1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 */ 85 86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <linux/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/udp.h> 111 #include <linux/init.h> 112 #include <linux/highmem.h> 113 #include <linux/user_namespace.h> 114 #include <linux/static_key.h> 115 #include <linux/memcontrol.h> 116 #include <linux/prefetch.h> 117 #include <linux/compat.h> 118 #include <linux/mroute.h> 119 #include <linux/mroute6.h> 120 #include <linux/icmpv6.h> 121 122 #include <linux/uaccess.h> 123 124 #include <linux/netdevice.h> 125 #include <net/protocol.h> 126 #include <linux/skbuff.h> 127 #include <linux/skbuff_ref.h> 128 #include <net/net_namespace.h> 129 #include <net/request_sock.h> 130 #include <net/sock.h> 131 #include <net/proto_memory.h> 132 #include <linux/net_tstamp.h> 133 #include <net/xfrm.h> 134 #include <linux/ipsec.h> 135 #include <net/cls_cgroup.h> 136 #include <net/netprio_cgroup.h> 137 #include <linux/sock_diag.h> 138 139 #include <linux/filter.h> 140 #include <net/sock_reuseport.h> 141 #include <net/bpf_sk_storage.h> 142 143 #include <trace/events/sock.h> 144 145 #include <net/psp.h> 146 #include <net/tcp.h> 147 #include <net/busy_poll.h> 148 #include <net/phonet/phonet.h> 149 150 #include <linux/ethtool.h> 151 152 #include <uapi/linux/pidfd.h> 153 154 #include "dev.h" 155 156 static DEFINE_MUTEX(proto_list_mutex); 157 static LIST_HEAD(proto_list); 158 159 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc); 160 static void sock_def_write_space(struct sock *sk); 161 162 /** 163 * sk_ns_capable - General socket capability test 164 * @sk: Socket to use a capability on or through 165 * @user_ns: The user namespace of the capability to use 166 * @cap: The capability to use 167 * 168 * Test to see if the opener of the socket had when the socket was 169 * created and the current process has the capability @cap in the user 170 * namespace @user_ns. 171 */ 172 bool sk_ns_capable(const struct sock *sk, 173 struct user_namespace *user_ns, int cap) 174 { 175 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 176 ns_capable(user_ns, cap); 177 } 178 EXPORT_SYMBOL(sk_ns_capable); 179 180 /** 181 * sk_capable - Socket global capability test 182 * @sk: Socket to use a capability on or through 183 * @cap: The global capability to use 184 * 185 * Test to see if the opener of the socket had when the socket was 186 * created and the current process has the capability @cap in all user 187 * namespaces. 188 */ 189 bool sk_capable(const struct sock *sk, int cap) 190 { 191 return sk_ns_capable(sk, &init_user_ns, cap); 192 } 193 EXPORT_SYMBOL(sk_capable); 194 195 /** 196 * sk_net_capable - Network namespace socket capability test 197 * @sk: Socket to use a capability on or through 198 * @cap: The capability to use 199 * 200 * Test to see if the opener of the socket had when the socket was created 201 * and the current process has the capability @cap over the network namespace 202 * the socket is a member of. 203 */ 204 bool sk_net_capable(const struct sock *sk, int cap) 205 { 206 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 207 } 208 EXPORT_SYMBOL(sk_net_capable); 209 210 /* 211 * Each address family might have different locking rules, so we have 212 * one slock key per address family and separate keys for internal and 213 * userspace sockets. 214 */ 215 static struct lock_class_key af_family_keys[AF_MAX]; 216 static struct lock_class_key af_family_kern_keys[AF_MAX]; 217 static struct lock_class_key af_family_slock_keys[AF_MAX]; 218 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 219 220 /* 221 * Make lock validator output more readable. (we pre-construct these 222 * strings build-time, so that runtime initialization of socket 223 * locks is fast): 224 */ 225 226 #define _sock_locks(x) \ 227 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 228 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 229 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 230 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 231 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 232 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 233 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 234 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 235 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 236 x "27" , x "28" , x "AF_CAN" , \ 237 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 238 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 239 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 240 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 241 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 242 x "AF_MCTP" , \ 243 x "AF_MAX" 244 245 static const char *const af_family_key_strings[AF_MAX+1] = { 246 _sock_locks("sk_lock-") 247 }; 248 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 249 _sock_locks("slock-") 250 }; 251 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 252 _sock_locks("clock-") 253 }; 254 255 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 256 _sock_locks("k-sk_lock-") 257 }; 258 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 259 _sock_locks("k-slock-") 260 }; 261 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 262 _sock_locks("k-clock-") 263 }; 264 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 265 _sock_locks("rlock-") 266 }; 267 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 268 _sock_locks("wlock-") 269 }; 270 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 271 _sock_locks("elock-") 272 }; 273 274 /* 275 * sk_callback_lock and sk queues locking rules are per-address-family, 276 * so split the lock classes by using a per-AF key: 277 */ 278 static struct lock_class_key af_callback_keys[AF_MAX]; 279 static struct lock_class_key af_rlock_keys[AF_MAX]; 280 static struct lock_class_key af_wlock_keys[AF_MAX]; 281 static struct lock_class_key af_elock_keys[AF_MAX]; 282 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 283 284 /* Run time adjustable parameters. */ 285 __u32 sysctl_wmem_max __read_mostly = 4 << 20; 286 EXPORT_SYMBOL(sysctl_wmem_max); 287 __u32 sysctl_rmem_max __read_mostly = 4 << 20; 288 EXPORT_SYMBOL(sysctl_rmem_max); 289 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_DEFAULT; 290 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_DEFAULT; 291 292 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 293 EXPORT_SYMBOL_GPL(memalloc_socks_key); 294 295 /** 296 * sk_set_memalloc - sets %SOCK_MEMALLOC 297 * @sk: socket to set it on 298 * 299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 300 * It's the responsibility of the admin to adjust min_free_kbytes 301 * to meet the requirements 302 */ 303 void sk_set_memalloc(struct sock *sk) 304 { 305 sock_set_flag(sk, SOCK_MEMALLOC); 306 sk->sk_allocation |= __GFP_MEMALLOC; 307 static_branch_inc(&memalloc_socks_key); 308 } 309 EXPORT_SYMBOL_GPL(sk_set_memalloc); 310 311 void sk_clear_memalloc(struct sock *sk) 312 { 313 sock_reset_flag(sk, SOCK_MEMALLOC); 314 sk->sk_allocation &= ~__GFP_MEMALLOC; 315 static_branch_dec(&memalloc_socks_key); 316 317 /* 318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 319 * progress of swapping. SOCK_MEMALLOC may be cleared while 320 * it has rmem allocations due to the last swapfile being deactivated 321 * but there is a risk that the socket is unusable due to exceeding 322 * the rmem limits. Reclaim the reserves and obey rmem limits again. 323 */ 324 sk_mem_reclaim(sk); 325 } 326 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 327 328 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 329 { 330 int ret; 331 unsigned int noreclaim_flag; 332 333 /* these should have been dropped before queueing */ 334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 335 336 noreclaim_flag = memalloc_noreclaim_save(); 337 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 338 tcp_v6_do_rcv, 339 tcp_v4_do_rcv, 340 sk, skb); 341 memalloc_noreclaim_restore(noreclaim_flag); 342 343 return ret; 344 } 345 EXPORT_SYMBOL(__sk_backlog_rcv); 346 347 void sk_error_report(struct sock *sk) 348 { 349 sk->sk_error_report(sk); 350 351 switch (sk->sk_family) { 352 case AF_INET: 353 fallthrough; 354 case AF_INET6: 355 trace_inet_sk_error_report(sk); 356 break; 357 default: 358 break; 359 } 360 } 361 EXPORT_SYMBOL(sk_error_report); 362 363 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 364 { 365 struct __kernel_sock_timeval tv; 366 367 if (timeo == MAX_SCHEDULE_TIMEOUT) { 368 tv.tv_sec = 0; 369 tv.tv_usec = 0; 370 } else { 371 tv.tv_sec = timeo / HZ; 372 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 373 } 374 375 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 376 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 377 *(struct old_timeval32 *)optval = tv32; 378 return sizeof(tv32); 379 } 380 381 if (old_timeval) { 382 struct __kernel_old_timeval old_tv; 383 old_tv.tv_sec = tv.tv_sec; 384 old_tv.tv_usec = tv.tv_usec; 385 *(struct __kernel_old_timeval *)optval = old_tv; 386 return sizeof(old_tv); 387 } 388 389 *(struct __kernel_sock_timeval *)optval = tv; 390 return sizeof(tv); 391 } 392 EXPORT_SYMBOL(sock_get_timeout); 393 394 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 395 sockptr_t optval, int optlen, bool old_timeval) 396 { 397 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 398 struct old_timeval32 tv32; 399 400 if (optlen < sizeof(tv32)) 401 return -EINVAL; 402 403 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 404 return -EFAULT; 405 tv->tv_sec = tv32.tv_sec; 406 tv->tv_usec = tv32.tv_usec; 407 } else if (old_timeval) { 408 struct __kernel_old_timeval old_tv; 409 410 if (optlen < sizeof(old_tv)) 411 return -EINVAL; 412 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 413 return -EFAULT; 414 tv->tv_sec = old_tv.tv_sec; 415 tv->tv_usec = old_tv.tv_usec; 416 } else { 417 if (optlen < sizeof(*tv)) 418 return -EINVAL; 419 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 420 return -EFAULT; 421 } 422 423 return 0; 424 } 425 EXPORT_SYMBOL(sock_copy_user_timeval); 426 427 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 428 bool old_timeval) 429 { 430 struct __kernel_sock_timeval tv; 431 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 432 long val; 433 434 if (err) 435 return err; 436 437 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 438 return -EDOM; 439 440 if (tv.tv_sec < 0) { 441 static int warned __read_mostly; 442 443 WRITE_ONCE(*timeo_p, 0); 444 if (warned < 10 && net_ratelimit()) { 445 warned++; 446 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 447 __func__, current->comm, task_pid_nr(current)); 448 } 449 return 0; 450 } 451 val = MAX_SCHEDULE_TIMEOUT; 452 if ((tv.tv_sec || tv.tv_usec) && 453 (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1))) 454 val = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, 455 USEC_PER_SEC / HZ); 456 WRITE_ONCE(*timeo_p, val); 457 return 0; 458 } 459 460 static bool sk_set_prio_allowed(const struct sock *sk, int val) 461 { 462 return ((val >= TC_PRIO_BESTEFFORT && val <= TC_PRIO_INTERACTIVE) || 463 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 464 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)); 465 } 466 467 static bool sock_needs_netstamp(const struct sock *sk) 468 { 469 switch (sk->sk_family) { 470 case AF_UNSPEC: 471 case AF_UNIX: 472 return false; 473 default: 474 return true; 475 } 476 } 477 478 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 479 { 480 if (sk->sk_flags & flags) { 481 sk->sk_flags &= ~flags; 482 if (sock_needs_netstamp(sk) && 483 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 484 net_disable_timestamp(); 485 } 486 } 487 488 489 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 490 { 491 unsigned long flags; 492 struct sk_buff_head *list = &sk->sk_receive_queue; 493 494 if (atomic_read(&sk->sk_rmem_alloc) >= READ_ONCE(sk->sk_rcvbuf)) { 495 sk_drops_inc(sk); 496 trace_sock_rcvqueue_full(sk, skb); 497 return -ENOMEM; 498 } 499 500 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 501 sk_drops_inc(sk); 502 return -ENOBUFS; 503 } 504 505 skb->dev = NULL; 506 skb_set_owner_r(skb, sk); 507 508 /* we escape from rcu protected region, make sure we dont leak 509 * a norefcounted dst 510 */ 511 skb_dst_force(skb); 512 513 spin_lock_irqsave(&list->lock, flags); 514 sock_skb_set_dropcount(sk, skb); 515 __skb_queue_tail(list, skb); 516 spin_unlock_irqrestore(&list->lock, flags); 517 518 if (!sock_flag(sk, SOCK_DEAD)) 519 sk->sk_data_ready(sk); 520 return 0; 521 } 522 EXPORT_SYMBOL(__sock_queue_rcv_skb); 523 524 enum skb_drop_reason 525 sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb) 526 { 527 enum skb_drop_reason drop_reason; 528 int err; 529 530 drop_reason = sk_filter_reason(sk, skb); 531 if (drop_reason) 532 return drop_reason; 533 534 err = __sock_queue_rcv_skb(sk, skb); 535 switch (err) { 536 case -ENOMEM: 537 return SKB_DROP_REASON_SOCKET_RCVBUFF; 538 case -ENOBUFS: 539 return SKB_DROP_REASON_PROTO_MEM; 540 } 541 return SKB_NOT_DROPPED_YET; 542 } 543 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 544 545 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 546 const int nested, unsigned int trim_cap, bool refcounted) 547 { 548 enum skb_drop_reason reason; 549 int rc = NET_RX_SUCCESS; 550 int err; 551 552 reason = sk_filter_trim_cap(sk, skb, trim_cap); 553 if (reason) 554 goto discard_and_relse; 555 556 skb->dev = NULL; 557 558 if (sk_rcvqueues_full(sk, READ_ONCE(sk->sk_rcvbuf))) { 559 sk_drops_inc(sk); 560 reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 561 goto discard_and_relse; 562 } 563 if (nested) 564 bh_lock_sock_nested(sk); 565 else 566 bh_lock_sock(sk); 567 if (!sock_owned_by_user(sk)) { 568 /* 569 * trylock + unlock semantics: 570 */ 571 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 572 573 rc = sk_backlog_rcv(sk, skb); 574 575 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 576 } else if ((err = sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf)))) { 577 bh_unlock_sock(sk); 578 if (err == -ENOMEM) 579 reason = SKB_DROP_REASON_PFMEMALLOC; 580 if (err == -ENOBUFS) 581 reason = SKB_DROP_REASON_SOCKET_BACKLOG; 582 sk_drops_inc(sk); 583 goto discard_and_relse; 584 } 585 586 bh_unlock_sock(sk); 587 out: 588 if (refcounted) 589 sock_put(sk); 590 return rc; 591 discard_and_relse: 592 sk_skb_reason_drop(sk, skb, reason); 593 goto out; 594 } 595 EXPORT_SYMBOL(__sk_receive_skb); 596 597 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 598 u32)); 599 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 600 u32)); 601 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 602 { 603 struct dst_entry *dst = __sk_dst_get(sk); 604 605 if (dst && READ_ONCE(dst->obsolete) && 606 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 607 dst, cookie) == NULL) { 608 sk_tx_queue_clear(sk); 609 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 610 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 611 dst_release(dst); 612 return NULL; 613 } 614 615 return dst; 616 } 617 EXPORT_SYMBOL(__sk_dst_check); 618 619 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 620 { 621 struct dst_entry *dst = sk_dst_get(sk); 622 623 if (dst && READ_ONCE(dst->obsolete) && 624 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 625 dst, cookie) == NULL) { 626 sk_dst_reset(sk); 627 dst_release(dst); 628 return NULL; 629 } 630 631 return dst; 632 } 633 EXPORT_SYMBOL(sk_dst_check); 634 635 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 636 { 637 int ret = -ENOPROTOOPT; 638 #ifdef CONFIG_NETDEVICES 639 struct net *net = sock_net(sk); 640 641 /* Sorry... */ 642 ret = -EPERM; 643 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 644 goto out; 645 646 ret = -EINVAL; 647 if (ifindex < 0) 648 goto out; 649 650 /* Paired with all READ_ONCE() done locklessly. */ 651 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 652 653 if (sk->sk_prot->rehash) 654 sk->sk_prot->rehash(sk); 655 sk_dst_reset(sk); 656 657 ret = 0; 658 659 out: 660 #endif 661 662 return ret; 663 } 664 665 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 666 { 667 int ret; 668 669 if (lock_sk) 670 lock_sock(sk); 671 ret = sock_bindtoindex_locked(sk, ifindex); 672 if (lock_sk) 673 release_sock(sk); 674 675 return ret; 676 } 677 EXPORT_SYMBOL(sock_bindtoindex); 678 679 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 680 { 681 int ret = -ENOPROTOOPT; 682 #ifdef CONFIG_NETDEVICES 683 struct net *net = sock_net(sk); 684 char devname[IFNAMSIZ]; 685 int index; 686 687 ret = -EINVAL; 688 if (optlen < 0) 689 goto out; 690 691 /* Bind this socket to a particular device like "eth0", 692 * as specified in the passed interface name. If the 693 * name is "" or the option length is zero the socket 694 * is not bound. 695 */ 696 if (optlen > IFNAMSIZ - 1) 697 optlen = IFNAMSIZ - 1; 698 memset(devname, 0, sizeof(devname)); 699 700 ret = -EFAULT; 701 if (copy_from_sockptr(devname, optval, optlen)) 702 goto out; 703 704 index = 0; 705 if (devname[0] != '\0') { 706 struct net_device *dev; 707 708 rcu_read_lock(); 709 dev = dev_get_by_name_rcu(net, devname); 710 if (dev) 711 index = dev->ifindex; 712 rcu_read_unlock(); 713 ret = -ENODEV; 714 if (!dev) 715 goto out; 716 } 717 718 sockopt_lock_sock(sk); 719 ret = sock_bindtoindex_locked(sk, index); 720 sockopt_release_sock(sk); 721 out: 722 #endif 723 724 return ret; 725 } 726 727 static int sock_getbindtodevice(struct sock *sk, sockptr_t optval, 728 sockptr_t optlen, int len) 729 { 730 int ret = -ENOPROTOOPT; 731 #ifdef CONFIG_NETDEVICES 732 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 733 struct net *net = sock_net(sk); 734 char devname[IFNAMSIZ]; 735 736 if (bound_dev_if == 0) { 737 len = 0; 738 goto zero; 739 } 740 741 ret = -EINVAL; 742 if (len < IFNAMSIZ) 743 goto out; 744 745 ret = netdev_get_name(net, devname, bound_dev_if); 746 if (ret) 747 goto out; 748 749 len = strlen(devname) + 1; 750 751 ret = -EFAULT; 752 if (copy_to_sockptr(optval, devname, len)) 753 goto out; 754 755 zero: 756 ret = -EFAULT; 757 if (copy_to_sockptr(optlen, &len, sizeof(int))) 758 goto out; 759 760 ret = 0; 761 762 out: 763 #endif 764 765 return ret; 766 } 767 768 bool sk_mc_loop(const struct sock *sk) 769 { 770 if (dev_recursion_level()) 771 return false; 772 if (!sk) 773 return true; 774 /* IPV6_ADDRFORM can change sk->sk_family under us. */ 775 switch (READ_ONCE(sk->sk_family)) { 776 case AF_INET: 777 return inet_test_bit(MC_LOOP, sk); 778 #if IS_ENABLED(CONFIG_IPV6) 779 case AF_INET6: 780 return inet6_test_bit(MC6_LOOP, sk); 781 #endif 782 } 783 return true; 784 } 785 EXPORT_SYMBOL(sk_mc_loop); 786 787 void sock_set_reuseaddr(struct sock *sk) 788 { 789 lock_sock(sk); 790 sk->sk_reuse = SK_CAN_REUSE; 791 release_sock(sk); 792 } 793 EXPORT_SYMBOL(sock_set_reuseaddr); 794 795 void sock_set_reuseport(struct sock *sk) 796 { 797 lock_sock(sk); 798 sk->sk_reuseport = true; 799 release_sock(sk); 800 } 801 EXPORT_SYMBOL(sock_set_reuseport); 802 803 void sock_no_linger(struct sock *sk) 804 { 805 lock_sock(sk); 806 WRITE_ONCE(sk->sk_lingertime, 0); 807 sock_set_flag(sk, SOCK_LINGER); 808 release_sock(sk); 809 } 810 EXPORT_SYMBOL(sock_no_linger); 811 812 void sock_set_priority(struct sock *sk, u32 priority) 813 { 814 WRITE_ONCE(sk->sk_priority, priority); 815 } 816 EXPORT_SYMBOL(sock_set_priority); 817 818 void sock_set_sndtimeo(struct sock *sk, s64 secs) 819 { 820 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 821 WRITE_ONCE(sk->sk_sndtimeo, secs * HZ); 822 else 823 WRITE_ONCE(sk->sk_sndtimeo, MAX_SCHEDULE_TIMEOUT); 824 } 825 EXPORT_SYMBOL(sock_set_sndtimeo); 826 827 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 828 { 829 sock_valbool_flag(sk, SOCK_RCVTSTAMP, val); 830 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, val && ns); 831 if (val) { 832 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 833 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 834 } 835 } 836 837 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 838 { 839 switch (optname) { 840 case SO_TIMESTAMP_OLD: 841 __sock_set_timestamps(sk, valbool, false, false); 842 break; 843 case SO_TIMESTAMP_NEW: 844 __sock_set_timestamps(sk, valbool, true, false); 845 break; 846 case SO_TIMESTAMPNS_OLD: 847 __sock_set_timestamps(sk, valbool, false, true); 848 break; 849 case SO_TIMESTAMPNS_NEW: 850 __sock_set_timestamps(sk, valbool, true, true); 851 break; 852 } 853 } 854 855 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 856 { 857 struct net *net = sock_net(sk); 858 struct net_device *dev = NULL; 859 bool match = false; 860 int *vclock_index; 861 int i, num; 862 863 if (sk->sk_bound_dev_if) 864 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 865 866 if (!dev) { 867 pr_err("%s: sock not bind to device\n", __func__); 868 return -EOPNOTSUPP; 869 } 870 871 num = ethtool_get_phc_vclocks(dev, &vclock_index); 872 dev_put(dev); 873 874 for (i = 0; i < num; i++) { 875 if (*(vclock_index + i) == phc_index) { 876 match = true; 877 break; 878 } 879 } 880 881 if (num > 0) 882 kfree(vclock_index); 883 884 if (!match) 885 return -EINVAL; 886 887 WRITE_ONCE(sk->sk_bind_phc, phc_index); 888 889 return 0; 890 } 891 892 int sock_set_timestamping(struct sock *sk, int optname, 893 struct so_timestamping timestamping) 894 { 895 int val = timestamping.flags; 896 int ret; 897 898 if (val & ~SOF_TIMESTAMPING_MASK) 899 return -EINVAL; 900 901 if (val & SOF_TIMESTAMPING_OPT_ID_TCP && 902 !(val & SOF_TIMESTAMPING_OPT_ID)) 903 return -EINVAL; 904 905 if (val & SOF_TIMESTAMPING_OPT_ID && 906 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 907 if (sk_is_tcp(sk)) { 908 if ((1 << sk->sk_state) & 909 (TCPF_CLOSE | TCPF_LISTEN)) 910 return -EINVAL; 911 if (val & SOF_TIMESTAMPING_OPT_ID_TCP) 912 atomic_set(&sk->sk_tskey, tcp_sk(sk)->write_seq); 913 else 914 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 915 } else { 916 atomic_set(&sk->sk_tskey, 0); 917 } 918 } 919 920 if (val & SOF_TIMESTAMPING_OPT_STATS && 921 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 922 return -EINVAL; 923 924 if (val & SOF_TIMESTAMPING_BIND_PHC) { 925 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 926 if (ret) 927 return ret; 928 } 929 930 WRITE_ONCE(sk->sk_tsflags, val); 931 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 932 sock_valbool_flag(sk, SOCK_TIMESTAMPING_ANY, !!(val & TSFLAGS_ANY)); 933 934 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 935 sock_enable_timestamp(sk, 936 SOCK_TIMESTAMPING_RX_SOFTWARE); 937 else 938 sock_disable_timestamp(sk, 939 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 940 return 0; 941 } 942 943 #if defined(CONFIG_CGROUP_BPF) 944 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 945 { 946 struct bpf_sock_ops_kern sock_ops; 947 948 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp)); 949 sock_ops.op = op; 950 sock_ops.is_fullsock = 1; 951 sock_ops.sk = sk; 952 bpf_skops_init_skb(&sock_ops, skb, 0); 953 __cgroup_bpf_run_filter_sock_ops(sk, &sock_ops, CGROUP_SOCK_OPS); 954 } 955 #endif 956 957 void sock_set_keepalive(struct sock *sk) 958 { 959 lock_sock(sk); 960 if (sk->sk_prot->keepalive) 961 sk->sk_prot->keepalive(sk, true); 962 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 963 release_sock(sk); 964 } 965 EXPORT_SYMBOL(sock_set_keepalive); 966 967 static void __sock_set_rcvbuf(struct sock *sk, int val) 968 { 969 struct socket *sock = sk->sk_socket; 970 971 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 972 * as a negative value. 973 */ 974 val = min_t(int, val, INT_MAX / 2); 975 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 976 977 /* We double it on the way in to account for "struct sk_buff" etc. 978 * overhead. Applications assume that the SO_RCVBUF setting they make 979 * will allow that much actual data to be received on that socket. 980 * 981 * Applications are unaware that "struct sk_buff" and other overheads 982 * allocate from the receive buffer during socket buffer allocation. 983 * 984 * And after considering the possible alternatives, returning the value 985 * we actually used in getsockopt is the most desirable behavior. 986 */ 987 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 988 989 if (sock) { 990 const struct proto_ops *ops = READ_ONCE(sock->ops); 991 992 if (ops->set_rcvbuf) 993 ops->set_rcvbuf(sk, sk->sk_rcvbuf); 994 } 995 } 996 997 void sock_set_rcvbuf(struct sock *sk, int val) 998 { 999 lock_sock(sk); 1000 __sock_set_rcvbuf(sk, val); 1001 release_sock(sk); 1002 } 1003 EXPORT_SYMBOL(sock_set_rcvbuf); 1004 1005 static void __sock_set_mark(struct sock *sk, u32 val) 1006 { 1007 if (val != sk->sk_mark) { 1008 WRITE_ONCE(sk->sk_mark, val); 1009 sk_dst_reset(sk); 1010 } 1011 } 1012 1013 void sock_set_mark(struct sock *sk, u32 val) 1014 { 1015 lock_sock(sk); 1016 __sock_set_mark(sk, val); 1017 release_sock(sk); 1018 } 1019 EXPORT_SYMBOL(sock_set_mark); 1020 1021 static void sock_release_reserved_memory(struct sock *sk, int bytes) 1022 { 1023 /* Round down bytes to multiple of pages */ 1024 bytes = round_down(bytes, PAGE_SIZE); 1025 1026 WARN_ON(bytes > sk->sk_reserved_mem); 1027 WRITE_ONCE(sk->sk_reserved_mem, sk->sk_reserved_mem - bytes); 1028 sk_mem_reclaim(sk); 1029 } 1030 1031 static int sock_reserve_memory(struct sock *sk, int bytes) 1032 { 1033 long allocated; 1034 bool charged; 1035 int pages; 1036 1037 if (!mem_cgroup_sk_enabled(sk) || !sk_is_tcp(sk)) 1038 return -EOPNOTSUPP; 1039 1040 if (!bytes) 1041 return 0; 1042 1043 pages = sk_mem_pages(bytes); 1044 1045 /* pre-charge to memcg */ 1046 charged = mem_cgroup_sk_charge(sk, pages, 1047 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1048 if (!charged) 1049 return -ENOMEM; 1050 1051 if (sk->sk_bypass_prot_mem) 1052 goto success; 1053 1054 /* pre-charge to forward_alloc */ 1055 sk_memory_allocated_add(sk, pages); 1056 allocated = sk_memory_allocated(sk); 1057 1058 /* If the system goes into memory pressure with this 1059 * precharge, give up and return error. 1060 */ 1061 if (allocated > sk_prot_mem_limits(sk, 1)) { 1062 sk_memory_allocated_sub(sk, pages); 1063 mem_cgroup_sk_uncharge(sk, pages); 1064 return -ENOMEM; 1065 } 1066 1067 success: 1068 sk_forward_alloc_add(sk, pages << PAGE_SHIFT); 1069 1070 WRITE_ONCE(sk->sk_reserved_mem, 1071 sk->sk_reserved_mem + (pages << PAGE_SHIFT)); 1072 1073 return 0; 1074 } 1075 1076 #ifdef CONFIG_PAGE_POOL 1077 1078 /* This is the number of tokens and frags that the user can SO_DEVMEM_DONTNEED 1079 * in 1 syscall. The limit exists to limit the amount of memory the kernel 1080 * allocates to copy these tokens, and to prevent looping over the frags for 1081 * too long. 1082 */ 1083 #define MAX_DONTNEED_TOKENS 128 1084 #define MAX_DONTNEED_FRAGS 1024 1085 1086 static noinline_for_stack int 1087 sock_devmem_dontneed(struct sock *sk, sockptr_t optval, unsigned int optlen) 1088 { 1089 unsigned int num_tokens, i, j, k, netmem_num = 0; 1090 struct dmabuf_token *tokens; 1091 int ret = 0, num_frags = 0; 1092 netmem_ref netmems[16]; 1093 1094 if (!sk_is_tcp(sk)) 1095 return -EBADF; 1096 1097 if (optlen % sizeof(*tokens) || 1098 optlen > sizeof(*tokens) * MAX_DONTNEED_TOKENS) 1099 return -EINVAL; 1100 1101 num_tokens = optlen / sizeof(*tokens); 1102 tokens = kvmalloc_objs(*tokens, num_tokens); 1103 if (!tokens) 1104 return -ENOMEM; 1105 1106 if (copy_from_sockptr(tokens, optval, optlen)) { 1107 kvfree(tokens); 1108 return -EFAULT; 1109 } 1110 1111 xa_lock_bh(&sk->sk_user_frags); 1112 for (i = 0; i < num_tokens; i++) { 1113 for (j = 0; j < tokens[i].token_count; j++) { 1114 if (++num_frags > MAX_DONTNEED_FRAGS) 1115 goto frag_limit_reached; 1116 1117 netmem_ref netmem = (__force netmem_ref)__xa_erase( 1118 &sk->sk_user_frags, tokens[i].token_start + j); 1119 1120 if (!netmem || WARN_ON_ONCE(!netmem_is_net_iov(netmem))) 1121 continue; 1122 1123 netmems[netmem_num++] = netmem; 1124 if (netmem_num == ARRAY_SIZE(netmems)) { 1125 xa_unlock_bh(&sk->sk_user_frags); 1126 for (k = 0; k < netmem_num; k++) 1127 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1128 netmem_num = 0; 1129 xa_lock_bh(&sk->sk_user_frags); 1130 } 1131 ret++; 1132 } 1133 } 1134 1135 frag_limit_reached: 1136 xa_unlock_bh(&sk->sk_user_frags); 1137 for (k = 0; k < netmem_num; k++) 1138 WARN_ON_ONCE(!napi_pp_put_page(netmems[k])); 1139 1140 kvfree(tokens); 1141 return ret; 1142 } 1143 #endif 1144 1145 void sockopt_lock_sock(struct sock *sk) 1146 { 1147 /* When current->bpf_ctx is set, the setsockopt is called from 1148 * a bpf prog. bpf has ensured the sk lock has been 1149 * acquired before calling setsockopt(). 1150 */ 1151 if (has_current_bpf_ctx()) 1152 return; 1153 1154 lock_sock(sk); 1155 } 1156 EXPORT_SYMBOL(sockopt_lock_sock); 1157 1158 void sockopt_release_sock(struct sock *sk) 1159 { 1160 if (has_current_bpf_ctx()) 1161 return; 1162 1163 release_sock(sk); 1164 } 1165 EXPORT_SYMBOL(sockopt_release_sock); 1166 1167 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1168 { 1169 return has_current_bpf_ctx() || ns_capable(ns, cap); 1170 } 1171 EXPORT_SYMBOL(sockopt_ns_capable); 1172 1173 bool sockopt_capable(int cap) 1174 { 1175 return has_current_bpf_ctx() || capable(cap); 1176 } 1177 EXPORT_SYMBOL(sockopt_capable); 1178 1179 static int sockopt_validate_clockid(__kernel_clockid_t value) 1180 { 1181 switch (value) { 1182 case CLOCK_REALTIME: 1183 case CLOCK_MONOTONIC: 1184 case CLOCK_TAI: 1185 return 0; 1186 } 1187 return -EINVAL; 1188 } 1189 1190 /* 1191 * This is meant for all protocols to use and covers goings on 1192 * at the socket level. Everything here is generic. 1193 */ 1194 1195 int sk_setsockopt(struct sock *sk, int level, int optname, 1196 sockptr_t optval, unsigned int optlen) 1197 { 1198 struct so_timestamping timestamping; 1199 struct socket *sock = sk->sk_socket; 1200 struct sock_txtime sk_txtime; 1201 int val; 1202 int valbool; 1203 struct linger ling; 1204 int ret = 0; 1205 1206 /* 1207 * Options without arguments 1208 */ 1209 1210 if (optname == SO_BINDTODEVICE) 1211 return sock_setbindtodevice(sk, optval, optlen); 1212 1213 if (optlen < sizeof(int)) 1214 return -EINVAL; 1215 1216 if (copy_from_sockptr(&val, optval, sizeof(val))) 1217 return -EFAULT; 1218 1219 valbool = val ? 1 : 0; 1220 1221 /* handle options which do not require locking the socket. */ 1222 switch (optname) { 1223 case SO_PRIORITY: 1224 if (sk_set_prio_allowed(sk, val)) { 1225 sock_set_priority(sk, val); 1226 return 0; 1227 } 1228 return -EPERM; 1229 case SO_TYPE: 1230 case SO_PROTOCOL: 1231 case SO_DOMAIN: 1232 case SO_ERROR: 1233 return -ENOPROTOOPT; 1234 #ifdef CONFIG_NET_RX_BUSY_POLL 1235 case SO_BUSY_POLL: 1236 if (val < 0) 1237 return -EINVAL; 1238 WRITE_ONCE(sk->sk_ll_usec, val); 1239 return 0; 1240 case SO_PREFER_BUSY_POLL: 1241 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1242 return -EPERM; 1243 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1244 return 0; 1245 case SO_BUSY_POLL_BUDGET: 1246 if (val > READ_ONCE(sk->sk_busy_poll_budget) && 1247 !sockopt_capable(CAP_NET_ADMIN)) 1248 return -EPERM; 1249 if (val < 0 || val > U16_MAX) 1250 return -EINVAL; 1251 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1252 return 0; 1253 #endif 1254 case SO_MAX_PACING_RATE: 1255 { 1256 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1257 unsigned long pacing_rate; 1258 1259 if (sizeof(ulval) != sizeof(val) && 1260 optlen >= sizeof(ulval) && 1261 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1262 return -EFAULT; 1263 } 1264 if (ulval != ~0UL) 1265 cmpxchg(&sk->sk_pacing_status, 1266 SK_PACING_NONE, 1267 SK_PACING_NEEDED); 1268 /* Pairs with READ_ONCE() from sk_getsockopt() */ 1269 WRITE_ONCE(sk->sk_max_pacing_rate, ulval); 1270 pacing_rate = READ_ONCE(sk->sk_pacing_rate); 1271 if (ulval < pacing_rate) 1272 WRITE_ONCE(sk->sk_pacing_rate, ulval); 1273 return 0; 1274 } 1275 case SO_TXREHASH: 1276 if (!sk_is_tcp(sk)) 1277 return -EOPNOTSUPP; 1278 if (val < -1 || val > 1) 1279 return -EINVAL; 1280 if ((u8)val == SOCK_TXREHASH_DEFAULT) 1281 val = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); 1282 /* Paired with READ_ONCE() in tcp_rtx_synack() 1283 * and sk_getsockopt(). 1284 */ 1285 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1286 return 0; 1287 case SO_PEEK_OFF: 1288 { 1289 int (*set_peek_off)(struct sock *sk, int val); 1290 1291 set_peek_off = READ_ONCE(sock->ops)->set_peek_off; 1292 if (set_peek_off) 1293 ret = set_peek_off(sk, val); 1294 else 1295 ret = -EOPNOTSUPP; 1296 return ret; 1297 } 1298 #ifdef CONFIG_PAGE_POOL 1299 case SO_DEVMEM_DONTNEED: 1300 return sock_devmem_dontneed(sk, optval, optlen); 1301 #endif 1302 case SO_SNDTIMEO_OLD: 1303 case SO_SNDTIMEO_NEW: 1304 return sock_set_timeout(&sk->sk_sndtimeo, optval, 1305 optlen, optname == SO_SNDTIMEO_OLD); 1306 case SO_RCVTIMEO_OLD: 1307 case SO_RCVTIMEO_NEW: 1308 return sock_set_timeout(&sk->sk_rcvtimeo, optval, 1309 optlen, optname == SO_RCVTIMEO_OLD); 1310 } 1311 1312 sockopt_lock_sock(sk); 1313 1314 switch (optname) { 1315 case SO_DEBUG: 1316 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1317 ret = -EACCES; 1318 else 1319 sock_valbool_flag(sk, SOCK_DBG, valbool); 1320 break; 1321 case SO_REUSEADDR: 1322 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1323 break; 1324 case SO_REUSEPORT: 1325 if (valbool && !sk_is_inet(sk)) 1326 ret = -EOPNOTSUPP; 1327 else 1328 sk->sk_reuseport = valbool; 1329 break; 1330 case SO_DONTROUTE: 1331 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1332 sk_dst_reset(sk); 1333 break; 1334 case SO_BROADCAST: 1335 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1336 break; 1337 case SO_SNDBUF: 1338 /* Don't error on this BSD doesn't and if you think 1339 * about it this is right. Otherwise apps have to 1340 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1341 * are treated in BSD as hints 1342 */ 1343 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 1344 set_sndbuf: 1345 /* Ensure val * 2 fits into an int, to prevent max_t() 1346 * from treating it as a negative value. 1347 */ 1348 val = min_t(int, val, INT_MAX / 2); 1349 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1350 WRITE_ONCE(sk->sk_sndbuf, 1351 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1352 /* Wake up sending tasks if we upped the value. */ 1353 sk->sk_write_space(sk); 1354 break; 1355 1356 case SO_SNDBUFFORCE: 1357 if (!sockopt_capable(CAP_NET_ADMIN)) { 1358 ret = -EPERM; 1359 break; 1360 } 1361 1362 /* No negative values (to prevent underflow, as val will be 1363 * multiplied by 2). 1364 */ 1365 if (val < 0) 1366 val = 0; 1367 goto set_sndbuf; 1368 1369 case SO_RCVBUF: 1370 /* Don't error on this BSD doesn't and if you think 1371 * about it this is right. Otherwise apps have to 1372 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1373 * are treated in BSD as hints 1374 */ 1375 __sock_set_rcvbuf(sk, min_t(u32, val, READ_ONCE(sysctl_rmem_max))); 1376 break; 1377 1378 case SO_RCVBUFFORCE: 1379 if (!sockopt_capable(CAP_NET_ADMIN)) { 1380 ret = -EPERM; 1381 break; 1382 } 1383 1384 /* No negative values (to prevent underflow, as val will be 1385 * multiplied by 2). 1386 */ 1387 __sock_set_rcvbuf(sk, max(val, 0)); 1388 break; 1389 1390 case SO_KEEPALIVE: 1391 if (sk->sk_prot->keepalive) 1392 sk->sk_prot->keepalive(sk, valbool); 1393 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1394 break; 1395 1396 case SO_OOBINLINE: 1397 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1398 break; 1399 1400 case SO_NO_CHECK: 1401 sk->sk_no_check_tx = valbool; 1402 break; 1403 1404 case SO_LINGER: 1405 if (optlen < sizeof(ling)) { 1406 ret = -EINVAL; /* 1003.1g */ 1407 break; 1408 } 1409 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1410 ret = -EFAULT; 1411 break; 1412 } 1413 if (!ling.l_onoff) { 1414 sock_reset_flag(sk, SOCK_LINGER); 1415 } else { 1416 unsigned long t_sec = ling.l_linger; 1417 1418 if (t_sec >= MAX_SCHEDULE_TIMEOUT / HZ) 1419 WRITE_ONCE(sk->sk_lingertime, MAX_SCHEDULE_TIMEOUT); 1420 else 1421 WRITE_ONCE(sk->sk_lingertime, t_sec * HZ); 1422 sock_set_flag(sk, SOCK_LINGER); 1423 } 1424 break; 1425 1426 case SO_BSDCOMPAT: 1427 break; 1428 1429 case SO_TIMESTAMP_OLD: 1430 case SO_TIMESTAMP_NEW: 1431 case SO_TIMESTAMPNS_OLD: 1432 case SO_TIMESTAMPNS_NEW: 1433 sock_set_timestamp(sk, optname, valbool); 1434 break; 1435 1436 case SO_TIMESTAMPING_NEW: 1437 case SO_TIMESTAMPING_OLD: 1438 if (optlen == sizeof(timestamping)) { 1439 if (copy_from_sockptr(×tamping, optval, 1440 sizeof(timestamping))) { 1441 ret = -EFAULT; 1442 break; 1443 } 1444 } else { 1445 memset(×tamping, 0, sizeof(timestamping)); 1446 timestamping.flags = val; 1447 } 1448 ret = sock_set_timestamping(sk, optname, timestamping); 1449 break; 1450 1451 case SO_RCVLOWAT: 1452 { 1453 int (*set_rcvlowat)(struct sock *sk, int val) = NULL; 1454 1455 if (val < 0) 1456 val = INT_MAX; 1457 if (sock) 1458 set_rcvlowat = READ_ONCE(sock->ops)->set_rcvlowat; 1459 if (set_rcvlowat) 1460 ret = set_rcvlowat(sk, val); 1461 else 1462 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1463 break; 1464 } 1465 case SO_ATTACH_FILTER: { 1466 struct sock_fprog fprog; 1467 1468 if (sk_is_tcp(sk) && 1469 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1470 ret = -EPERM; 1471 break; 1472 } 1473 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1474 if (!ret) 1475 ret = sk_attach_filter(&fprog, sk); 1476 break; 1477 } 1478 case SO_ATTACH_BPF: 1479 ret = -EINVAL; 1480 if (optlen == sizeof(u32)) { 1481 u32 ufd; 1482 1483 ret = -EFAULT; 1484 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1485 break; 1486 1487 ret = sk_attach_bpf(ufd, sk); 1488 } 1489 break; 1490 1491 case SO_ATTACH_REUSEPORT_CBPF: { 1492 struct sock_fprog fprog; 1493 1494 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1495 if (!ret) 1496 ret = sk_reuseport_attach_filter(&fprog, sk); 1497 break; 1498 } 1499 case SO_ATTACH_REUSEPORT_EBPF: 1500 ret = -EINVAL; 1501 if (optlen == sizeof(u32)) { 1502 u32 ufd; 1503 1504 ret = -EFAULT; 1505 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1506 break; 1507 1508 ret = sk_reuseport_attach_bpf(ufd, sk); 1509 } 1510 break; 1511 1512 case SO_DETACH_REUSEPORT_BPF: 1513 ret = reuseport_detach_prog(sk); 1514 break; 1515 1516 case SO_DETACH_FILTER: 1517 ret = sk_detach_filter(sk); 1518 break; 1519 1520 case SO_LOCK_FILTER: 1521 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1522 ret = -EPERM; 1523 else 1524 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1525 break; 1526 1527 case SO_MARK: 1528 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1529 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1530 ret = -EPERM; 1531 break; 1532 } 1533 1534 __sock_set_mark(sk, val); 1535 break; 1536 case SO_RCVMARK: 1537 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1538 break; 1539 1540 case SO_RCVPRIORITY: 1541 sock_valbool_flag(sk, SOCK_RCVPRIORITY, valbool); 1542 break; 1543 1544 case SO_RXQ_OVFL: 1545 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1546 break; 1547 1548 case SO_WIFI_STATUS: 1549 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1550 break; 1551 1552 case SO_NOFCS: 1553 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1554 break; 1555 1556 case SO_SELECT_ERR_QUEUE: 1557 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1558 break; 1559 1560 case SO_PASSCRED: 1561 if (sk_may_scm_recv(sk)) 1562 sk->sk_scm_credentials = valbool; 1563 else 1564 ret = -EOPNOTSUPP; 1565 break; 1566 1567 case SO_PASSSEC: 1568 if (IS_ENABLED(CONFIG_SECURITY_NETWORK) && sk_may_scm_recv(sk)) 1569 sk->sk_scm_security = valbool; 1570 else 1571 ret = -EOPNOTSUPP; 1572 break; 1573 1574 case SO_PASSPIDFD: 1575 if (sk_is_unix(sk)) 1576 sk->sk_scm_pidfd = valbool; 1577 else 1578 ret = -EOPNOTSUPP; 1579 break; 1580 1581 case SO_PASSPIDFD_THREAD: 1582 if (sk_is_unix(sk)) 1583 sk->sk_scm_pidfd_thread = valbool; 1584 else 1585 ret = -EOPNOTSUPP; 1586 break; 1587 1588 case SO_PASSRIGHTS: 1589 if (sk_is_unix(sk)) 1590 sk->sk_scm_rights = valbool; 1591 else 1592 ret = -EOPNOTSUPP; 1593 break; 1594 1595 case SO_INCOMING_CPU: 1596 reuseport_update_incoming_cpu(sk, val); 1597 break; 1598 1599 case SO_CNX_ADVICE: 1600 if (val == 1) 1601 dst_negative_advice(sk); 1602 break; 1603 1604 case SO_ZEROCOPY: 1605 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1606 if (!(sk_is_tcp(sk) || 1607 (sk->sk_type == SOCK_DGRAM && 1608 sk->sk_protocol == IPPROTO_UDP))) 1609 ret = -EOPNOTSUPP; 1610 } else if (sk->sk_family != PF_RDS) { 1611 ret = -EOPNOTSUPP; 1612 } 1613 if (!ret) { 1614 if (val < 0 || val > 1) 1615 ret = -EINVAL; 1616 else 1617 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1618 } 1619 break; 1620 1621 case SO_TXTIME: 1622 if (optlen != sizeof(struct sock_txtime)) { 1623 ret = -EINVAL; 1624 break; 1625 } else if (copy_from_sockptr(&sk_txtime, optval, 1626 sizeof(struct sock_txtime))) { 1627 ret = -EFAULT; 1628 break; 1629 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1630 ret = -EINVAL; 1631 break; 1632 } 1633 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1634 * scheduler has enough safe guards. 1635 */ 1636 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1637 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1638 ret = -EPERM; 1639 break; 1640 } 1641 1642 ret = sockopt_validate_clockid(sk_txtime.clockid); 1643 if (ret) 1644 break; 1645 1646 sock_valbool_flag(sk, SOCK_TXTIME, true); 1647 sk->sk_clockid = sk_txtime.clockid; 1648 sk->sk_txtime_deadline_mode = 1649 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1650 sk->sk_txtime_report_errors = 1651 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1652 break; 1653 1654 case SO_BINDTOIFINDEX: 1655 ret = sock_bindtoindex_locked(sk, val); 1656 break; 1657 1658 case SO_BUF_LOCK: 1659 if (val & ~SOCK_BUF_LOCK_MASK) { 1660 ret = -EINVAL; 1661 break; 1662 } 1663 sk->sk_userlocks = val | (sk->sk_userlocks & 1664 ~SOCK_BUF_LOCK_MASK); 1665 break; 1666 1667 case SO_RESERVE_MEM: 1668 { 1669 int delta; 1670 1671 if (val < 0 || val > SZ_1G) { 1672 ret = -EINVAL; 1673 break; 1674 } 1675 1676 delta = val - sk->sk_reserved_mem; 1677 if (delta < 0) 1678 sock_release_reserved_memory(sk, -delta); 1679 else 1680 ret = sock_reserve_memory(sk, delta); 1681 break; 1682 } 1683 1684 default: 1685 ret = -ENOPROTOOPT; 1686 break; 1687 } 1688 sockopt_release_sock(sk); 1689 return ret; 1690 } 1691 1692 int sock_setsockopt(struct socket *sock, int level, int optname, 1693 sockptr_t optval, unsigned int optlen) 1694 { 1695 return sk_setsockopt(sock->sk, level, optname, 1696 optval, optlen); 1697 } 1698 EXPORT_SYMBOL(sock_setsockopt); 1699 1700 static const struct cred *sk_get_peer_cred(struct sock *sk) 1701 { 1702 const struct cred *cred; 1703 1704 spin_lock(&sk->sk_peer_lock); 1705 cred = get_cred(sk->sk_peer_cred); 1706 spin_unlock(&sk->sk_peer_lock); 1707 1708 return cred; 1709 } 1710 1711 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1712 struct ucred *ucred) 1713 { 1714 ucred->pid = pid_vnr(pid); 1715 ucred->uid = ucred->gid = -1; 1716 if (cred) { 1717 struct user_namespace *current_ns = current_user_ns(); 1718 1719 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1720 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1721 } 1722 } 1723 1724 static int groups_to_user(sockptr_t dst, const struct group_info *src) 1725 { 1726 struct user_namespace *user_ns = current_user_ns(); 1727 int i; 1728 1729 for (i = 0; i < src->ngroups; i++) { 1730 gid_t gid = from_kgid_munged(user_ns, src->gid[i]); 1731 1732 if (copy_to_sockptr_offset(dst, i * sizeof(gid), &gid, sizeof(gid))) 1733 return -EFAULT; 1734 } 1735 1736 return 0; 1737 } 1738 1739 /* Hand out a pidfd for @type of the socket's peer via SO_PEERPIDFD*. */ 1740 static int sk_getsockopt_peerpidfd(struct sock *sk, enum pid_type type, 1741 sockptr_t optval, sockptr_t optlen, int len) 1742 { 1743 struct file *pidfd_file = NULL; 1744 unsigned int flags = 0; 1745 struct pid *peer_pid; 1746 int pidfd; 1747 1748 if (len > sizeof(pidfd)) 1749 len = sizeof(pidfd); 1750 1751 spin_lock(&sk->sk_peer_lock); 1752 peer_pid = get_pid(sk->sk_peer_pid[type]); 1753 spin_unlock(&sk->sk_peer_lock); 1754 1755 if (!peer_pid) 1756 return -ENODATA; 1757 1758 /* The use of PIDFD_STALE requires stashing of struct pid on pidfs 1759 * with pidfs_register_pid() and only AF_UNIX is prepared for this. 1760 */ 1761 if (sk_is_unix(sk)) 1762 flags |= PIDFD_STALE; 1763 if (type == PIDTYPE_PID) 1764 flags |= PIDFD_THREAD; 1765 1766 pidfd = pidfd_prepare(peer_pid, flags, &pidfd_file); 1767 put_pid(peer_pid); 1768 if (pidfd < 0) 1769 return pidfd; 1770 1771 if (copy_to_sockptr(optval, &pidfd, len) || 1772 copy_to_sockptr(optlen, &len, sizeof(int))) { 1773 put_unused_fd(pidfd); 1774 fput(pidfd_file); 1775 1776 return -EFAULT; 1777 } 1778 1779 fd_install(pidfd, pidfd_file); 1780 return 0; 1781 } 1782 1783 int sk_getsockopt(struct sock *sk, int level, int optname, 1784 sockptr_t optval, sockptr_t optlen) 1785 { 1786 struct socket *sock = sk->sk_socket; 1787 1788 union { 1789 int val; 1790 u64 val64; 1791 unsigned long ulval; 1792 struct linger ling; 1793 struct old_timeval32 tm32; 1794 struct __kernel_old_timeval tm; 1795 struct __kernel_sock_timeval stm; 1796 struct sock_txtime txtime; 1797 struct so_timestamping timestamping; 1798 } v; 1799 1800 int lv = sizeof(int); 1801 int len; 1802 1803 if (copy_from_sockptr(&len, optlen, sizeof(int))) 1804 return -EFAULT; 1805 if (len < 0) 1806 return -EINVAL; 1807 1808 memset(&v, 0, sizeof(v)); 1809 1810 switch (optname) { 1811 case SO_DEBUG: 1812 v.val = sock_flag(sk, SOCK_DBG); 1813 break; 1814 1815 case SO_DONTROUTE: 1816 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1817 break; 1818 1819 case SO_BROADCAST: 1820 v.val = sock_flag(sk, SOCK_BROADCAST); 1821 break; 1822 1823 case SO_SNDBUF: 1824 v.val = READ_ONCE(sk->sk_sndbuf); 1825 break; 1826 1827 case SO_RCVBUF: 1828 v.val = READ_ONCE(sk->sk_rcvbuf); 1829 break; 1830 1831 case SO_REUSEADDR: 1832 v.val = sk->sk_reuse; 1833 break; 1834 1835 case SO_REUSEPORT: 1836 v.val = sk->sk_reuseport; 1837 break; 1838 1839 case SO_KEEPALIVE: 1840 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1841 break; 1842 1843 case SO_TYPE: 1844 v.val = sk->sk_type; 1845 break; 1846 1847 case SO_PROTOCOL: 1848 v.val = sk->sk_protocol; 1849 break; 1850 1851 case SO_DOMAIN: 1852 v.val = sk->sk_family; 1853 break; 1854 1855 case SO_ERROR: 1856 v.val = -sock_error(sk); 1857 if (v.val == 0) 1858 v.val = xchg(&sk->sk_err_soft, 0); 1859 break; 1860 1861 case SO_OOBINLINE: 1862 v.val = sock_flag(sk, SOCK_URGINLINE); 1863 break; 1864 1865 case SO_NO_CHECK: 1866 v.val = sk->sk_no_check_tx; 1867 break; 1868 1869 case SO_PRIORITY: 1870 v.val = READ_ONCE(sk->sk_priority); 1871 break; 1872 1873 case SO_LINGER: 1874 lv = sizeof(v.ling); 1875 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1876 v.ling.l_linger = READ_ONCE(sk->sk_lingertime) / HZ; 1877 break; 1878 1879 case SO_BSDCOMPAT: 1880 break; 1881 1882 case SO_TIMESTAMP_OLD: 1883 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1884 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1885 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1886 break; 1887 1888 case SO_TIMESTAMPNS_OLD: 1889 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1890 break; 1891 1892 case SO_TIMESTAMP_NEW: 1893 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1894 break; 1895 1896 case SO_TIMESTAMPNS_NEW: 1897 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1898 break; 1899 1900 case SO_TIMESTAMPING_OLD: 1901 case SO_TIMESTAMPING_NEW: 1902 lv = sizeof(v.timestamping); 1903 /* For the later-added case SO_TIMESTAMPING_NEW: Be strict about only 1904 * returning the flags when they were set through the same option. 1905 * Don't change the beviour for the old case SO_TIMESTAMPING_OLD. 1906 */ 1907 if (optname == SO_TIMESTAMPING_OLD || sock_flag(sk, SOCK_TSTAMP_NEW)) { 1908 v.timestamping.flags = READ_ONCE(sk->sk_tsflags); 1909 v.timestamping.bind_phc = READ_ONCE(sk->sk_bind_phc); 1910 } 1911 break; 1912 1913 case SO_RCVTIMEO_OLD: 1914 case SO_RCVTIMEO_NEW: 1915 lv = sock_get_timeout(READ_ONCE(sk->sk_rcvtimeo), &v, 1916 SO_RCVTIMEO_OLD == optname); 1917 break; 1918 1919 case SO_SNDTIMEO_OLD: 1920 case SO_SNDTIMEO_NEW: 1921 lv = sock_get_timeout(READ_ONCE(sk->sk_sndtimeo), &v, 1922 SO_SNDTIMEO_OLD == optname); 1923 break; 1924 1925 case SO_RCVLOWAT: 1926 v.val = READ_ONCE(sk->sk_rcvlowat); 1927 break; 1928 1929 case SO_SNDLOWAT: 1930 v.val = 1; 1931 break; 1932 1933 case SO_PASSCRED: 1934 if (!sk_may_scm_recv(sk)) 1935 return -EOPNOTSUPP; 1936 1937 v.val = sk->sk_scm_credentials; 1938 break; 1939 1940 case SO_PASSPIDFD: 1941 if (!sk_is_unix(sk)) 1942 return -EOPNOTSUPP; 1943 1944 v.val = sk->sk_scm_pidfd; 1945 break; 1946 1947 case SO_PASSPIDFD_THREAD: 1948 if (!sk_is_unix(sk)) 1949 return -EOPNOTSUPP; 1950 1951 v.val = sk->sk_scm_pidfd_thread; 1952 break; 1953 1954 case SO_PASSRIGHTS: 1955 if (!sk_is_unix(sk)) 1956 return -EOPNOTSUPP; 1957 1958 v.val = sk->sk_scm_rights; 1959 break; 1960 1961 case SO_PEERCRED: 1962 { 1963 struct ucred peercred; 1964 if (len > sizeof(peercred)) 1965 len = sizeof(peercred); 1966 1967 spin_lock(&sk->sk_peer_lock); 1968 cred_to_ucred(sk->sk_peer_pid[PIDTYPE_TGID], sk->sk_peer_cred, 1969 &peercred); 1970 spin_unlock(&sk->sk_peer_lock); 1971 1972 if (copy_to_sockptr(optval, &peercred, len)) 1973 return -EFAULT; 1974 goto lenout; 1975 } 1976 1977 case SO_PEERPIDFD: 1978 return sk_getsockopt_peerpidfd(sk, PIDTYPE_TGID, optval, optlen, len); 1979 1980 case SO_PEERPIDFD_THREAD: 1981 /* Only AF_UNIX records the peer's connecting thread. */ 1982 if (!sk_is_unix(sk)) 1983 return -EOPNOTSUPP; 1984 1985 return sk_getsockopt_peerpidfd(sk, PIDTYPE_PID, optval, optlen, len); 1986 1987 case SO_PEERGROUPS: 1988 { 1989 const struct cred *cred; 1990 int ret, n; 1991 1992 cred = sk_get_peer_cred(sk); 1993 if (!cred) 1994 return -ENODATA; 1995 1996 n = cred->group_info->ngroups; 1997 if (len < n * sizeof(gid_t)) { 1998 len = n * sizeof(gid_t); 1999 put_cred(cred); 2000 return copy_to_sockptr(optlen, &len, sizeof(int)) ? -EFAULT : -ERANGE; 2001 } 2002 len = n * sizeof(gid_t); 2003 2004 ret = groups_to_user(optval, cred->group_info); 2005 put_cred(cred); 2006 if (ret) 2007 return ret; 2008 goto lenout; 2009 } 2010 2011 case SO_PEERNAME: 2012 { 2013 struct sockaddr_storage address; 2014 2015 lv = READ_ONCE(sock->ops)->getname(sock, (struct sockaddr *)&address, 2); 2016 if (lv < 0) 2017 return -ENOTCONN; 2018 if (lv < len) 2019 return -EINVAL; 2020 if (copy_to_sockptr(optval, &address, len)) 2021 return -EFAULT; 2022 goto lenout; 2023 } 2024 2025 /* Dubious BSD thing... Probably nobody even uses it, but 2026 * the UNIX standard wants it for whatever reason... -DaveM 2027 */ 2028 case SO_ACCEPTCONN: 2029 v.val = sk->sk_state == TCP_LISTEN; 2030 break; 2031 2032 case SO_PASSSEC: 2033 if (!IS_ENABLED(CONFIG_SECURITY_NETWORK) || !sk_may_scm_recv(sk)) 2034 return -EOPNOTSUPP; 2035 2036 v.val = sk->sk_scm_security; 2037 break; 2038 2039 case SO_PEERSEC: 2040 return security_socket_getpeersec_stream(sock, 2041 optval, optlen, len); 2042 2043 case SO_MARK: 2044 v.val = READ_ONCE(sk->sk_mark); 2045 break; 2046 2047 case SO_RCVMARK: 2048 v.val = sock_flag(sk, SOCK_RCVMARK); 2049 break; 2050 2051 case SO_RCVPRIORITY: 2052 v.val = sock_flag(sk, SOCK_RCVPRIORITY); 2053 break; 2054 2055 case SO_RXQ_OVFL: 2056 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 2057 break; 2058 2059 case SO_WIFI_STATUS: 2060 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 2061 break; 2062 2063 case SO_PEEK_OFF: 2064 if (!READ_ONCE(sock->ops)->set_peek_off) 2065 return -EOPNOTSUPP; 2066 2067 v.val = READ_ONCE(sk->sk_peek_off); 2068 break; 2069 case SO_NOFCS: 2070 v.val = sock_flag(sk, SOCK_NOFCS); 2071 break; 2072 2073 case SO_BINDTODEVICE: 2074 return sock_getbindtodevice(sk, optval, optlen, len); 2075 2076 case SO_GET_FILTER: 2077 len = sk_get_filter(sk, optval, len); 2078 if (len < 0) 2079 return len; 2080 2081 goto lenout; 2082 2083 case SO_LOCK_FILTER: 2084 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 2085 break; 2086 2087 case SO_BPF_EXTENSIONS: 2088 v.val = bpf_tell_extensions(); 2089 break; 2090 2091 case SO_SELECT_ERR_QUEUE: 2092 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 2093 break; 2094 2095 #ifdef CONFIG_NET_RX_BUSY_POLL 2096 case SO_BUSY_POLL: 2097 v.val = READ_ONCE(sk->sk_ll_usec); 2098 break; 2099 case SO_PREFER_BUSY_POLL: 2100 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 2101 break; 2102 #endif 2103 2104 case SO_MAX_PACING_RATE: 2105 /* The READ_ONCE() pair with the WRITE_ONCE() in sk_setsockopt() */ 2106 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 2107 lv = sizeof(v.ulval); 2108 v.ulval = READ_ONCE(sk->sk_max_pacing_rate); 2109 } else { 2110 /* 32bit version */ 2111 v.val = min_t(unsigned long, ~0U, 2112 READ_ONCE(sk->sk_max_pacing_rate)); 2113 } 2114 break; 2115 2116 case SO_INCOMING_CPU: 2117 v.val = READ_ONCE(sk->sk_incoming_cpu); 2118 break; 2119 2120 case SO_MEMINFO: 2121 { 2122 u32 meminfo[SK_MEMINFO_VARS]; 2123 2124 sk_get_meminfo(sk, meminfo); 2125 2126 len = min_t(unsigned int, len, sizeof(meminfo)); 2127 if (copy_to_sockptr(optval, &meminfo, len)) 2128 return -EFAULT; 2129 2130 goto lenout; 2131 } 2132 2133 #ifdef CONFIG_NET_RX_BUSY_POLL 2134 case SO_INCOMING_NAPI_ID: 2135 v.val = READ_ONCE(sk->sk_napi_id); 2136 2137 /* aggregate non-NAPI IDs down to 0 */ 2138 if (!napi_id_valid(v.val)) 2139 v.val = 0; 2140 2141 break; 2142 #endif 2143 2144 case SO_COOKIE: 2145 lv = sizeof(u64); 2146 if (len < lv) 2147 return -EINVAL; 2148 v.val64 = sock_gen_cookie(sk); 2149 break; 2150 2151 case SO_ZEROCOPY: 2152 v.val = sock_flag(sk, SOCK_ZEROCOPY); 2153 break; 2154 2155 case SO_TXTIME: 2156 lv = sizeof(v.txtime); 2157 v.txtime.clockid = sk->sk_clockid; 2158 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 2159 SOF_TXTIME_DEADLINE_MODE : 0; 2160 v.txtime.flags |= sk->sk_txtime_report_errors ? 2161 SOF_TXTIME_REPORT_ERRORS : 0; 2162 break; 2163 2164 case SO_BINDTOIFINDEX: 2165 v.val = READ_ONCE(sk->sk_bound_dev_if); 2166 break; 2167 2168 case SO_NETNS_COOKIE: 2169 lv = sizeof(u64); 2170 if (len != lv) 2171 return -EINVAL; 2172 v.val64 = sock_net(sk)->net_cookie; 2173 break; 2174 2175 case SO_BUF_LOCK: 2176 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 2177 break; 2178 2179 case SO_RESERVE_MEM: 2180 v.val = READ_ONCE(sk->sk_reserved_mem); 2181 break; 2182 2183 case SO_TXREHASH: 2184 if (!sk_is_tcp(sk)) 2185 return -EOPNOTSUPP; 2186 2187 /* Paired with WRITE_ONCE() in sk_setsockopt() */ 2188 v.val = READ_ONCE(sk->sk_txrehash); 2189 break; 2190 2191 default: 2192 /* We implement the SO_SNDLOWAT etc to not be settable 2193 * (1003.1g 7). 2194 */ 2195 return -ENOPROTOOPT; 2196 } 2197 2198 if (len > lv) 2199 len = lv; 2200 if (copy_to_sockptr(optval, &v, len)) 2201 return -EFAULT; 2202 lenout: 2203 if (copy_to_sockptr(optlen, &len, sizeof(int))) 2204 return -EFAULT; 2205 return 0; 2206 } 2207 2208 /* 2209 * Initialize an sk_lock. 2210 * 2211 * (We also register the sk_lock with the lock validator.) 2212 */ 2213 static inline void sock_lock_init(struct sock *sk) 2214 { 2215 sk_owner_clear(sk); 2216 2217 if (sk->sk_kern_sock) 2218 sock_lock_init_class_and_name( 2219 sk, 2220 af_family_kern_slock_key_strings[sk->sk_family], 2221 af_family_kern_slock_keys + sk->sk_family, 2222 af_family_kern_key_strings[sk->sk_family], 2223 af_family_kern_keys + sk->sk_family); 2224 else 2225 sock_lock_init_class_and_name( 2226 sk, 2227 af_family_slock_key_strings[sk->sk_family], 2228 af_family_slock_keys + sk->sk_family, 2229 af_family_key_strings[sk->sk_family], 2230 af_family_keys + sk->sk_family); 2231 } 2232 2233 /* 2234 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 2235 * even temporarily, because of RCU lookups. sk_node should also be left as is. 2236 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 2237 */ 2238 static void sock_copy(struct sock *nsk, const struct sock *osk) 2239 { 2240 const struct proto *prot = READ_ONCE(osk->sk_prot); 2241 #ifdef CONFIG_SECURITY_NETWORK 2242 void *sptr = nsk->sk_security; 2243 #endif 2244 2245 /* If we move sk_tx_queue_mapping out of the private section, 2246 * we must check if sk_tx_queue_clear() is called after 2247 * sock_copy() in sk_clone_lock(). 2248 */ 2249 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 2250 offsetof(struct sock, sk_dontcopy_begin) || 2251 offsetof(struct sock, sk_tx_queue_mapping) >= 2252 offsetof(struct sock, sk_dontcopy_end)); 2253 2254 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 2255 2256 unsafe_memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 2257 prot->obj_size - offsetof(struct sock, sk_dontcopy_end), 2258 /* alloc is larger than struct, see sk_prot_alloc() */); 2259 2260 #ifdef CONFIG_SECURITY_NETWORK 2261 nsk->sk_security = sptr; 2262 security_sk_clone(osk, nsk); 2263 #endif 2264 } 2265 2266 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2267 int family) 2268 { 2269 struct sock *sk; 2270 struct kmem_cache *slab; 2271 2272 slab = prot->slab; 2273 if (slab != NULL) { 2274 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2275 if (!sk) 2276 return sk; 2277 if (want_init_on_alloc(priority)) 2278 sk_prot_clear_nulls(sk, prot->obj_size); 2279 } else 2280 sk = kmalloc(prot->obj_size, priority); 2281 2282 if (sk != NULL) { 2283 if (security_sk_alloc(sk, family, priority)) 2284 goto out_free; 2285 2286 if (!try_module_get(prot->owner)) 2287 goto out_free_sec; 2288 } 2289 2290 return sk; 2291 2292 out_free_sec: 2293 security_sk_free(sk); 2294 out_free: 2295 if (slab != NULL) 2296 kmem_cache_free(slab, sk); 2297 else 2298 kfree(sk); 2299 return NULL; 2300 } 2301 2302 static void sk_prot_free(struct proto *prot, struct sock *sk) 2303 { 2304 struct kmem_cache *slab; 2305 struct module *owner; 2306 2307 owner = prot->owner; 2308 slab = prot->slab; 2309 2310 cgroup_sk_free(&sk->sk_cgrp_data); 2311 mem_cgroup_sk_free(sk); 2312 security_sk_free(sk); 2313 2314 sk_owner_put(sk); 2315 2316 if (slab != NULL) 2317 kmem_cache_free(slab, sk); 2318 else 2319 kfree(sk); 2320 module_put(owner); 2321 } 2322 2323 /** 2324 * sk_alloc - All socket objects are allocated here 2325 * @net: the applicable net namespace 2326 * @family: protocol family 2327 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2328 * @prot: struct proto associated with this new sock instance 2329 * @kern: is this to be a kernel socket? 2330 */ 2331 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2332 struct proto *prot, int kern) 2333 { 2334 struct sock *sk; 2335 2336 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2337 if (sk) { 2338 sk->sk_family = family; 2339 /* 2340 * See comment in struct sock definition to understand 2341 * why we need sk_prot_creator -acme 2342 */ 2343 sk->sk_prot = sk->sk_prot_creator = prot; 2344 2345 if (READ_ONCE(net->core.sysctl_bypass_prot_mem)) 2346 sk->sk_bypass_prot_mem = 1; 2347 2348 sk->sk_kern_sock = kern; 2349 sock_lock_init(sk); 2350 2351 sk->sk_net_refcnt = kern ? 0 : 1; 2352 if (likely(sk->sk_net_refcnt)) { 2353 get_net_track(net, &sk->ns_tracker, priority); 2354 sock_inuse_add(net, 1); 2355 } else { 2356 net_passive_inc(net); 2357 __netns_tracker_alloc(net, &sk->ns_tracker, 2358 false, priority); 2359 } 2360 2361 sock_net_set(sk, net); 2362 refcount_set(&sk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS); 2363 2364 mem_cgroup_sk_alloc(sk); 2365 cgroup_sk_alloc(&sk->sk_cgrp_data); 2366 sock_update_classid(&sk->sk_cgrp_data); 2367 sock_update_netprioidx(&sk->sk_cgrp_data); 2368 sk_tx_queue_clear(sk); 2369 } 2370 2371 return sk; 2372 } 2373 EXPORT_SYMBOL(sk_alloc); 2374 2375 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2376 * grace period. This is the case for UDP sockets and TCP listeners. 2377 */ 2378 static void __sk_destruct(struct rcu_head *head) 2379 { 2380 struct sock *sk = container_of(head, struct sock, sk_rcu); 2381 struct net *net = sock_net(sk); 2382 struct sk_filter *filter; 2383 2384 if (sk->sk_destruct) 2385 sk->sk_destruct(sk); 2386 2387 filter = rcu_dereference_check(sk->sk_filter, 2388 refcount_read(&sk->sk_wmem_alloc) == 0); 2389 if (filter) { 2390 sk_filter_uncharge(sk, filter); 2391 RCU_INIT_POINTER(sk->sk_filter, NULL); 2392 } 2393 2394 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2395 2396 #ifdef CONFIG_BPF_SYSCALL 2397 bpf_sk_storage_free(sk); 2398 #endif 2399 2400 if (atomic_read(&sk->sk_omem_alloc)) 2401 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2402 __func__, atomic_read(&sk->sk_omem_alloc)); 2403 2404 if (sk->sk_frag.page) { 2405 put_page(sk->sk_frag.page); 2406 sk->sk_frag.page = NULL; 2407 } 2408 2409 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2410 put_cred(sk->sk_peer_cred); 2411 put_pids(sk->sk_peer_pid); 2412 2413 if (likely(sk->sk_net_refcnt)) { 2414 put_net_track(net, &sk->ns_tracker); 2415 } else { 2416 __netns_tracker_free(net, &sk->ns_tracker, false); 2417 net_passive_dec(net); 2418 } 2419 sk_prot_free(sk->sk_prot_creator, sk); 2420 } 2421 2422 void sk_net_refcnt_upgrade(struct sock *sk) 2423 { 2424 struct net *net = sock_net(sk); 2425 2426 WARN_ON_ONCE(sk->sk_net_refcnt); 2427 __netns_tracker_free(net, &sk->ns_tracker, false); 2428 net_passive_dec(net); 2429 sk->sk_net_refcnt = 1; 2430 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 2431 sock_inuse_add(net, 1); 2432 } 2433 EXPORT_SYMBOL_GPL(sk_net_refcnt_upgrade); 2434 2435 void sk_destruct(struct sock *sk) 2436 { 2437 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2438 2439 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2440 reuseport_detach_sock(sk); 2441 use_call_rcu = true; 2442 } 2443 2444 if (use_call_rcu) 2445 call_rcu(&sk->sk_rcu, __sk_destruct); 2446 else 2447 __sk_destruct(&sk->sk_rcu); 2448 } 2449 2450 static void __sk_free(struct sock *sk) 2451 { 2452 if (likely(sk->sk_net_refcnt)) 2453 sock_inuse_add(sock_net(sk), -1); 2454 2455 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2456 sock_diag_broadcast_destroy(sk); 2457 else 2458 sk_destruct(sk); 2459 } 2460 2461 void sk_free(struct sock *sk) 2462 { 2463 /* 2464 * We subtract one from sk_wmem_alloc and can know if 2465 * some packets are still in some tx queue. 2466 * If not null, sock_wfree() will call __sk_free(sk) later 2467 */ 2468 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2469 __sk_free(sk); 2470 } 2471 EXPORT_SYMBOL(sk_free); 2472 2473 static void sk_init_common(struct sock *sk) 2474 { 2475 skb_queue_head_init(&sk->sk_receive_queue); 2476 skb_queue_head_init(&sk->sk_write_queue); 2477 skb_queue_head_init(&sk->sk_error_queue); 2478 2479 rwlock_init(&sk->sk_callback_lock); 2480 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2481 af_rlock_keys + sk->sk_family, 2482 af_family_rlock_key_strings[sk->sk_family]); 2483 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2484 af_wlock_keys + sk->sk_family, 2485 af_family_wlock_key_strings[sk->sk_family]); 2486 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2487 af_elock_keys + sk->sk_family, 2488 af_family_elock_key_strings[sk->sk_family]); 2489 if (sk->sk_kern_sock) 2490 lockdep_set_class_and_name(&sk->sk_callback_lock, 2491 af_kern_callback_keys + sk->sk_family, 2492 af_family_kern_clock_key_strings[sk->sk_family]); 2493 else 2494 lockdep_set_class_and_name(&sk->sk_callback_lock, 2495 af_callback_keys + sk->sk_family, 2496 af_family_clock_key_strings[sk->sk_family]); 2497 } 2498 2499 /** 2500 * sk_clone - clone a socket 2501 * @sk: the socket to clone 2502 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2503 * @lock: if true, lock the cloned sk 2504 * 2505 * If @lock is true, the clone is locked by bh_lock_sock(), and 2506 * caller must unlock socket even in error path by bh_unlock_sock(). 2507 */ 2508 struct sock *sk_clone(const struct sock *sk, const gfp_t priority, 2509 bool lock) 2510 { 2511 struct proto *prot = READ_ONCE(sk->sk_prot); 2512 struct sk_filter *filter; 2513 bool is_charged = true; 2514 struct sock *newsk; 2515 2516 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2517 if (!newsk) 2518 goto out; 2519 2520 sock_copy(newsk, sk); 2521 2522 newsk->sk_prot_creator = prot; 2523 #ifdef CONFIG_BPF_SYSCALL 2524 RCU_INIT_POINTER(newsk->sk_bpf_storage, NULL); 2525 #endif 2526 #if IS_ENABLED(CONFIG_INET_PSP) 2527 RCU_INIT_POINTER(newsk->psp_assoc, NULL); 2528 #endif 2529 2530 /* SANITY */ 2531 if (likely(newsk->sk_net_refcnt)) { 2532 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2533 sock_inuse_add(sock_net(newsk), 1); 2534 } else { 2535 /* Kernel sockets are not elevating the struct net refcount. 2536 * Instead, use a tracker to more easily detect if a layer 2537 * is not properly dismantling its kernel sockets at netns 2538 * destroy time. 2539 */ 2540 net_passive_inc(sock_net(newsk)); 2541 __netns_tracker_alloc(sock_net(newsk), &newsk->ns_tracker, 2542 false, priority); 2543 } 2544 2545 sk_node_init(&newsk->sk_node); 2546 sock_lock_init(newsk); 2547 2548 if (lock) 2549 bh_lock_sock(newsk); 2550 2551 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2552 newsk->sk_backlog.len = 0; 2553 2554 atomic_set(&newsk->sk_rmem_alloc, 0); 2555 2556 refcount_set(&newsk->sk_wmem_alloc, SK_WMEM_ALLOC_BIAS); 2557 2558 atomic_set(&newsk->sk_omem_alloc, 0); 2559 sk_init_common(newsk); 2560 2561 newsk->sk_dst_cache = NULL; 2562 newsk->sk_dst_pending_confirm = 0; 2563 newsk->sk_wmem_queued = 0; 2564 newsk->sk_forward_alloc = 0; 2565 newsk->sk_reserved_mem = 0; 2566 DEBUG_NET_WARN_ON_ONCE(newsk->sk_drop_counters); 2567 sk_drops_reset(newsk); 2568 newsk->sk_send_head = NULL; 2569 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2570 atomic_set(&newsk->sk_zckey, 0); 2571 2572 sock_reset_flag(newsk, SOCK_DONE); 2573 2574 #ifdef CONFIG_MEMCG 2575 /* sk->sk_memcg will be populated at accept() time */ 2576 newsk->sk_memcg = NULL; 2577 #endif 2578 2579 cgroup_sk_clone(&newsk->sk_cgrp_data); 2580 2581 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2582 2583 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2584 net_enable_timestamp(); 2585 2586 rcu_read_lock(); 2587 filter = rcu_dereference(sk->sk_filter); 2588 if (filter != NULL) 2589 /* though it's an empty new sock, the charging may fail 2590 * if sysctl_optmem_max was changed between creation of 2591 * original socket and cloning 2592 */ 2593 is_charged = sk_filter_charge(newsk, filter); 2594 RCU_INIT_POINTER(newsk->sk_filter, filter); 2595 rcu_read_unlock(); 2596 2597 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2598 /* We need to make sure that we don't uncharge the new 2599 * socket if we couldn't charge it in the first place 2600 * as otherwise we uncharge the parent's filter. 2601 */ 2602 if (!is_charged) 2603 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2604 2605 goto free; 2606 } 2607 2608 if (bpf_sk_storage_clone(sk, newsk)) 2609 goto free; 2610 2611 /* Clear sk_user_data if parent had the pointer tagged 2612 * as not suitable for copying when cloning. 2613 */ 2614 if (sk_user_data_is_nocopy(newsk)) 2615 newsk->sk_user_data = NULL; 2616 2617 newsk->sk_err = 0; 2618 newsk->sk_err_soft = 0; 2619 newsk->sk_priority = 0; 2620 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2621 2622 /* Before updating sk_refcnt, we must commit prior changes to memory 2623 * (Documentation/RCU/rculist_nulls.rst for details) 2624 */ 2625 smp_wmb(); 2626 refcount_set(&newsk->sk_refcnt, 2); 2627 2628 sk_set_socket(newsk, NULL); 2629 sk_tx_queue_clear(newsk); 2630 sk_rx_queue_clear(newsk); 2631 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2632 2633 if (newsk->sk_prot->sockets_allocated) 2634 sk_sockets_allocated_inc(newsk); 2635 out: 2636 return newsk; 2637 free: 2638 /* It is still raw copy of parent, so invalidate 2639 * destructor and make plain sk_free() 2640 */ 2641 newsk->sk_destruct = NULL; 2642 if (lock) 2643 bh_unlock_sock(newsk); 2644 sk_free(newsk); 2645 newsk = NULL; 2646 goto out; 2647 } 2648 EXPORT_SYMBOL_GPL(sk_clone); 2649 2650 static u32 sk_dst_gso_max_size(struct sock *sk, const struct net_device *dev) 2651 { 2652 bool is_ipv6 = false; 2653 u32 max_size; 2654 2655 #if IS_ENABLED(CONFIG_IPV6) 2656 is_ipv6 = (sk->sk_family == AF_INET6 && 2657 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)); 2658 #endif 2659 /* pairs with the WRITE_ONCE() in netif_set_gso(_ipv4)_max_size() */ 2660 max_size = is_ipv6 ? READ_ONCE(dev->gso_max_size) : 2661 READ_ONCE(dev->gso_ipv4_max_size); 2662 if (max_size > GSO_LEGACY_MAX_SIZE && !sk_is_tcp(sk)) 2663 max_size = GSO_LEGACY_MAX_SIZE; 2664 2665 return max_size - (MAX_TCP_HEADER + 1); 2666 } 2667 2668 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2669 { 2670 const struct net_device *dev; 2671 u32 max_segs = 1; 2672 2673 rcu_read_lock(); 2674 dev = dst_dev_rcu(dst); 2675 sk->sk_route_caps = dev->features; 2676 if (sk_is_tcp(sk)) { 2677 struct inet_connection_sock *icsk = inet_csk(sk); 2678 2679 sk->sk_route_caps |= NETIF_F_GSO; 2680 icsk->icsk_ack.dst_quick_ack = dst_metric(dst, RTAX_QUICKACK); 2681 } 2682 if (sk->sk_route_caps & NETIF_F_GSO) 2683 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2684 if (unlikely(sk->sk_gso_disabled)) 2685 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2686 if (sk_can_gso(sk)) { 2687 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2688 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2689 } else { 2690 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2691 sk->sk_gso_max_size = sk_dst_gso_max_size(sk, dev); 2692 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2693 max_segs = max_t(u32, READ_ONCE(dev->gso_max_segs), 1); 2694 } 2695 } 2696 sk->sk_gso_max_segs = max_segs; 2697 sk_dst_set(sk, dst); 2698 rcu_read_unlock(); 2699 } 2700 EXPORT_SYMBOL_GPL(sk_setup_caps); 2701 2702 bool sk_has_decrypt_user(const struct sock *sk) 2703 { 2704 return psp_sk_assoc(sk) || 2705 (sk_is_inet(sk) && inet_csk_has_ulp(sk)); /* for tls */ 2706 } 2707 2708 /* 2709 * Simple resource managers for sockets. 2710 */ 2711 2712 2713 /* 2714 * Write buffer destructor automatically called from kfree_skb. 2715 */ 2716 void sock_wfree(struct sk_buff *skb) 2717 { 2718 unsigned int len = skb->truesize; 2719 struct sock *sk = skb->sk; 2720 bool free; 2721 int old; 2722 2723 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2724 void (*sk_write_space)(struct sock *sk); 2725 2726 sk_write_space = READ_ONCE(sk->sk_write_space); 2727 2728 if (sock_flag(sk, SOCK_RCU_FREE) && 2729 sk_write_space == sock_def_write_space) { 2730 rcu_read_lock(); 2731 free = __refcount_sub_and_test(len, &sk->sk_wmem_alloc, 2732 &old); 2733 sock_def_write_space_wfree(sk, old - len); 2734 rcu_read_unlock(); 2735 if (unlikely(free)) 2736 __sk_free(sk); 2737 return; 2738 } 2739 2740 /* 2741 * Keep a reference on sk_wmem_alloc, this will be released 2742 * after sk_write_space() call 2743 */ 2744 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2745 sk_write_space(sk); 2746 len = 1; 2747 } 2748 /* 2749 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2750 * could not do because of in-flight packets 2751 */ 2752 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2753 __sk_free(sk); 2754 } 2755 EXPORT_SYMBOL(sock_wfree); 2756 2757 /* This variant of sock_wfree() is used by TCP, 2758 * since it sets SOCK_USE_WRITE_QUEUE. 2759 */ 2760 #ifdef CONFIG_INET 2761 void __sock_wfree(struct sk_buff *skb) 2762 { 2763 struct sock *sk = skb->sk; 2764 2765 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2766 __sk_free(sk); 2767 } 2768 EXPORT_SYMBOL_GPL(__sock_wfree); 2769 #endif 2770 2771 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2772 { 2773 int old_wmem; 2774 2775 skb_orphan(skb); 2776 #ifdef CONFIG_INET 2777 if (unlikely(!sk_fullsock(sk))) 2778 return skb_set_owner_edemux(skb, sk); 2779 #endif 2780 skb->sk = sk; 2781 skb->destructor = sock_wfree; 2782 skb_set_hash_from_sk(skb, sk); 2783 /* 2784 * We used to take a refcount on sk, but following operation 2785 * is enough to guarantee sk_free() won't free this sock until 2786 * all in-flight packets are completed 2787 */ 2788 __refcount_add(skb->truesize, &sk->sk_wmem_alloc, &old_wmem); 2789 2790 /* (old_wmem == SK_WMEM_ALLOC_BIAS) if no other TX packet for this socket 2791 * is in a host queue (qdisc, NIC queue). 2792 * Set skb->ooo_okay so that netdev_pick_tx() can choose a TX queue 2793 * based on XPS for better performance. 2794 * Otherwise clear ooo_okay to not risk Out Of Order delivery. 2795 */ 2796 skb->ooo_okay = (old_wmem == SK_WMEM_ALLOC_BIAS); 2797 } 2798 EXPORT_SYMBOL(skb_set_owner_w); 2799 2800 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2801 { 2802 /* Drivers depend on in-order delivery for crypto offload, 2803 * partial orphan breaks out-of-order-OK logic. 2804 */ 2805 if (skb_is_decrypted(skb)) 2806 return false; 2807 2808 return (skb->destructor == sock_wfree || 2809 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2810 } 2811 2812 /* This helper is used by netem, as it can hold packets in its 2813 * delay queue. We want to allow the owner socket to send more 2814 * packets, as if they were already TX completed by a typical driver. 2815 * But we also want to keep skb->sk set because some packet schedulers 2816 * rely on it (sch_fq for example). 2817 */ 2818 void skb_orphan_partial(struct sk_buff *skb) 2819 { 2820 if (skb_is_tcp_pure_ack(skb)) 2821 return; 2822 2823 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2824 return; 2825 2826 skb_orphan(skb); 2827 } 2828 EXPORT_SYMBOL(skb_orphan_partial); 2829 2830 /* 2831 * Read buffer destructor automatically called from kfree_skb. 2832 */ 2833 void sock_rfree(struct sk_buff *skb) 2834 { 2835 struct sock *sk = skb->sk; 2836 unsigned int len = skb->truesize; 2837 2838 atomic_sub(len, &sk->sk_rmem_alloc); 2839 sk_mem_uncharge(sk, len); 2840 } 2841 EXPORT_SYMBOL(sock_rfree); 2842 2843 /* 2844 * Buffer destructor for skbs that are not used directly in read or write 2845 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2846 */ 2847 void sock_efree(struct sk_buff *skb) 2848 { 2849 sock_put(skb->sk); 2850 } 2851 EXPORT_SYMBOL(sock_efree); 2852 2853 /* Buffer destructor for prefetch/receive path where reference count may 2854 * not be held, e.g. for listen sockets. 2855 */ 2856 #ifdef CONFIG_INET 2857 void sock_pfree(struct sk_buff *skb) 2858 { 2859 struct sock *sk = skb->sk; 2860 2861 if (!sk_is_refcounted(sk)) 2862 return; 2863 2864 if (sk->sk_state == TCP_NEW_SYN_RECV && inet_reqsk(sk)->syncookie) { 2865 inet_reqsk(sk)->rsk_listener = NULL; 2866 reqsk_free(inet_reqsk(sk)); 2867 return; 2868 } 2869 2870 sock_gen_put(sk); 2871 } 2872 EXPORT_SYMBOL(sock_pfree); 2873 #endif /* CONFIG_INET */ 2874 2875 /* 2876 * Allocate a skb from the socket's send buffer. 2877 */ 2878 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2879 gfp_t priority) 2880 { 2881 if (force || 2882 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2883 struct sk_buff *skb = alloc_skb(size, priority); 2884 2885 if (skb) { 2886 skb_set_owner_w(skb, sk); 2887 return skb; 2888 } 2889 } 2890 return NULL; 2891 } 2892 EXPORT_SYMBOL(sock_wmalloc); 2893 2894 static void sock_ofree(struct sk_buff *skb) 2895 { 2896 struct sock *sk = skb->sk; 2897 2898 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2899 } 2900 2901 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2902 gfp_t priority) 2903 { 2904 struct sk_buff *skb; 2905 2906 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2907 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2908 READ_ONCE(sock_net(sk)->core.sysctl_optmem_max)) 2909 return NULL; 2910 2911 skb = alloc_skb(size, priority); 2912 if (!skb) 2913 return NULL; 2914 2915 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2916 skb->sk = sk; 2917 skb->destructor = sock_ofree; 2918 return skb; 2919 } 2920 2921 /* 2922 * Allocate a memory block from the socket's option memory buffer. 2923 */ 2924 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2925 { 2926 int optmem_max = READ_ONCE(sock_net(sk)->core.sysctl_optmem_max); 2927 2928 if ((unsigned int)size <= optmem_max && 2929 atomic_read(&sk->sk_omem_alloc) + size < optmem_max) { 2930 void *mem; 2931 /* First do the add, to avoid the race if kmalloc 2932 * might sleep. 2933 */ 2934 atomic_add(size, &sk->sk_omem_alloc); 2935 mem = kmalloc(size, priority); 2936 if (mem) 2937 return mem; 2938 atomic_sub(size, &sk->sk_omem_alloc); 2939 } 2940 return NULL; 2941 } 2942 EXPORT_SYMBOL(sock_kmalloc); 2943 2944 /* 2945 * Duplicate the input "src" memory block using the socket's 2946 * option memory buffer. 2947 */ 2948 void *sock_kmemdup(struct sock *sk, const void *src, 2949 int size, gfp_t priority) 2950 { 2951 void *mem; 2952 2953 mem = sock_kmalloc(sk, size, priority); 2954 if (mem) 2955 memcpy(mem, src, size); 2956 return mem; 2957 } 2958 EXPORT_SYMBOL(sock_kmemdup); 2959 2960 /* Free an option memory block. Note, we actually want the inline 2961 * here as this allows gcc to detect the nullify and fold away the 2962 * condition entirely. 2963 */ 2964 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2965 const bool nullify) 2966 { 2967 if (WARN_ON_ONCE(!mem)) 2968 return; 2969 if (nullify) 2970 kfree_sensitive(mem); 2971 else 2972 kfree(mem); 2973 atomic_sub(size, &sk->sk_omem_alloc); 2974 } 2975 2976 void sock_kfree_s(struct sock *sk, void *mem, int size) 2977 { 2978 __sock_kfree_s(sk, mem, size, false); 2979 } 2980 EXPORT_SYMBOL(sock_kfree_s); 2981 2982 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2983 { 2984 __sock_kfree_s(sk, mem, size, true); 2985 } 2986 EXPORT_SYMBOL(sock_kzfree_s); 2987 2988 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2989 I think, these locks should be removed for datagram sockets. 2990 */ 2991 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2992 { 2993 DEFINE_WAIT(wait); 2994 2995 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2996 for (;;) { 2997 if (!timeo) 2998 break; 2999 if (signal_pending(current)) 3000 break; 3001 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 3002 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 3003 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 3004 break; 3005 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 3006 break; 3007 if (READ_ONCE(sk->sk_err)) 3008 break; 3009 timeo = schedule_timeout(timeo); 3010 } 3011 finish_wait(sk_sleep(sk), &wait); 3012 return timeo; 3013 } 3014 3015 3016 /* 3017 * Generic send/receive buffer handlers 3018 */ 3019 3020 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 3021 unsigned long data_len, int noblock, 3022 int *errcode, int max_page_order) 3023 { 3024 struct sk_buff *skb; 3025 long timeo; 3026 int err; 3027 3028 timeo = sock_sndtimeo(sk, noblock); 3029 for (;;) { 3030 err = sock_error(sk); 3031 if (err != 0) 3032 goto failure; 3033 3034 err = -EPIPE; 3035 if (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) 3036 goto failure; 3037 3038 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 3039 break; 3040 3041 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 3042 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 3043 err = -EAGAIN; 3044 if (!timeo) 3045 goto failure; 3046 if (signal_pending(current)) 3047 goto interrupted; 3048 timeo = sock_wait_for_wmem(sk, timeo); 3049 } 3050 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 3051 errcode, sk->sk_allocation); 3052 if (skb) 3053 skb_set_owner_w(skb, sk); 3054 return skb; 3055 3056 interrupted: 3057 err = sock_intr_errno(timeo); 3058 failure: 3059 *errcode = err; 3060 return NULL; 3061 } 3062 EXPORT_SYMBOL(sock_alloc_send_pskb); 3063 3064 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 3065 struct sockcm_cookie *sockc) 3066 { 3067 u32 tsflags; 3068 3069 BUILD_BUG_ON(SOF_TIMESTAMPING_LAST == (1 << 31)); 3070 3071 switch (cmsg->cmsg_type) { 3072 case SO_MARK: 3073 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 3074 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 3075 return -EPERM; 3076 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3077 return -EINVAL; 3078 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 3079 break; 3080 case SO_TIMESTAMPING_OLD: 3081 case SO_TIMESTAMPING_NEW: 3082 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3083 return -EINVAL; 3084 3085 tsflags = *(u32 *)CMSG_DATA(cmsg); 3086 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 3087 return -EINVAL; 3088 3089 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 3090 sockc->tsflags |= tsflags; 3091 break; 3092 case SCM_TXTIME: 3093 { 3094 ktime_t tmin; 3095 u64 txtime; 3096 3097 if (!sock_flag(sk, SOCK_TXTIME)) 3098 return -EINVAL; 3099 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 3100 return -EINVAL; 3101 3102 txtime = get_unaligned((u64 *)CMSG_DATA(cmsg)); 3103 3104 /* Allow sending without a delivery time: zero special case */ 3105 if (!txtime) { 3106 sockc->transmit_time = 0; 3107 break; 3108 } 3109 3110 switch (sk->sk_clockid) { 3111 case CLOCK_MONOTONIC: 3112 tmin = 1; 3113 break; 3114 case CLOCK_REALTIME: 3115 tmin = max(ktime_mono_to_real(0), 1); 3116 break; 3117 case CLOCK_TAI: 3118 tmin = max(ktime_mono_to_any(0, TK_OFFS_TAI), 1); 3119 break; 3120 default: 3121 tmin = 1; 3122 WARN_ON_ONCE(1); 3123 break; 3124 } 3125 3126 sockc->transmit_time = max_t(ktime_t, txtime, tmin); 3127 break; 3128 } 3129 case SCM_TS_OPT_ID: 3130 if (sk_is_tcp(sk)) 3131 return -EINVAL; 3132 tsflags = READ_ONCE(sk->sk_tsflags); 3133 if (!(tsflags & SOF_TIMESTAMPING_OPT_ID)) 3134 return -EINVAL; 3135 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3136 return -EINVAL; 3137 sockc->ts_opt_id = *(u32 *)CMSG_DATA(cmsg); 3138 sockc->tsflags |= SOCKCM_FLAG_TS_OPT_ID; 3139 break; 3140 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 3141 case SCM_RIGHTS: 3142 case SCM_CREDENTIALS: 3143 break; 3144 case SO_PRIORITY: 3145 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3146 return -EINVAL; 3147 if (!sk_set_prio_allowed(sk, *(u32 *)CMSG_DATA(cmsg))) 3148 return -EPERM; 3149 sockc->priority = *(u32 *)CMSG_DATA(cmsg); 3150 break; 3151 case SCM_DEVMEM_DMABUF: 3152 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 3153 return -EINVAL; 3154 sockc->dmabuf_id = *(u32 *)CMSG_DATA(cmsg); 3155 break; 3156 default: 3157 return -EINVAL; 3158 } 3159 return 0; 3160 } 3161 EXPORT_SYMBOL(__sock_cmsg_send); 3162 3163 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 3164 struct sockcm_cookie *sockc) 3165 { 3166 struct cmsghdr *cmsg; 3167 int ret; 3168 3169 for_each_cmsghdr(cmsg, msg) { 3170 if (!CMSG_OK(msg, cmsg)) 3171 return -EINVAL; 3172 if (cmsg->cmsg_level != SOL_SOCKET) 3173 continue; 3174 ret = __sock_cmsg_send(sk, cmsg, sockc); 3175 if (ret) 3176 return ret; 3177 } 3178 return 0; 3179 } 3180 EXPORT_SYMBOL(sock_cmsg_send); 3181 3182 static void sk_enter_memory_pressure(struct sock *sk) 3183 { 3184 if (!sk->sk_prot->enter_memory_pressure) 3185 return; 3186 3187 sk->sk_prot->enter_memory_pressure(sk); 3188 } 3189 3190 static void sk_leave_memory_pressure(struct sock *sk) 3191 { 3192 if (sk->sk_prot->leave_memory_pressure) { 3193 INDIRECT_CALL_INET_1(sk->sk_prot->leave_memory_pressure, 3194 tcp_leave_memory_pressure, sk); 3195 } else { 3196 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 3197 3198 if (memory_pressure && READ_ONCE(*memory_pressure)) 3199 WRITE_ONCE(*memory_pressure, 0); 3200 } 3201 } 3202 3203 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3204 3205 /** 3206 * skb_page_frag_refill - check that a page_frag contains enough room 3207 * @sz: minimum size of the fragment we want to get 3208 * @pfrag: pointer to page_frag 3209 * @gfp: priority for memory allocation 3210 * 3211 * Note: While this allocator tries to use high order pages, there is 3212 * no guarantee that allocations succeed. Therefore, @sz MUST be 3213 * less or equal than PAGE_SIZE. 3214 */ 3215 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 3216 { 3217 if (pfrag->page) { 3218 if (page_ref_count(pfrag->page) == 1) { 3219 pfrag->offset = 0; 3220 return true; 3221 } 3222 if (pfrag->offset + sz <= pfrag->size) 3223 return true; 3224 put_page(pfrag->page); 3225 } 3226 3227 pfrag->offset = 0; 3228 if (SKB_FRAG_PAGE_ORDER && 3229 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 3230 /* Avoid direct reclaim but allow kswapd to wake */ 3231 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 3232 __GFP_COMP | __GFP_NOWARN | 3233 __GFP_NORETRY, 3234 SKB_FRAG_PAGE_ORDER); 3235 if (likely(pfrag->page)) { 3236 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 3237 return true; 3238 } 3239 } 3240 pfrag->page = alloc_page(gfp); 3241 if (likely(pfrag->page)) { 3242 pfrag->size = PAGE_SIZE; 3243 return true; 3244 } 3245 return false; 3246 } 3247 EXPORT_SYMBOL(skb_page_frag_refill); 3248 3249 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 3250 { 3251 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 3252 return true; 3253 3254 if (!sk->sk_bypass_prot_mem) 3255 sk_enter_memory_pressure(sk); 3256 3257 sk_stream_moderate_sndbuf(sk); 3258 3259 return false; 3260 } 3261 EXPORT_SYMBOL(sk_page_frag_refill); 3262 3263 static void __lock_sock(struct sock *sk) 3264 __releases(&sk->sk_lock.slock) 3265 __acquires(&sk->sk_lock.slock) 3266 { 3267 DEFINE_WAIT(wait); 3268 3269 for (;;) { 3270 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 3271 TASK_UNINTERRUPTIBLE); 3272 spin_unlock_bh(&sk->sk_lock.slock); 3273 schedule(); 3274 spin_lock_bh(&sk->sk_lock.slock); 3275 if (!sock_owned_by_user(sk)) 3276 break; 3277 } 3278 finish_wait(&sk->sk_lock.wq, &wait); 3279 } 3280 3281 void __release_sock(struct sock *sk) 3282 __releases(&sk->sk_lock.slock) 3283 __acquires(&sk->sk_lock.slock) 3284 { 3285 struct sk_buff *skb, *next; 3286 int nb = 0; 3287 3288 while ((skb = sk->sk_backlog.head) != NULL) { 3289 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 3290 3291 spin_unlock_bh(&sk->sk_lock.slock); 3292 3293 while (1) { 3294 next = skb->next; 3295 prefetch(next); 3296 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 3297 skb_mark_not_on_list(skb); 3298 sk_backlog_rcv(sk, skb); 3299 3300 skb = next; 3301 if (!skb) 3302 break; 3303 3304 if (!(++nb & 15)) 3305 cond_resched(); 3306 } 3307 3308 spin_lock_bh(&sk->sk_lock.slock); 3309 } 3310 3311 /* 3312 * Doing the zeroing here guarantee we can not loop forever 3313 * while a wild producer attempts to flood us. 3314 */ 3315 sk->sk_backlog.len = 0; 3316 } 3317 3318 void __sk_flush_backlog(struct sock *sk) 3319 { 3320 spin_lock_bh(&sk->sk_lock.slock); 3321 __release_sock(sk); 3322 3323 if (sk->sk_prot->release_cb) 3324 INDIRECT_CALL_INET_1(sk->sk_prot->release_cb, 3325 tcp_release_cb, sk); 3326 3327 spin_unlock_bh(&sk->sk_lock.slock); 3328 } 3329 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 3330 3331 /** 3332 * sk_wait_data - wait for data to arrive at sk_receive_queue 3333 * @sk: sock to wait on 3334 * @timeo: for how long 3335 * @skb: last skb seen on sk_receive_queue 3336 * 3337 * Now socket state including sk->sk_err is changed only under lock, 3338 * hence we may omit checks after joining wait queue. 3339 * We check receive queue before schedule() only as optimization; 3340 * it is very likely that release_sock() added new data. 3341 */ 3342 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 3343 { 3344 DEFINE_WAIT_FUNC(wait, woken_wake_function); 3345 int rc; 3346 3347 add_wait_queue(sk_sleep(sk), &wait); 3348 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3349 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 3350 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 3351 remove_wait_queue(sk_sleep(sk), &wait); 3352 return rc; 3353 } 3354 EXPORT_SYMBOL(sk_wait_data); 3355 3356 /** 3357 * __sk_mem_raise_allocated - increase memory_allocated 3358 * @sk: socket 3359 * @size: memory size to allocate 3360 * @amt: pages to allocate 3361 * @kind: allocation type 3362 * 3363 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc. 3364 * 3365 * Unlike the globally shared limits among the sockets under same protocol, 3366 * consuming the budget of a memcg won't have direct effect on other ones. 3367 * So be optimistic about memcg's tolerance, and leave the callers to decide 3368 * whether or not to raise allocated through sk_under_memory_pressure() or 3369 * its variants. 3370 */ 3371 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 3372 { 3373 bool memcg_enabled = false, charged = false; 3374 struct proto *prot = sk->sk_prot; 3375 long allocated = 0; 3376 3377 if (!sk->sk_bypass_prot_mem) { 3378 sk_memory_allocated_add(sk, amt); 3379 allocated = sk_memory_allocated(sk); 3380 } 3381 3382 if (mem_cgroup_sk_enabled(sk)) { 3383 memcg_enabled = true; 3384 charged = mem_cgroup_sk_charge(sk, amt, gfp_memcg_charge()); 3385 if (!charged) 3386 goto suppress_allocation; 3387 } 3388 3389 if (!allocated) 3390 return 1; 3391 3392 /* Under limit. */ 3393 if (allocated <= sk_prot_mem_limits(sk, 0)) { 3394 sk_leave_memory_pressure(sk); 3395 return 1; 3396 } 3397 3398 /* Under pressure. */ 3399 if (allocated > sk_prot_mem_limits(sk, 1)) 3400 sk_enter_memory_pressure(sk); 3401 3402 /* Over hard limit. */ 3403 if (allocated > sk_prot_mem_limits(sk, 2)) 3404 goto suppress_allocation; 3405 3406 /* Guarantee minimum buffer size under pressure (either global 3407 * or memcg) to make sure features described in RFC 7323 (TCP 3408 * Extensions for High Performance) work properly. 3409 * 3410 * This rule does NOT stand when exceeds global or memcg's hard 3411 * limit, or else a DoS attack can be taken place by spawning 3412 * lots of sockets whose usage are under minimum buffer size. 3413 */ 3414 if (kind == SK_MEM_RECV) { 3415 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 3416 return 1; 3417 3418 } else { /* SK_MEM_SEND */ 3419 int wmem0 = sk_get_wmem0(sk, prot); 3420 3421 if (sk->sk_type == SOCK_STREAM) { 3422 if (sk->sk_wmem_queued < wmem0) 3423 return 1; 3424 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 3425 return 1; 3426 } 3427 } 3428 3429 if (sk_has_memory_pressure(sk)) { 3430 u64 alloc; 3431 3432 /* The following 'average' heuristic is within the 3433 * scope of global accounting, so it only makes 3434 * sense for global memory pressure. 3435 */ 3436 if (!sk_under_global_memory_pressure(sk)) 3437 return 1; 3438 3439 /* Try to be fair among all the sockets under global 3440 * pressure by allowing the ones that below average 3441 * usage to raise. 3442 */ 3443 alloc = sk_sockets_allocated_read_positive(sk); 3444 if (sk_prot_mem_limits(sk, 2) > alloc * 3445 sk_mem_pages(sk->sk_wmem_queued + 3446 atomic_read(&sk->sk_rmem_alloc) + 3447 sk->sk_forward_alloc)) 3448 return 1; 3449 } 3450 3451 suppress_allocation: 3452 3453 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3454 sk_stream_moderate_sndbuf(sk); 3455 3456 /* Fail only if socket is _under_ its sndbuf. 3457 * In this case we cannot block, so that we have to fail. 3458 */ 3459 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3460 /* Force charge with __GFP_NOFAIL */ 3461 if (memcg_enabled && !charged) 3462 mem_cgroup_sk_charge(sk, amt, 3463 gfp_memcg_charge() | __GFP_NOFAIL); 3464 return 1; 3465 } 3466 } 3467 3468 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3469 3470 if (allocated) 3471 sk_memory_allocated_sub(sk, amt); 3472 3473 if (charged) 3474 mem_cgroup_sk_uncharge(sk, amt); 3475 3476 return 0; 3477 } 3478 3479 /** 3480 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3481 * @sk: socket 3482 * @size: memory size to allocate 3483 * @kind: allocation type 3484 * 3485 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3486 * rmem allocation. This function assumes that protocols which have 3487 * memory_pressure use sk_wmem_queued as write buffer accounting. 3488 */ 3489 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3490 { 3491 int ret, amt = sk_mem_pages(size); 3492 3493 sk_forward_alloc_add(sk, amt << PAGE_SHIFT); 3494 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3495 if (!ret) 3496 sk_forward_alloc_add(sk, -(amt << PAGE_SHIFT)); 3497 return ret; 3498 } 3499 EXPORT_SYMBOL(__sk_mem_schedule); 3500 3501 /** 3502 * __sk_mem_reduce_allocated - reclaim memory_allocated 3503 * @sk: socket 3504 * @amount: number of quanta 3505 * 3506 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3507 */ 3508 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3509 { 3510 if (mem_cgroup_sk_enabled(sk)) 3511 mem_cgroup_sk_uncharge(sk, amount); 3512 3513 if (sk->sk_bypass_prot_mem) 3514 return; 3515 3516 sk_memory_allocated_sub(sk, amount); 3517 3518 if (sk_under_global_memory_pressure(sk) && 3519 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3520 sk_leave_memory_pressure(sk); 3521 } 3522 3523 /** 3524 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3525 * @sk: socket 3526 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3527 */ 3528 void __sk_mem_reclaim(struct sock *sk, int amount) 3529 { 3530 amount >>= PAGE_SHIFT; 3531 sk_forward_alloc_add(sk, -(amount << PAGE_SHIFT)); 3532 __sk_mem_reduce_allocated(sk, amount); 3533 } 3534 EXPORT_SYMBOL(__sk_mem_reclaim); 3535 3536 void __sk_charge(struct sock *sk, gfp_t gfp) 3537 { 3538 int amt; 3539 3540 gfp |= __GFP_NOFAIL; 3541 if (mem_cgroup_from_sk(sk)) { 3542 /* The socket has not been accepted yet, no need 3543 * to look at newsk->sk_wmem_queued. 3544 */ 3545 amt = sk_mem_pages(sk->sk_forward_alloc + 3546 atomic_read(&sk->sk_rmem_alloc)); 3547 if (amt) 3548 mem_cgroup_sk_charge(sk, amt, gfp); 3549 } 3550 3551 kmem_cache_charge(sk, gfp); 3552 } 3553 3554 int sk_set_peek_off(struct sock *sk, int val) 3555 { 3556 WRITE_ONCE(sk->sk_peek_off, val); 3557 return 0; 3558 } 3559 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3560 3561 /* 3562 * Set of default routines for initialising struct proto_ops when 3563 * the protocol does not support a particular function. In certain 3564 * cases where it makes no sense for a protocol to have a "do nothing" 3565 * function, some default processing is provided. 3566 */ 3567 3568 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len) 3569 { 3570 return -EOPNOTSUPP; 3571 } 3572 EXPORT_SYMBOL(sock_no_bind); 3573 3574 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr, 3575 int len, int flags) 3576 { 3577 return -EOPNOTSUPP; 3578 } 3579 EXPORT_SYMBOL(sock_no_connect); 3580 3581 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3582 { 3583 return -EOPNOTSUPP; 3584 } 3585 EXPORT_SYMBOL(sock_no_socketpair); 3586 3587 int sock_no_accept(struct socket *sock, struct socket *newsock, 3588 struct proto_accept_arg *arg) 3589 { 3590 return -EOPNOTSUPP; 3591 } 3592 EXPORT_SYMBOL(sock_no_accept); 3593 3594 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3595 int peer) 3596 { 3597 return -EOPNOTSUPP; 3598 } 3599 EXPORT_SYMBOL(sock_no_getname); 3600 3601 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3602 { 3603 return -EOPNOTSUPP; 3604 } 3605 EXPORT_SYMBOL(sock_no_ioctl); 3606 3607 int sock_no_listen(struct socket *sock, int backlog) 3608 { 3609 return -EOPNOTSUPP; 3610 } 3611 EXPORT_SYMBOL(sock_no_listen); 3612 3613 int sock_no_shutdown(struct socket *sock, int how) 3614 { 3615 return -EOPNOTSUPP; 3616 } 3617 EXPORT_SYMBOL(sock_no_shutdown); 3618 3619 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3620 { 3621 return -EOPNOTSUPP; 3622 } 3623 EXPORT_SYMBOL(sock_no_sendmsg); 3624 3625 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3626 { 3627 return -EOPNOTSUPP; 3628 } 3629 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3630 3631 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3632 int flags) 3633 { 3634 return -EOPNOTSUPP; 3635 } 3636 EXPORT_SYMBOL(sock_no_recvmsg); 3637 3638 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3639 { 3640 /* Mirror missing mmap method error code */ 3641 return -ENODEV; 3642 } 3643 EXPORT_SYMBOL(sock_no_mmap); 3644 3645 /* 3646 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3647 * various sock-based usage counts. 3648 */ 3649 void __receive_sock(struct file *file) 3650 { 3651 struct socket *sock; 3652 3653 sock = sock_from_file(file); 3654 if (sock) { 3655 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3656 sock_update_classid(&sock->sk->sk_cgrp_data); 3657 } 3658 } 3659 3660 /* 3661 * Default Socket Callbacks 3662 */ 3663 3664 static void sock_def_wakeup(struct sock *sk) 3665 { 3666 struct socket_wq *wq; 3667 3668 rcu_read_lock(); 3669 wq = rcu_dereference(sk->sk_wq); 3670 if (skwq_has_sleeper(wq)) 3671 wake_up_interruptible_all(&wq->wait); 3672 rcu_read_unlock(); 3673 } 3674 3675 static void sock_def_error_report(struct sock *sk) 3676 { 3677 struct socket_wq *wq; 3678 3679 rcu_read_lock(); 3680 wq = rcu_dereference(sk->sk_wq); 3681 if (skwq_has_sleeper(wq)) 3682 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3683 sk_wake_async_rcu(sk, SOCK_WAKE_IO, POLL_ERR); 3684 rcu_read_unlock(); 3685 } 3686 3687 void sock_def_readable(struct sock *sk) 3688 { 3689 struct socket_wq *wq; 3690 3691 trace_sk_data_ready(sk); 3692 3693 rcu_read_lock(); 3694 wq = rcu_dereference(sk->sk_wq); 3695 if (skwq_has_sleeper(wq)) 3696 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3697 EPOLLRDNORM | EPOLLRDBAND); 3698 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN); 3699 rcu_read_unlock(); 3700 } 3701 3702 static void sock_def_write_space(struct sock *sk) 3703 { 3704 struct socket_wq *wq; 3705 3706 rcu_read_lock(); 3707 3708 /* Do not wake up a writer until he can make "significant" 3709 * progress. --DaveM 3710 */ 3711 if (sock_writeable(sk)) { 3712 wq = rcu_dereference(sk->sk_wq); 3713 if (skwq_has_sleeper(wq)) 3714 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3715 EPOLLWRNORM | EPOLLWRBAND); 3716 3717 /* Should agree with poll, otherwise some programs break */ 3718 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3719 } 3720 3721 rcu_read_unlock(); 3722 } 3723 3724 /* An optimised version of sock_def_write_space(), should only be called 3725 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3726 * ->sk_wmem_alloc. 3727 */ 3728 static void sock_def_write_space_wfree(struct sock *sk, int wmem_alloc) 3729 { 3730 /* Do not wake up a writer until he can make "significant" 3731 * progress. --DaveM 3732 */ 3733 if (__sock_writeable(sk, wmem_alloc)) { 3734 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3735 3736 /* rely on refcount_sub from sock_wfree() */ 3737 smp_mb__after_atomic(); 3738 if (wq && waitqueue_active(&wq->wait)) 3739 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3740 EPOLLWRNORM | EPOLLWRBAND); 3741 3742 /* Should agree with poll, otherwise some programs break */ 3743 sk_wake_async_rcu(sk, SOCK_WAKE_SPACE, POLL_OUT); 3744 } 3745 } 3746 3747 static void sock_def_destruct(struct sock *sk) 3748 { 3749 } 3750 3751 void sk_send_sigurg(struct sock *sk) 3752 { 3753 if (sk->sk_socket && sk->sk_socket->file) 3754 if (send_sigurg(sk->sk_socket->file)) 3755 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3756 } 3757 EXPORT_SYMBOL(sk_send_sigurg); 3758 3759 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3760 unsigned long expires) 3761 { 3762 if (!mod_timer(timer, expires)) 3763 sock_hold(sk); 3764 } 3765 EXPORT_SYMBOL(sk_reset_timer); 3766 3767 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3768 { 3769 if (timer_delete(timer)) 3770 __sock_put(sk); 3771 } 3772 EXPORT_SYMBOL(sk_stop_timer); 3773 3774 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3775 { 3776 if (timer_delete_sync(timer)) 3777 __sock_put(sk); 3778 } 3779 EXPORT_SYMBOL(sk_stop_timer_sync); 3780 3781 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid) 3782 { 3783 sk_init_common(sk); 3784 sk->sk_send_head = NULL; 3785 3786 timer_setup(&sk->sk_timer, NULL, 0); 3787 3788 sk->sk_allocation = GFP_KERNEL; 3789 sk->sk_rcvbuf = READ_ONCE(sysctl_rmem_default); 3790 sk->sk_sndbuf = READ_ONCE(sysctl_wmem_default); 3791 sk->sk_state = TCP_CLOSE; 3792 sk->sk_use_task_frag = true; 3793 sk_set_socket(sk, sock); 3794 3795 sock_set_flag(sk, SOCK_ZAPPED); 3796 3797 if (sock) { 3798 sk->sk_type = sock->type; 3799 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3800 sock->sk = sk; 3801 } else { 3802 RCU_INIT_POINTER(sk->sk_wq, NULL); 3803 } 3804 sk->sk_uid = uid; 3805 3806 sk->sk_state_change = sock_def_wakeup; 3807 sk->sk_data_ready = sock_def_readable; 3808 sk->sk_write_space = sock_def_write_space; 3809 sk->sk_error_report = sock_def_error_report; 3810 sk->sk_destruct = sock_def_destruct; 3811 3812 sk->sk_frag.page = NULL; 3813 sk->sk_frag.offset = 0; 3814 sk->sk_peek_off = -1; 3815 3816 memset(sk->sk_peer_pid, 0, sizeof(sk->sk_peer_pid)); 3817 sk->sk_peer_cred = NULL; 3818 spin_lock_init(&sk->sk_peer_lock); 3819 3820 sk->sk_write_pending = 0; 3821 sk->sk_rcvlowat = 1; 3822 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3823 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3824 3825 sk->sk_stamp = SK_DEFAULT_STAMP; 3826 #if BITS_PER_LONG==32 3827 seqlock_init(&sk->sk_stamp_seq); 3828 #endif 3829 atomic_set(&sk->sk_zckey, 0); 3830 3831 #ifdef CONFIG_NET_RX_BUSY_POLL 3832 sk->sk_napi_id = 0; 3833 sk->sk_ll_usec = READ_ONCE(sysctl_net_busy_read); 3834 #endif 3835 3836 sk->sk_max_pacing_rate = ~0UL; 3837 sk->sk_pacing_rate = ~0UL; 3838 WRITE_ONCE(sk->sk_pacing_shift, 10); 3839 sk->sk_incoming_cpu = -1; 3840 3841 sk_rx_queue_clear(sk); 3842 /* 3843 * Before updating sk_refcnt, we must commit prior changes to memory 3844 * (Documentation/RCU/rculist_nulls.rst for details) 3845 */ 3846 smp_wmb(); 3847 refcount_set(&sk->sk_refcnt, 1); 3848 sk_drops_reset(sk); 3849 } 3850 EXPORT_SYMBOL(sock_init_data_uid); 3851 3852 void sock_init_data(struct socket *sock, struct sock *sk) 3853 { 3854 kuid_t uid = sock ? 3855 SOCK_INODE(sock)->i_uid : 3856 make_kuid(sock_net(sk)->user_ns, 0); 3857 3858 sock_init_data_uid(sock, sk, uid); 3859 } 3860 EXPORT_SYMBOL(sock_init_data); 3861 3862 void noinline lock_sock_nested(struct sock *sk, int subclass) 3863 { 3864 /* The sk_lock has mutex_lock() semantics here. */ 3865 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3866 3867 might_sleep(); 3868 #ifdef CONFIG_64BIT 3869 if (sizeof(struct slock_owned) == sizeof(long)) { 3870 socket_lock_t tmp = { 3871 .slock = __SPIN_LOCK_UNLOCKED(tmp.slock), 3872 .owned = 1, 3873 }; 3874 socket_lock_t old = { 3875 .slock = __SPIN_LOCK_UNLOCKED(old.slock), 3876 .owned = 0, 3877 }; 3878 3879 if (likely(try_cmpxchg(&sk->sk_lock.combined, 3880 &old.combined, tmp.combined))) 3881 return; 3882 } 3883 #endif 3884 spin_lock_bh(&sk->sk_lock.slock); 3885 if (unlikely(sock_owned_by_user_nocheck(sk))) 3886 __lock_sock(sk); 3887 sk->sk_lock.owned = 1; 3888 spin_unlock_bh(&sk->sk_lock.slock); 3889 } 3890 EXPORT_SYMBOL(lock_sock_nested); 3891 3892 void release_sock(struct sock *sk) 3893 { 3894 spin_lock_bh(&sk->sk_lock.slock); 3895 3896 if (unlikely(sk->sk_backlog.tail)) 3897 __release_sock(sk); 3898 3899 if (sk->sk_prot->release_cb) { 3900 if (!tcp_release_cb_cond(sk)) 3901 sk->sk_prot->release_cb(sk); 3902 } 3903 sock_release_ownership(sk); 3904 if (unlikely(waitqueue_active(&sk->sk_lock.wq))) 3905 wake_up(&sk->sk_lock.wq); 3906 3907 spin_unlock_bh(&sk->sk_lock.slock); 3908 } 3909 EXPORT_SYMBOL(release_sock); 3910 3911 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3912 { 3913 might_sleep(); 3914 spin_lock_bh(&sk->sk_lock.slock); 3915 3916 if (likely(!sock_owned_by_user_nocheck(sk))) { 3917 /* 3918 * Fast path return with bottom halves disabled and 3919 * sock::sk_lock.slock held. 3920 * 3921 * The 'mutex' is not contended and holding 3922 * sock::sk_lock.slock prevents all other lockers to 3923 * proceed so the corresponding unlock_sock_fast() can 3924 * avoid the slow path of release_sock() completely and 3925 * just release slock. 3926 * 3927 * From a semantical POV this is equivalent to 'acquiring' 3928 * the 'mutex', hence the corresponding lockdep 3929 * mutex_release() has to happen in the fast path of 3930 * unlock_sock_fast(). 3931 */ 3932 return false; 3933 } 3934 3935 __lock_sock(sk); 3936 sk->sk_lock.owned = 1; 3937 __acquire(&sk->sk_lock.slock); 3938 spin_unlock_bh(&sk->sk_lock.slock); 3939 return true; 3940 } 3941 EXPORT_SYMBOL(__lock_sock_fast); 3942 3943 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3944 bool timeval, bool time32) 3945 { 3946 struct sock *sk = sock->sk; 3947 struct timespec64 ts; 3948 3949 /* sk->sk_flags must only be changed under the socket lock, 3950 * because sock_set_flag() uses non atomic operations. 3951 */ 3952 if (!sock_flag(sk, SOCK_TIMESTAMP)) { 3953 lock_sock(sk); 3954 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3955 release_sock(sk); 3956 } 3957 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3958 if (ts.tv_sec == -1) 3959 return -ENOENT; 3960 if (ts.tv_sec == 0) { 3961 ktime_t kt = ktime_get_real(); 3962 sock_write_timestamp(sk, kt); 3963 ts = ktime_to_timespec64(kt); 3964 } 3965 3966 if (timeval) 3967 ts.tv_nsec /= 1000; 3968 3969 #ifdef CONFIG_COMPAT_32BIT_TIME 3970 if (time32) 3971 return put_old_timespec32(&ts, userstamp); 3972 #endif 3973 #ifdef CONFIG_SPARC64 3974 /* beware of padding in sparc64 timeval */ 3975 if (timeval && !in_compat_syscall()) { 3976 struct __kernel_old_timeval __user tv = { 3977 .tv_sec = ts.tv_sec, 3978 .tv_usec = ts.tv_nsec, 3979 }; 3980 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3981 return -EFAULT; 3982 return 0; 3983 } 3984 #endif 3985 return put_timespec64(&ts, userstamp); 3986 } 3987 EXPORT_SYMBOL(sock_gettstamp); 3988 3989 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3990 { 3991 if (!sock_flag(sk, flag)) { 3992 unsigned long previous_flags = sk->sk_flags; 3993 3994 sock_set_flag(sk, flag); 3995 /* 3996 * we just set one of the two flags which require net 3997 * time stamping, but time stamping might have been on 3998 * already because of the other one 3999 */ 4000 if (sock_needs_netstamp(sk) && 4001 !(previous_flags & SK_FLAGS_TIMESTAMP)) 4002 net_enable_timestamp(); 4003 } 4004 } 4005 4006 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 4007 int level, int type) 4008 { 4009 struct sock_extended_err ee; 4010 struct sk_buff *skb; 4011 int copied, err; 4012 4013 err = -EAGAIN; 4014 skb = sock_dequeue_err_skb(sk); 4015 if (skb == NULL) 4016 goto out; 4017 4018 copied = skb->len; 4019 if (copied > len) { 4020 msg->msg_flags |= MSG_TRUNC; 4021 copied = len; 4022 } 4023 err = skb_copy_datagram_msg(skb, 0, msg, copied); 4024 if (err) 4025 goto out_free_skb; 4026 4027 sock_recv_timestamp(msg, sk, skb); 4028 4029 /* We must use a bounce buffer for CONFIG_HARDENED_USERCOPY=y */ 4030 ee = SKB_EXT_ERR(skb)->ee; 4031 put_cmsg(msg, level, type, sizeof(ee), &ee); 4032 4033 msg->msg_flags |= MSG_ERRQUEUE; 4034 err = copied; 4035 4036 out_free_skb: 4037 kfree_skb(skb); 4038 out: 4039 return err; 4040 } 4041 EXPORT_SYMBOL(sock_recv_errqueue); 4042 4043 /* 4044 * Get a socket option on an socket. 4045 * 4046 * FIX: POSIX 1003.1g is very ambiguous here. It states that 4047 * asynchronous errors should be reported by getsockopt. We assume 4048 * this means if you specify SO_ERROR (otherwise what is the point of it). 4049 */ 4050 int sock_common_getsockopt(struct socket *sock, int level, int optname, 4051 char __user *optval, int __user *optlen) 4052 { 4053 struct sock *sk = sock->sk; 4054 4055 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 4056 return READ_ONCE(sk->sk_prot)->getsockopt(sk, level, optname, optval, optlen); 4057 } 4058 EXPORT_SYMBOL(sock_common_getsockopt); 4059 4060 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 4061 int flags) 4062 { 4063 struct sock *sk = sock->sk; 4064 4065 return sk->sk_prot->recvmsg(sk, msg, size, flags); 4066 } 4067 EXPORT_SYMBOL(sock_common_recvmsg); 4068 4069 /* 4070 * Set socket options on an inet socket. 4071 */ 4072 int sock_common_setsockopt(struct socket *sock, int level, int optname, 4073 sockptr_t optval, unsigned int optlen) 4074 { 4075 struct sock *sk = sock->sk; 4076 4077 /* IPV6_ADDRFORM can change sk->sk_prot under us. */ 4078 return READ_ONCE(sk->sk_prot)->setsockopt(sk, level, optname, optval, optlen); 4079 } 4080 EXPORT_SYMBOL(sock_common_setsockopt); 4081 4082 void sk_common_release(struct sock *sk) 4083 { 4084 if (sk->sk_prot->destroy) 4085 sk->sk_prot->destroy(sk); 4086 4087 /* 4088 * Observation: when sk_common_release is called, processes have 4089 * no access to socket. But net still has. 4090 * Step one, detach it from networking: 4091 * 4092 * A. Remove from hash tables. 4093 */ 4094 4095 sk->sk_prot->unhash(sk); 4096 4097 /* 4098 * In this point socket cannot receive new packets, but it is possible 4099 * that some packets are in flight because some CPU runs receiver and 4100 * did hash table lookup before we unhashed socket. They will achieve 4101 * receive queue and will be purged by socket destructor. 4102 * 4103 * Also we still have packets pending on receive queue and probably, 4104 * our own packets waiting in device queues. sock_destroy will drain 4105 * receive queue, but transmitted packets will delay socket destruction 4106 * until the last reference will be released. 4107 */ 4108 4109 sock_orphan(sk); 4110 4111 xfrm_sk_free_policy(sk); 4112 4113 sock_put(sk); 4114 } 4115 EXPORT_SYMBOL(sk_common_release); 4116 4117 void sk_get_meminfo(const struct sock *sk, u32 *mem) 4118 { 4119 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 4120 4121 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 4122 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 4123 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 4124 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 4125 mem[SK_MEMINFO_FWD_ALLOC] = READ_ONCE(sk->sk_forward_alloc); 4126 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 4127 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 4128 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 4129 mem[SK_MEMINFO_DROPS] = sk_drops_read(sk); 4130 } 4131 4132 #ifdef CONFIG_PROC_FS 4133 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 4134 4135 int sock_prot_inuse_get(struct net *net, struct proto *prot) 4136 { 4137 int cpu, idx = prot->inuse_idx; 4138 int res = 0; 4139 4140 for_each_possible_cpu(cpu) 4141 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 4142 4143 return res >= 0 ? res : 0; 4144 } 4145 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 4146 4147 int sock_inuse_get(struct net *net) 4148 { 4149 int cpu, res = 0; 4150 4151 for_each_possible_cpu(cpu) 4152 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 4153 4154 return res; 4155 } 4156 4157 EXPORT_SYMBOL_GPL(sock_inuse_get); 4158 4159 static int __net_init sock_inuse_init_net(struct net *net) 4160 { 4161 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 4162 if (net->core.prot_inuse == NULL) 4163 return -ENOMEM; 4164 return 0; 4165 } 4166 4167 static void __net_exit sock_inuse_exit_net(struct net *net) 4168 { 4169 free_percpu(net->core.prot_inuse); 4170 } 4171 4172 static struct pernet_operations net_inuse_ops = { 4173 .init = sock_inuse_init_net, 4174 .exit = sock_inuse_exit_net, 4175 }; 4176 4177 static __init int net_inuse_init(void) 4178 { 4179 if (register_pernet_subsys(&net_inuse_ops)) 4180 panic("Cannot initialize net inuse counters"); 4181 4182 return 0; 4183 } 4184 4185 core_initcall(net_inuse_init); 4186 4187 static int assign_proto_idx(struct proto *prot) 4188 { 4189 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 4190 4191 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR)) { 4192 pr_err("PROTO_INUSE_NR exhausted\n"); 4193 return -ENOSPC; 4194 } 4195 4196 set_bit(prot->inuse_idx, proto_inuse_idx); 4197 return 0; 4198 } 4199 4200 static void release_proto_idx(struct proto *prot) 4201 { 4202 if (prot->inuse_idx != PROTO_INUSE_NR) 4203 clear_bit(prot->inuse_idx, proto_inuse_idx); 4204 } 4205 #else 4206 static inline int assign_proto_idx(struct proto *prot) 4207 { 4208 return 0; 4209 } 4210 4211 static inline void release_proto_idx(struct proto *prot) 4212 { 4213 } 4214 4215 #endif 4216 4217 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 4218 { 4219 if (!twsk_prot) 4220 return; 4221 kfree(twsk_prot->twsk_slab_name); 4222 twsk_prot->twsk_slab_name = NULL; 4223 kmem_cache_destroy(twsk_prot->twsk_slab); 4224 twsk_prot->twsk_slab = NULL; 4225 } 4226 4227 static int tw_prot_init(const struct proto *prot) 4228 { 4229 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 4230 4231 if (!twsk_prot) 4232 return 0; 4233 4234 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 4235 prot->name); 4236 if (!twsk_prot->twsk_slab_name) 4237 return -ENOMEM; 4238 4239 twsk_prot->twsk_slab = 4240 kmem_cache_create(twsk_prot->twsk_slab_name, 4241 twsk_prot->twsk_obj_size, 0, 4242 SLAB_ACCOUNT | prot->slab_flags, 4243 NULL); 4244 if (!twsk_prot->twsk_slab) { 4245 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 4246 prot->name); 4247 return -ENOMEM; 4248 } 4249 4250 return 0; 4251 } 4252 4253 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 4254 { 4255 if (!rsk_prot) 4256 return; 4257 kfree(rsk_prot->slab_name); 4258 rsk_prot->slab_name = NULL; 4259 kmem_cache_destroy(rsk_prot->slab); 4260 rsk_prot->slab = NULL; 4261 } 4262 4263 static int req_prot_init(const struct proto *prot) 4264 { 4265 struct request_sock_ops *rsk_prot = prot->rsk_prot; 4266 4267 if (!rsk_prot) 4268 return 0; 4269 4270 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 4271 prot->name); 4272 if (!rsk_prot->slab_name) 4273 return -ENOMEM; 4274 4275 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 4276 rsk_prot->obj_size, 0, 4277 SLAB_ACCOUNT | prot->slab_flags, 4278 NULL); 4279 4280 if (!rsk_prot->slab) { 4281 pr_crit("%s: Can't create request sock SLAB cache!\n", 4282 prot->name); 4283 return -ENOMEM; 4284 } 4285 return 0; 4286 } 4287 4288 int proto_register(struct proto *prot, int alloc_slab) 4289 { 4290 int ret = -ENOBUFS; 4291 4292 if (prot->memory_allocated && !prot->sysctl_mem) { 4293 pr_err("%s: missing sysctl_mem\n", prot->name); 4294 return -EINVAL; 4295 } 4296 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 4297 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 4298 return -EINVAL; 4299 } 4300 if (alloc_slab) { 4301 struct kmem_cache_args args = { 4302 .useroffset = prot->useroffset, 4303 .usersize = prot->usersize, 4304 .freeptr_offset = prot->freeptr_offset, 4305 .use_freeptr_offset = !!prot->freeptr_offset, 4306 }; 4307 4308 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 4309 &args, 4310 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 4311 prot->slab_flags); 4312 if (prot->slab == NULL) { 4313 pr_crit("%s: Can't create sock SLAB cache!\n", 4314 prot->name); 4315 goto out; 4316 } 4317 4318 if (req_prot_init(prot)) 4319 goto out_free_request_sock_slab; 4320 4321 if (tw_prot_init(prot)) 4322 goto out_free_timewait_sock_slab; 4323 } 4324 4325 mutex_lock(&proto_list_mutex); 4326 ret = assign_proto_idx(prot); 4327 if (ret) { 4328 mutex_unlock(&proto_list_mutex); 4329 goto out_free_timewait_sock_slab; 4330 } 4331 list_add(&prot->node, &proto_list); 4332 mutex_unlock(&proto_list_mutex); 4333 return ret; 4334 4335 out_free_timewait_sock_slab: 4336 if (alloc_slab) 4337 tw_prot_cleanup(prot->twsk_prot); 4338 out_free_request_sock_slab: 4339 if (alloc_slab) { 4340 req_prot_cleanup(prot->rsk_prot); 4341 4342 kmem_cache_destroy(prot->slab); 4343 prot->slab = NULL; 4344 } 4345 out: 4346 return ret; 4347 } 4348 EXPORT_SYMBOL(proto_register); 4349 4350 void proto_unregister(struct proto *prot) 4351 { 4352 mutex_lock(&proto_list_mutex); 4353 release_proto_idx(prot); 4354 list_del(&prot->node); 4355 mutex_unlock(&proto_list_mutex); 4356 4357 kmem_cache_destroy(prot->slab); 4358 prot->slab = NULL; 4359 4360 req_prot_cleanup(prot->rsk_prot); 4361 tw_prot_cleanup(prot->twsk_prot); 4362 } 4363 EXPORT_SYMBOL(proto_unregister); 4364 4365 int sock_load_diag_module(int family, int protocol) 4366 { 4367 if (!protocol) { 4368 if (!sock_is_registered(family)) 4369 return -ENOENT; 4370 4371 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 4372 NETLINK_SOCK_DIAG, family); 4373 } 4374 4375 #ifdef CONFIG_INET 4376 if (family == AF_INET && 4377 protocol != IPPROTO_RAW && 4378 protocol < MAX_INET_PROTOS && 4379 !rcu_access_pointer(inet_protos[protocol])) 4380 return -ENOENT; 4381 #endif 4382 4383 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 4384 NETLINK_SOCK_DIAG, family, protocol); 4385 } 4386 EXPORT_SYMBOL(sock_load_diag_module); 4387 4388 #ifdef CONFIG_PROC_FS 4389 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 4390 __acquires(proto_list_mutex) 4391 { 4392 mutex_lock(&proto_list_mutex); 4393 return seq_list_start_head(&proto_list, *pos); 4394 } 4395 4396 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 4397 { 4398 return seq_list_next(v, &proto_list, pos); 4399 } 4400 4401 static void proto_seq_stop(struct seq_file *seq, void *v) 4402 __releases(proto_list_mutex) 4403 { 4404 mutex_unlock(&proto_list_mutex); 4405 } 4406 4407 static char proto_method_implemented(const void *method) 4408 { 4409 return method == NULL ? 'n' : 'y'; 4410 } 4411 static long sock_prot_memory_allocated(struct proto *proto) 4412 { 4413 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 4414 } 4415 4416 static const char *sock_prot_memory_pressure(struct proto *proto) 4417 { 4418 return proto->memory_pressure != NULL ? 4419 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 4420 } 4421 4422 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 4423 { 4424 4425 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 4426 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 4427 proto->name, 4428 proto->obj_size, 4429 sock_prot_inuse_get(seq_file_net(seq), proto), 4430 sock_prot_memory_allocated(proto), 4431 sock_prot_memory_pressure(proto), 4432 proto->max_header, 4433 proto->slab == NULL ? "no" : "yes", 4434 module_name(proto->owner), 4435 proto_method_implemented(proto->close), 4436 proto_method_implemented(proto->connect), 4437 proto_method_implemented(proto->disconnect), 4438 proto_method_implemented(proto->accept), 4439 proto_method_implemented(proto->ioctl), 4440 proto_method_implemented(proto->init), 4441 proto_method_implemented(proto->destroy), 4442 proto_method_implemented(proto->shutdown), 4443 proto_method_implemented(proto->setsockopt), 4444 proto_method_implemented(proto->getsockopt), 4445 proto_method_implemented(proto->sendmsg), 4446 proto_method_implemented(proto->recvmsg), 4447 proto_method_implemented(proto->bind), 4448 proto_method_implemented(proto->backlog_rcv), 4449 proto_method_implemented(proto->hash), 4450 proto_method_implemented(proto->unhash), 4451 proto_method_implemented(proto->get_port), 4452 proto_method_implemented(proto->enter_memory_pressure)); 4453 } 4454 4455 static int proto_seq_show(struct seq_file *seq, void *v) 4456 { 4457 if (v == &proto_list) 4458 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4459 "protocol", 4460 "size", 4461 "sockets", 4462 "memory", 4463 "press", 4464 "maxhdr", 4465 "slab", 4466 "module", 4467 "cl co di ac io in de sh ss gs se re bi br ha uh gp em\n"); 4468 else 4469 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4470 return 0; 4471 } 4472 4473 static const struct seq_operations proto_seq_ops = { 4474 .start = proto_seq_start, 4475 .next = proto_seq_next, 4476 .stop = proto_seq_stop, 4477 .show = proto_seq_show, 4478 }; 4479 4480 static __net_init int proto_init_net(struct net *net) 4481 { 4482 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4483 sizeof(struct seq_net_private))) 4484 return -ENOMEM; 4485 4486 return 0; 4487 } 4488 4489 static __net_exit void proto_exit_net(struct net *net) 4490 { 4491 remove_proc_entry("protocols", net->proc_net); 4492 } 4493 4494 4495 static __net_initdata struct pernet_operations proto_net_ops = { 4496 .init = proto_init_net, 4497 .exit = proto_exit_net, 4498 }; 4499 4500 static int __init proto_init(void) 4501 { 4502 return register_pernet_subsys(&proto_net_ops); 4503 } 4504 4505 subsys_initcall(proto_init); 4506 4507 #endif /* PROC_FS */ 4508 4509 #ifdef CONFIG_NET_RX_BUSY_POLL 4510 bool sk_busy_loop_end(void *p, unsigned long start_time) 4511 { 4512 struct sock *sk = p; 4513 4514 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 4515 return true; 4516 4517 if (sk_is_udp(sk) && 4518 !skb_queue_empty_lockless(&udp_sk(sk)->reader_queue)) 4519 return true; 4520 4521 return sk_busy_loop_timeout(sk, start_time); 4522 } 4523 EXPORT_SYMBOL(sk_busy_loop_end); 4524 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4525 4526 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len) 4527 { 4528 if (!sk->sk_prot->bind_add) 4529 return -EOPNOTSUPP; 4530 return sk->sk_prot->bind_add(sk, addr, addr_len); 4531 } 4532 EXPORT_SYMBOL(sock_bind_add); 4533 4534 /* Copy 'size' bytes from userspace and return `size` back to userspace */ 4535 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 4536 void __user *arg, void *karg, size_t size) 4537 { 4538 int ret; 4539 4540 if (copy_from_user(karg, arg, size)) 4541 return -EFAULT; 4542 4543 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg); 4544 if (ret) 4545 return ret; 4546 4547 if (copy_to_user(arg, karg, size)) 4548 return -EFAULT; 4549 4550 return 0; 4551 } 4552 EXPORT_SYMBOL(sock_ioctl_inout); 4553 4554 /* This is the most common ioctl prep function, where the result (4 bytes) is 4555 * copied back to userspace if the ioctl() returns successfully. No input is 4556 * copied from userspace as input argument. 4557 */ 4558 static int sock_ioctl_out(struct sock *sk, unsigned int cmd, void __user *arg) 4559 { 4560 int ret, karg = 0; 4561 4562 ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, &karg); 4563 if (ret) 4564 return ret; 4565 4566 return put_user(karg, (int __user *)arg); 4567 } 4568 4569 /* A wrapper around sock ioctls, which copies the data from userspace 4570 * (depending on the protocol/ioctl), and copies back the result to userspace. 4571 * The main motivation for this function is to pass kernel memory to the 4572 * protocol ioctl callbacks, instead of userspace memory. 4573 */ 4574 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg) 4575 { 4576 int rc = 1; 4577 4578 if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET) 4579 rc = ipmr_sk_ioctl(sk, cmd, arg); 4580 else if (sk->sk_type == SOCK_RAW && sk->sk_family == AF_INET6) 4581 rc = ip6mr_sk_ioctl(sk, cmd, arg); 4582 else if (sk_is_phonet(sk)) 4583 rc = phonet_sk_ioctl(sk, cmd, arg); 4584 4585 /* If ioctl was processed, returns its value */ 4586 if (rc <= 0) 4587 return rc; 4588 4589 /* Otherwise call the default handler */ 4590 return sock_ioctl_out(sk, cmd, arg); 4591 } 4592 EXPORT_SYMBOL(sk_ioctl); 4593 4594 static int __init sock_struct_check(void) 4595 { 4596 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_drops); 4597 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_peek_off); 4598 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_error_queue); 4599 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_receive_queue); 4600 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rx, sk_backlog); 4601 4602 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst); 4603 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_ifindex); 4604 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rx_dst_cookie); 4605 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvbuf); 4606 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_filter); 4607 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_wq); 4608 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_data_ready); 4609 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvtimeo); 4610 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rx, sk_rcvlowat); 4611 4612 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_err); 4613 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_socket); 4614 #ifdef CONFIG_MEMCG 4615 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_rxtx, sk_memcg); 4616 #endif 4617 4618 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_lock); 4619 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_reserved_mem); 4620 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_forward_alloc); 4621 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_rxtx, sk_tsflags); 4622 4623 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4624 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_omem_alloc); 4625 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_err_soft); 4626 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_queued); 4627 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_wmem_alloc); 4628 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tsq_flags); 4629 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_send_head); 4630 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_queue); 4631 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_write_pending); 4632 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_frag); 4633 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_timer); 4634 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_pacing_rate); 4635 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_zckey); 4636 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_write_tx, sk_tskey); 4637 4638 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_pending_confirm); 4639 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_status); 4640 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_max_pacing_rate); 4641 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndtimeo); 4642 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_priority); 4643 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_mark); 4644 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_uid); 4645 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_protocol); 4646 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_dst_cache); 4647 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_route_caps); 4648 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_type); 4649 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_size); 4650 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_allocation); 4651 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_txhash); 4652 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_sndbuf); 4653 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_gso_max_segs); 4654 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_pacing_shift); 4655 CACHELINE_ASSERT_GROUP_MEMBER(struct sock, sock_read_tx, sk_use_task_frag); 4656 return 0; 4657 } 4658 4659 core_initcall(sock_struct_check); 4660