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