 sk_prot_clear_nulls(sk, prot->obj_size); 2251 } else 2252 sk = kmalloc(prot->obj_size, priority); 2253 2254 if (sk != NULL) { 2255 if (security_sk_alloc(sk, family, priority)) 2256 goto out_free; 2257 2258 if (!try_module_get(prot->owner)) 2259 goto out_free_sec; 2260 } 2261 2262 return sk; 2263 2264 out_free_sec: 2265 security_sk_free(sk); 2266 out_free: 2267 if (slab != NULL) 2268 kmem_cache_free(slab, sk); 2269 else 2270 kfree(sk); 2271 return NULL; 2272 } 2273 2274 static void sk_prot_free(struct proto *prot, struct sock *sk) 2275 { 2276 struct kmem_cache *slab; 2277 struct module *owner; 2278 2279 owner = prot->owner; 2280 slab = prot->slab; 2281 2282 cgroup_sk_free(&sk->sk_cgrp_data); 2283 mem_cgroup_sk_free(sk); 2284 security_sk_free(sk); 2285 2286 sk_owner_put(sk); 2287 2288 if (slab != NULL) 2289 kmem_cache_free(slab, sk); 2290 else 2291 kfree(sk); 2292 module_put(owner); 2293 } 2294 2295 /** 2296 * sk_alloc - All socket objects are allocated here 2297 * @net: the applicable net namespace 2298 * @family: protocol family 2299 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2300 * @prot: struct proto associated with this new sock instance 2301 * @kern: is this to be a kernel socket? 2302 */ 2303 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2304 struct proto *prot, int kern) 2305 { 2306 struct sock *sk; 2307 2308 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2309 if (sk) { 2310 sk->sk_family = family; 2311 /* 2312 * See comment in struct sock definition to understand 2313 * why we need sk_prot_creator -acme 2314 */ 2315 sk->sk_prot = sk->sk_prot_creator = prot; 2316 2317 if (READ_ONCE(net->core.sysctl_bypass_prot_mem)) 2318 sk->sk_bypass_prot_mem = 1; 2319 2320 sk->sk_kern_sock = kern; 2321 sock_lock_init(sk); 2322 2323 sk->sk_net_refcnt = kern ? 0 : 1; 2324 if (likely(sk->sk_net_refcnt)) { 2325 get_net_track(net, &sk->ns_tracker, priority); 2326 sock_inuse_add(net, 1); 2327 } else { 2328 net_passive_inc(net); 2329 __netns_tracker_alloc(net, &sk->ns_tracker, 2330 false, priority); 2331 } 2332 2333 sock_net_set(sk, net); 2334 refcount_set(&sk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS); 2335 2336 mem_cgroup_sk_alloc(sk); 2337 cgroup_sk_alloc(&sk->sk_cgrp_data); 2338 sock_update_classid(&sk->sk_cgrp_data); 2339 sock_update_netprioidx(&sk->sk_cgrp_data); 2340 sk_tx_queue_clear(sk); 2341 } 2342 2343 return sk; 2344 } 2345 EXPORT_SYMBOL(sk_alloc); 2346 2347 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2348 * grace period. This is the case for UDP sockets and TCP listeners. 2349 */ 2350 static void __sk_destruct(struct rcu_head *head) 2351 { 2352 struct sock *sk = container_of(head, struct sock, sk_rcu); 2353 struct net *net = sock_net(sk); 2354 struct sk_filter *filter; 2355 2356 if (sk->sk_destruct) 2357 sk->sk_destruct(sk); 2358 2359 filter = rcu_dereference_check(sk->sk_filter, 2360 refcount_read(&sk->sk_wmem_alloc) == 0); 2361 if (filter) { 2362 sk_filter_uncharge(sk, filter); 2363 RCU_INIT_POINTER(sk->sk_filter, NULL); 2364 } 2365 2366 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2367 2368 #ifdef CONFIG_BPF_SYSCALL 2369 bpf_sk_storage_free(sk); 2370 #endif 2371 2372 if (atomic_read(&sk->sk_omem_alloc)) 2373 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2374 __func__, atomic_read(&sk->sk_omem_alloc)); 2375 2376 if (sk->sk_frag.page) { 2377 put_page(sk->sk_frag.page); 2378 sk->sk_frag.page = NULL; 2379 } 2380 2381 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2382 put_cred(sk->sk_peer_cred); 2383 put_pid(sk->sk_peer_pid); 2384 2385 if (likely(sk->sk_net_refcnt)) { 2386 put_net_track(net, &sk->ns_tracker); 2387 } else { 2388 __netns_tracker_free(net, &sk->ns_tracker, false); 2389 net_passive_dec(net); 2390 } 2391 sk_prot_free(sk->sk_prot_creator, sk); 2392 } 2393 2394 void sk_net_refcnt_upgrade(struct sock *sk) 2395 { 2396 struct net *net = sock_net(sk); 2397 2398 WARN_ON_ONCE(sk->sk_net_refcnt); 2399 __netns_tracker_free(net, &sk->ns_tracker, false); 2400 net_passive_dec(net); 2401 sk->sk_net_refcnt = 1; 2402 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 2403 sock_inuse_add(net, 1); 2404 } 2405 EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade); 2406 2407 void sk_destruct(struct sock *sk) 2408 { 2409 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2410 2411 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2412 reuseport_detach_sock(sk); 2413 use_call_rcu = true; 2414 } 2415 2416 if (use_call_rcu) 2417 call_rcu(&sk->sk_rcu, __sk_destruct); 2418 else 2419 __sk_destruct(&sk->sk_rcu); 2420 } 2421 2422 static void __sk_free(struct sock *sk) 2423 { 2424 if (likely(sk->sk_net_refcnt)) 2425 sock_inuse_add(sock_net(sk), -1); 2426 2427 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2428 sock_diag_broadcast_destroy(sk); 2429 else 2430 sk_destruct(sk); 2431 } 2432 2433 void sk_free(struct sock *sk) 2434 { 2435 /* 2436 * We subtract one from sk_wmem_alloc and can know if 2437 * some packets are still in some tx queue. 2438 * If not null, sock_wfree() will call __sk_free(sk) later 2439 */ 2440 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2441 __sk_free(sk); 2442 } 2443 EXPORT_SYMBOL(sk_free); 2444 2445 static void sk_init_common(struct sock *sk) 2446 { 2447 skb_queue_head_init(&sk->sk_receive_queue); 2448 skb_queue_head_init(&sk->sk_write_queue); 2449 skb_queue_head_init(&sk->sk_error_queue); 2450 2451 rwlock_init(&sk->sk_callback_lock); 2452 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2453 af_rlock_keys + sk->sk_family, 2454 af_family_rlock_key_strings[sk->sk_family]); 2455 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2456 af_wlock_keys + sk->sk_family, 2457 af_family_wlock_key_strings[sk->sk_family]); 2458 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2459 af_elock_keys + sk->sk_family, 2460 af_family_elock_key_strings[sk->sk_family]); 2461 if (sk->sk_kern_sock) 2462 lockdep_set_class_and_name(&sk->sk_callback_lock, 2463 af_kern_callback_keys + sk->sk_family, 2464 af_family_kern_clock_key_strings[sk->sk_family]); 2465 else 2466 lockdep_set_class_and_name(&sk->sk_callback_lock, 2467 af_callback_keys + sk->sk_family, 2468 af_family_clock_key_strings[sk->sk_family]); 2469 } 2470 2471 /** 2472 * sk_clone - clone a socket 2473 * @sk: the socket to clone 2474 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2475 * @lock: if true, lock the cloned sk 2476 * 2477 * If @lock is true, the clone is locked by bh_lock_sock(), and 2478 * caller must unlock socket even in error path by bh_unlock_sock(). 2479 */ 2480 struct sock *sk_clone(const struct sock *sk, const gfp_t priority, 2481 bool lock) 2482 { 2483 struct proto *prot = READ_ONCE(sk->sk_prot); 2484 struct sk_filter *filter; 2485 bool is_charged = true; 2486 struct sock *newsk; 2487 2488 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2489 if (!newsk) 2490 goto out; 2491 2492 sock_copy(newsk, sk); 2493 2494 newsk->sk_prot_creator = prot; 2495 2496 /* SANITY */ 2497 if (likely(newsk->sk_net_refcnt)) { 2498 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2499 sock_inuse_add(sock_net(newsk), 1); 2500 } else { 2501 /* Kernel sockets are not elevating the struct net refcount. 2502 * Instead, use a tracker to more easily detect if a layer 2503 * is not properly dismantling its kernel sockets at netns 2504 * destroy time. 2505 */ 2506 net_passive_inc(sock_net(newsk)); 2507 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2508 false, priority); 2509 } 2510 2511 sk_node_init(&newsk->sk_node); 2512 sock_lock_init(newsk); 2513 2514 if (lock) 2515 bh_lock_sock(newsk); 2516 2517 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2518 newsk->sk_backlog.len = 0; 2519 2520 atomic_set(&newsk->sk_rmem_alloc, 0); 2521 2522 refcount_set(&newsk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS); 2523 2524 atomic_set(&newsk->sk_omem_alloc, 0); 2525 sk_init_common(newsk); 2526 2527 newsk->sk_dst_cache = NULL; 2528 newsk->sk_dst_pending_confirm = 0; 2529 newsk->sk_wmem_queued = 0; 2530 newsk->sk_forward_alloc = 0; 2531 newsk->sk_reserved_mem = 0; 2532 DEBUG_NET_WARN_ON_ONCE(newsk->sk_drop_counters); 2533 sk_drops_reset(newsk); 2534 newsk->sk_send_head = NULL; 2535 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2536 atomic_set(&newsk->sk_zckey, 0); 2537 2538 sock_reset_flag(newsk, SOCK_DONE); 2539 2540 #ifdef CONFIG_MEMCG 2541 /* sk->sk_memcg will be populated at accept() time */ 2542 newsk->sk_memcg = NULL; 2543 #endif 2544 2545 cgroup_sk_clone(&newsk->sk_cgrp_data); 2546 2547 rcu_read_lock(); 2548 filter = rcu_dereference(sk->sk_filter); 2549 if (filter != NULL) 2550 /* though it's an empty new sock, the charging may fail 2551 * if sysctl_optmem_max was changed between creation of 2552 * original socket and cloning 2553 */ 2554 is_charged = sk_filter_charge(newsk, filter); 2555 RCU_INIT_POINTER(newsk->sk_filter, filter); 2556 rcu_read_unlock(); 2557 2558 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2559 /* We need to make sure that we don't uncharge the new 2560 * socket if we couldn't charge it in the first place 2561 * as otherwise we uncharge the parent's filter. 2562 */ 2563 if (!is_charged) 2564 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2565 2566 goto free; 2567 } 2568 2569 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2570 2571 if (bpf_sk_storage_clone(sk, newsk)) 2572 goto free; 2573 2574 /* Clear sk_user_data if parent had the pointer tagged 2575 * as not suitable for copying when cloning. 2576 */ 2577 if (sk_user_data_is_nocopy(newsk)) 2578 newsk->sk_user_data = NULL; 2579 2580 newsk->sk_err = 0; 2581 newsk->sk_err_soft = 0; 2582 newsk->sk_priority = 0; 2583 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2584 2585 /* Before updating sk_refcnt, we must commit prior changes to memory 2586 * (Documentation/RCU/rculist_nulls.rst for details) 2587 */ 2588 smp_wmb(); 2589 refcount_set(&newsk->sk_refcnt, 2); 2590 2591 sk_set_socket(newsk, NULL); 2592 sk_tx_queue_clear(newsk); 2593 sk_rx_queue_clear(newsk); 2594 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2595 2596 if (newsk->sk_prot->sockets_allocated) 2597 sk_sockets_allocated_inc(newsk); 2598 2599 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2600 net_enable_timestamp(); 2601 out: 2602 return newsk; 2603 free: 2604 /* It is still raw copy of parent, so invalidate 2605 * destructor and make plain sk_free() 2606 */ 2607 newsk->sk_destruct = NULL; 2608 if (lock) 2609 bh_unlock_sock(newsk); 2610 sk_free(newsk); 2611 newsk = NULL; 2612 goto out; 2613 } 2614 EXPORT_SYMBOL_GPL(sk_clone); 2615 2616 static u32 sk_dst_gso_max_size(struct sock *sk, const struct net_device *dev) 2617 { 2618 bool is_ipv6 = false; 2619 u32 max_size; 2620 2621 #if IS_ENABLED(CONFIG_IPV6) 2622 is_ipv6 = (sk->sk_family == AF_INET6 && 2623 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2624 #endif 2625 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2626 max_size = is_ipv6 ? READ_ONCE(dev->gso_max_size) : 2627 READ_ONCE(dev->gso_ipv4_max_size); 2628 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2629 max_size = GSO_LEGACY_MAX_SIZE; 2630 2631 return max_size - (MAX_TCP_HEADER + 1); 2632 } 2633 2634 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2635 { 2636 const struct net_device *dev; 2637 u32 max_segs = 1; 2638 2639 rcu_read_lock(); 2640 dev = dst_dev_rcu(dst); 2641 sk->sk_route_caps = dev->features; 2642 if (sk_is_tcp(sk)) { 2643 struct inet_connection_sock *icsk = inet_csk(sk); 2644 2645 sk->sk_route_caps |= NETIF_F_GSO; 2646 icsk->icsk_ack.dst_quick_ack = dst_metric(dst, RTAX_QUICKACK); 2647 } 2648 if (sk->sk_route_caps & NETIF_F_GSO) 2649 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2650 if (unlikely(sk->sk_gso_disabled)) 2651 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2652 if (sk_can_gso(sk)) { 2653 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2654 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2655 } else { 2656 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2657 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dev); 2658 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2659 max_segs = max_t(u32, READ_ONCE(dev->gso_max_segs), 1); 2660 } 2661 } 2662 sk->sk_gso_max_segs = max_segs; 2663 sk_dst_set(sk, dst); 2664 rcu_read_unlock(); 2665 } 2666 EXPORT_SYMBOL_GPL(sk_setup_caps); 2667 2668 /* 2669 * Simple resource managers for sockets. 2670 */ 2671 2672 2673 /* 2674 * Write buffer destructor automatically called from kfree_skb. 2675 */ 2676 void sock_wfree(struct sk_buff *skb) 2677 { 2678 unsigned int len = skb->truesize; 2679 struct sock *sk = skb->sk; 2680 bool free; 2681 int old; 2682 2683 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2684 void (*sk_write_space)(struct sock *sk); 2685 2686 sk_write_space = READ_ONCE(sk->sk_write_space); 2687 2688 if (sock_flag(sk, SOCK_RCU_FREE) && 2689 sk_write_space == sock_def_write_space) { 2690 rcu_read_lock(); 2691 free = __refcount_sub_and_test(len, &sk->sk_wmem_alloc, 2692 &old); 2693 sock_def_write_space_wfree(sk, old - len); 2694 rcu_read_unlock(); 2695 if (unlikely(free)) 2696 __sk_free(sk); 2697 return; 2698 } 2699 2700 /* 2701 * Keep a reference on sk_wmem_alloc, this will be released 2702 * after sk_write_space() call 2703 */ 2704 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2705 sk_write_space(sk); 2706 len = 1; 2707 } 2708 /* 2709 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2710 * could not do because of in-flight packets 2711 */ 2712 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2713 __sk_free(sk); 2714 } 2715 EXPORT_SYMBOL(sock_wfree); 2716 2717 /* This variant of sock_wfree() is used by TCP, 2718 * since it sets SOCK_USE_WRITE_QUEUE. 2719 */ 2720 void __sock_wfree(struct sk_buff *skb) 2721 { 2722 struct sock *sk = skb->sk; 2723 2724 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2725 __sk_free(sk); 2726 } 2727 2728 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2729 { 2730 int old_wmem; 2731 2732 skb_orphan(skb); 2733 #ifdef CONFIG_INET 2734 if (unlikely(!sk_fullsock(sk))) 2735 return skb_set_owner_edemux(skb, sk); 2736 #endif 2737 skb->sk = sk; 2738 skb->destructor = sock_wfree; 2739 skb_set_hash_from_sk(skb, sk); 2740 /* 2741 * We used to take a refcount on sk, but following operation 2742 * is enough to guarantee sk_free() won't free this sock until 2743 * all in-flight packets are completed 2744 */ 2745 __refcount_add(skb->truesize, &sk->sk_wmem_alloc, &old_wmem); 2746 2747 /* (old_wmem == SK_WMEM_ALLOC_BIAS) if no other TX packet for this socket 2748 * is in a host queue (qdisc, NIC queue). 2749 * Set skb->ooo_okay so that netdev_pick_tx() can choose a TX queue 2750 * based on XPS for better performance. 2751 * Otherwise clear ooo_okay to not risk Out Of Order delivery. 2752 */ 2753 skb->ooo_okay = (old_wmem == SK_WMEM_ALLOC_BIAS); 2754 } 2755 EXPORT_SYMBOL(skb_set_owner_w); 2756 2757 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2758 { 2759 /* Drivers depend on in-order delivery for crypto offload, 2760 * partial orphan breaks out-of-order-OK logic. 2761 */ 2762 if (skb_is_decrypted(skb)) 2763 return false; 2764 2765 return (skb->destructor == sock_wfree || 2766 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2767 } 2768 2769 /* This helper is used by netem, as it can hold packets in its 2770 * delay queue. We want to allow the owner socket to send more 2771 * packets, as if they were already TX completed by a typical driver. 2772 * But we also want to keep skb->sk set because some packet schedulers 2773 * rely on it (sch_fq for example). 2774 */ 2775 void skb_orphan_partial(struct sk_buff *skb) 2776 { 2777 if (skb_is_tcp_pure_ack(skb)) 2778 return; 2779 2780 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2781 return; 2782 2783 skb_orphan(skb); 2784 } 2785 EXPORT_SYMBOL(skb_orphan_partial); 2786 2787 /* 2788 * Read buffer destructor automatically called from kfree_skb. 2789 */ 2790 void sock_rfree(struct sk_buff *skb) 2791 { 2792 struct sock *sk = skb->sk; 2793 unsigned int len = skb->truesize; 2794 2795 atomic_sub(len, &sk->sk_rmem_alloc); 2796 sk_mem_uncharge(sk, len); 2797 } 2798 EXPORT_SYMBOL(sock_rfree); 2799 2800 /* 2801 * Buffer destructor for skbs that are not used directly in read or write 2802 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2803 */ 2804 void sock_efree(struct sk_buff *skb) 2805 { 2806 sock_put(skb->sk); 2807 } 2808 EXPORT_SYMBOL(sock_efree); 2809 2810 /* Buffer destructor for prefetch/receive path where reference count may 2811 * not be held, e.g. for listen sockets. 2812 */ 2813 #ifdef CONFIG_INET 2814 void sock_pfree(struct sk_buff *skb) 2815 { 2816 struct sock *sk = skb->sk; 2817 2818 if (!sk_is_refcounted(sk)) 2819 return; 2820 2821 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2822 inet_reqsk(sk)->rsk_listener = NULL; 2823 reqsk_free(inet_reqsk(sk)); 2824 return; 2825 } 2826 2827 sock_gen_put(sk); 2828 } 2829 EXPORT_SYMBOL(sock_pfree); 2830 #endif /* CONFIG_INET */ 2831 2832 /* 2833 * Allocate a skb from the socket's send buffer. 2834 */ 2835 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2836 gfp_t priority) 2837 { 2838 if (force || 2839 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2840 struct sk_buff *skb = alloc_skb(size, priority); 2841 2842 if (skb) { 2843 skb_set_owner_w(skb, sk); 2844 return skb; 2845 } 2846 } 2847 return NULL; 2848 } 2849 EXPORT_SYMBOL(sock_wmalloc); 2850 2851 static void sock_ofree(struct sk_buff *skb) 2852 { 2853 struct sock *sk = skb->sk; 2854 2855 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2856 } 2857 2858 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2859 gfp_t priority) 2860 { 2861 struct sk_buff *skb; 2862 2863 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2864 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2865 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2866 return NULL; 2867 2868 skb = alloc_skb(size, priority); 2869 if (!skb) 2870 return NULL; 2871 2872 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2873 skb->sk = sk; 2874 skb->destructor = sock_ofree; 2875 return skb; 2876 } 2877 2878 /* 2879 * Allocate a memory block from the socket's option memory buffer. 2880 */ 2881 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2882 { 2883 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2884 2885 if ((unsigned int)size <= optmem_max && 2886 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2887 void *mem; 2888 /* First do the add, to avoid the race if kmalloc 2889 * might sleep. 2890 */ 2891 atomic_add(size, &sk->sk_omem_alloc); 2892 mem = kmalloc(size, priority); 2893 if (mem) 2894 return mem; 2895 atomic_sub(size, &sk->sk_omem_alloc); 2896 } 2897 return NULL; 2898 } 2899 EXPORT_SYMBOL(sock_kmalloc); 2900 2901 /* 2902 * Duplicate the input "src" memory block using the socket's 2903 * option memory buffer. 2904 */ 2905 void *sock_kmemdup(struct sock *sk, const void *src, 2906 int size, gfp_t priority) 2907 { 2908 void *mem; 2909 2910 mem = sock_kmalloc(sk, size, priority); 2911 if (mem) 2912 memcpy(mem, src, size); 2913 return mem; 2914 } 2915 EXPORT_SYMBOL(sock_kmemdup); 2916 2917 /* Free an option memory block. Note, we actually want the inline 2918 * here as this allows gcc to detect the nullify and fold away the 2919 * condition entirely. 2920 */ 2921 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2922 const bool nullify) 2923 { 2924 if (WARN_ON_ONCE(!mem)) 2925 return; 2926 if (nullify) 2927 kfree_sensitive(mem); 2928 else 2929 kfree(mem); 2930 atomic_sub(size, &sk->sk_omem_alloc); 2931 } 2932 2933 void sock_kfree_s(struct sock *sk, void *mem, int size) 2934 { 2935 __sock_kfree_s(sk, mem, size, false); 2936 } 2937 EXPORT_SYMBOL(sock_kfree_s); 2938 2939 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2940 { 2941 __sock_kfree_s(sk, mem, size, true); 2942 } 2943 EXPORT_SYMBOL(sock_kzfree_s); 2944 2945 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2946 I think, these locks should be removed for datagram sockets. 2947 */ 2948 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2949 { 2950 DEFINE_WAIT(wait); 2951 2952 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2953 for (;;) { 2954 if (!timeo) 2955 break; 2956 if (signal_pending(current)) 2957 break; 2958 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2959 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2960 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2961 break; 2962 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2963 break; 2964 if (READ_ONCE(sk->sk_err)) 2965 break; 2966 timeo = schedule_timeout(timeo); 2967 } 2968 finish_wait(sk_sleep(sk), &wait); 2969 return timeo; 2970 } 2971 2972 2973 /* 2974 * Generic send/receive buffer handlers 2975 */ 2976 2977 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2978 unsigned long data_len, int noblock, 2979 int *errcode, int max_page_order) 2980 { 2981 struct sk_buff *skb; 2982 long timeo; 2983 int err; 2984 2985 timeo = sock_sndtimeo(sk, noblock); 2986 for (;;) { 2987 err = sock_error(sk); 2988 if (err != 0) 2989 goto failure; 2990 2991 err = -EPIPE; 2992 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 2993 goto failure; 2994 2995 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2996 break; 2997 2998 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2999 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 3000 err = -EAGAIN; 3001 if (!timeo) 3002 goto failure; 3003 if (signal_pending(current)) 3004 goto interrupted; 3005 timeo = sock_wait_for_wmem(sk, timeo); 3006 } 3007 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 3008 errcode, sk->sk_allocation); 3009 if (skb) 3010 skb_set_owner_w(skb, sk); 3011 return skb; 3012 3013 interrupted: 3014 err = sock_intr_errno(timeo); 3015 failure: 3016 *errcode = err; 3017 return NULL; 3018 } 3019 EXPORT_SYMBOL(sock_alloc_send_pskb); 3020 3021 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 3022 struct sockcm_cookie *sockc) 3023 { 3024 u32 tsflags; 3025 3026 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 3027 3028 switch (cmsg->cmsg_type) { 3029 case SO_MARK: 3030 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 3031 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 3032 return -EPERM; 3033 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3034 return -EINVAL; 3035 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 3036 break; 3037 case SO_TIMESTAMPING_OLD: 3038 case SO_TIMESTAMPING_NEW: 3039 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3040 return -EINVAL; 3041 3042 tsflags = *(u32 *)CMSG_DATA(cmsg); 3043 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 3044 return -EINVAL; 3045 3046 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 3047 sockc->tsflags |= tsflags; 3048 break; 3049 case SCM_TXTIME: 3050 if (!sock_flag(sk, SOCK_TXTIME)) 3051 return -EINVAL; 3052 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 3053 return -EINVAL; 3054 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 3055 break; 3056 case SCM_TS_OPT_ID: 3057 if (sk_is_tcp(sk)) 3058 return -EINVAL; 3059 tsflags = READ_ONCE(sk->sk_tsflags); 3060 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 3061 return -EINVAL; 3062 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3063 return -EINVAL; 3064 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 3065 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 3066 break; 3067 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 3068 case SCM_RIGHTS: 3069 case SCM_CREDENTIALS: 3070 break; 3071 case SO_PRIORITY: 3072 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3073 return -EINVAL; 3074 if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg))) 3075 return -EPERM; 3076 sockc->priority = *(u32 *)CMSG_DATA(cmsg); 3077 break; 3078 case SCM_DEVMEM_DMABUF: 3079 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3080 return -EINVAL; 3081 sockc->dmabuf_id = *(u32 *)CMSG_DATA(cmsg); 3082 break; 3083 default: 3084 return -EINVAL; 3085 } 3086 return 0; 3087 } 3088 EXPORT_SYMBOL(__sock_cmsg_send); 3089 3090 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 3091 struct sockcm_cookie *sockc) 3092 { 3093 struct cmsghdr *cmsg; 3094 int ret; 3095 3096 for_each_cmsghdr(cmsg, msg) { 3097 if (!CMSG_OK(msg, cmsg)) 3098 return -EINVAL; 3099 if (cmsg->cmsg_level != SOL_SOCKET) 3100 continue; 3101 ret = __sock_cmsg_send(sk, cmsg, sockc); 3102 if (ret) 3103 return ret; 3104 } 3105 return 0; 3106 } 3107 EXPORT_SYMBOL(sock_cmsg_send); 3108 3109 static void sk_enter_memory_pressure(struct sock *sk) 3110 { 3111 if (!sk->sk_prot->enter_memory_pressure) 3112 return; 3113 3114 sk->sk_prot->enter_memory_pressure(sk); 3115 } 3116 3117 static void sk_leave_memory_pressure(struct sock *sk) 3118 { 3119 if (sk->sk_prot->leave_memory_pressure) { 3120 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 3121 tcp_leave_memory_pressure, sk); 3122 } else { 3123 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 3124 3125 if (memory_pressure && READ_ONCE(*memory_pressure)) 3126 WRITE_ONCE(*memory_pressure, 0); 3127 } 3128 } 3129 3130 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3131 3132 /** 3133 * skb_page_frag_refill - check that a page_frag contains enough room 3134 * @sz: minimum size of the fragment we want to get 3135 * @pfrag: pointer to page_frag 3136 * @gfp: priority for memory allocation 3137 * 3138 * Note: While this allocator tries to use high order pages, there is 3139 * no guarantee that allocations succeed. Therefore, @sz MUST be 3140 * less or equal than PAGE_SIZE. 3141 */ 3142 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 3143 { 3144 if (pfrag->page) { 3145 if (page_ref_count(pfrag->page) == 1) { 3146 pfrag->offset = 0; 3147 return true; 3148 } 3149 if (pfrag->offset + sz <= pfrag->size) 3150 return true; 3151 put_page(pfrag->page); 3152 } 3153 3154 pfrag->offset = 0; 3155 if (SKB_FRAG_PAGE_ORDER && 3156 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3157 /* Avoid direct reclaim but allow kswapd to wake */ 3158 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3159 __GFP_COMP | __GFP_NOWARN | 3160 __GFP_NORETRY, 3161 SKB_FRAG_PAGE_ORDER); 3162 if (likely(pfrag->page)) { 3163 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3164 return true; 3165 } 3166 } 3167 pfrag->page = alloc_page(gfp); 3168 if (likely(pfrag->page)) { 3169 pfrag->size = PAGE_SIZE; 3170 return true; 3171 } 3172 return false; 3173 } 3174 EXPORT_SYMBOL(skb_page_frag_refill); 3175 3176 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3177 { 3178 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3179 return true; 3180 3181 if (!sk->sk_bypass_prot_mem) 3182 sk_enter_memory_pressure(sk); 3183 3184 sk_stream_moderate_sndbuf(sk); 3185 3186 return false; 3187 } 3188 EXPORT_SYMBOL(sk_page_frag_refill); 3189 3190 static void __lock_sock(struct sock *sk) 3191 __releases(&sk->sk_lock.slock) 3192 __acquires(&sk->sk_lock.slock) 3193 { 3194 DEFINE_WAIT(wait); 3195 3196 for (;;) { 3197 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3198 TASK_UNINTERRUPTIBLE); 3199 spin_unlock_bh(&sk->sk_lock.slock); 3200 schedule(); 3201 spin_lock_bh(&sk->sk_lock.slock); 3202 if (!sock_owned_by_user(sk)) 3203 break; 3204 } 3205 finish_wait(&sk->sk_lock.wq, &wait); 3206 } 3207 3208 void __release_sock(struct sock *sk) 3209 __releases(&sk->sk_lock.slock) 3210 __acquires(&sk->sk_lock.slock) 3211 { 3212 struct sk_buff *skb, *next; 3213 int nb = 0; 3214 3215 while ((skb = sk->sk_backlog.head) != NULL) { 3216 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3217 3218 spin_unlock_bh(&sk->sk_lock.slock); 3219 3220 while (1) { 3221 next = skb->next; 3222 prefetch(next); 3223 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3224 skb_mark_not_on_list(skb); 3225 sk_backlog_rcv(sk, skb); 3226 3227 skb = next; 3228 if (!skb) 3229 break; 3230 3231 if (!(++nb & 15)) 3232 cond_resched(); 3233 } 3234 3235 spin_lock_bh(&sk->sk_lock.slock); 3236 } 3237 3238 /* 3239 * Doing the zeroing here guarantee we can not loop forever 3240 * while a wild producer attempts to flood us. 3241 */ 3242 sk->sk_backlog.len = 0; 3243 } 3244 3245 void __sk_flush_backlog(struct sock *sk) 3246 { 3247 spin_lock_bh(&sk->sk_lock.slock); 3248 __release_sock(sk); 3249 3250 if (sk->sk_prot->release_cb) 3251 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3252 tcp_release_cb, sk); 3253 3254 spin_unlock_bh(&sk->sk_lock.slock); 3255 } 3256 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3257 3258 /** 3259 * sk_wait_data - wait for data to arrive at sk_receive_queue 3260 * @sk: sock to wait on 3261 * @timeo: for how long 3262 * @skb: last skb seen on sk_receive_queue 3263 * 3264 * Now socket state including sk->sk_err is changed only under lock, 3265 * hence we may omit checks after joining wait queue. 3266 * We check receive queue before schedule() only as optimization; 3267 * it is very likely that release_sock() added new data. 3268 */ 3269 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3270 { 3271 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3272 int rc; 3273 3274 add_wait_queue(sk_sleep(sk), &wait); 3275 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3276 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3277 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3278 remove_wait_queue(sk_sleep(sk), &wait); 3279 return rc; 3280 } 3281 EXPORT_SYMBOL(sk_wait_data); 3282 3283 /** 3284 * __sk_mem_raise_allocated - increase memory_allocated 3285 * @sk: socket 3286 * @size: memory size to allocate 3287 * @amt: pages to allocate 3288 * @kind: allocation type 3289 * 3290 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3291 * 3292 * Unlike the globally shared limits among the sockets under same protocol, 3293 * consuming the budget of a memcg won't have direct effect on other ones. 3294 * So be optimistic about memcg's tolerance, and leave the callers to decide 3295 * whether or not to raise allocated through sk_under_memory_pressure() or 3296 * its variants. 3297 */ 3298 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3299 { 3300 bool memcg_enabled = false, charged = false; 3301 struct proto *prot = sk->sk_prot; 3302 long allocated = 0; 3303 3304 if (!sk->sk_bypass_prot_mem) { 3305 sk_memory_allocated_add(sk, amt); 3306 allocated = sk_memory_allocated(sk); 3307 } 3308 3309 if (mem_cgroup_sk_enabled(sk)) { 3310 memcg_enabled = true; 3311 charged = mem_cgroup_sk_charge(sk, amt, gfp_memcg_charge()); 3312 if (!charged) 3313 goto suppress_allocation; 3314 } 3315 3316 if (!allocated) 3317 return 1; 3318 3319 /* Under limit. */ 3320 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3321 sk_leave_memory_pressure(sk); 3322 return 1; 3323 } 3324 3325 /* Under pressure. */ 3326 if (allocated > sk_prot_mem_limits(sk, 1)) 3327 sk_enter_memory_pressure(sk); 3328 3329 /* Over hard limit. */ 3330 if (allocated > sk_prot_mem_limits(sk, 2)) 3331 goto suppress_allocation; 3332 3333 /* Guarantee minimum buffer size under pressure (either global 3334 * or memcg) to make sure features described in RFC 7323 (TCP 3335 * Extensions for High Performance) work properly. 3336 * 3337 * This rule does NOT stand when exceeds global or memcg's hard 3338 * limit, or else a DoS attack can be taken place by spawning 3339 * lots of sockets whose usage are under minimum buffer size. 3340 */ 3341 if (kind == SK_MEM_RECV) { 3342 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3343 return 1; 3344 3345 } else { /* SK_MEM_SEND */ 3346 int wmem0 = sk_get_wmem0(sk, prot); 3347 3348 if (sk->sk_type == SOCK_STREAM) { 3349 if (sk->sk_wmem_queued < wmem0) 3350 return 1; 3351 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3352 return 1; 3353 } 3354 } 3355 3356 if (sk_has_memory_pressure(sk)) { 3357 u64 alloc; 3358 3359 /* The following 'average' heuristic is within the 3360 * scope of global accounting, so it only makes 3361 * sense for global memory pressure. 3362 */ 3363 if (!sk_under_global_memory_pressure(sk)) 3364 return 1; 3365 3366 /* Try to be fair among all the sockets under global 3367 * pressure by allowing the ones that below average 3368 * usage to raise. 3369 */ 3370 alloc = sk_sockets_allocated_read_positive(sk); 3371 if (sk_prot_mem_limits(sk, 2) > alloc * 3372 sk_mem_pages(sk->sk_wmem_queued + 3373 atomic_read(&sk->sk_rmem_alloc) + 3374 sk->sk_forward_alloc)) 3375 return 1; 3376 } 3377 3378 suppress_allocation: 3379 3380 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3381 sk_stream_moderate_sndbuf(sk); 3382 3383 /* Fail only if socket is _under_ its sndbuf. 3384 * In this case we cannot block, so that we have to fail. 3385 */ 3386 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3387 /* Force charge with __GFP_NOFAIL */ 3388 if (memcg_enabled && !charged) 3389 mem_cgroup_sk_charge(sk, amt, 3390 gfp_memcg_charge() | __GFP_NOFAIL); 3391 return 1; 3392 } 3393 } 3394 3395 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3396 3397 if (allocated) 3398 sk_memory_allocated_sub(sk, amt); 3399 3400 if (charged) 3401 mem_cgroup_sk_uncharge(sk, amt); 3402 3403 return 0; 3404 } 3405 3406 /** 3407 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3408 * @sk: socket 3409 * @size: memory size to allocate 3410 * @kind: allocation type 3411 * 3412 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3413 * rmem allocation. This function assumes that protocols which have 3414 * memory_pressure use sk_wmem_queued as write buffer accounting. 3415 */ 3416 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3417 { 3418 int ret, amt = sk_mem_pages(size); 3419 3420 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3421 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3422 if (!ret) 3423 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3424 return ret; 3425 } 3426 EXPORT_SYMBOL(__sk_mem_schedule); 3427 3428 /** 3429 * __sk_mem_reduce_allocated - reclaim memory_allocated 3430 * @sk: socket 3431 * @amount: number of quanta 3432 * 3433 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3434 */ 3435 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3436 { 3437 if (mem_cgroup_sk_enabled(sk)) 3438 mem_cgroup_sk_uncharge(sk, amount); 3439 3440 if (sk->sk_bypass_prot_mem) 3441 return; 3442 3443 sk_memory_allocated_sub(sk, amount); 3444 3445 if (sk_under_global_memory_pressure(sk) && 3446 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3447 sk_leave_memory_pressure(sk); 3448 } 3449 3450 /** 3451 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3452 * @sk: socket 3453 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3454 */ 3455 void __sk_mem_reclaim(struct sock *sk, int amount) 3456 { 3457 amount >>= PAGE_SHIFT; 3458 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3459 __sk_mem_reduce_allocated(sk, amount); 3460 } 3461 EXPORT_SYMBOL(__sk_mem_reclaim); 3462 3463 void __sk_charge(struct sock *sk, gfp_t gfp) 3464 { 3465 int amt; 3466 3467 gfp |= __GFP_NOFAIL; 3468 if (mem_cgroup_from_sk(sk)) { 3469 /* The socket has not been accepted yet, no need 3470 * to look at newsk->sk_wmem_queued. 3471 */ 3472 amt = sk_mem_pages(sk->sk_forward_alloc + 3473 atomic_read(&sk->sk_rmem_alloc)); 3474 if (amt) 3475 mem_cgroup_sk_charge(sk, amt, gfp); 3476 } 3477 3478 kmem_cache_charge(sk, gfp); 3479 } 3480 3481 int sk_set_peek_off(struct sock *sk, int val) 3482 { 3483 WRITE_ONCE(sk->sk_peek_off, val); 3484 return 0; 3485 } 3486 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3487 3488 /* 3489 * Set of default routines for initialising struct proto_ops when 3490 * the protocol does not support a particular function. In certain 3491 * cases where it makes no sense for a protocol to have a "do nothing" 3492 * function, some default processing is provided. 3493 */ 3494 3495 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len) 3496 { 3497 return -EOPNOTSUPP; 3498 } 3499 EXPORT_SYMBOL(sock_no_bind); 3500 3501 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr, 3502 int len, int flags) 3503 { 3504 return -EOPNOTSUPP; 3505 } 3506 EXPORT_SYMBOL(sock_no_connect); 3507 3508 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3509 { 3510 return -EOPNOTSUPP; 3511 } 3512 EXPORT_SYMBOL(sock_no_socketpair); 3513 3514 int sock_no_accept(struct socket *sock, struct socket *newsock, 3515 struct proto_accept_arg *arg) 3516 { 3517 return -EOPNOTSUPP; 3518 } 3519 EXPORT_SYMBOL(sock_no_accept); 3520 3521 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3522 int peer) 3523 { 3524 return -EOPNOTSUPP; 3525 } 3526 EXPORT_SYMBOL(sock_no_getname); 3527 3528 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3529 { 3530 return -EOPNOTSUPP; 3531 } 3532 EXPORT_SYMBOL(sock_no_ioctl); 3533 3534 int sock_no_listen(struct socket *sock, int backlog) 3535 { 3536 return -EOPNOTSUPP; 3537 } 3538 EXPORT_SYMBOL(sock_no_listen); 3539 3540 int sock_no_shutdown(struct socket *sock, int how) 3541 { 3542 return -EOPNOTSUPP; 3543 } 3544 EXPORT_SYMBOL(sock_no_shutdown); 3545 3546 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3547 { 3548 return -EOPNOTSUPP; 3549 } 3550 EXPORT_SYMBOL(sock_no_sendmsg); 3551 3552 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3553 { 3554 return -EOPNOTSUPP; 3555 } 3556 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3557 3558 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3559 int flags) 3560 { 3561 return -EOPNOTSUPP; 3562 } 3563 EXPORT_SYMBOL(sock_no_recvmsg); 3564 3565 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3566 { 3567 /* Mirror missing mmap method error code */ 3568 return -ENODEV; 3569 } 3570 EXPORT_SYMBOL(sock_no_mmap); 3571 3572 /* 3573 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3574 * various sock-based usage counts. 3575 */ 3576 void __receive_sock(struct file *file) 3577 { 3578 struct socket *sock; 3579 3580 sock = sock_from_file(file); 3581 if (sock) { 3582 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3583 sock_update_classid(&sock->sk->sk_cgrp_data); 3584 } 3585 } 3586 3587 /* 3588 * Default Socket Callbacks 3589 */ 3590 3591 static void sock_def_wakeup(struct sock *sk) 3592 { 3593 struct socket_wq *wq; 3594 3595 rcu_read_lock(); 3596 wq = rcu_dereference(sk->sk_wq); 3597 if (skwq_has_sleeper(wq)) 3598 wake_up_interruptible_all(&wq->wait); 3599 rcu_read_unlock(); 3600 } 3601 3602 static void sock_def_error_report(struct sock *sk) 3603 { 3604 struct socket_wq *wq; 3605 3606 rcu_read_lock(); 3607 wq = rcu_dereference(sk->sk_wq); 3608 if (skwq_has_sleeper(wq)) 3609 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3610 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3611 rcu_read_unlock(); 3612 } 3613 3614 void sock_def_readable(struct sock *sk) 3615 { 3616 struct socket_wq *wq; 3617 3618 trace_sk_data_ready(sk); 3619 3620 rcu_read_lock(); 3621 wq = rcu_dereference(sk->sk_wq); 3622 if (skwq_has_sleeper(wq)) 3623 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3624 EPOLLRDNORM | EPOLLRDBAND); 3625 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3626 rcu_read_unlock(); 3627 } 3628 3629 static void sock_def_write_space(struct sock *sk) 3630 { 3631 struct socket_wq *wq; 3632 3633 rcu_read_lock(); 3634 3635 /* Do not wake up a writer until he can make "significant" 3636 * progress. --DaveM 3637 */ 3638 if (sock_writeable(sk)) { 3639 wq = rcu_dereference(sk->sk_wq); 3640 if (skwq_has_sleeper(wq)) 3641 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3642 EPOLLWRNORM | EPOLLWRBAND); 3643 3644 /* Should agree with poll, otherwise some programs break */ 3645 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3646 } 3647 3648 rcu_read_unlock(); 3649 } 3650 3651 /* An optimised version of sock_def_write_space(), should only be called 3652 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3653 * ->sk_wmem_alloc. 3654 */ 3655 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc) 3656 { 3657 /* Do not wake up a writer until he can make "significant" 3658 * progress. --DaveM 3659 */ 3660 if (__sock_writeable(sk, wmem_alloc)) { 3661 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3662 3663 /* rely on refcount_sub from sock_wfree() */ 3664 smp_mb__after_atomic(); 3665 if (wq && waitqueue_active(&wq->wait)) 3666 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3667 EPOLLWRNORM | EPOLLWRBAND); 3668 3669 /* Should agree with poll, otherwise some programs break */ 3670 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3671 } 3672 } 3673 3674 static void sock_def_destruct(struct sock *sk) 3675 { 3676 } 3677 3678 void sk_send_sigurg(struct sock *sk) 3679 { 3680 if (sk->sk_socket && sk->sk_socket->file) 3681 if (send_sigurg(sk->sk_socket->file)) 3682 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3683 } 3684 EXPORT_SYMBOL(sk_send_sigurg); 3685 3686 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3687 unsigned long expires) 3688 { 3689 if (!mod_timer(timer, expires)) 3690 sock_hold(sk); 3691 } 3692 EXPORT_SYMBOL(sk_reset_timer); 3693 3694 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3695 { 3696 if (timer_delete(timer)) 3697 __sock_put(sk); 3698 } 3699 EXPORT_SYMBOL(sk_stop_timer); 3700 3701 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3702 { 3703 if (timer_delete_sync(timer)) 3704 __sock_put(sk); 3705 } 3706 EXPORT_SYMBOL(sk_stop_timer_sync); 3707 3708 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3709 { 3710 sk_init_common(sk); 3711 sk->sk_send_head = NULL; 3712 3713 timer_setup(&sk->sk_timer, NULL, 0); 3714 3715 sk->sk_allocation = GFP_KERNEL; 3716 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3717 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3718 sk->sk_state = TCP_CLOSE; 3719 sk->sk_use_task_frag = true; 3720 sk_set_socket(sk, sock); 3721 3722 sock_set_flag(sk, SOCK_ZAPPED); 3723 3724 if (sock) { 3725 sk->sk_type = sock->type; 3726 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3727 sock->sk = sk; 3728 } else { 3729 RCU_INIT_POINTER(sk->sk_wq, NULL); 3730 } 3731 sk->sk_uid = uid; 3732 3733 sk->sk_state_change = sock_def_wakeup; 3734 sk->sk_data_ready = sock_def_readable; 3735 sk->sk_write_space = sock_def_write_space; 3736 sk->sk_error_report = sock_def_error_report; 3737 sk->sk_destruct = sock_def_destruct; 3738 3739 sk->sk_frag.page = NULL; 3740 sk->sk_frag.offset = 0; 3741 sk->sk_peek_off = -1; 3742 3743 sk->sk_peer_pid = NULL; 3744 sk->sk_peer_cred = NULL; 3745 spin_lock_init(&sk->sk_peer_lock); 3746 3747 sk->sk_write_pending = 0; 3748 sk->sk_rcvlowat = 1; 3749 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3750 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3751 3752 sk->sk_stamp = SK_DEFAULT_STAMP; 3753 #if BITS_PER_LONG==32 3754 seqlock_init(&sk->sk_stamp_seq); 3755 #endif 3756 atomic_set(&sk->sk_zckey, 0); 3757 3758 #ifdef CONFIG_NET_RX_BUSY_POLL 3759 sk->sk_napi_id = 0; 3760 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3761 #endif 3762 3763 sk->sk_max_pacing_rate = ~0UL; 3764 sk->sk_pacing_rate = ~0UL; 3765 WRITE_ONCE(sk->sk_pacing_shift, 10); 3766 sk->sk_incoming_cpu = -1; 3767 3768 sk_rx_queue_clear(sk); 3769 /* 3770 * Before updating sk_refcnt, we must commit prior changes to memory 3771 * (Documentation/RCU/rculist_nulls.rst for details) 3772 */ 3773 smp_wmb(); 3774 refcount_set(&sk->sk_refcnt, 1); 3775 sk_drops_reset(sk); 3776 } 3777 EXPORT_SYMBOL(sock_init_data_uid); 3778 3779 void sock_init_data(struct socket *sock, struct sock *sk) 3780 { 3781 kuid_t uid = sock ? 3782 SOCK_INODE(sock)->i_uid : 3783 make_kuid(sock_net(sk)->user_ns, 0); 3784 3785 sock_init_data_uid(sock, sk, uid); 3786 } 3787 EXPORT_SYMBOL(sock_init_data); 3788 3789 void noinline lock_sock_nested(struct sock *sk, int subclass) 3790 { 3791 /* The sk_lock has mutex_lock() semantics here. */ 3792 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3793 3794 might_sleep(); 3795 #ifdef CONFIG_64BIT 3796 if (sizeof(struct slock_owned) == sizeof(long)) { 3797 socket_lock_t tmp = { 3798 .slock = __SPIN_LOCK_UNLOCKED(tmp.slock), 3799 .owned = 1, 3800 }; 3801 socket_lock_t old = { 3802 .slock = __SPIN_LOCK_UNLOCKED(old.slock), 3803 .owned = 0, 3804 }; 3805 3806 if (likely(try_cmpxchg(&sk->sk_lock.combined, 3807 &old.combined, tmp.combined))) 3808 return; 3809 } 3810 #endif 3811 spin_lock_bh(&sk->sk_lock.slock); 3812 if (unlikely(sock_owned_by_user_nocheck(sk))) 3813 __lock_sock(sk); 3814 sk->sk_lock.owned = 1; 3815 spin_unlock_bh(&sk->sk_lock.slock); 3816 } 3817 EXPORT_SYMBOL(lock_sock_nested); 3818 3819 void release_sock(struct sock *sk) 3820 { 3821 spin_lock_bh(&sk->sk_lock.slock); 3822 3823 if (unlikely(sk->sk_backlog.tail)) 3824 __release_sock(sk); 3825 3826 if (sk->sk_prot->release_cb) { 3827 if (!tcp_release_cb_cond(sk)) 3828 sk->sk_prot->release_cb(sk); 3829 } 3830 sock_release_ownership(sk); 3831 if (unlikely(waitqueue_active(&sk->sk_lock.wq))) 3832 wake_up(&sk->sk_lock.wq); 3833 3834 spin_unlock_bh(&sk->sk_lock.slock); 3835 } 3836 EXPORT_SYMBOL(release_sock); 3837 3838 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3839 { 3840 might_sleep(); 3841 spin_lock_bh(&sk->sk_lock.slock); 3842 3843 if (likely(!sock_owned_by_user_nocheck(sk))) { 3844 /* 3845 * Fast path return with bottom halves disabled and 3846 * sock::sk_lock.slock held. 3847 * 3848 * The 'mutex' is not contended and holding 3849 * sock::sk_lock.slock prevents all other lockers to 3850 * proceed so the corresponding unlock_sock_fast() can 3851 * avoid the slow path of release_sock() completely and 3852 * just release slock. 3853 * 3854 * From a semantical POV this is equivalent to 'acquiring' 3855 * the 'mutex', hence the corresponding lockdep 3856 * mutex_release() has to happen in the fast path of 3857 * unlock_sock_fast(). 3858 */ 3859 return false; 3860 } 3861 3862 __lock_sock(sk); 3863 sk->sk_lock.owned = 1; 3864 __acquire(&sk->sk_lock.slock); 3865 spin_unlock_bh(&sk->sk_lock.slock); 3866 return true; 3867 } 3868 EXPORT_SYMBOL(__lock_sock_fast); 3869 3870 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3871 bool timeval, bool time32) 3872 { 3873 struct sock *sk = sock->sk; 3874 struct timespec64 ts; 3875 3876 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3877 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3878 if (ts.tv_sec == -1) 3879 return -ENOENT; 3880 if (ts.tv_sec == 0) { 3881 ktime_t kt = ktime_get_real(); 3882 sock_write_timestamp(sk, kt); 3883 ts = ktime_to_timespec64(kt); 3884 } 3885 3886 if (timeval) 3887 ts.tv_nsec /= 1000; 3888 3889 #ifdef CONFIG_COMPAT_32BIT_TIME 3890 if (time32) 3891 return put_old_timespec32(&ts, userstamp); 3892 #endif 3893 #ifdef CONFIG_SPARC64 3894 /* beware of padding in sparc64 timeval */ 3895 if (timeval && !in_compat_syscall()) { 3896 struct __kernel_old_timeval __user tv = { 3897 .tv_sec = ts.tv_sec, 3898 .tv_usec = ts.tv_nsec, 3899 }; 3900 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3901 return -EFAULT; 3902 return 0; 3903 } 3904 #endif 3905 return put_timespec64(&ts, userstamp); 3906 } 3907 EXPORT_SYMBOL(sock_gettstamp); 3908 3909 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3910 { 3911 if (!sock_flag(sk, flag)) { 3912 unsigned long previous_flags = sk->sk_flags; 3913 3914 sock_set_flag(sk, flag); 3915 /* 3916 * we just set one of the two flags which require net 3917 * time stamping, but time stamping might have been on 3918 * already because of the other one 3919 */ 3920 if (sock_needs_netstamp(sk) && 3921 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3922 net_enable_timestamp(); 3923 } 3924 } 3925 3926 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3927 int level, int type) 3928 { 3929 struct sock_extended_err ee; 3930 struct sk_buff *skb; 3931 int copied, err; 3932 3933 err = -EAGAIN; 3934 skb = sock_dequeue_err_skb(sk); 3935 if (skb == NULL) 3936 goto out; 3937 3938 copied = skb->len; 3939 if (copied > len) { 3940 msg->msg_flags |= MSG_TRUNC; 3941 copied = len; 3942 } 3943 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3944 if (err) 3945 goto out_free_skb; 3946 3947 sock_recv_timestamp(msg, sk, skb); 3948 3949 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */ 3950 ee = SKB_EXT_ERR(skb)->ee; 3951 put_cmsg(msg, level, type, sizeof(ee), &ee); 3952 3953 msg->msg_flags |= MSG_ERRQUEUE; 3954 err = copied; 3955 3956 out_free_skb: 3957 kfree_skb(skb); 3958 out: 3959 return err; 3960 } 3961 EXPORT_SYMBOL(sock_recv_errqueue); 3962 3963 /* 3964 * Get a socket option on an socket. 3965 * 3966 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3967 * asynchronous errors should be reported by getsockopt. We assume 3968 * this means if you specify SO_ERROR (otherwise what is the point of it). 3969 */ 3970 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3971 char __user *optval, int __user *optlen) 3972 { 3973 struct sock *sk = sock->sk; 3974 3975 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3976 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 3977 } 3978 EXPORT_SYMBOL(sock_common_getsockopt); 3979 3980 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3981 int flags) 3982 { 3983 struct sock *sk = sock->sk; 3984 3985 return sk->sk_prot->recvmsg(sk, msg, size, flags); 3986 } 3987 EXPORT_SYMBOL(sock_common_recvmsg); 3988 3989 /* 3990 * Set socket options on an inet socket. 3991 */ 3992 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3993 sockptr_t optval, unsigned int optlen) 3994 { 3995 struct sock *sk = sock->sk; 3996 3997 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 3998 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 3999 } 4000 EXPORT_SYMBOL(sock_common_setsockopt); 4001 4002 void sk_common_release(struct sock *sk) 4003 { 4004 if (sk->sk_prot->destroy) 4005 sk->sk_prot->destroy(sk); 4006 4007 /* 4008 * Observation: when sk_common_release is called, processes have 4009 * no access to socket. But net still has. 4010 * Step one, detach it from networking: 4011 * 4012 * A. Remove from hash tables. 4013 */ 4014 4015 sk->sk_prot->unhash(sk); 4016 4017 /* 4018 * In this point socket cannot receive new packets, but it is possible 4019 * that some packets are in flight because some CPU runs receiver and 4020 * did hash table lookup before we unhashed socket. They will achieve 4021 * receive queue and will be purged by socket destructor. 4022 * 4023 * Also we still have packets pending on receive queue and probably, 4024 * our own packets waiting in device queues. sock_destroy will drain 4025 * receive queue, but transmitted packets will delay socket destruction 4026 * until the last reference will be released. 4027 */ 4028 4029 sock_orphan(sk); 4030 4031 xfrm_sk_free_policy(sk); 4032 4033 sock_put(sk); 4034 } 4035 EXPORT_SYMBOL(sk_common_release); 4036 4037 void sk_get_meminfo(const struct sock *sk, u32 *mem) 4038 { 4039 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 4040 4041 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 4042 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 4043 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 4044 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 4045 mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc); 4046 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 4047 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 4048 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 4049 mem[SK_MEMINFO_DROPS] = sk_drops_read(sk); 4050 } 4051 4052 #ifdef CONFIG_PROC_FS 4053 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 4054 4055 int sock_prot_inuse_get(struct net *net, struct proto *prot) 4056 { 4057 int cpu, idx = prot->inuse_idx; 4058 int res = 0; 4059 4060 for_each_possible_cpu(cpu) 4061 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 4062 4063 return res >= 0 ? res : 0; 4064 } 4065 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 4066 4067 int sock_inuse_get(struct net *net) 4068 { 4069 int cpu, res = 0; 4070 4071 for_each_possible_cpu(cpu) 4072 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 4073 4074 return res; 4075 } 4076 4077 EXPORT_SYMBOL_GPL(sock_inuse_get); 4078 4079 static int __net_init sock_inuse_init_net(struct net *net) 4080 { 4081 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 4082 if (net->core.prot_inuse == NULL) 4083 return -ENOMEM; 4084 return 0; 4085 } 4086 4087 static void __net_exit sock_inuse_exit_net(struct net *net) 4088 { 4089 free_percpu(net->core.prot_inuse); 4090 } 4091 4092 static struct pernet_operations net_inuse_ops = { 4093 .init = sock_inuse_init_net, 4094 .exit = sock_inuse_exit_net, 4095 }; 4096 4097 static __init int net_inuse_init(void) 4098 { 4099 if (register_pernet_subsys(&net_inuse_ops)) 4100 panic("Cannot initialize net inuse counters"); 4101 4102 return 0; 4103 } 4104 4105 core_initcall(net_inuse_init); 4106 4107 static int assign_proto_idx(struct proto *prot) 4108 { 4109 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 4110 4111 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR)) { 4112 pr_err("PROTO_INUSE_NR exhausted\n"); 4113 return -ENOSPC; 4114 } 4115 4116 set_bit(prot->inuse_idx, proto_inuse_idx); 4117 return 0; 4118 } 4119 4120 static void release_proto_idx(struct proto *prot) 4121 { 4122 if (prot->inuse_idx != PROTO_INUSE_NR) 4123 clear_bit(prot->inuse_idx, proto_inuse_idx); 4124 } 4125 #else 4126 static inline int assign_proto_idx(struct proto *prot) 4127 { 4128 return 0; 4129 } 4130 4131 static inline void release_proto_idx(struct proto *prot) 4132 { 4133 } 4134 4135 #endif 4136 4137 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 4138 { 4139 if (!twsk_prot) 4140 return; 4141 kfree(twsk_prot->twsk_slab_name); 4142 twsk_prot->twsk_slab_name = NULL; 4143 kmem_cache_destroy(twsk_prot->twsk_slab); 4144 twsk_prot->twsk_slab = NULL; 4145 } 4146 4147 static int tw_prot_init(const struct proto *prot) 4148 { 4149 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 4150 4151 if (!twsk_prot) 4152 return 0; 4153 4154 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 4155 prot->name); 4156 if (!twsk_prot->twsk_slab_name) 4157 return -ENOMEM; 4158 4159 twsk_prot->twsk_slab = 4160 kmem_cache_create(twsk_prot->twsk_slab_name, 4161 twsk_prot->twsk_obj_size, 0, 4162 SLAB_ACCOUNT | prot->slab_flags, 4163 NULL); 4164 if (!twsk_prot->twsk_slab) { 4165 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 4166 prot->name); 4167 return -ENOMEM; 4168 } 4169 4170 return 0; 4171 } 4172 4173 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 4174 { 4175 if (!rsk_prot) 4176 return; 4177 kfree(rsk_prot->slab_name); 4178 rsk_prot->slab_name = NULL; 4179 kmem_cache_destroy(rsk_prot->slab); 4180 rsk_prot->slab = NULL; 4181 } 4182 4183 static int req_prot_init(const struct proto *prot) 4184 { 4185 struct request_sock_ops *rsk_prot = prot->rsk_prot; 4186 4187 if (!rsk_prot) 4188 return 0; 4189 4190 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4191 prot->name); 4192 if (!rsk_prot->slab_name) 4193 return -ENOMEM; 4194 4195 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4196 rsk_prot->obj_size, 0, 4197 SLAB_ACCOUNT | prot->slab_flags, 4198 NULL); 4199 4200 if (!rsk_prot->slab) { 4201 pr_crit("%s: Can't create request sock SLAB cache!\n", 4202 prot->name); 4203 return -ENOMEM; 4204 } 4205 return 0; 4206 } 4207 4208 int proto_register(struct proto *prot, int alloc_slab) 4209 { 4210 int ret = -ENOBUFS; 4211 4212 if (prot->memory_allocated && !prot->sysctl_mem) { 4213 pr_err("%s: missing sysctl_mem\n", prot->name); 4214 return -EINVAL; 4215 } 4216 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4217 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4218 return -EINVAL; 4219 } 4220 if (alloc_slab) { 4221 struct kmem_cache_args args = { 4222 .useroffset = prot->useroffset, 4223 .usersize = prot->usersize, 4224 .freeptr_offset = prot->freeptr_offset, 4225 .use_freeptr_offset = !!prot->freeptr_offset, 4226 }; 4227 4228 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 4229 &args, 4230 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4231 prot->slab_flags); 4232 if (prot->slab == NULL) { 4233 pr_crit("%s: Can't create sock SLAB cache!\n", 4234 prot->name); 4235 goto out; 4236 } 4237 4238 if (req_prot_init(prot)) 4239 goto out_free_request_sock_slab; 4240 4241 if (tw_prot_init(prot)) 4242 goto out_free_timewait_sock_slab; 4243 } 4244 4245 mutex_lock(&proto_list_mutex); 4246 ret = assign_proto_idx(prot); 4247 if (ret) { 4248 mutex_unlock(&proto_list_mutex); 4249 goto out_free_timewait_sock_slab; 4250 } 4251 list_add(&prot->node, &proto_list); 4252 mutex_unlock(&proto_list_mutex); 4253 return ret; 4254 4255 out_free_timewait_sock_slab: 4256 if (alloc_slab) 4257 tw_prot_cleanup(prot->twsk_prot); 4258 out_free_request_sock_slab: 4259 if (alloc_slab) { 4260 req_prot_cleanup(prot->rsk_prot); 4261 4262 kmem_cache_destroy(prot->slab); 4263 prot->slab = NULL; 4264 } 4265 out: 4266 return ret; 4267 } 4268 EXPORT_SYMBOL(proto_register); 4269 4270 void proto_unregister(struct proto *prot) 4271 { 4272 mutex_lock(&proto_list_mutex); 4273 release_proto_idx(prot); 4274 list_del(&prot->node); 4275 mutex_unlock(&proto_list_mutex); 4276 4277 kmem_cache_destroy(prot->slab); 4278 prot->slab = NULL; 4279 4280 req_prot_cleanup(prot->rsk_prot); 4281 tw_prot_cleanup(prot->twsk_prot); 4282 } 4283 EXPORT_SYMBOL(proto_unregister); 4284 4285 int sock_load_diag_module(int family, int protocol) 4286 { 4287 if (!protocol) { 4288 if (!sock_is_registered(family)) 4289 return -ENOENT; 4290 4291 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4292 NETLINK_SOCK_DIAG, family); 4293 } 4294 4295 #ifdef CONFIG_INET 4296 if (family == AF_INET && 4297 protocol != IPPROTO_RAW && 4298 protocol < MAX_INET_PROTOS && 4299 !rcu_access_pointer(inet_protos[protocol])) 4300 return -ENOENT; 4301 #endif 4302 4303 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4304 NETLINK_SOCK_DIAG, family, protocol); 4305 } 4306 EXPORT_SYMBOL(sock_load_diag_module); 4307 4308 #ifdef CONFIG_PROC_FS 4309 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4310 __acquires(proto_list_mutex) 4311 { 4312 mutex_lock(&proto_list_mutex); 4313 return seq_list_start_head(&proto_list, *pos); 4314 } 4315 4316 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4317 { 4318 return seq_list_next(v, &proto_list, pos); 4319 } 4320 4321 static void proto_seq_stop(struct seq_file *seq, void *v) 4322 __releases(proto_list_mutex) 4323 { 4324 mutex_unlock(&proto_list_mutex); 4325 } 4326 4327 static char proto_method_implemented(const void *method) 4328 { 4329 return method == NULL ? 'n' : 'y'; 4330 } 4331 static long sock_prot_memory_allocated(struct proto *proto) 4332 { 4333 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4334 } 4335 4336 static const char *sock_prot_memory_pressure(struct proto *proto) 4337 { 4338 return proto->memory_pressure != NULL ? 4339 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4340 } 4341 4342 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4343 { 4344 4345 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4346 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4347 proto->name, 4348 proto->obj_size, 4349 sock_prot_inuse_get(seq_file_net(seq), proto), 4350 sock_prot_memory_allocated(proto), 4351 sock_prot_memory_pressure(proto), 4352 proto->max_header, 4353 proto->slab == NULL ? "no" : "yes", 4354 module_name(proto->owner), 4355 proto_method_implemented(proto->close), 4356 proto_method_implemented(proto->connect), 4357 proto_method_implemented(proto->disconnect), 4358 proto_method_implemented(proto->accept), 4359 proto_method_implemented(proto->ioctl), 4360 proto_method_implemented(proto->init), 4361 proto_method_implemented(proto->destroy), 4362 proto_method_implemented(proto->shutdown), 4363 proto_method_implemented(proto->setsockopt), 4364 proto_method_implemented(proto->getsockopt), 4365 proto_method_implemented(proto->sendmsg), 4366 proto_method_implemented(proto->recvmsg), 4367 proto_method_implemented(proto->bind), 4368 proto_method_implemented(proto->backlog_rcv), 4369 proto_method_implemented(proto->hash), 4370 proto_method_implemented(proto->unhash), 4371 proto_method_implemented(proto->get_port), 4372 proto_method_implemented(proto->enter_memory_pressure)); 4373 } 4374 4375 static int proto_seq_show(struct seq_file *seq, void *v) 4376 { 4377 if (v == &proto_list) 4378 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4379 "protocol", 4380 "size", 4381 "sockets", 4382 "memory", 4383 "press", 4384 "maxhdr", 4385 "slab", 4386 "module", 4387 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4388 else 4389 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4390 return 0; 4391 } 4392 4393 static const struct seq_operations proto_seq_ops = { 4394 .start = proto_seq_start, 4395 .next = proto_seq_next, 4396 .stop = proto_seq_stop, 4397 .show = proto_seq_show, 4398 }; 4399 4400 static __net_init int proto_init_net(struct net *net) 4401 { 4402 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4403 sizeof(struct seq_net_private))) 4404 return -ENOMEM; 4405 4406 return 0; 4407 } 4408 4409 static __net_exit void proto_exit_net(struct net *net) 4410 { 4411 remove_proc_entry("protocols", net->proc_net); 4412 } 4413 4414 4415 static __net_initdata struct pernet_operations proto_net_ops = { 4416 .init = proto_init_net, 4417 .exit = proto_exit_net, 4418 }; 4419 4420 static int __init proto_init(void) 4421 { 4422 return register_pernet_subsys(&proto_net_ops); 4423 } 4424 4425 subsys_initcall(proto_init); 4426 4427 #endif /* PROC_FS */ 4428 4429 #ifdef CONFIG_NET_RX_BUSY_POLL 4430 bool sk_busy_loop_end(void *p, unsigned long start_time) 4431 { 4432 struct sock *sk = p; 4433 4434 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4435 return true; 4436 4437 if (sk_is_udp(sk) && 4438 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4439 return true; 4440 4441 return sk_busy_loop_timeout(sk, start_time); 4442 } 4443 EXPORT_SYMBOL(sk_busy_loop_end); 4444 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4445 4446 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len) 4447 { 4448 if (!sk->sk_prot->bind_add) 4449 return -EOPNOTSUPP; 4450 return sk->sk_prot->bind_add(sk, addr, addr_len); 4451 } 4452 EXPORT_SYMBOL(sock_bind_add); 4453 4454 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4455 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4456 void __user *arg, void *karg, size_t size) 4457 { 4458 int ret; 4459 4460 if (copy_from_user(karg, arg, size)) 4461 return -EFAULT; 4462 4463 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4464 if (ret) 4465 return ret; 4466 4467 if (copy_to_user(arg, karg, size)) 4468 return -EFAULT; 4469 4470 return 0; 4471 } 4472 EXPORT_SYMBOL(sock_ioctl_inout); 4473 4474 /* This is the most common ioctl prep function, where the result (4 bytes) is 4475 * copied back to userspace if the ioctl() returns successfully. No input is 4476 * copied from userspace as input argument. 4477 */ 4478 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4479 { 4480 int ret, karg = 0; 4481 4482 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4483 if (ret) 4484 return ret; 4485 4486 return put_user(karg, (int __user *)arg); 4487 } 4488 4489 /* A wrapper around sock ioctls, which copies the data from userspace 4490 * (depending on the protocol/ioctl), and copies back the result to userspace. 4491 * The main motivation for this function is to pass kernel memory to the 4492 * protocol ioctl callbacks, instead of userspace memory. 4493 */ 4494 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4495 { 4496 int rc = 1; 4497 4498 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4499 rc = ipmr_sk_ioctl(sk, cmd, arg); 4500 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4501 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4502 else if (sk_is_phonet(sk)) 4503 rc = phonet_sk_ioctl(sk, cmd, arg); 4504 4505 /* If ioctl was processed, returns its value */ 4506 if (rc <= 0) 4507 return rc; 4508 4509 /* Otherwise call the default handler */ 4510 return sock_ioctl_out(sk, cmd, arg); 4511 } 4512 EXPORT_SYMBOL(sk_ioctl); 4513 4514 static int __init sock_struct_check(void) 4515 { 4516 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4517 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4518 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4519 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4520 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4521 4522 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4523 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4524 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4525 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4526 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4527 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4528 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4529 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4530 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4531 4532 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4533 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4534 #ifdef CONFIG_MEMCG 4535 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4536 #endif 4537 4538 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4539 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4540 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4541 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4542 4543 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4544 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4545 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_err_soft); 4546 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4547 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4548 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4549 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4550 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4551 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4552 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4553 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4554 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4555 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4556 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4557 4558 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_pending_confirm); 4559 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_status); 4560 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4561 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4562 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4563 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4564 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_uid); 4565 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_protocol); 4566 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4567 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4568 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4569 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4570 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4571 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4572 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndbuf); 4573 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4574 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4575 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4576 return 0; 4577 } 4578 4579 core_initcall(sock_struct_check); 4580