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 <net/page_pool/types.h> 133 #include <linux/net_tstamp.h> 134 #include <net/xfrm.h> 135 #include <linux/ipsec.h> 136 #include <net/cls_cgroup.h> 137 #include <net/netprio_cgroup.h> 138 #include <linux/sock_diag.h> 139 140 #include <linux/filter.h> 141 #include <net/sock_reuseport.h> 142 #include <net/bpf_sk_storage.h> 143 144 #include <trace/events/sock.h> 145 146 #include <net/psp.h> 147 #include <net/tcp.h> 148 #include <net/busy_poll.h> 149 #include <net/phonet/phonet.h> 150 151 #include <linux/ethtool.h> 152 153 #include <uapi/linux/pidfd.h> 154 155 #include "dev.h" 156 157 static DEFINE_MUTEX(proto_list_mutex); 158 static LIST_HEAD(proto_list); 159 160 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc); 161 static void sock_def_write_space(struct sock *sk); 162 163 /** 164 * sk_ns_capable - General socket capability test 165 * @sk: Socket to use a capability on or through 166 * @user_ns: The user namespace of the capability to use 167 * @cap: The capability to use 168 * 169 * Test to see if the opener of the socket had when the socket was 170 * created and the current process has the capability @cap in the user 171 * namespace @user_ns. 172 */ 173 bool sk_ns_capable(const struct sock *sk, 174 struct user_namespace *user_ns, int cap) 175 { 176 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 177 ns_capable(user_ns, cap); 178 } 179 EXPORT_SYMBOL(sk_ns_capable); 180 181 /** 182 * sk_capable - Socket global capability test 183 * @sk: Socket to use a capability on or through 184 * @cap: The global capability to use 185 * 186 * Test to see if the opener of the socket had when the socket was 187 * created and the current process has the capability @cap in all user 188 * namespaces. 189 */ 190 bool sk_capable(const struct sock *sk, int cap) 191 { 192 return sk_ns_capable(sk, &init_user_ns, cap); 193 } 194 EXPORT_SYMBOL(sk_capable); 195 196 /** 197 * sk_net_capable - Network namespace socket capability test 198 * @sk: Socket to use a capability on or through 199 * @cap: The capability to use 200 * 201 * Test to see if the opener of the socket had when the socket was created 202 * and the current process has the capability @cap over the network namespace 203 * the socket is a member of. 204 */ 205 bool sk_net_capable(const struct sock *sk, int cap) 206 { 207 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 208 } 209 EXPORT_SYMBOL(sk_net_capable); 210 211 /* 212 * Each address family might have different locking rules, so we have 213 * one slock key per address family and separate keys for internal and 214 * userspace sockets. 215 */ 216 static struct lock_class_key af_family_keys[AF_MAX]; 217 static struct lock_class_key af_family_kern_keys[AF_MAX]; 218 static struct lock_class_key af_family_slock_keys[AF_MAX]; 219 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 220 221 /* 222 * Make lock validator output more readable. (we pre-construct these 223 * strings build-time, so that runtime initialization of socket 224 * locks is fast): 225 */ 226 227 #define _sock_locks(x) \ 228 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 229 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 230 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 231 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 232 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 233 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 234 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 235 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 236 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 237 x "27" , x "28" , x "AF_CAN" , \ 238 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 239 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 240 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 241 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 242 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 243 x "AF_MCTP" , \ 244 x "AF_MAX" 245 246 static const char *const af_family_key_strings[AF_MAX+1] = { 247 _sock_locks("sk_lock-") 248 }; 249 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 250 _sock_locks("slock-") 251 }; 252 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 253 _sock_locks("clock-") 254 }; 255 256 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 257 _sock_locks("k-sk_lock-") 258 }; 259 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 260 _sock_locks("k-slock-") 261 }; 262 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 263 _sock_locks("k-clock-") 264 }; 265 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 266 _sock_locks("rlock-") 267 }; 268 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 269 _sock_locks("wlock-") 270 }; 271 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 272 _sock_locks("elock-") 273 }; 274 275 /* 276 * sk_callback_lock and sk queues locking rules are per-address-family, 277 * so split the lock classes by using a per-AF key: 278 */ 279 static struct lock_class_key af_callback_keys[AF_MAX]; 280 static struct lock_class_key af_rlock_keys[AF_MAX]; 281 static struct lock_class_key af_wlock_keys[AF_MAX]; 282 static struct lock_class_key af_elock_keys[AF_MAX]; 283 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 284 285 /* Run time adjustable parameters. */ 286 __u32 sysctl_wmem_max __read_mostly = 4 << 20; 287 EXPORT_SYMBOL(sysctl_wmem_max); 288 __u32 sysctl_rmem_max __read_mostly = 4 << 20; 289 EXPORT_SYMBOL(sysctl_rmem_max); 290 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_DEFAULT; 291 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_DEFAULT; 292 293 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 294 EXPORT_SYMBOL_GPL(memalloc_socks_key); 295 296 /** 297 * sk_set_memalloc - sets %SOCK_MEMALLOC 298 * @sk: socket to set it on 299 * 300 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 301 * It's the responsibility of the admin to adjust min_free_kbytes 302 * to meet the requirements 303 */ 304 void sk_set_memalloc(struct sock *sk) 305 { 306 sock_set_flag(sk, SOCK_MEMALLOC); 307 sk->sk_allocation |= __GFP_MEMALLOC; 308 static_branch_inc(&memalloc_socks_key); 309 } 310 EXPORT_SYMBOL_GPL(sk_set_memalloc); 311 312 void sk_clear_memalloc(struct sock *sk) 313 { 314 sock_reset_flag(sk, SOCK_MEMALLOC); 315 sk->sk_allocation &= ~__GFP_MEMALLOC; 316 static_branch_dec(&memalloc_socks_key); 317 318 /* 319 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 320 * progress of swapping. SOCK_MEMALLOC may be cleared while 321 * it has rmem allocations due to the last swapfile being deactivated 322 * but there is a risk that the socket is unusable due to exceeding 323 * the rmem limits. Reclaim the reserves and obey rmem limits again. 324 */ 325 sk_mem_reclaim(sk); 326 } 327 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 328 329 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 330 { 331 int ret; 332 unsigned int noreclaim_flag; 333 334 /* these should have been dropped before queueing */ 335 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 336 337 noreclaim_flag = memalloc_noreclaim_save(); 338 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 339 tcp_v6_do_rcv, 340 tcp_v4_do_rcv, 341 sk, skb); 342 memalloc_noreclaim_restore(noreclaim_flag); 343 344 return ret; 345 } 346 EXPORT_SYMBOL(__sk_backlog_rcv); 347 348 void sk_error_report(struct sock *sk) 349 { 350 sk->sk_error_report(sk); 351 352 switch (sk->sk_family) { 353 case AF_INET: 354 fallthrough; 355 case AF_INET6: 356 trace_inet_sk_error_report(sk); 357 break; 358 default: 359 break; 360 } 361 } 362 EXPORT_SYMBOL(sk_error_report); 363 364 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 365 { 366 struct __kernel_sock_timeval tv; 367 368 if (timeo == MAX_SCHEDULE_TIMEOUT) { 369 tv.tv_sec = 0; 370 tv.tv_usec = 0; 371 } else { 372 tv.tv_sec = timeo / HZ; 373 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 374 } 375 376 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 377 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 378 *(struct old_timeval32 *)optval = tv32; 379 return sizeof(tv32); 380 } 381 382 if (old_timeval) { 383 struct __kernel_old_timeval old_tv; 384 old_tv.tv_sec = tv.tv_sec; 385 old_tv.tv_usec = tv.tv_usec; 386 *(struct __kernel_old_timeval *)optval = old_tv; 387 return sizeof(old_tv); 388 } 389 390 *(struct __kernel_sock_timeval *)optval = tv; 391 return sizeof(tv); 392 } 393 EXPORT_SYMBOL(sock_get_timeout); 394 395 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 396 sockptr_t optval, int optlen, bool old_timeval) 397 { 398 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 399 struct old_timeval32 tv32; 400 401 if (optlen < sizeof(tv32)) 402 return -EINVAL; 403 404 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 405 return -EFAULT; 406 tv->tv_sec = tv32.tv_sec; 407 tv->tv_usec = tv32.tv_usec; 408 } else if (old_timeval) { 409 struct __kernel_old_timeval old_tv; 410 411 if (optlen < sizeof(old_tv)) 412 return -EINVAL; 413 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 414 return -EFAULT; 415 tv->tv_sec = old_tv.tv_sec; 416 tv->tv_usec = old_tv.tv_usec; 417 } else { 418 if (optlen < sizeof(*tv)) 419 return -EINVAL; 420 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 421 return -EFAULT; 422 } 423 424 return 0; 425 } 426 EXPORT_SYMBOL(sock_copy_user_timeval); 427 428 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 429 bool old_timeval) 430 { 431 struct __kernel_sock_timeval tv; 432 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 433 long val; 434 435 if (err) 436 return err; 437 438 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 439 return -EDOM; 440 441 if (tv.tv_sec < 0) { 442 static int warned __read_mostly; 443 444 WRITE_ONCE(*timeo_p, 0); 445 if (warned < 10 && net_ratelimit()) { 446 warned++; 447 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 448 __func__, current->comm, task_pid_nr(current)); 449 } 450 return 0; 451 } 452 val = MAX_SCHEDULE_TIMEOUT; 453 if ((tv.tv_sec || tv.tv_usec) && 454 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 455 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 456 USEC_PER_SEC / HZ); 457 WRITE_ONCE(*timeo_p, val); 458 return 0; 459 } 460 461 static bool sk_set_prio_allowed(const struct sock *sk, int val) 462 { 463 return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) || 464 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 465 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)); 466 } 467 468 static bool sock_needs_netstamp(const struct sock *sk) 469 { 470 switch (sk->sk_family) { 471 case AF_UNSPEC: 472 case AF_UNIX: 473 return false; 474 default: 475 return true; 476 } 477 } 478 479 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 480 { 481 if (sk->sk_flags & flags) { 482 sk->sk_flags &= ~flags; 483 if (sock_needs_netstamp(sk) && 484 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 485 net_disable_timestamp(); 486 } 487 } 488 489 490 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 491 { 492 unsigned long flags; 493 struct sk_buff_head *list = &sk->sk_receive_queue; 494 495 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) { 496 sk_drops_inc(sk); 497 trace_sock_rcvqueue_full(sk, skb); 498 return -ENOMEM; 499 } 500 501 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 502 sk_drops_inc(sk); 503 return -ENOBUFS; 504 } 505 506 skb->dev = NULL; 507 skb_set_owner_r(skb, sk); 508 509 /* we escape from rcu protected region, make sure we dont leak 510 * a norefcounted dst 511 */ 512 skb_dst_force(skb); 513 514 spin_lock_irqsave(&list->lock, flags); 515 sock_skb_set_dropcount(sk, skb); 516 __skb_queue_tail(list, skb); 517 spin_unlock_irqrestore(&list->lock, flags); 518 519 if (!sock_flag(sk, SOCK_DEAD)) 520 sk->sk_data_ready(sk); 521 return 0; 522 } 523 EXPORT_SYMBOL(__sock_queue_rcv_skb); 524 525 enum skb_drop_reason 526 sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb) 527 { 528 enum skb_drop_reason drop_reason; 529 int err; 530 531 drop_reason = sk_filter_reason(sk, skb); 532 if (drop_reason) 533 return drop_reason; 534 535 err = __sock_queue_rcv_skb(sk, skb); 536 switch (err) { 537 case -ENOMEM: 538 return SKB_DROP_REASON_SOCKET_RCVBUFF; 539 case -ENOBUFS: 540 return SKB_DROP_REASON_PROTO_MEM; 541 } 542 return SKB_NOT_DROPPED_YET; 543 } 544 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 545 546 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 547 const int nested, unsigned int trim_cap, bool refcounted) 548 { 549 enum skb_drop_reason reason; 550 int rc = NET_RX_SUCCESS; 551 int err; 552 553 reason = sk_filter_trim_cap(sk, skb, trim_cap); 554 if (reason) 555 goto discard_and_relse; 556 557 skb->dev = NULL; 558 559 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) { 560 sk_drops_inc(sk); 561 reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 562 goto discard_and_relse; 563 } 564 if (nested) 565 bh_lock_sock_nested(sk); 566 else 567 bh_lock_sock(sk); 568 if (!sock_owned_by_user(sk)) { 569 /* 570 * trylock + unlock semantics: 571 */ 572 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 573 574 rc = sk_backlog_rcv(sk, skb); 575 576 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 577 } else if ((err = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))) { 578 bh_unlock_sock(sk); 579 if (err == -ENOMEM) 580 reason = SKB_DROP_REASON_PFMEMALLOC; 581 if (err == -ENOBUFS) 582 reason = SKB_DROP_REASON_SOCKET_BACKLOG; 583 sk_drops_inc(sk); 584 goto discard_and_relse; 585 } 586 587 bh_unlock_sock(sk); 588 out: 589 if (refcounted) 590 sock_put(sk); 591 return rc; 592 discard_and_relse: 593 sk_skb_reason_drop(sk, skb, reason); 594 goto out; 595 } 596 EXPORT_SYMBOL(__sk_receive_skb); 597 598 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 599 u32)); 600 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 601 u32)); 602 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 603 { 604 struct dst_entry *dst = __sk_dst_get(sk); 605 606 if (dst && READ_ONCE(dst->obsolete) && 607 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 608 dst, cookie) == NULL) { 609 sk_tx_queue_clear(sk); 610 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 611 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 612 dst_release(dst); 613 return NULL; 614 } 615 616 return dst; 617 } 618 EXPORT_SYMBOL(__sk_dst_check); 619 620 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 621 { 622 struct dst_entry *dst = sk_dst_get(sk); 623 624 if (dst && READ_ONCE(dst->obsolete) && 625 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 626 dst, cookie) == NULL) { 627 sk_dst_reset(sk); 628 dst_release(dst); 629 return NULL; 630 } 631 632 return dst; 633 } 634 EXPORT_SYMBOL(sk_dst_check); 635 636 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 637 { 638 int ret = -ENOPROTOOPT; 639 #ifdef CONFIG_NETDEVICES 640 struct net *net = sock_net(sk); 641 642 /* Sorry... */ 643 ret = -EPERM; 644 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 645 goto out; 646 647 ret = -EINVAL; 648 if (ifindex < 0) 649 goto out; 650 651 /* Paired with all READ_ONCE() done locklessly. */ 652 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 653 654 if (sk->sk_prot->rehash) 655 sk->sk_prot->rehash(sk); 656 sk_dst_reset(sk); 657 658 ret = 0; 659 660 out: 661 #endif 662 663 return ret; 664 } 665 666 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 667 { 668 int ret; 669 670 if (lock_sk) 671 lock_sock(sk); 672 ret = sock_bindtoindex_locked(sk, ifindex); 673 if (lock_sk) 674 release_sock(sk); 675 676 return ret; 677 } 678 EXPORT_SYMBOL(sock_bindtoindex); 679 680 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 681 { 682 int ret = -ENOPROTOOPT; 683 #ifdef CONFIG_NETDEVICES 684 struct net *net = sock_net(sk); 685 char devname[IFNAMSIZ]; 686 int index; 687 688 ret = -EINVAL; 689 if (optlen < 0) 690 goto out; 691 692 /* Bind this socket to a particular device like "eth0", 693 * as specified in the passed interface name. If the 694 * name is "" or the option length is zero the socket 695 * is not bound. 696 */ 697 if (optlen > IFNAMSIZ - 1) 698 optlen = IFNAMSIZ - 1; 699 memset(devname, 0, sizeof(devname)); 700 701 ret = -EFAULT; 702 if (copy_from_sockptr(devname, optval, optlen)) 703 goto out; 704 705 index = 0; 706 if (devname[0] != '\0') { 707 struct net_device *dev; 708 709 rcu_read_lock(); 710 dev = dev_get_by_name_rcu(net, devname); 711 if (dev) 712 index = dev->ifindex; 713 rcu_read_unlock(); 714 ret = -ENODEV; 715 if (!dev) 716 goto out; 717 } 718 719 sockopt_lock_sock(sk); 720 ret = sock_bindtoindex_locked(sk, index); 721 sockopt_release_sock(sk); 722 out: 723 #endif 724 725 return ret; 726 } 727 728 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 729 sockptr_t optlen, int len) 730 { 731 int ret = -ENOPROTOOPT; 732 #ifdef CONFIG_NETDEVICES 733 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 734 struct net *net = sock_net(sk); 735 char devname[IFNAMSIZ]; 736 737 if (bound_dev_if == 0) { 738 len = 0; 739 goto zero; 740 } 741 742 ret = -EINVAL; 743 if (len < IFNAMSIZ) 744 goto out; 745 746 ret = netdev_get_name(net, devname, bound_dev_if); 747 if (ret) 748 goto out; 749 750 len = strlen(devname) + 1; 751 752 ret = -EFAULT; 753 if (copy_to_sockptr(optval, devname, len)) 754 goto out; 755 756 zero: 757 ret = -EFAULT; 758 if (copy_to_sockptr(optlen, &len, sizeof(int))) 759 goto out; 760 761 ret = 0; 762 763 out: 764 #endif 765 766 return ret; 767 } 768 769 bool sk_mc_loop(const struct sock *sk) 770 { 771 if (dev_recursion_level()) 772 return false; 773 if (!sk) 774 return true; 775 776 switch (READ_ONCE(sk->sk_family)) { 777 case AF_INET: 778 return inet_test_bit(MC_LOOP, sk); 779 #if IS_ENABLED(CONFIG_IPV6) 780 case AF_INET6: 781 return inet6_test_bit(MC6_LOOP, sk); 782 #endif 783 } 784 return true; 785 } 786 EXPORT_SYMBOL(sk_mc_loop); 787 788 void sock_set_reuseaddr(struct sock *sk) 789 { 790 lock_sock(sk); 791 sk->sk_reuse = SK_CAN_REUSE; 792 release_sock(sk); 793 } 794 EXPORT_SYMBOL(sock_set_reuseaddr); 795 796 void sock_set_reuseport(struct sock *sk) 797 { 798 lock_sock(sk); 799 sk->sk_reuseport = true; 800 release_sock(sk); 801 } 802 EXPORT_SYMBOL(sock_set_reuseport); 803 804 void sock_no_linger(struct sock *sk) 805 { 806 lock_sock(sk); 807 WRITE_ONCE(sk->sk_lingertime, 0); 808 sock_set_flag(sk, SOCK_LINGER); 809 release_sock(sk); 810 } 811 EXPORT_SYMBOL(sock_no_linger); 812 813 void sock_set_priority(struct sock *sk, u32 priority) 814 { 815 WRITE_ONCE(sk->sk_priority, priority); 816 } 817 EXPORT_SYMBOL(sock_set_priority); 818 819 void sock_set_sndtimeo(struct sock *sk, s64 secs) 820 { 821 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 822 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 823 else 824 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 825 } 826 EXPORT_SYMBOL(sock_set_sndtimeo); 827 828 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 829 { 830 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val); 831 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns); 832 if (val) { 833 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 834 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 835 } 836 } 837 838 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 839 { 840 switch (optname) { 841 case SO_TIMESTAMP_OLD: 842 __sock_set_timestamps(sk, valbool, false, false); 843 break; 844 case SO_TIMESTAMP_NEW: 845 __sock_set_timestamps(sk, valbool, true, false); 846 break; 847 case SO_TIMESTAMPNS_OLD: 848 __sock_set_timestamps(sk, valbool, false, true); 849 break; 850 case SO_TIMESTAMPNS_NEW: 851 __sock_set_timestamps(sk, valbool, true, true); 852 break; 853 } 854 } 855 856 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 857 { 858 struct net *net = sock_net(sk); 859 struct net_device *dev = NULL; 860 bool match = false; 861 int *vclock_index; 862 int i, num; 863 864 if (sk->sk_bound_dev_if) 865 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 866 867 if (!dev) { 868 pr_err("%s: sock not bind to device\n", __func__); 869 return -EOPNOTSUPP; 870 } 871 872 num = ethtool_get_phc_vclocks(dev, &vclock_index); 873 dev_put(dev); 874 875 for (i = 0; i < num; i++) { 876 if (*(vclock_index + i) == phc_index) { 877 match = true; 878 break; 879 } 880 } 881 882 if (num > 0) 883 kfree(vclock_index); 884 885 if (!match) 886 return -EINVAL; 887 888 WRITE_ONCE(sk->sk_bind_phc, phc_index); 889 890 return 0; 891 } 892 893 int sock_set_timestamping(struct sock *sk, int optname, 894 struct so_timestamping timestamping) 895 { 896 int val = timestamping.flags; 897 int ret; 898 899 if (val & ~SOF_TIMESTAMPING_MASK) 900 return -EINVAL; 901 902 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 903 !(val & SOF_TIMESTAMPING_OPT_ID)) 904 return -EINVAL; 905 906 if (val & SOF_TIMESTAMPING_OPT_ID && 907 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 908 if (sk_is_tcp(sk)) { 909 if ((1 << sk->sk_state) & 910 (TCPF_CLOSE | TCPF_LISTEN)) 911 return -EINVAL; 912 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 913 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 914 else 915 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 916 } else { 917 atomic_set(&sk->sk_tskey, 0); 918 } 919 } 920 921 if (val & SOF_TIMESTAMPING_OPT_STATS && 922 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 923 return -EINVAL; 924 925 if (val & SOF_TIMESTAMPING_BIND_PHC) { 926 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 927 if (ret) 928 return ret; 929 } 930 931 WRITE_ONCE(sk->sk_tsflags, val); 932 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 933 sock_valbool_flag(sk, SOCK_TIMESTAMPING_ANY, !!(val & TSFLAGS_ANY)); 934 935 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 936 sock_enable_timestamp(sk, 937 SOCK_TIMESTAMPING_RX_SOFTWARE); 938 else 939 sock_disable_timestamp(sk, 940 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 941 return 0; 942 } 943 944 #if defined(CONFIG_CGROUP_BPF) 945 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 946 { 947 struct bpf_sock_ops_kern sock_ops; 948 949 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 950 sock_ops.op = op; 951 sock_ops.is_fullsock = 1; 952 sock_ops.sk = sk; 953 bpf_skops_init_skb(&sock_ops, skb, 0); 954 __cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS); 955 } 956 #endif 957 958 void sock_set_keepalive(struct sock *sk) 959 { 960 lock_sock(sk); 961 if (sk->sk_prot->keepalive) 962 sk->sk_prot->keepalive(sk, true); 963 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 964 release_sock(sk); 965 } 966 EXPORT_SYMBOL(sock_set_keepalive); 967 968 static void __sock_set_rcvbuf(struct sock *sk, int val) 969 { 970 struct socket *sock = sk->sk_socket; 971 972 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 973 * as a negative value. 974 */ 975 val = min_t(int, val, INT_MAX / 2); 976 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 977 978 /* We double it on the way in to account for "struct sk_buff" etc. 979 * overhead. Applications assume that the SO_RCVBUF setting they make 980 * will allow that much actual data to be received on that socket. 981 * 982 * Applications are unaware that "struct sk_buff" and other overheads 983 * allocate from the receive buffer during socket buffer allocation. 984 * 985 * And after considering the possible alternatives, returning the value 986 * we actually used in getsockopt is the most desirable behavior. 987 */ 988 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 989 990 if (sock) { 991 const struct proto_ops *ops = READ_ONCE(sock->ops); 992 993 if (ops->set_rcvbuf) 994 ops->set_rcvbuf(sk, sk->sk_rcvbuf); 995 } 996 } 997 998 void sock_set_rcvbuf(struct sock *sk, int val) 999 { 1000 lock_sock(sk); 1001 __sock_set_rcvbuf(sk, val); 1002 release_sock(sk); 1003 } 1004 EXPORT_SYMBOL(sock_set_rcvbuf); 1005 1006 static void __sock_set_mark(struct sock *sk, u32 val) 1007 { 1008 if (val != sk->sk_mark) { 1009 WRITE_ONCE(sk->sk_mark, val); 1010 sk_dst_reset(sk); 1011 } 1012 } 1013 1014 void sock_set_mark(struct sock *sk, u32 val) 1015 { 1016 lock_sock(sk); 1017 __sock_set_mark(sk, val); 1018 release_sock(sk); 1019 } 1020 EXPORT_SYMBOL(sock_set_mark); 1021 1022 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1023 { 1024 /* Round down bytes to multiple of pages */ 1025 bytes = round_down(bytes, PAGE_SIZE); 1026 1027 WARN_ON(bytes > sk->sk_reserved_mem); 1028 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1029 sk_mem_reclaim(sk); 1030 } 1031 1032 static int sock_reserve_memory(struct sock *sk, int bytes) 1033 { 1034 long allocated; 1035 bool charged; 1036 int pages; 1037 1038 if (!mem_cgroup_sk_enabled(sk) || !sk_is_tcp(sk)) 1039 return -EOPNOTSUPP; 1040 1041 if (!bytes) 1042 return 0; 1043 1044 pages = sk_mem_pages(bytes); 1045 1046 /* pre-charge to memcg */ 1047 charged = mem_cgroup_sk_charge(sk, pages, 1048 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1049 if (!charged) 1050 return -ENOMEM; 1051 1052 if (sk->sk_bypass_prot_mem) 1053 goto success; 1054 1055 /* pre-charge to forward_alloc */ 1056 sk_memory_allocated_add(sk, pages); 1057 allocated = sk_memory_allocated(sk); 1058 1059 /* If the system goes into memory pressure with this 1060 * precharge, give up and return error. 1061 */ 1062 if (allocated > sk_prot_mem_limits(sk, 1)) { 1063 sk_memory_allocated_sub(sk, pages); 1064 mem_cgroup_sk_uncharge(sk, pages); 1065 return -ENOMEM; 1066 } 1067 1068 success: 1069 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1070 1071 WRITE_ONCE(sk->sk_reserved_mem, 1072 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1073 1074 return 0; 1075 } 1076 1077 #ifdef CONFIG_PAGE_POOL 1078 1079 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED 1080 * in 1 syscall. The limit exists to limit the amount of memory the kernel 1081 * allocates to copy these tokens, and to prevent looping over the frags for 1082 * too long. 1083 */ 1084 #define MAX_DONTNEED_TOKENS 128 1085 #define MAX_DONTNEED_FRAGS 1024 1086 1087 static noinline_for_stack int 1088 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1089 { 1090 unsigned int num_tokens, i, j, netmem_num = 0; 1091 struct dmabuf_token *tokens; 1092 int ret = 0, num_frags = 0; 1093 netmem_ref netmems[16]; 1094 1095 if (!sk_is_tcp(sk)) 1096 return -EBADF; 1097 1098 if (optlen % sizeof(*tokens) || 1099 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1100 return -EINVAL; 1101 1102 num_tokens = optlen / sizeof(*tokens); 1103 tokens = kvmalloc_objs(*tokens, num_tokens); 1104 if (!tokens) 1105 return -ENOMEM; 1106 1107 if (copy_from_sockptr(tokens, optval, optlen)) { 1108 kvfree(tokens); 1109 return -EFAULT; 1110 } 1111 1112 xa_lock_bh(&sk->sk_user_frags); 1113 for (i = 0; i < num_tokens; i++) { 1114 for (j = 0; j < tokens[i].token_count; j++) { 1115 if (++num_frags > MAX_DONTNEED_FRAGS) 1116 goto frag_limit_reached; 1117 1118 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1119 &sk->sk_user_frags, tokens[i].token_start + j); 1120 1121 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem))) 1122 continue; 1123 1124 netmems[netmem_num++] = netmem; 1125 if (netmem_num == ARRAY_SIZE(netmems)) { 1126 xa_unlock_bh(&sk->sk_user_frags); 1127 page_pool_put_netmem_bulk(netmems, 1128 ARRAY_SIZE(netmems)); 1129 netmem_num = 0; 1130 xa_lock_bh(&sk->sk_user_frags); 1131 } 1132 ret++; 1133 } 1134 } 1135 1136 frag_limit_reached: 1137 xa_unlock_bh(&sk->sk_user_frags); 1138 page_pool_put_netmem_bulk(netmems, netmem_num); 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 || val > SZ_1G) { 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 #ifdef CONFIG_BPF_SYSCALL 2496 RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL); 2497 #endif 2498 #if IS_ENABLED(CONFIG_INET_PSP) 2499 DEBUG_NET_WARN_ON_ONCE(rcu_access_pointer(sk->psp_assoc)); 2500 #endif 2501 2502 /* SANITY */ 2503 if (likely(newsk->sk_net_refcnt)) { 2504 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2505 sock_inuse_add(sock_net(newsk), 1); 2506 } else { 2507 /* Kernel sockets are not elevating the struct net refcount. 2508 * Instead, use a tracker to more easily detect if a layer 2509 * is not properly dismantling its kernel sockets at netns 2510 * destroy time. 2511 */ 2512 net_passive_inc(sock_net(newsk)); 2513 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2514 false, priority); 2515 } 2516 2517 sk_node_init(&newsk->sk_node); 2518 sock_lock_init(newsk); 2519 2520 if (lock) 2521 bh_lock_sock(newsk); 2522 2523 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2524 newsk->sk_backlog.len = 0; 2525 2526 atomic_set(&newsk->sk_rmem_alloc, 0); 2527 2528 refcount_set(&newsk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS); 2529 2530 atomic_set(&newsk->sk_omem_alloc, 0); 2531 sk_init_common(newsk); 2532 2533 newsk->sk_dst_cache = NULL; 2534 newsk->sk_dst_pending_confirm = 0; 2535 newsk->sk_wmem_queued = 0; 2536 newsk->sk_forward_alloc = 0; 2537 newsk->sk_reserved_mem = 0; 2538 DEBUG_NET_WARN_ON_ONCE(newsk->sk_drop_counters); 2539 sk_drops_reset(newsk); 2540 newsk->sk_send_head = NULL; 2541 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2542 atomic_set(&newsk->sk_zckey, 0); 2543 2544 sock_reset_flag(newsk, SOCK_DONE); 2545 2546 #ifdef CONFIG_MEMCG 2547 /* sk->sk_memcg will be populated at accept() time */ 2548 newsk->sk_memcg = NULL; 2549 #endif 2550 2551 cgroup_sk_clone(&newsk->sk_cgrp_data); 2552 2553 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2554 2555 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2556 net_enable_timestamp(); 2557 2558 rcu_read_lock(); 2559 filter = rcu_dereference(sk->sk_filter); 2560 if (filter != NULL) 2561 /* though it's an empty new sock, the charging may fail 2562 * if sysctl_optmem_max was changed between creation of 2563 * original socket and cloning 2564 */ 2565 is_charged = sk_filter_charge(newsk, filter); 2566 RCU_INIT_POINTER(newsk->sk_filter, filter); 2567 rcu_read_unlock(); 2568 2569 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2570 /* We need to make sure that we don't uncharge the new 2571 * socket if we couldn't charge it in the first place 2572 * as otherwise we uncharge the parent's filter. 2573 */ 2574 if (!is_charged) 2575 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2576 2577 goto free; 2578 } 2579 2580 if (bpf_sk_storage_clone(sk, newsk)) 2581 goto free; 2582 2583 /* Clear sk_user_data if parent had the pointer tagged 2584 * as not suitable for copying when cloning. 2585 */ 2586 if (sk_user_data_is_nocopy(newsk)) 2587 newsk->sk_user_data = NULL; 2588 2589 newsk->sk_err = 0; 2590 newsk->sk_err_soft = 0; 2591 newsk->sk_priority = 0; 2592 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2593 2594 /* Before updating sk_refcnt, we must commit prior changes to memory 2595 * (Documentation/RCU/rculist_nulls.rst for details) 2596 */ 2597 smp_wmb(); 2598 refcount_set(&newsk->sk_refcnt, 2); 2599 2600 sk_set_socket(newsk, NULL); 2601 sk_tx_queue_clear(newsk); 2602 sk_rx_queue_clear(newsk); 2603 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2604 2605 if (newsk->sk_prot->sockets_allocated) 2606 sk_sockets_allocated_inc(newsk); 2607 out: 2608 return newsk; 2609 free: 2610 /* It is still raw copy of parent, so invalidate 2611 * destructor and make plain sk_free() 2612 */ 2613 newsk->sk_destruct = NULL; 2614 if (lock) 2615 bh_unlock_sock(newsk); 2616 sk_free(newsk); 2617 newsk = NULL; 2618 goto out; 2619 } 2620 EXPORT_SYMBOL_GPL(sk_clone); 2621 2622 static u32 sk_dst_gso_max_size(struct sock *sk, const struct net_device *dev) 2623 { 2624 bool is_ipv6 = false; 2625 u32 max_size; 2626 2627 #if IS_ENABLED(CONFIG_IPV6) 2628 is_ipv6 = (sk->sk_family == AF_INET6 && 2629 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2630 #endif 2631 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2632 max_size = is_ipv6 ? READ_ONCE(dev->gso_max_size) : 2633 READ_ONCE(dev->gso_ipv4_max_size); 2634 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2635 max_size = GSO_LEGACY_MAX_SIZE; 2636 2637 return max_size - (MAX_TCP_HEADER + 1); 2638 } 2639 2640 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2641 { 2642 const struct net_device *dev; 2643 u32 max_segs = 1; 2644 2645 rcu_read_lock(); 2646 dev = dst_dev_rcu(dst); 2647 sk->sk_route_caps = dev->features; 2648 if (sk_is_tcp(sk)) { 2649 struct inet_connection_sock *icsk = inet_csk(sk); 2650 2651 sk->sk_route_caps |= NETIF_F_GSO; 2652 icsk->icsk_ack.dst_quick_ack = dst_metric(dst, RTAX_QUICKACK); 2653 } 2654 if (sk->sk_route_caps & NETIF_F_GSO) 2655 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2656 if (unlikely(sk->sk_gso_disabled)) 2657 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2658 if (sk_can_gso(sk)) { 2659 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2660 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2661 } else { 2662 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2663 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dev); 2664 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2665 max_segs = max_t(u32, READ_ONCE(dev->gso_max_segs), 1); 2666 } 2667 } 2668 sk->sk_gso_max_segs = max_segs; 2669 sk_dst_set(sk, dst); 2670 rcu_read_unlock(); 2671 } 2672 EXPORT_SYMBOL_GPL(sk_setup_caps); 2673 2674 bool sk_has_decrypt_user(const struct sock *sk) 2675 { 2676 return psp_sk_assoc(sk) || 2677 (sk_is_inet(sk) && inet_csk_has_ulp(sk)); /* for tls */ 2678 } 2679 2680 /* 2681 * Simple resource managers for sockets. 2682 */ 2683 2684 2685 /* 2686 * Write buffer destructor automatically called from kfree_skb. 2687 */ 2688 void sock_wfree(struct sk_buff *skb) 2689 { 2690 unsigned int len = skb->truesize; 2691 struct sock *sk = skb->sk; 2692 bool free; 2693 int old; 2694 2695 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2696 void (*sk_write_space)(struct sock *sk); 2697 2698 sk_write_space = READ_ONCE(sk->sk_write_space); 2699 2700 if (sock_flag(sk, SOCK_RCU_FREE) && 2701 sk_write_space == sock_def_write_space) { 2702 rcu_read_lock(); 2703 free = __refcount_sub_and_test(len, &sk->sk_wmem_alloc, 2704 &old); 2705 sock_def_write_space_wfree(sk, old - len); 2706 rcu_read_unlock(); 2707 if (unlikely(free)) 2708 __sk_free(sk); 2709 return; 2710 } 2711 2712 /* 2713 * Keep a reference on sk_wmem_alloc, this will be released 2714 * after sk_write_space() call 2715 */ 2716 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2717 sk_write_space(sk); 2718 len = 1; 2719 } 2720 /* 2721 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2722 * could not do because of in-flight packets 2723 */ 2724 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2725 __sk_free(sk); 2726 } 2727 EXPORT_SYMBOL(sock_wfree); 2728 2729 /* This variant of sock_wfree() is used by TCP, 2730 * since it sets SOCK_USE_WRITE_QUEUE. 2731 */ 2732 #ifdef CONFIG_INET 2733 void __sock_wfree(struct sk_buff *skb) 2734 { 2735 struct sock *sk = skb->sk; 2736 2737 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2738 __sk_free(sk); 2739 } 2740 EXPORT_SYMBOL_GPL(__sock_wfree); 2741 #endif 2742 2743 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2744 { 2745 int old_wmem; 2746 2747 skb_orphan(skb); 2748 #ifdef CONFIG_INET 2749 if (unlikely(!sk_fullsock(sk))) 2750 return skb_set_owner_edemux(skb, sk); 2751 #endif 2752 skb->sk = sk; 2753 skb->destructor = sock_wfree; 2754 skb_set_hash_from_sk(skb, sk); 2755 /* 2756 * We used to take a refcount on sk, but following operation 2757 * is enough to guarantee sk_free() won't free this sock until 2758 * all in-flight packets are completed 2759 */ 2760 __refcount_add(skb->truesize, &sk->sk_wmem_alloc, &old_wmem); 2761 2762 /* (old_wmem == SK_WMEM_ALLOC_BIAS) if no other TX packet for this socket 2763 * is in a host queue (qdisc, NIC queue). 2764 * Set skb->ooo_okay so that netdev_pick_tx() can choose a TX queue 2765 * based on XPS for better performance. 2766 * Otherwise clear ooo_okay to not risk Out Of Order delivery. 2767 */ 2768 skb->ooo_okay = (old_wmem == SK_WMEM_ALLOC_BIAS); 2769 } 2770 EXPORT_SYMBOL(skb_set_owner_w); 2771 2772 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2773 { 2774 /* Drivers depend on in-order delivery for crypto offload, 2775 * partial orphan breaks out-of-order-OK logic. 2776 */ 2777 if (skb_is_decrypted(skb)) 2778 return false; 2779 2780 return (skb->destructor == sock_wfree || 2781 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2782 } 2783 2784 /* This helper is used by netem, as it can hold packets in its 2785 * delay queue. We want to allow the owner socket to send more 2786 * packets, as if they were already TX completed by a typical driver. 2787 * But we also want to keep skb->sk set because some packet schedulers 2788 * rely on it (sch_fq for example). 2789 */ 2790 void skb_orphan_partial(struct sk_buff *skb) 2791 { 2792 if (skb_is_tcp_pure_ack(skb)) 2793 return; 2794 2795 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2796 return; 2797 2798 skb_orphan(skb); 2799 } 2800 EXPORT_SYMBOL(skb_orphan_partial); 2801 2802 /* 2803 * Read buffer destructor automatically called from kfree_skb. 2804 */ 2805 void sock_rfree(struct sk_buff *skb) 2806 { 2807 struct sock *sk = skb->sk; 2808 unsigned int len = skb->truesize; 2809 2810 atomic_sub(len, &sk->sk_rmem_alloc); 2811 sk_mem_uncharge(sk, len); 2812 } 2813 EXPORT_SYMBOL(sock_rfree); 2814 2815 /* 2816 * Buffer destructor for skbs that are not used directly in read or write 2817 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2818 */ 2819 void sock_efree(struct sk_buff *skb) 2820 { 2821 sock_put(skb->sk); 2822 } 2823 EXPORT_SYMBOL(sock_efree); 2824 2825 /* Buffer destructor for prefetch/receive path where reference count may 2826 * not be held, e.g. for listen sockets. 2827 */ 2828 #ifdef CONFIG_INET 2829 void sock_pfree(struct sk_buff *skb) 2830 { 2831 struct sock *sk = skb->sk; 2832 2833 if (!sk_is_refcounted(sk)) 2834 return; 2835 2836 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2837 inet_reqsk(sk)->rsk_listener = NULL; 2838 reqsk_free(inet_reqsk(sk)); 2839 return; 2840 } 2841 2842 sock_gen_put(sk); 2843 } 2844 EXPORT_SYMBOL(sock_pfree); 2845 #endif /* CONFIG_INET */ 2846 2847 /* 2848 * Allocate a skb from the socket's send buffer. 2849 */ 2850 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2851 gfp_t priority) 2852 { 2853 if (force || 2854 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2855 struct sk_buff *skb = alloc_skb(size, priority); 2856 2857 if (skb) { 2858 skb_set_owner_w(skb, sk); 2859 return skb; 2860 } 2861 } 2862 return NULL; 2863 } 2864 EXPORT_SYMBOL(sock_wmalloc); 2865 2866 static void sock_ofree(struct sk_buff *skb) 2867 { 2868 struct sock *sk = skb->sk; 2869 2870 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2871 } 2872 2873 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2874 gfp_t priority) 2875 { 2876 struct sk_buff *skb; 2877 2878 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2879 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2880 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2881 return NULL; 2882 2883 skb = alloc_skb(size, priority); 2884 if (!skb) 2885 return NULL; 2886 2887 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2888 skb->sk = sk; 2889 skb->destructor = sock_ofree; 2890 return skb; 2891 } 2892 2893 /* 2894 * Allocate a memory block from the socket's option memory buffer. 2895 */ 2896 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2897 { 2898 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2899 2900 if ((unsigned int)size <= optmem_max && 2901 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2902 void *mem; 2903 /* First do the add, to avoid the race if kmalloc 2904 * might sleep. 2905 */ 2906 atomic_add(size, &sk->sk_omem_alloc); 2907 mem = kmalloc(size, priority); 2908 if (mem) 2909 return mem; 2910 atomic_sub(size, &sk->sk_omem_alloc); 2911 } 2912 return NULL; 2913 } 2914 EXPORT_SYMBOL(sock_kmalloc); 2915 2916 /* 2917 * Duplicate the input "src" memory block using the socket's 2918 * option memory buffer. 2919 */ 2920 void *sock_kmemdup(struct sock *sk, const void *src, 2921 int size, gfp_t priority) 2922 { 2923 void *mem; 2924 2925 mem = sock_kmalloc(sk, size, priority); 2926 if (mem) 2927 memcpy(mem, src, size); 2928 return mem; 2929 } 2930 EXPORT_SYMBOL(sock_kmemdup); 2931 2932 /* Free an option memory block. Note, we actually want the inline 2933 * here as this allows gcc to detect the nullify and fold away the 2934 * condition entirely. 2935 */ 2936 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2937 const bool nullify) 2938 { 2939 if (WARN_ON_ONCE(!mem)) 2940 return; 2941 if (nullify) 2942 kfree_sensitive(mem); 2943 else 2944 kfree(mem); 2945 atomic_sub(size, &sk->sk_omem_alloc); 2946 } 2947 2948 void sock_kfree_s(struct sock *sk, void *mem, int size) 2949 { 2950 __sock_kfree_s(sk, mem, size, false); 2951 } 2952 EXPORT_SYMBOL(sock_kfree_s); 2953 2954 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2955 { 2956 __sock_kfree_s(sk, mem, size, true); 2957 } 2958 EXPORT_SYMBOL(sock_kzfree_s); 2959 2960 /** 2961 * sk_set_nospace - tell the transport a writer is waiting for space 2962 * @sk: socket 2963 * 2964 * Must be called before the final check of the available send space, 2965 * so that the transport can not miss the request and forget to call 2966 * sk->sk_write_space() once space is available again. 2967 */ 2968 void sk_set_nospace(struct sock *sk) 2969 { 2970 struct socket *sock = sk->sk_socket; 2971 2972 if (!sock) 2973 return; 2974 /* Set SOCK_NOSPACE before tp->tcp_nospace (paired with 2975 * sk_clear_nospace() clearing tp->tcp_nospace before SOCK_NOSPACE) 2976 * so a concurrent clear cannot leave SOCK_NOSPACE set with 2977 * tp->tcp_nospace cleared. 2978 */ 2979 set_bit(SOCK_NOSPACE, &sock->flags); 2980 tcp_set_nospace(sk); 2981 } 2982 EXPORT_SYMBOL(sk_set_nospace); 2983 2984 /** 2985 * sk_clear_nospace - tell the transport no writer is waiting for space 2986 * @sk: socket 2987 * 2988 * Called from ->sk_write_space() handlers, once send space has been 2989 * made available to writers. 2990 */ 2991 void sk_clear_nospace(struct sock *sk) 2992 { 2993 struct socket *sock = sk->sk_socket; 2994 2995 if (!sock) 2996 return; 2997 tcp_clear_nospace(sk); 2998 clear_bit(SOCK_NOSPACE, &sock->flags); 2999 } 3000 EXPORT_SYMBOL(sk_clear_nospace); 3001 3002 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 3003 I think, these locks should be removed for datagram sockets. 3004 */ 3005 static long sock_wait_for_wmem(struct sock *sk, long timeo) 3006 { 3007 DEFINE_WAIT(wait); 3008 3009 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 3010 for (;;) { 3011 if (!timeo) 3012 break; 3013 if (signal_pending(current)) 3014 break; 3015 sk_set_nospace(sk); 3016 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 3017 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 3018 break; 3019 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 3020 break; 3021 if (READ_ONCE(sk->sk_err)) 3022 break; 3023 timeo = schedule_timeout(timeo); 3024 } 3025 finish_wait(sk_sleep(sk), &wait); 3026 return timeo; 3027 } 3028 3029 3030 /* 3031 * Generic send/receive buffer handlers 3032 */ 3033 3034 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 3035 unsigned long data_len, int noblock, 3036 int *errcode, int max_page_order) 3037 { 3038 struct sk_buff *skb; 3039 long timeo; 3040 int err; 3041 3042 timeo = sock_sndtimeo(sk, noblock); 3043 for (;;) { 3044 err = sock_error(sk); 3045 if (err != 0) 3046 goto failure; 3047 3048 err = -EPIPE; 3049 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 3050 goto failure; 3051 3052 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 3053 break; 3054 3055 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 3056 sk_set_nospace(sk); 3057 err = -EAGAIN; 3058 if (!timeo) 3059 goto failure; 3060 if (signal_pending(current)) 3061 goto interrupted; 3062 timeo = sock_wait_for_wmem(sk, timeo); 3063 } 3064 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 3065 errcode, sk->sk_allocation); 3066 if (skb) 3067 skb_set_owner_w(skb, sk); 3068 return skb; 3069 3070 interrupted: 3071 err = sock_intr_errno(timeo); 3072 failure: 3073 *errcode = err; 3074 return NULL; 3075 } 3076 EXPORT_SYMBOL(sock_alloc_send_pskb); 3077 3078 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 3079 struct sockcm_cookie *sockc) 3080 { 3081 u32 tsflags; 3082 3083 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 3084 3085 switch (cmsg->cmsg_type) { 3086 case SO_MARK: 3087 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 3088 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 3089 return -EPERM; 3090 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3091 return -EINVAL; 3092 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 3093 break; 3094 case SO_TIMESTAMPING_OLD: 3095 case SO_TIMESTAMPING_NEW: 3096 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3097 return -EINVAL; 3098 3099 tsflags = *(u32 *)CMSG_DATA(cmsg); 3100 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 3101 return -EINVAL; 3102 3103 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 3104 sockc->tsflags |= tsflags; 3105 break; 3106 case SCM_TXTIME: 3107 { 3108 ktime_t tmin; 3109 u64 txtime; 3110 3111 if (!sock_flag(sk, SOCK_TXTIME)) 3112 return -EINVAL; 3113 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 3114 return -EINVAL; 3115 3116 txtime = get_unaligned((u64 *)CMSG_DATA(cmsg)); 3117 3118 /* Allow sending without a delivery time: zero special case */ 3119 if (!txtime) { 3120 sockc->transmit_time = 0; 3121 break; 3122 } 3123 3124 switch (sk->sk_clockid) { 3125 case CLOCK_MONOTONIC: 3126 tmin = 1; 3127 break; 3128 case CLOCK_REALTIME: 3129 tmin = max(ktime_mono_to_real(0), 1); 3130 break; 3131 case CLOCK_TAI: 3132 tmin = max(ktime_mono_to_any(0, TK_OFFS_TAI), 1); 3133 break; 3134 default: 3135 tmin = 1; 3136 WARN_ON_ONCE(1); 3137 break; 3138 } 3139 3140 sockc->transmit_time = max_t(ktime_t, txtime, tmin); 3141 break; 3142 } 3143 case SCM_TS_OPT_ID: 3144 if (sk_is_tcp(sk)) 3145 return -EINVAL; 3146 tsflags = READ_ONCE(sk->sk_tsflags); 3147 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 3148 return -EINVAL; 3149 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3150 return -EINVAL; 3151 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 3152 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 3153 break; 3154 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 3155 case SCM_RIGHTS: 3156 case SCM_CREDENTIALS: 3157 break; 3158 case SO_PRIORITY: 3159 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3160 return -EINVAL; 3161 if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg))) 3162 return -EPERM; 3163 sockc->priority = *(u32 *)CMSG_DATA(cmsg); 3164 break; 3165 case SCM_DEVMEM_DMABUF: 3166 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3167 return -EINVAL; 3168 sockc->dmabuf_id = *(u32 *)CMSG_DATA(cmsg); 3169 break; 3170 default: 3171 return -EINVAL; 3172 } 3173 return 0; 3174 } 3175 EXPORT_SYMBOL(__sock_cmsg_send); 3176 3177 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 3178 struct sockcm_cookie *sockc) 3179 { 3180 struct cmsghdr *cmsg; 3181 int ret; 3182 3183 for_each_cmsghdr(cmsg, msg) { 3184 if (!CMSG_OK(msg, cmsg)) 3185 return -EINVAL; 3186 if (cmsg->cmsg_level != SOL_SOCKET) 3187 continue; 3188 ret = __sock_cmsg_send(sk, cmsg, sockc); 3189 if (ret) 3190 return ret; 3191 } 3192 return 0; 3193 } 3194 EXPORT_SYMBOL(sock_cmsg_send); 3195 3196 static void sk_enter_memory_pressure(struct sock *sk) 3197 { 3198 if (!sk->sk_prot->enter_memory_pressure) 3199 return; 3200 3201 sk->sk_prot->enter_memory_pressure(sk); 3202 } 3203 3204 static void sk_leave_memory_pressure(struct sock *sk) 3205 { 3206 if (sk->sk_prot->leave_memory_pressure) { 3207 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 3208 tcp_leave_memory_pressure, sk); 3209 } else { 3210 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 3211 3212 if (memory_pressure && READ_ONCE(*memory_pressure)) 3213 WRITE_ONCE(*memory_pressure, 0); 3214 } 3215 } 3216 3217 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3218 3219 /** 3220 * skb_page_frag_refill - check that a page_frag contains enough room 3221 * @sz: minimum size of the fragment we want to get 3222 * @pfrag: pointer to page_frag 3223 * @gfp: priority for memory allocation 3224 * 3225 * Note: While this allocator tries to use high order pages, there is 3226 * no guarantee that allocations succeed. Therefore, @sz MUST be 3227 * less or equal than PAGE_SIZE. 3228 */ 3229 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 3230 { 3231 if (pfrag->page) { 3232 if (page_ref_count(pfrag->page) == 1) { 3233 pfrag->offset = 0; 3234 return true; 3235 } 3236 if (pfrag->offset + sz <= pfrag->size) 3237 return true; 3238 put_page(pfrag->page); 3239 } 3240 3241 pfrag->offset = 0; 3242 if (SKB_FRAG_PAGE_ORDER && 3243 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3244 /* Avoid direct reclaim but allow kswapd to wake */ 3245 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3246 __GFP_COMP | __GFP_NOWARN | 3247 __GFP_NORETRY, 3248 SKB_FRAG_PAGE_ORDER); 3249 if (likely(pfrag->page)) { 3250 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3251 return true; 3252 } 3253 } 3254 pfrag->page = alloc_page(gfp); 3255 if (likely(pfrag->page)) { 3256 pfrag->size = PAGE_SIZE; 3257 return true; 3258 } 3259 return false; 3260 } 3261 EXPORT_SYMBOL(skb_page_frag_refill); 3262 3263 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3264 { 3265 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3266 return true; 3267 3268 if (!sk->sk_bypass_prot_mem) 3269 sk_enter_memory_pressure(sk); 3270 3271 sk_stream_moderate_sndbuf(sk); 3272 3273 return false; 3274 } 3275 EXPORT_SYMBOL(sk_page_frag_refill); 3276 3277 static void __lock_sock(struct sock *sk) 3278 __releases(&sk->sk_lock.slock) 3279 __acquires(&sk->sk_lock.slock) 3280 { 3281 DEFINE_WAIT(wait); 3282 3283 for (;;) { 3284 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3285 TASK_UNINTERRUPTIBLE); 3286 spin_unlock_bh(&sk->sk_lock.slock); 3287 schedule(); 3288 spin_lock_bh(&sk->sk_lock.slock); 3289 if (!sock_owned_by_user(sk)) 3290 break; 3291 } 3292 finish_wait(&sk->sk_lock.wq, &wait); 3293 } 3294 3295 void __release_sock(struct sock *sk) 3296 __releases(&sk->sk_lock.slock) 3297 __acquires(&sk->sk_lock.slock) 3298 { 3299 struct sk_buff *skb, *next; 3300 int nb = 0; 3301 3302 while ((skb = sk->sk_backlog.head) != NULL) { 3303 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3304 3305 spin_unlock_bh(&sk->sk_lock.slock); 3306 3307 while (1) { 3308 next = skb->next; 3309 prefetch(next); 3310 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3311 skb_mark_not_on_list(skb); 3312 sk_backlog_rcv(sk, skb); 3313 3314 skb = next; 3315 if (!skb) 3316 break; 3317 3318 if (!(++nb & 15)) 3319 cond_resched(); 3320 } 3321 3322 spin_lock_bh(&sk->sk_lock.slock); 3323 } 3324 3325 /* 3326 * Doing the zeroing here guarantee we can not loop forever 3327 * while a wild producer attempts to flood us. 3328 */ 3329 sk->sk_backlog.len = 0; 3330 } 3331 3332 void __sk_flush_backlog(struct sock *sk) 3333 { 3334 spin_lock_bh(&sk->sk_lock.slock); 3335 __release_sock(sk); 3336 3337 if (sk->sk_prot->release_cb) 3338 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3339 tcp_release_cb, sk); 3340 3341 spin_unlock_bh(&sk->sk_lock.slock); 3342 } 3343 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3344 3345 /** 3346 * sk_wait_data - wait for data to arrive at sk_receive_queue 3347 * @sk: sock to wait on 3348 * @timeo: for how long 3349 * @skb: last skb seen on sk_receive_queue 3350 * 3351 * Now socket state including sk->sk_err is changed only under lock, 3352 * hence we may omit checks after joining wait queue. 3353 * We check receive queue before schedule() only as optimization; 3354 * it is very likely that release_sock() added new data. 3355 */ 3356 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3357 { 3358 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3359 int rc; 3360 3361 add_wait_queue(sk_sleep(sk), &wait); 3362 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3363 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3364 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3365 remove_wait_queue(sk_sleep(sk), &wait); 3366 return rc; 3367 } 3368 EXPORT_SYMBOL(sk_wait_data); 3369 3370 /** 3371 * __sk_mem_raise_allocated - increase memory_allocated 3372 * @sk: socket 3373 * @size: memory size to allocate 3374 * @amt: pages to allocate 3375 * @kind: allocation type 3376 * 3377 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3378 * 3379 * Unlike the globally shared limits among the sockets under same protocol, 3380 * consuming the budget of a memcg won't have direct effect on other ones. 3381 * So be optimistic about memcg's tolerance, and leave the callers to decide 3382 * whether or not to raise allocated through sk_under_memory_pressure() or 3383 * its variants. 3384 */ 3385 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3386 { 3387 bool memcg_enabled = false, charged = false; 3388 struct proto *prot = sk->sk_prot; 3389 long allocated = 0; 3390 3391 if (!sk->sk_bypass_prot_mem) { 3392 sk_memory_allocated_add(sk, amt); 3393 allocated = sk_memory_allocated(sk); 3394 } 3395 3396 if (mem_cgroup_sk_enabled(sk)) { 3397 memcg_enabled = true; 3398 charged = mem_cgroup_sk_charge(sk, amt, gfp_memcg_charge()); 3399 if (!charged) 3400 goto suppress_allocation; 3401 } 3402 3403 if (!allocated) 3404 return 1; 3405 3406 /* Under limit. */ 3407 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3408 sk_leave_memory_pressure(sk); 3409 return 1; 3410 } 3411 3412 /* Under pressure. */ 3413 if (allocated > sk_prot_mem_limits(sk, 1)) 3414 sk_enter_memory_pressure(sk); 3415 3416 /* Over hard limit. */ 3417 if (allocated > sk_prot_mem_limits(sk, 2)) 3418 goto suppress_allocation; 3419 3420 /* Guarantee minimum buffer size under pressure (either global 3421 * or memcg) to make sure features described in RFC 7323 (TCP 3422 * Extensions for High Performance) work properly. 3423 * 3424 * This rule does NOT stand when exceeds global or memcg's hard 3425 * limit, or else a DoS attack can be taken place by spawning 3426 * lots of sockets whose usage are under minimum buffer size. 3427 */ 3428 if (kind == SK_MEM_RECV) { 3429 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3430 return 1; 3431 3432 } else { /* SK_MEM_SEND */ 3433 int wmem0 = sk_get_wmem0(sk, prot); 3434 3435 if (sk->sk_type == SOCK_STREAM) { 3436 if (sk->sk_wmem_queued < wmem0) 3437 return 1; 3438 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3439 return 1; 3440 } 3441 } 3442 3443 if (sk_has_memory_pressure(sk)) { 3444 u64 alloc; 3445 3446 /* The following 'average' heuristic is within the 3447 * scope of global accounting, so it only makes 3448 * sense for global memory pressure. 3449 */ 3450 if (!sk_under_global_memory_pressure(sk)) 3451 return 1; 3452 3453 /* Try to be fair among all the sockets under global 3454 * pressure by allowing the ones that below average 3455 * usage to raise. 3456 */ 3457 alloc = sk_sockets_allocated_read_positive(sk); 3458 if (sk_prot_mem_limits(sk, 2) > alloc * 3459 sk_mem_pages(sk->sk_wmem_queued + 3460 atomic_read(&sk->sk_rmem_alloc) + 3461 sk->sk_forward_alloc)) 3462 return 1; 3463 } 3464 3465 suppress_allocation: 3466 3467 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3468 sk_stream_moderate_sndbuf(sk); 3469 3470 /* Fail only if socket is _under_ its sndbuf. 3471 * In this case we cannot block, so that we have to fail. 3472 */ 3473 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3474 /* Force charge with __GFP_NOFAIL */ 3475 if (memcg_enabled && !charged) 3476 mem_cgroup_sk_charge(sk, amt, 3477 gfp_memcg_charge() | __GFP_NOFAIL); 3478 return 1; 3479 } 3480 } 3481 3482 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3483 3484 if (allocated) 3485 sk_memory_allocated_sub(sk, amt); 3486 3487 if (charged) 3488 mem_cgroup_sk_uncharge(sk, amt); 3489 3490 return 0; 3491 } 3492 3493 /** 3494 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3495 * @sk: socket 3496 * @size: memory size to allocate 3497 * @kind: allocation type 3498 * 3499 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3500 * rmem allocation. This function assumes that protocols which have 3501 * memory_pressure use sk_wmem_queued as write buffer accounting. 3502 */ 3503 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3504 { 3505 int ret, amt = sk_mem_pages(size); 3506 3507 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3508 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3509 if (!ret) 3510 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3511 return ret; 3512 } 3513 EXPORT_SYMBOL(__sk_mem_schedule); 3514 3515 /** 3516 * __sk_mem_reduce_allocated - reclaim memory_allocated 3517 * @sk: socket 3518 * @amount: number of quanta 3519 * 3520 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3521 */ 3522 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3523 { 3524 if (mem_cgroup_sk_enabled(sk)) 3525 mem_cgroup_sk_uncharge(sk, amount); 3526 3527 if (sk->sk_bypass_prot_mem) 3528 return; 3529 3530 sk_memory_allocated_sub(sk, amount); 3531 3532 if (sk_under_global_memory_pressure(sk) && 3533 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3534 sk_leave_memory_pressure(sk); 3535 } 3536 3537 /** 3538 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3539 * @sk: socket 3540 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3541 */ 3542 void __sk_mem_reclaim(struct sock *sk, int amount) 3543 { 3544 amount >>= PAGE_SHIFT; 3545 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3546 __sk_mem_reduce_allocated(sk, amount); 3547 } 3548 EXPORT_SYMBOL(__sk_mem_reclaim); 3549 3550 void __sk_charge(struct sock *sk, gfp_t gfp) 3551 { 3552 int amt; 3553 3554 gfp |= __GFP_NOFAIL; 3555 if (mem_cgroup_from_sk(sk)) { 3556 /* The socket has not been accepted yet, no need 3557 * to look at newsk->sk_wmem_queued. 3558 */ 3559 amt = sk_mem_pages(sk->sk_forward_alloc + 3560 atomic_read(&sk->sk_rmem_alloc)); 3561 if (amt) 3562 mem_cgroup_sk_charge(sk, amt, gfp); 3563 } 3564 3565 kmem_cache_charge(sk, gfp); 3566 } 3567 3568 int sk_set_peek_off(struct sock *sk, int val) 3569 { 3570 WRITE_ONCE(sk->sk_peek_off, val); 3571 return 0; 3572 } 3573 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3574 3575 /* 3576 * Set of default routines for initialising struct proto_ops when 3577 * the protocol does not support a particular function. In certain 3578 * cases where it makes no sense for a protocol to have a "do nothing" 3579 * function, some default processing is provided. 3580 */ 3581 3582 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len) 3583 { 3584 return -EOPNOTSUPP; 3585 } 3586 EXPORT_SYMBOL(sock_no_bind); 3587 3588 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr, 3589 int len, int flags) 3590 { 3591 return -EOPNOTSUPP; 3592 } 3593 EXPORT_SYMBOL(sock_no_connect); 3594 3595 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3596 { 3597 return -EOPNOTSUPP; 3598 } 3599 EXPORT_SYMBOL(sock_no_socketpair); 3600 3601 int sock_no_accept(struct socket *sock, struct socket *newsock, 3602 struct proto_accept_arg *arg) 3603 { 3604 return -EOPNOTSUPP; 3605 } 3606 EXPORT_SYMBOL(sock_no_accept); 3607 3608 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3609 int peer) 3610 { 3611 return -EOPNOTSUPP; 3612 } 3613 EXPORT_SYMBOL(sock_no_getname); 3614 3615 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3616 { 3617 return -EOPNOTSUPP; 3618 } 3619 EXPORT_SYMBOL(sock_no_ioctl); 3620 3621 int sock_no_listen(struct socket *sock, int backlog) 3622 { 3623 return -EOPNOTSUPP; 3624 } 3625 EXPORT_SYMBOL(sock_no_listen); 3626 3627 int sock_no_shutdown(struct socket *sock, int how) 3628 { 3629 return -EOPNOTSUPP; 3630 } 3631 EXPORT_SYMBOL(sock_no_shutdown); 3632 3633 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3634 { 3635 return -EOPNOTSUPP; 3636 } 3637 EXPORT_SYMBOL(sock_no_sendmsg); 3638 3639 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3640 { 3641 return -EOPNOTSUPP; 3642 } 3643 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3644 3645 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3646 int flags) 3647 { 3648 return -EOPNOTSUPP; 3649 } 3650 EXPORT_SYMBOL(sock_no_recvmsg); 3651 3652 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3653 { 3654 /* Mirror missing mmap method error code */ 3655 return -ENODEV; 3656 } 3657 EXPORT_SYMBOL(sock_no_mmap); 3658 3659 /* 3660 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3661 * various sock-based usage counts. 3662 */ 3663 void __receive_sock(struct file *file) 3664 { 3665 struct socket *sock; 3666 3667 sock = sock_from_file(file); 3668 if (sock) { 3669 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3670 sock_update_classid(&sock->sk->sk_cgrp_data); 3671 } 3672 } 3673 3674 /* 3675 * Default Socket Callbacks 3676 */ 3677 3678 static void sock_def_wakeup(struct sock *sk) 3679 { 3680 struct socket_wq *wq; 3681 3682 rcu_read_lock(); 3683 wq = rcu_dereference(sk->sk_wq); 3684 if (skwq_has_sleeper(wq)) 3685 wake_up_interruptible_all(&wq->wait); 3686 rcu_read_unlock(); 3687 } 3688 3689 static void sock_def_error_report(struct sock *sk) 3690 { 3691 struct socket_wq *wq; 3692 3693 rcu_read_lock(); 3694 wq = rcu_dereference(sk->sk_wq); 3695 if (skwq_has_sleeper(wq)) 3696 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3697 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3698 rcu_read_unlock(); 3699 } 3700 3701 void sock_def_readable(struct sock *sk) 3702 { 3703 struct socket_wq *wq; 3704 3705 trace_sk_data_ready(sk); 3706 3707 rcu_read_lock(); 3708 wq = rcu_dereference(sk->sk_wq); 3709 if (skwq_has_sleeper(wq)) 3710 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3711 EPOLLRDNORM | EPOLLRDBAND); 3712 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3713 rcu_read_unlock(); 3714 } 3715 3716 static void sock_def_write_space(struct sock *sk) 3717 { 3718 struct socket_wq *wq; 3719 3720 rcu_read_lock(); 3721 3722 /* Do not wake up a writer until he can make "significant" 3723 * progress. --DaveM 3724 */ 3725 if (sock_writeable(sk)) { 3726 wq = rcu_dereference(sk->sk_wq); 3727 if (skwq_has_sleeper(wq)) 3728 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3729 EPOLLWRNORM | EPOLLWRBAND); 3730 3731 /* Should agree with poll, otherwise some programs break */ 3732 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3733 } 3734 3735 rcu_read_unlock(); 3736 } 3737 3738 /* An optimised version of sock_def_write_space(), should only be called 3739 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3740 * ->sk_wmem_alloc. 3741 */ 3742 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc) 3743 { 3744 /* Do not wake up a writer until he can make "significant" 3745 * progress. --DaveM 3746 */ 3747 if (__sock_writeable(sk, wmem_alloc)) { 3748 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3749 3750 /* rely on refcount_sub from sock_wfree() */ 3751 smp_mb__after_atomic(); 3752 if (wq && waitqueue_active(&wq->wait)) 3753 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3754 EPOLLWRNORM | EPOLLWRBAND); 3755 3756 /* Should agree with poll, otherwise some programs break */ 3757 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3758 } 3759 } 3760 3761 static void sock_def_destruct(struct sock *sk) 3762 { 3763 } 3764 3765 void sk_send_sigurg(struct sock *sk) 3766 { 3767 if (sk->sk_socket && sk->sk_socket->file) 3768 if (send_sigurg(sk->sk_socket->file)) 3769 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3770 } 3771 EXPORT_SYMBOL(sk_send_sigurg); 3772 3773 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3774 unsigned long expires) 3775 { 3776 if (!mod_timer(timer, expires)) 3777 sock_hold(sk); 3778 } 3779 EXPORT_SYMBOL(sk_reset_timer); 3780 3781 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3782 { 3783 if (timer_delete(timer)) 3784 __sock_put(sk); 3785 } 3786 EXPORT_SYMBOL(sk_stop_timer); 3787 3788 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3789 { 3790 if (timer_delete_sync(timer)) 3791 __sock_put(sk); 3792 } 3793 EXPORT_SYMBOL(sk_stop_timer_sync); 3794 3795 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3796 { 3797 sk_init_common(sk); 3798 sk->sk_send_head = NULL; 3799 3800 timer_setup(&sk->sk_timer, NULL, 0); 3801 3802 sk->sk_allocation = GFP_KERNEL; 3803 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3804 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3805 sk->sk_state = TCP_CLOSE; 3806 sk->sk_use_task_frag = true; 3807 sk_set_socket(sk, sock); 3808 3809 sock_set_flag(sk, SOCK_ZAPPED); 3810 3811 if (sock) { 3812 sk->sk_type = sock->type; 3813 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3814 sock->sk = sk; 3815 } else { 3816 RCU_INIT_POINTER(sk->sk_wq, NULL); 3817 } 3818 sk->sk_uid = uid; 3819 3820 sk->sk_state_change = sock_def_wakeup; 3821 sk->sk_data_ready = sock_def_readable; 3822 sk->sk_write_space = sock_def_write_space; 3823 sk->sk_error_report = sock_def_error_report; 3824 sk->sk_destruct = sock_def_destruct; 3825 3826 sk->sk_frag.page = NULL; 3827 sk->sk_frag.offset = 0; 3828 sk->sk_peek_off = -1; 3829 3830 sk->sk_peer_pid = NULL; 3831 sk->sk_peer_cred = NULL; 3832 spin_lock_init(&sk->sk_peer_lock); 3833 3834 sk->sk_write_pending = 0; 3835 sk->sk_rcvlowat = 1; 3836 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3837 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3838 3839 sk->sk_stamp = SK_DEFAULT_STAMP; 3840 #if BITS_PER_LONG==32 3841 seqlock_init(&sk->sk_stamp_seq); 3842 #endif 3843 atomic_set(&sk->sk_zckey, 0); 3844 3845 #ifdef CONFIG_NET_RX_BUSY_POLL 3846 sk->sk_napi_id = 0; 3847 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3848 #endif 3849 3850 sk->sk_max_pacing_rate = ~0UL; 3851 sk->sk_pacing_rate = ~0UL; 3852 WRITE_ONCE(sk->sk_pacing_shift, 10); 3853 sk->sk_incoming_cpu = -1; 3854 3855 sk_rx_queue_clear(sk); 3856 /* 3857 * Before updating sk_refcnt, we must commit prior changes to memory 3858 * (Documentation/RCU/rculist_nulls.rst for details) 3859 */ 3860 smp_wmb(); 3861 refcount_set(&sk->sk_refcnt, 1); 3862 sk_drops_reset(sk); 3863 } 3864 EXPORT_SYMBOL(sock_init_data_uid); 3865 3866 void sock_init_data(struct socket *sock, struct sock *sk) 3867 { 3868 kuid_t uid = sock ? 3869 SOCK_INODE(sock)->i_uid : 3870 make_kuid(sock_net(sk)->user_ns, 0); 3871 3872 sock_init_data_uid(sock, sk, uid); 3873 } 3874 EXPORT_SYMBOL(sock_init_data); 3875 3876 void noinline lock_sock_nested(struct sock *sk, int subclass) 3877 { 3878 /* The sk_lock has mutex_lock() semantics here. */ 3879 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3880 3881 might_sleep(); 3882 #ifdef CONFIG_64BIT 3883 if (sizeof(struct slock_owned) == sizeof(long)) { 3884 socket_lock_t tmp = { 3885 .slock = __SPIN_LOCK_UNLOCKED(tmp.slock), 3886 .owned = 1, 3887 }; 3888 socket_lock_t old = { 3889 .slock = __SPIN_LOCK_UNLOCKED(old.slock), 3890 .owned = 0, 3891 }; 3892 3893 if (likely(try_cmpxchg(&sk->sk_lock.combined, 3894 &old.combined, tmp.combined))) 3895 return; 3896 } 3897 #endif 3898 spin_lock_bh(&sk->sk_lock.slock); 3899 if (unlikely(sock_owned_by_user_nocheck(sk))) 3900 __lock_sock(sk); 3901 sk->sk_lock.owned = 1; 3902 spin_unlock_bh(&sk->sk_lock.slock); 3903 } 3904 EXPORT_SYMBOL(lock_sock_nested); 3905 3906 void release_sock(struct sock *sk) 3907 { 3908 spin_lock_bh(&sk->sk_lock.slock); 3909 3910 if (unlikely(sk->sk_backlog.tail)) 3911 __release_sock(sk); 3912 3913 if (sk->sk_prot->release_cb) { 3914 if (!tcp_release_cb_cond(sk)) 3915 sk->sk_prot->release_cb(sk); 3916 } 3917 sock_release_ownership(sk); 3918 if (unlikely(waitqueue_active(&sk->sk_lock.wq))) 3919 wake_up(&sk->sk_lock.wq); 3920 3921 spin_unlock_bh(&sk->sk_lock.slock); 3922 } 3923 EXPORT_SYMBOL(release_sock); 3924 3925 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3926 { 3927 might_sleep(); 3928 spin_lock_bh(&sk->sk_lock.slock); 3929 3930 if (likely(!sock_owned_by_user_nocheck(sk))) { 3931 /* 3932 * Fast path return with bottom halves disabled and 3933 * sock::sk_lock.slock held. 3934 * 3935 * The 'mutex' is not contended and holding 3936 * sock::sk_lock.slock prevents all other lockers to 3937 * proceed so the corresponding unlock_sock_fast() can 3938 * avoid the slow path of release_sock() completely and 3939 * just release slock. 3940 * 3941 * From a semantical POV this is equivalent to 'acquiring' 3942 * the 'mutex', hence the corresponding lockdep 3943 * mutex_release() has to happen in the fast path of 3944 * unlock_sock_fast(). 3945 */ 3946 return false; 3947 } 3948 3949 __lock_sock(sk); 3950 sk->sk_lock.owned = 1; 3951 __acquire(&sk->sk_lock.slock); 3952 spin_unlock_bh(&sk->sk_lock.slock); 3953 return true; 3954 } 3955 EXPORT_SYMBOL(__lock_sock_fast); 3956 3957 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3958 bool timeval, bool time32) 3959 { 3960 struct sock *sk = sock->sk; 3961 struct timespec64 ts; 3962 3963 /* sk->sk_flags must only be changed under the socket lock, 3964 * because sock_set_flag() uses non atomic operations. 3965 */ 3966 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 3967 lock_sock(sk); 3968 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3969 release_sock(sk); 3970 } 3971 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3972 if (ts.tv_sec == -1) 3973 return -ENOENT; 3974 if (ts.tv_sec == 0) { 3975 ktime_t kt = ktime_get_real(); 3976 sock_write_timestamp(sk, kt); 3977 ts = ktime_to_timespec64(kt); 3978 } 3979 3980 if (timeval) 3981 ts.tv_nsec /= 1000; 3982 3983 #ifdef CONFIG_COMPAT_32BIT_TIME 3984 if (time32) 3985 return put_old_timespec32(&ts, userstamp); 3986 #endif 3987 #ifdef CONFIG_SPARC64 3988 /* beware of padding in sparc64 timeval */ 3989 if (timeval && !in_compat_syscall()) { 3990 struct __kernel_old_timeval __user tv = { 3991 .tv_sec = ts.tv_sec, 3992 .tv_usec = ts.tv_nsec, 3993 }; 3994 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3995 return -EFAULT; 3996 return 0; 3997 } 3998 #endif 3999 return put_timespec64(&ts, userstamp); 4000 } 4001 EXPORT_SYMBOL(sock_gettstamp); 4002 4003 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 4004 { 4005 if (!sock_flag(sk, flag)) { 4006 unsigned long previous_flags = sk->sk_flags; 4007 4008 sock_set_flag(sk, flag); 4009 /* 4010 * we just set one of the two flags which require net 4011 * time stamping, but time stamping might have been on 4012 * already because of the other one 4013 */ 4014 if (sock_needs_netstamp(sk) && 4015 !(previous_flags & SK_FLAGS_TIMESTAMP)) 4016 net_enable_timestamp(); 4017 } 4018 } 4019 4020 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 4021 int level, int type) 4022 { 4023 struct sock_extended_err ee; 4024 struct sk_buff *skb; 4025 int copied, err; 4026 4027 err = -EAGAIN; 4028 skb = sock_dequeue_err_skb(sk); 4029 if (skb == NULL) 4030 goto out; 4031 4032 copied = skb->len; 4033 if (copied > len) { 4034 msg->msg_flags |= MSG_TRUNC; 4035 copied = len; 4036 } 4037 err = skb_copy_datagram_msg(skb, 0, msg, copied); 4038 if (err) 4039 goto out_free_skb; 4040 4041 sock_recv_timestamp(msg, sk, skb); 4042 4043 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */ 4044 ee = SKB_EXT_ERR(skb)->ee; 4045 put_cmsg(msg, level, type, sizeof(ee), &ee); 4046 4047 msg->msg_flags |= MSG_ERRQUEUE; 4048 err = copied; 4049 4050 out_free_skb: 4051 kfree_skb(skb); 4052 out: 4053 return err; 4054 } 4055 EXPORT_SYMBOL(sock_recv_errqueue); 4056 4057 /* 4058 * Get a socket option on an socket. 4059 * 4060 * FIX: POSIX 1003.1g is very ambiguous here. It states that 4061 * asynchronous errors should be reported by getsockopt. We assume 4062 * this means if you specify SO_ERROR (otherwise what is the point of it). 4063 */ 4064 int sock_common_getsockopt(struct socket *sock, int level, int optname, 4065 char __user *optval, int __user *optlen) 4066 { 4067 struct sock *sk = sock->sk; 4068 4069 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 4070 } 4071 EXPORT_SYMBOL(sock_common_getsockopt); 4072 4073 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 4074 int flags) 4075 { 4076 struct sock *sk = sock->sk; 4077 4078 return sk->sk_prot->recvmsg(sk, msg, size, flags); 4079 } 4080 EXPORT_SYMBOL(sock_common_recvmsg); 4081 4082 /* 4083 * Set socket options on an inet socket. 4084 */ 4085 int sock_common_setsockopt(struct socket *sock, int level, int optname, 4086 sockptr_t optval, unsigned int optlen) 4087 { 4088 struct sock *sk = sock->sk; 4089 4090 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 4091 } 4092 EXPORT_SYMBOL(sock_common_setsockopt); 4093 4094 void sk_common_release(struct sock *sk) 4095 { 4096 if (sk->sk_prot->destroy) 4097 sk->sk_prot->destroy(sk); 4098 4099 /* 4100 * Observation: when sk_common_release is called, processes have 4101 * no access to socket. But net still has. 4102 * Step one, detach it from networking: 4103 * 4104 * A. Remove from hash tables. 4105 */ 4106 4107 sk->sk_prot->unhash(sk); 4108 4109 /* 4110 * In this point socket cannot receive new packets, but it is possible 4111 * that some packets are in flight because some CPU runs receiver and 4112 * did hash table lookup before we unhashed socket. They will achieve 4113 * receive queue and will be purged by socket destructor. 4114 * 4115 * Also we still have packets pending on receive queue and probably, 4116 * our own packets waiting in device queues. sock_destroy will drain 4117 * receive queue, but transmitted packets will delay socket destruction 4118 * until the last reference will be released. 4119 */ 4120 4121 sock_orphan(sk); 4122 4123 xfrm_sk_free_policy(sk); 4124 4125 sock_put(sk); 4126 } 4127 EXPORT_SYMBOL(sk_common_release); 4128 4129 void sk_get_meminfo(const struct sock *sk, u32 *mem) 4130 { 4131 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 4132 4133 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 4134 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 4135 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 4136 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 4137 mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc); 4138 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 4139 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 4140 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 4141 mem[SK_MEMINFO_DROPS] = sk_drops_read(sk); 4142 } 4143 4144 #ifdef CONFIG_PROC_FS 4145 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 4146 4147 int sock_prot_inuse_get(struct net *net, struct proto *prot) 4148 { 4149 int cpu, idx = prot->inuse_idx; 4150 int res = 0; 4151 4152 for_each_possible_cpu(cpu) 4153 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 4154 4155 return res >= 0 ? res : 0; 4156 } 4157 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 4158 4159 int sock_inuse_get(struct net *net) 4160 { 4161 int cpu, res = 0; 4162 4163 for_each_possible_cpu(cpu) 4164 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 4165 4166 return res; 4167 } 4168 4169 EXPORT_SYMBOL_GPL(sock_inuse_get); 4170 4171 static int __net_init sock_inuse_init_net(struct net *net) 4172 { 4173 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 4174 if (net->core.prot_inuse == NULL) 4175 return -ENOMEM; 4176 return 0; 4177 } 4178 4179 static void __net_exit sock_inuse_exit_net(struct net *net) 4180 { 4181 free_percpu(net->core.prot_inuse); 4182 } 4183 4184 static struct pernet_operations net_inuse_ops = { 4185 .init = sock_inuse_init_net, 4186 .exit = sock_inuse_exit_net, 4187 }; 4188 4189 static __init int net_inuse_init(void) 4190 { 4191 if (register_pernet_subsys(&net_inuse_ops)) 4192 panic("Cannot initialize net inuse counters"); 4193 4194 return 0; 4195 } 4196 4197 core_initcall(net_inuse_init); 4198 4199 static int assign_proto_idx(struct proto *prot) 4200 { 4201 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 4202 4203 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR)) { 4204 pr_err("PROTO_INUSE_NR exhausted\n"); 4205 return -ENOSPC; 4206 } 4207 4208 set_bit(prot->inuse_idx, proto_inuse_idx); 4209 return 0; 4210 } 4211 4212 static void release_proto_idx(struct proto *prot) 4213 { 4214 if (prot->inuse_idx != PROTO_INUSE_NR) 4215 clear_bit(prot->inuse_idx, proto_inuse_idx); 4216 } 4217 #else 4218 static inline int assign_proto_idx(struct proto *prot) 4219 { 4220 return 0; 4221 } 4222 4223 static inline void release_proto_idx(struct proto *prot) 4224 { 4225 } 4226 4227 #endif 4228 4229 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 4230 { 4231 if (!twsk_prot) 4232 return; 4233 kfree(twsk_prot->twsk_slab_name); 4234 twsk_prot->twsk_slab_name = NULL; 4235 kmem_cache_destroy(twsk_prot->twsk_slab); 4236 twsk_prot->twsk_slab = NULL; 4237 } 4238 4239 static int tw_prot_init(const struct proto *prot) 4240 { 4241 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 4242 4243 if (!twsk_prot) 4244 return 0; 4245 4246 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 4247 prot->name); 4248 if (!twsk_prot->twsk_slab_name) 4249 return -ENOMEM; 4250 4251 twsk_prot->twsk_slab = 4252 kmem_cache_create(twsk_prot->twsk_slab_name, 4253 twsk_prot->twsk_obj_size, 0, 4254 SLAB_ACCOUNT | prot->slab_flags, 4255 NULL); 4256 if (!twsk_prot->twsk_slab) { 4257 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 4258 prot->name); 4259 return -ENOMEM; 4260 } 4261 4262 return 0; 4263 } 4264 4265 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 4266 { 4267 if (!rsk_prot) 4268 return; 4269 kfree(rsk_prot->slab_name); 4270 rsk_prot->slab_name = NULL; 4271 kmem_cache_destroy(rsk_prot->slab); 4272 rsk_prot->slab = NULL; 4273 } 4274 4275 static int req_prot_init(const struct proto *prot) 4276 { 4277 struct request_sock_ops *rsk_prot = prot->rsk_prot; 4278 4279 if (!rsk_prot) 4280 return 0; 4281 4282 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4283 prot->name); 4284 if (!rsk_prot->slab_name) 4285 return -ENOMEM; 4286 4287 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4288 rsk_prot->obj_size, 0, 4289 SLAB_ACCOUNT | prot->slab_flags, 4290 NULL); 4291 4292 if (!rsk_prot->slab) { 4293 pr_crit("%s: Can't create request sock SLAB cache!\n", 4294 prot->name); 4295 return -ENOMEM; 4296 } 4297 return 0; 4298 } 4299 4300 int proto_register(struct proto *prot, int alloc_slab) 4301 { 4302 int ret = -ENOBUFS; 4303 4304 if (prot->memory_allocated && !prot->sysctl_mem) { 4305 pr_err("%s: missing sysctl_mem\n", prot->name); 4306 return -EINVAL; 4307 } 4308 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4309 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4310 return -EINVAL; 4311 } 4312 if (alloc_slab) { 4313 struct kmem_cache_args args = { 4314 .useroffset = prot->useroffset, 4315 .usersize = prot->usersize, 4316 .freeptr_offset = prot->freeptr_offset, 4317 .use_freeptr_offset = !!prot->freeptr_offset, 4318 }; 4319 4320 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 4321 &args, 4322 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4323 prot->slab_flags); 4324 if (prot->slab == NULL) { 4325 pr_crit("%s: Can't create sock SLAB cache!\n", 4326 prot->name); 4327 goto out; 4328 } 4329 4330 if (req_prot_init(prot)) 4331 goto out_free_request_sock_slab; 4332 4333 if (tw_prot_init(prot)) 4334 goto out_free_timewait_sock_slab; 4335 } 4336 4337 mutex_lock(&proto_list_mutex); 4338 ret = assign_proto_idx(prot); 4339 if (ret) { 4340 mutex_unlock(&proto_list_mutex); 4341 goto out_free_timewait_sock_slab; 4342 } 4343 list_add(&prot->node, &proto_list); 4344 mutex_unlock(&proto_list_mutex); 4345 return ret; 4346 4347 out_free_timewait_sock_slab: 4348 if (alloc_slab) 4349 tw_prot_cleanup(prot->twsk_prot); 4350 out_free_request_sock_slab: 4351 if (alloc_slab) { 4352 req_prot_cleanup(prot->rsk_prot); 4353 4354 kmem_cache_destroy(prot->slab); 4355 prot->slab = NULL; 4356 } 4357 out: 4358 return ret; 4359 } 4360 EXPORT_SYMBOL(proto_register); 4361 4362 void proto_unregister(struct proto *prot) 4363 { 4364 mutex_lock(&proto_list_mutex); 4365 release_proto_idx(prot); 4366 list_del(&prot->node); 4367 mutex_unlock(&proto_list_mutex); 4368 4369 kmem_cache_destroy(prot->slab); 4370 prot->slab = NULL; 4371 4372 req_prot_cleanup(prot->rsk_prot); 4373 tw_prot_cleanup(prot->twsk_prot); 4374 } 4375 EXPORT_SYMBOL(proto_unregister); 4376 4377 int sock_load_diag_module(int family, int protocol) 4378 { 4379 if (!protocol) { 4380 if (!sock_is_registered(family)) 4381 return -ENOENT; 4382 4383 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4384 NETLINK_SOCK_DIAG, family); 4385 } 4386 4387 #ifdef CONFIG_INET 4388 if (family == AF_INET && 4389 protocol != IPPROTO_RAW && 4390 protocol < MAX_INET_PROTOS && 4391 !rcu_access_pointer(inet_protos[protocol])) 4392 return -ENOENT; 4393 #endif 4394 4395 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4396 NETLINK_SOCK_DIAG, family, protocol); 4397 } 4398 EXPORT_SYMBOL(sock_load_diag_module); 4399 4400 #ifdef CONFIG_PROC_FS 4401 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4402 __acquires(proto_list_mutex) 4403 { 4404 mutex_lock(&proto_list_mutex); 4405 return seq_list_start_head(&proto_list, *pos); 4406 } 4407 4408 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4409 { 4410 return seq_list_next(v, &proto_list, pos); 4411 } 4412 4413 static void proto_seq_stop(struct seq_file *seq, void *v) 4414 __releases(proto_list_mutex) 4415 { 4416 mutex_unlock(&proto_list_mutex); 4417 } 4418 4419 static char proto_method_implemented(const void *method) 4420 { 4421 return method == NULL ? 'n' : 'y'; 4422 } 4423 static long sock_prot_memory_allocated(struct proto *proto) 4424 { 4425 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4426 } 4427 4428 static const char *sock_prot_memory_pressure(struct proto *proto) 4429 { 4430 return proto->memory_pressure != NULL ? 4431 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4432 } 4433 4434 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4435 { 4436 4437 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4438 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4439 proto->name, 4440 proto->obj_size, 4441 sock_prot_inuse_get(seq_file_net(seq), proto), 4442 sock_prot_memory_allocated(proto), 4443 sock_prot_memory_pressure(proto), 4444 proto->max_header, 4445 proto->slab == NULL ? "no" : "yes", 4446 module_name(proto->owner), 4447 proto_method_implemented(proto->close), 4448 proto_method_implemented(proto->connect), 4449 proto_method_implemented(proto->disconnect), 4450 proto_method_implemented(proto->accept), 4451 proto_method_implemented(proto->ioctl), 4452 proto_method_implemented(proto->init), 4453 proto_method_implemented(proto->destroy), 4454 proto_method_implemented(proto->shutdown), 4455 proto_method_implemented(proto->setsockopt), 4456 proto_method_implemented(proto->getsockopt), 4457 proto_method_implemented(proto->sendmsg), 4458 proto_method_implemented(proto->recvmsg), 4459 proto_method_implemented(proto->bind), 4460 proto_method_implemented(proto->backlog_rcv), 4461 proto_method_implemented(proto->hash), 4462 proto_method_implemented(proto->unhash), 4463 proto_method_implemented(proto->get_port), 4464 proto_method_implemented(proto->enter_memory_pressure)); 4465 } 4466 4467 static int proto_seq_show(struct seq_file *seq, void *v) 4468 { 4469 if (v == &proto_list) 4470 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4471 "protocol", 4472 "size", 4473 "sockets", 4474 "memory", 4475 "press", 4476 "maxhdr", 4477 "slab", 4478 "module", 4479 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4480 else 4481 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4482 return 0; 4483 } 4484 4485 static const struct seq_operations proto_seq_ops = { 4486 .start = proto_seq_start, 4487 .next = proto_seq_next, 4488 .stop = proto_seq_stop, 4489 .show = proto_seq_show, 4490 }; 4491 4492 static __net_init int proto_init_net(struct net *net) 4493 { 4494 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4495 sizeof(struct seq_net_private))) 4496 return -ENOMEM; 4497 4498 return 0; 4499 } 4500 4501 static __net_exit void proto_exit_net(struct net *net) 4502 { 4503 remove_proc_entry("protocols", net->proc_net); 4504 } 4505 4506 4507 static __net_initdata struct pernet_operations proto_net_ops = { 4508 .init = proto_init_net, 4509 .exit = proto_exit_net, 4510 }; 4511 4512 static int __init proto_init(void) 4513 { 4514 return register_pernet_subsys(&proto_net_ops); 4515 } 4516 4517 subsys_initcall(proto_init); 4518 4519 #endif /* PROC_FS */ 4520 4521 #ifdef CONFIG_NET_RX_BUSY_POLL 4522 bool sk_busy_loop_end(void *p, unsigned long start_time) 4523 { 4524 struct sock *sk = p; 4525 4526 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4527 return true; 4528 4529 if (sk_is_udp(sk) && 4530 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4531 return true; 4532 4533 return sk_busy_loop_timeout(sk, start_time); 4534 } 4535 EXPORT_SYMBOL(sk_busy_loop_end); 4536 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4537 4538 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len) 4539 { 4540 if (!sk->sk_prot->bind_add) 4541 return -EOPNOTSUPP; 4542 return sk->sk_prot->bind_add(sk, addr, addr_len); 4543 } 4544 EXPORT_SYMBOL(sock_bind_add); 4545 4546 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4547 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4548 void __user *arg, void *karg, size_t size) 4549 { 4550 int ret; 4551 4552 if (copy_from_user(karg, arg, size)) 4553 return -EFAULT; 4554 4555 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4556 if (ret) 4557 return ret; 4558 4559 if (copy_to_user(arg, karg, size)) 4560 return -EFAULT; 4561 4562 return 0; 4563 } 4564 EXPORT_SYMBOL(sock_ioctl_inout); 4565 4566 /* This is the most common ioctl prep function, where the result (4 bytes) is 4567 * copied back to userspace if the ioctl() returns successfully. No input is 4568 * copied from userspace as input argument. 4569 */ 4570 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4571 { 4572 int ret, karg = 0; 4573 4574 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4575 if (ret) 4576 return ret; 4577 4578 return put_user(karg, (int __user *)arg); 4579 } 4580 4581 /* A wrapper around sock ioctls, which copies the data from userspace 4582 * (depending on the protocol/ioctl), and copies back the result to userspace. 4583 * The main motivation for this function is to pass kernel memory to the 4584 * protocol ioctl callbacks, instead of userspace memory. 4585 */ 4586 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4587 { 4588 int rc = 1; 4589 4590 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4591 rc = ipmr_sk_ioctl(sk, cmd, arg); 4592 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4593 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4594 else if (sk_is_phonet(sk)) 4595 rc = phonet_sk_ioctl(sk, cmd, arg); 4596 4597 /* If ioctl was processed, returns its value */ 4598 if (rc <= 0) 4599 return rc; 4600 4601 /* Otherwise call the default handler */ 4602 return sock_ioctl_out(sk, cmd, arg); 4603 } 4604 EXPORT_SYMBOL(sk_ioctl); 4605 4606 static int __init sock_struct_check(void) 4607 { 4608 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4609 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4610 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4611 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4612 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4613 4614 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4615 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4616 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4617 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4618 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4619 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4620 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4621 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4622 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4623 4624 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4625 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4626 #ifdef CONFIG_MEMCG 4627 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4628 #endif 4629 4630 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4631 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4632 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4633 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4634 4635 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4636 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4637 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_err_soft); 4638 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4639 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4640 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4641 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4642 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4643 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4644 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4645 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4646 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4647 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4648 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4649 4650 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_pending_confirm); 4651 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_status); 4652 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4653 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4654 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4655 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4656 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_uid); 4657 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_protocol); 4658 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4659 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4660 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4661 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4662 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4663 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4664 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndbuf); 4665 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4666 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4667 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4668 return 0; 4669 } 4670 4671 core_initcall(sock_struct_check); 4672