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 <asm/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/init.h> 111 #include <linux/highmem.h> 112 #include <linux/user_namespace.h> 113 #include <linux/static_key.h> 114 #include <linux/memcontrol.h> 115 #include <linux/prefetch.h> 116 #include <linux/compat.h> 117 118 #include <linux/uaccess.h> 119 120 #include <linux/netdevice.h> 121 #include <net/protocol.h> 122 #include <linux/skbuff.h> 123 #include <net/net_namespace.h> 124 #include <net/request_sock.h> 125 #include <net/sock.h> 126 #include <linux/net_tstamp.h> 127 #include <net/xfrm.h> 128 #include <linux/ipsec.h> 129 #include <net/cls_cgroup.h> 130 #include <net/netprio_cgroup.h> 131 #include <linux/sock_diag.h> 132 133 #include <linux/filter.h> 134 #include <net/sock_reuseport.h> 135 #include <net/bpf_sk_storage.h> 136 137 #include <trace/events/sock.h> 138 139 #include <net/tcp.h> 140 #include <net/busy_poll.h> 141 142 #include <linux/ethtool.h> 143 144 #include "dev.h" 145 146 static DEFINE_MUTEX(proto_list_mutex); 147 static LIST_HEAD(proto_list); 148 149 static void sock_def_write_space_wfree(struct sock *sk); 150 static void sock_def_write_space(struct sock *sk); 151 152 /** 153 * sk_ns_capable - General socket capability test 154 * @sk: Socket to use a capability on or through 155 * @user_ns: The user namespace of the capability to use 156 * @cap: The capability to use 157 * 158 * Test to see if the opener of the socket had when the socket was 159 * created and the current process has the capability @cap in the user 160 * namespace @user_ns. 161 */ 162 bool sk_ns_capable(const struct sock *sk, 163 struct user_namespace *user_ns, int cap) 164 { 165 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 166 ns_capable(user_ns, cap); 167 } 168 EXPORT_SYMBOL(sk_ns_capable); 169 170 /** 171 * sk_capable - Socket global capability test 172 * @sk: Socket to use a capability on or through 173 * @cap: The global capability to use 174 * 175 * Test to see if the opener of the socket had when the socket was 176 * created and the current process has the capability @cap in all user 177 * namespaces. 178 */ 179 bool sk_capable(const struct sock *sk, int cap) 180 { 181 return sk_ns_capable(sk, &init_user_ns, cap); 182 } 183 EXPORT_SYMBOL(sk_capable); 184 185 /** 186 * sk_net_capable - Network namespace socket capability test 187 * @sk: Socket to use a capability on or through 188 * @cap: The capability to use 189 * 190 * Test to see if the opener of the socket had when the socket was created 191 * and the current process has the capability @cap over the network namespace 192 * the socket is a member of. 193 */ 194 bool sk_net_capable(const struct sock *sk, int cap) 195 { 196 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 197 } 198 EXPORT_SYMBOL(sk_net_capable); 199 200 /* 201 * Each address family might have different locking rules, so we have 202 * one slock key per address family and separate keys for internal and 203 * userspace sockets. 204 */ 205 static struct lock_class_key af_family_keys[AF_MAX]; 206 static struct lock_class_key af_family_kern_keys[AF_MAX]; 207 static struct lock_class_key af_family_slock_keys[AF_MAX]; 208 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 209 210 /* 211 * Make lock validator output more readable. (we pre-construct these 212 * strings build-time, so that runtime initialization of socket 213 * locks is fast): 214 */ 215 216 #define _sock_locks(x) \ 217 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 218 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 219 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 220 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 221 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 222 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 223 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 224 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 225 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 226 x "27" , x "28" , x "AF_CAN" , \ 227 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 228 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 229 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 230 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 231 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 232 x "AF_MCTP" , \ 233 x "AF_MAX" 234 235 static const char *const af_family_key_strings[AF_MAX+1] = { 236 _sock_locks("sk_lock-") 237 }; 238 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 239 _sock_locks("slock-") 240 }; 241 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 242 _sock_locks("clock-") 243 }; 244 245 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 246 _sock_locks("k-sk_lock-") 247 }; 248 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 249 _sock_locks("k-slock-") 250 }; 251 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 252 _sock_locks("k-clock-") 253 }; 254 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 255 _sock_locks("rlock-") 256 }; 257 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 258 _sock_locks("wlock-") 259 }; 260 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 261 _sock_locks("elock-") 262 }; 263 264 /* 265 * sk_callback_lock and sk queues locking rules are per-address-family, 266 * so split the lock classes by using a per-AF key: 267 */ 268 static struct lock_class_key af_callback_keys[AF_MAX]; 269 static struct lock_class_key af_rlock_keys[AF_MAX]; 270 static struct lock_class_key af_wlock_keys[AF_MAX]; 271 static struct lock_class_key af_elock_keys[AF_MAX]; 272 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 273 274 /* Run time adjustable parameters. */ 275 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 276 EXPORT_SYMBOL(sysctl_wmem_max); 277 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 278 EXPORT_SYMBOL(sysctl_rmem_max); 279 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 280 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 281 282 /* Maximal space eaten by iovec or ancillary data plus some space */ 283 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 284 EXPORT_SYMBOL(sysctl_optmem_max); 285 286 int sysctl_tstamp_allow_data __read_mostly = 1; 287 288 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 289 EXPORT_SYMBOL_GPL(memalloc_socks_key); 290 291 /** 292 * sk_set_memalloc - sets %SOCK_MEMALLOC 293 * @sk: socket to set it on 294 * 295 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 296 * It's the responsibility of the admin to adjust min_free_kbytes 297 * to meet the requirements 298 */ 299 void sk_set_memalloc(struct sock *sk) 300 { 301 sock_set_flag(sk, SOCK_MEMALLOC); 302 sk->sk_allocation |= __GFP_MEMALLOC; 303 static_branch_inc(&memalloc_socks_key); 304 } 305 EXPORT_SYMBOL_GPL(sk_set_memalloc); 306 307 void sk_clear_memalloc(struct sock *sk) 308 { 309 sock_reset_flag(sk, SOCK_MEMALLOC); 310 sk->sk_allocation &= ~__GFP_MEMALLOC; 311 static_branch_dec(&memalloc_socks_key); 312 313 /* 314 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 315 * progress of swapping. SOCK_MEMALLOC may be cleared while 316 * it has rmem allocations due to the last swapfile being deactivated 317 * but there is a risk that the socket is unusable due to exceeding 318 * the rmem limits. Reclaim the reserves and obey rmem limits again. 319 */ 320 sk_mem_reclaim(sk); 321 } 322 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 323 324 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 325 { 326 int ret; 327 unsigned int noreclaim_flag; 328 329 /* these should have been dropped before queueing */ 330 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 331 332 noreclaim_flag = memalloc_noreclaim_save(); 333 ret = INDIRECT_CALL_INET(sk->sk_backlog_rcv, 334 tcp_v6_do_rcv, 335 tcp_v4_do_rcv, 336 sk, skb); 337 memalloc_noreclaim_restore(noreclaim_flag); 338 339 return ret; 340 } 341 EXPORT_SYMBOL(__sk_backlog_rcv); 342 343 void sk_error_report(struct sock *sk) 344 { 345 sk->sk_error_report(sk); 346 347 switch (sk->sk_family) { 348 case AF_INET: 349 fallthrough; 350 case AF_INET6: 351 trace_inet_sk_error_report(sk); 352 break; 353 default: 354 break; 355 } 356 } 357 EXPORT_SYMBOL(sk_error_report); 358 359 int sock_get_timeout(long timeo, void *optval, bool old_timeval) 360 { 361 struct __kernel_sock_timeval tv; 362 363 if (timeo == MAX_SCHEDULE_TIMEOUT) { 364 tv.tv_sec = 0; 365 tv.tv_usec = 0; 366 } else { 367 tv.tv_sec = timeo / HZ; 368 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 369 } 370 371 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 372 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 373 *(struct old_timeval32 *)optval = tv32; 374 return sizeof(tv32); 375 } 376 377 if (old_timeval) { 378 struct __kernel_old_timeval old_tv; 379 old_tv.tv_sec = tv.tv_sec; 380 old_tv.tv_usec = tv.tv_usec; 381 *(struct __kernel_old_timeval *)optval = old_tv; 382 return sizeof(old_tv); 383 } 384 385 *(struct __kernel_sock_timeval *)optval = tv; 386 return sizeof(tv); 387 } 388 EXPORT_SYMBOL(sock_get_timeout); 389 390 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 391 sockptr_t optval, int optlen, bool old_timeval) 392 { 393 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 394 struct old_timeval32 tv32; 395 396 if (optlen < sizeof(tv32)) 397 return -EINVAL; 398 399 if (copy_from_sockptr(&tv32, optval, sizeof(tv32))) 400 return -EFAULT; 401 tv->tv_sec = tv32.tv_sec; 402 tv->tv_usec = tv32.tv_usec; 403 } else if (old_timeval) { 404 struct __kernel_old_timeval old_tv; 405 406 if (optlen < sizeof(old_tv)) 407 return -EINVAL; 408 if (copy_from_sockptr(&old_tv, optval, sizeof(old_tv))) 409 return -EFAULT; 410 tv->tv_sec = old_tv.tv_sec; 411 tv->tv_usec = old_tv.tv_usec; 412 } else { 413 if (optlen < sizeof(*tv)) 414 return -EINVAL; 415 if (copy_from_sockptr(tv, optval, sizeof(*tv))) 416 return -EFAULT; 417 } 418 419 return 0; 420 } 421 EXPORT_SYMBOL(sock_copy_user_timeval); 422 423 static int sock_set_timeout(long *timeo_p, sockptr_t optval, int optlen, 424 bool old_timeval) 425 { 426 struct __kernel_sock_timeval tv; 427 int err = sock_copy_user_timeval(&tv, optval, optlen, old_timeval); 428 429 if (err) 430 return err; 431 432 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 433 return -EDOM; 434 435 if (tv.tv_sec < 0) { 436 static int warned __read_mostly; 437 438 *timeo_p = 0; 439 if (warned < 10 && net_ratelimit()) { 440 warned++; 441 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 442 __func__, current->comm, task_pid_nr(current)); 443 } 444 return 0; 445 } 446 *timeo_p = MAX_SCHEDULE_TIMEOUT; 447 if (tv.tv_sec == 0 && tv.tv_usec == 0) 448 return 0; 449 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 450 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 451 return 0; 452 } 453 454 static bool sock_needs_netstamp(const struct sock *sk) 455 { 456 switch (sk->sk_family) { 457 case AF_UNSPEC: 458 case AF_UNIX: 459 return false; 460 default: 461 return true; 462 } 463 } 464 465 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 466 { 467 if (sk->sk_flags & flags) { 468 sk->sk_flags &= ~flags; 469 if (sock_needs_netstamp(sk) && 470 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 471 net_disable_timestamp(); 472 } 473 } 474 475 476 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 477 { 478 unsigned long flags; 479 struct sk_buff_head *list = &sk->sk_receive_queue; 480 481 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 482 atomic_inc(&sk->sk_drops); 483 trace_sock_rcvqueue_full(sk, skb); 484 return -ENOMEM; 485 } 486 487 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 488 atomic_inc(&sk->sk_drops); 489 return -ENOBUFS; 490 } 491 492 skb->dev = NULL; 493 skb_set_owner_r(skb, sk); 494 495 /* we escape from rcu protected region, make sure we dont leak 496 * a norefcounted dst 497 */ 498 skb_dst_force(skb); 499 500 spin_lock_irqsave(&list->lock, flags); 501 sock_skb_set_dropcount(sk, skb); 502 __skb_queue_tail(list, skb); 503 spin_unlock_irqrestore(&list->lock, flags); 504 505 if (!sock_flag(sk, SOCK_DEAD)) 506 sk->sk_data_ready(sk); 507 return 0; 508 } 509 EXPORT_SYMBOL(__sock_queue_rcv_skb); 510 511 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 512 enum skb_drop_reason *reason) 513 { 514 enum skb_drop_reason drop_reason; 515 int err; 516 517 err = sk_filter(sk, skb); 518 if (err) { 519 drop_reason = SKB_DROP_REASON_SOCKET_FILTER; 520 goto out; 521 } 522 err = __sock_queue_rcv_skb(sk, skb); 523 switch (err) { 524 case -ENOMEM: 525 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF; 526 break; 527 case -ENOBUFS: 528 drop_reason = SKB_DROP_REASON_PROTO_MEM; 529 break; 530 default: 531 drop_reason = SKB_NOT_DROPPED_YET; 532 break; 533 } 534 out: 535 if (reason) 536 *reason = drop_reason; 537 return err; 538 } 539 EXPORT_SYMBOL(sock_queue_rcv_skb_reason); 540 541 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 542 const int nested, unsigned int trim_cap, bool refcounted) 543 { 544 int rc = NET_RX_SUCCESS; 545 546 if (sk_filter_trim_cap(sk, skb, trim_cap)) 547 goto discard_and_relse; 548 549 skb->dev = NULL; 550 551 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 552 atomic_inc(&sk->sk_drops); 553 goto discard_and_relse; 554 } 555 if (nested) 556 bh_lock_sock_nested(sk); 557 else 558 bh_lock_sock(sk); 559 if (!sock_owned_by_user(sk)) { 560 /* 561 * trylock + unlock semantics: 562 */ 563 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 564 565 rc = sk_backlog_rcv(sk, skb); 566 567 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 568 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 569 bh_unlock_sock(sk); 570 atomic_inc(&sk->sk_drops); 571 goto discard_and_relse; 572 } 573 574 bh_unlock_sock(sk); 575 out: 576 if (refcounted) 577 sock_put(sk); 578 return rc; 579 discard_and_relse: 580 kfree_skb(skb); 581 goto out; 582 } 583 EXPORT_SYMBOL(__sk_receive_skb); 584 585 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ip6_dst_check(struct dst_entry *, 586 u32)); 587 INDIRECT_CALLABLE_DECLARE(struct dst_entry *ipv4_dst_check(struct dst_entry *, 588 u32)); 589 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 590 { 591 struct dst_entry *dst = __sk_dst_get(sk); 592 593 if (dst && dst->obsolete && 594 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 595 dst, cookie) == NULL) { 596 sk_tx_queue_clear(sk); 597 sk->sk_dst_pending_confirm = 0; 598 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 599 dst_release(dst); 600 return NULL; 601 } 602 603 return dst; 604 } 605 EXPORT_SYMBOL(__sk_dst_check); 606 607 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 608 { 609 struct dst_entry *dst = sk_dst_get(sk); 610 611 if (dst && dst->obsolete && 612 INDIRECT_CALL_INET(dst->ops->check, ip6_dst_check, ipv4_dst_check, 613 dst, cookie) == NULL) { 614 sk_dst_reset(sk); 615 dst_release(dst); 616 return NULL; 617 } 618 619 return dst; 620 } 621 EXPORT_SYMBOL(sk_dst_check); 622 623 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 624 { 625 int ret = -ENOPROTOOPT; 626 #ifdef CONFIG_NETDEVICES 627 struct net *net = sock_net(sk); 628 629 /* Sorry... */ 630 ret = -EPERM; 631 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 632 goto out; 633 634 ret = -EINVAL; 635 if (ifindex < 0) 636 goto out; 637 638 /* Paired with all READ_ONCE() done locklessly. */ 639 WRITE_ONCE(sk->sk_bound_dev_if, ifindex); 640 641 if (sk->sk_prot->rehash) 642 sk->sk_prot->rehash(sk); 643 sk_dst_reset(sk); 644 645 ret = 0; 646 647 out: 648 #endif 649 650 return ret; 651 } 652 653 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 654 { 655 int ret; 656 657 if (lock_sk) 658 lock_sock(sk); 659 ret = sock_bindtoindex_locked(sk, ifindex); 660 if (lock_sk) 661 release_sock(sk); 662 663 return ret; 664 } 665 EXPORT_SYMBOL(sock_bindtoindex); 666 667 static int sock_setbindtodevice(struct sock *sk, sockptr_t optval, int optlen) 668 { 669 int ret = -ENOPROTOOPT; 670 #ifdef CONFIG_NETDEVICES 671 struct net *net = sock_net(sk); 672 char devname[IFNAMSIZ]; 673 int index; 674 675 ret = -EINVAL; 676 if (optlen < 0) 677 goto out; 678 679 /* Bind this socket to a particular device like "eth0", 680 * as specified in the passed interface name. If the 681 * name is "" or the option length is zero the socket 682 * is not bound. 683 */ 684 if (optlen > IFNAMSIZ - 1) 685 optlen = IFNAMSIZ - 1; 686 memset(devname, 0, sizeof(devname)); 687 688 ret = -EFAULT; 689 if (copy_from_sockptr(devname, optval, optlen)) 690 goto out; 691 692 index = 0; 693 if (devname[0] != '\0') { 694 struct net_device *dev; 695 696 rcu_read_lock(); 697 dev = dev_get_by_name_rcu(net, devname); 698 if (dev) 699 index = dev->ifindex; 700 rcu_read_unlock(); 701 ret = -ENODEV; 702 if (!dev) 703 goto out; 704 } 705 706 sockopt_lock_sock(sk); 707 ret = sock_bindtoindex_locked(sk, index); 708 sockopt_release_sock(sk); 709 out: 710 #endif 711 712 return ret; 713 } 714 715 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 716 int __user *optlen, int len) 717 { 718 int ret = -ENOPROTOOPT; 719 #ifdef CONFIG_NETDEVICES 720 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 721 struct net *net = sock_net(sk); 722 char devname[IFNAMSIZ]; 723 724 if (bound_dev_if == 0) { 725 len = 0; 726 goto zero; 727 } 728 729 ret = -EINVAL; 730 if (len < IFNAMSIZ) 731 goto out; 732 733 ret = netdev_get_name(net, devname, bound_dev_if); 734 if (ret) 735 goto out; 736 737 len = strlen(devname) + 1; 738 739 ret = -EFAULT; 740 if (copy_to_user(optval, devname, len)) 741 goto out; 742 743 zero: 744 ret = -EFAULT; 745 if (put_user(len, optlen)) 746 goto out; 747 748 ret = 0; 749 750 out: 751 #endif 752 753 return ret; 754 } 755 756 bool sk_mc_loop(struct sock *sk) 757 { 758 if (dev_recursion_level()) 759 return false; 760 if (!sk) 761 return true; 762 switch (sk->sk_family) { 763 case AF_INET: 764 return inet_sk(sk)->mc_loop; 765 #if IS_ENABLED(CONFIG_IPV6) 766 case AF_INET6: 767 return inet6_sk(sk)->mc_loop; 768 #endif 769 } 770 WARN_ON_ONCE(1); 771 return true; 772 } 773 EXPORT_SYMBOL(sk_mc_loop); 774 775 void sock_set_reuseaddr(struct sock *sk) 776 { 777 lock_sock(sk); 778 sk->sk_reuse = SK_CAN_REUSE; 779 release_sock(sk); 780 } 781 EXPORT_SYMBOL(sock_set_reuseaddr); 782 783 void sock_set_reuseport(struct sock *sk) 784 { 785 lock_sock(sk); 786 sk->sk_reuseport = true; 787 release_sock(sk); 788 } 789 EXPORT_SYMBOL(sock_set_reuseport); 790 791 void sock_no_linger(struct sock *sk) 792 { 793 lock_sock(sk); 794 sk->sk_lingertime = 0; 795 sock_set_flag(sk, SOCK_LINGER); 796 release_sock(sk); 797 } 798 EXPORT_SYMBOL(sock_no_linger); 799 800 void sock_set_priority(struct sock *sk, u32 priority) 801 { 802 lock_sock(sk); 803 sk->sk_priority = priority; 804 release_sock(sk); 805 } 806 EXPORT_SYMBOL(sock_set_priority); 807 808 void sock_set_sndtimeo(struct sock *sk, s64 secs) 809 { 810 lock_sock(sk); 811 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 812 sk->sk_sndtimeo = secs * HZ; 813 else 814 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 815 release_sock(sk); 816 } 817 EXPORT_SYMBOL(sock_set_sndtimeo); 818 819 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 820 { 821 if (val) { 822 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 823 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 824 sock_set_flag(sk, SOCK_RCVTSTAMP); 825 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 826 } else { 827 sock_reset_flag(sk, SOCK_RCVTSTAMP); 828 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 829 } 830 } 831 832 void sock_enable_timestamps(struct sock *sk) 833 { 834 lock_sock(sk); 835 __sock_set_timestamps(sk, true, false, true); 836 release_sock(sk); 837 } 838 EXPORT_SYMBOL(sock_enable_timestamps); 839 840 void sock_set_timestamp(struct sock *sk, int optname, bool valbool) 841 { 842 switch (optname) { 843 case SO_TIMESTAMP_OLD: 844 __sock_set_timestamps(sk, valbool, false, false); 845 break; 846 case SO_TIMESTAMP_NEW: 847 __sock_set_timestamps(sk, valbool, true, false); 848 break; 849 case SO_TIMESTAMPNS_OLD: 850 __sock_set_timestamps(sk, valbool, false, true); 851 break; 852 case SO_TIMESTAMPNS_NEW: 853 __sock_set_timestamps(sk, valbool, true, true); 854 break; 855 } 856 } 857 858 static int sock_timestamping_bind_phc(struct sock *sk, int phc_index) 859 { 860 struct net *net = sock_net(sk); 861 struct net_device *dev = NULL; 862 bool match = false; 863 int *vclock_index; 864 int i, num; 865 866 if (sk->sk_bound_dev_if) 867 dev = dev_get_by_index(net, sk->sk_bound_dev_if); 868 869 if (!dev) { 870 pr_err("%s: sock not bind to device\n", __func__); 871 return -EOPNOTSUPP; 872 } 873 874 num = ethtool_get_phc_vclocks(dev, &vclock_index); 875 dev_put(dev); 876 877 for (i = 0; i < num; i++) { 878 if (*(vclock_index + i) == phc_index) { 879 match = true; 880 break; 881 } 882 } 883 884 if (num > 0) 885 kfree(vclock_index); 886 887 if (!match) 888 return -EINVAL; 889 890 sk->sk_bind_phc = phc_index; 891 892 return 0; 893 } 894 895 int sock_set_timestamping(struct sock *sk, int optname, 896 struct so_timestamping timestamping) 897 { 898 int val = timestamping.flags; 899 int ret; 900 901 if (val & ~SOF_TIMESTAMPING_MASK) 902 return -EINVAL; 903 904 if (val & SOF_TIMESTAMPING_OPT_ID && 905 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 906 if (sk_is_tcp(sk)) { 907 if ((1 << sk->sk_state) & 908 (TCPF_CLOSE | TCPF_LISTEN)) 909 return -EINVAL; 910 atomic_set(&sk->sk_tskey, tcp_sk(sk)->snd_una); 911 } else { 912 atomic_set(&sk->sk_tskey, 0); 913 } 914 } 915 916 if (val & SOF_TIMESTAMPING_OPT_STATS && 917 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) 918 return -EINVAL; 919 920 if (val & SOF_TIMESTAMPING_BIND_PHC) { 921 ret = sock_timestamping_bind_phc(sk, timestamping.bind_phc); 922 if (ret) 923 return ret; 924 } 925 926 sk->sk_tsflags = val; 927 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, optname == SO_TIMESTAMPING_NEW); 928 929 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 930 sock_enable_timestamp(sk, 931 SOCK_TIMESTAMPING_RX_SOFTWARE); 932 else 933 sock_disable_timestamp(sk, 934 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 935 return 0; 936 } 937 938 void sock_set_keepalive(struct sock *sk) 939 { 940 lock_sock(sk); 941 if (sk->sk_prot->keepalive) 942 sk->sk_prot->keepalive(sk, true); 943 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 944 release_sock(sk); 945 } 946 EXPORT_SYMBOL(sock_set_keepalive); 947 948 static void __sock_set_rcvbuf(struct sock *sk, int val) 949 { 950 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 951 * as a negative value. 952 */ 953 val = min_t(int, val, INT_MAX / 2); 954 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 955 956 /* We double it on the way in to account for "struct sk_buff" etc. 957 * overhead. Applications assume that the SO_RCVBUF setting they make 958 * will allow that much actual data to be received on that socket. 959 * 960 * Applications are unaware that "struct sk_buff" and other overheads 961 * allocate from the receive buffer during socket buffer allocation. 962 * 963 * And after considering the possible alternatives, returning the value 964 * we actually used in getsockopt is the most desirable behavior. 965 */ 966 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 967 } 968 969 void sock_set_rcvbuf(struct sock *sk, int val) 970 { 971 lock_sock(sk); 972 __sock_set_rcvbuf(sk, val); 973 release_sock(sk); 974 } 975 EXPORT_SYMBOL(sock_set_rcvbuf); 976 977 static void __sock_set_mark(struct sock *sk, u32 val) 978 { 979 if (val != sk->sk_mark) { 980 sk->sk_mark = val; 981 sk_dst_reset(sk); 982 } 983 } 984 985 void sock_set_mark(struct sock *sk, u32 val) 986 { 987 lock_sock(sk); 988 __sock_set_mark(sk, val); 989 release_sock(sk); 990 } 991 EXPORT_SYMBOL(sock_set_mark); 992 993 static void sock_release_reserved_memory(struct sock *sk, int bytes) 994 { 995 /* Round down bytes to multiple of pages */ 996 bytes = round_down(bytes, PAGE_SIZE); 997 998 WARN_ON(bytes > sk->sk_reserved_mem); 999 sk->sk_reserved_mem -= bytes; 1000 sk_mem_reclaim(sk); 1001 } 1002 1003 static int sock_reserve_memory(struct sock *sk, int bytes) 1004 { 1005 long allocated; 1006 bool charged; 1007 int pages; 1008 1009 if (!mem_cgroup_sockets_enabled || !sk->sk_memcg || !sk_has_account(sk)) 1010 return -EOPNOTSUPP; 1011 1012 if (!bytes) 1013 return 0; 1014 1015 pages = sk_mem_pages(bytes); 1016 1017 /* pre-charge to memcg */ 1018 charged = mem_cgroup_charge_skmem(sk->sk_memcg, pages, 1019 GFP_KERNEL | __GFP_RETRY_MAYFAIL); 1020 if (!charged) 1021 return -ENOMEM; 1022 1023 /* pre-charge to forward_alloc */ 1024 sk_memory_allocated_add(sk, pages); 1025 allocated = sk_memory_allocated(sk); 1026 /* If the system goes into memory pressure with this 1027 * precharge, give up and return error. 1028 */ 1029 if (allocated > sk_prot_mem_limits(sk, 1)) { 1030 sk_memory_allocated_sub(sk, pages); 1031 mem_cgroup_uncharge_skmem(sk->sk_memcg, pages); 1032 return -ENOMEM; 1033 } 1034 sk->sk_forward_alloc += pages << PAGE_SHIFT; 1035 1036 sk->sk_reserved_mem += pages << PAGE_SHIFT; 1037 1038 return 0; 1039 } 1040 1041 void sockopt_lock_sock(struct sock *sk) 1042 { 1043 /* When current->bpf_ctx is set, the setsockopt is called from 1044 * a bpf prog. bpf has ensured the sk lock has been 1045 * acquired before calling setsockopt(). 1046 */ 1047 if (has_current_bpf_ctx()) 1048 return; 1049 1050 lock_sock(sk); 1051 } 1052 EXPORT_SYMBOL(sockopt_lock_sock); 1053 1054 void sockopt_release_sock(struct sock *sk) 1055 { 1056 if (has_current_bpf_ctx()) 1057 return; 1058 1059 release_sock(sk); 1060 } 1061 EXPORT_SYMBOL(sockopt_release_sock); 1062 1063 bool sockopt_ns_capable(struct user_namespace *ns, int cap) 1064 { 1065 return has_current_bpf_ctx() || ns_capable(ns, cap); 1066 } 1067 EXPORT_SYMBOL(sockopt_ns_capable); 1068 1069 bool sockopt_capable(int cap) 1070 { 1071 return has_current_bpf_ctx() || capable(cap); 1072 } 1073 EXPORT_SYMBOL(sockopt_capable); 1074 1075 /* 1076 * This is meant for all protocols to use and covers goings on 1077 * at the socket level. Everything here is generic. 1078 */ 1079 1080 int sk_setsockopt(struct sock *sk, int level, int optname, 1081 sockptr_t optval, unsigned int optlen) 1082 { 1083 struct so_timestamping timestamping; 1084 struct socket *sock = sk->sk_socket; 1085 struct sock_txtime sk_txtime; 1086 int val; 1087 int valbool; 1088 struct linger ling; 1089 int ret = 0; 1090 1091 /* 1092 * Options without arguments 1093 */ 1094 1095 if (optname == SO_BINDTODEVICE) 1096 return sock_setbindtodevice(sk, optval, optlen); 1097 1098 if (optlen < sizeof(int)) 1099 return -EINVAL; 1100 1101 if (copy_from_sockptr(&val, optval, sizeof(val))) 1102 return -EFAULT; 1103 1104 valbool = val ? 1 : 0; 1105 1106 sockopt_lock_sock(sk); 1107 1108 switch (optname) { 1109 case SO_DEBUG: 1110 if (val && !sockopt_capable(CAP_NET_ADMIN)) 1111 ret = -EACCES; 1112 else 1113 sock_valbool_flag(sk, SOCK_DBG, valbool); 1114 break; 1115 case SO_REUSEADDR: 1116 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 1117 break; 1118 case SO_REUSEPORT: 1119 sk->sk_reuseport = valbool; 1120 break; 1121 case SO_TYPE: 1122 case SO_PROTOCOL: 1123 case SO_DOMAIN: 1124 case SO_ERROR: 1125 ret = -ENOPROTOOPT; 1126 break; 1127 case SO_DONTROUTE: 1128 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 1129 sk_dst_reset(sk); 1130 break; 1131 case SO_BROADCAST: 1132 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 1133 break; 1134 case SO_SNDBUF: 1135 /* Don't error on this BSD doesn't and if you think 1136 * about it this is right. Otherwise apps have to 1137 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1138 * are treated in BSD as hints 1139 */ 1140 val = min_t(u32, val, sysctl_wmem_max); 1141 set_sndbuf: 1142 /* Ensure val * 2 fits into an int, to prevent max_t() 1143 * from treating it as a negative value. 1144 */ 1145 val = min_t(int, val, INT_MAX / 2); 1146 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 1147 WRITE_ONCE(sk->sk_sndbuf, 1148 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 1149 /* Wake up sending tasks if we upped the value. */ 1150 sk->sk_write_space(sk); 1151 break; 1152 1153 case SO_SNDBUFFORCE: 1154 if (!sockopt_capable(CAP_NET_ADMIN)) { 1155 ret = -EPERM; 1156 break; 1157 } 1158 1159 /* No negative values (to prevent underflow, as val will be 1160 * multiplied by 2). 1161 */ 1162 if (val < 0) 1163 val = 0; 1164 goto set_sndbuf; 1165 1166 case SO_RCVBUF: 1167 /* Don't error on this BSD doesn't and if you think 1168 * about it this is right. Otherwise apps have to 1169 * play 'guess the biggest size' games. RCVBUF/SNDBUF 1170 * are treated in BSD as hints 1171 */ 1172 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max)); 1173 break; 1174 1175 case SO_RCVBUFFORCE: 1176 if (!sockopt_capable(CAP_NET_ADMIN)) { 1177 ret = -EPERM; 1178 break; 1179 } 1180 1181 /* No negative values (to prevent underflow, as val will be 1182 * multiplied by 2). 1183 */ 1184 __sock_set_rcvbuf(sk, max(val, 0)); 1185 break; 1186 1187 case SO_KEEPALIVE: 1188 if (sk->sk_prot->keepalive) 1189 sk->sk_prot->keepalive(sk, valbool); 1190 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 1191 break; 1192 1193 case SO_OOBINLINE: 1194 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 1195 break; 1196 1197 case SO_NO_CHECK: 1198 sk->sk_no_check_tx = valbool; 1199 break; 1200 1201 case SO_PRIORITY: 1202 if ((val >= 0 && val <= 6) || 1203 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) || 1204 sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 1205 sk->sk_priority = val; 1206 else 1207 ret = -EPERM; 1208 break; 1209 1210 case SO_LINGER: 1211 if (optlen < sizeof(ling)) { 1212 ret = -EINVAL; /* 1003.1g */ 1213 break; 1214 } 1215 if (copy_from_sockptr(&ling, optval, sizeof(ling))) { 1216 ret = -EFAULT; 1217 break; 1218 } 1219 if (!ling.l_onoff) 1220 sock_reset_flag(sk, SOCK_LINGER); 1221 else { 1222 #if (BITS_PER_LONG == 32) 1223 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 1224 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 1225 else 1226 #endif 1227 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 1228 sock_set_flag(sk, SOCK_LINGER); 1229 } 1230 break; 1231 1232 case SO_BSDCOMPAT: 1233 break; 1234 1235 case SO_PASSCRED: 1236 if (valbool) 1237 set_bit(SOCK_PASSCRED, &sock->flags); 1238 else 1239 clear_bit(SOCK_PASSCRED, &sock->flags); 1240 break; 1241 1242 case SO_TIMESTAMP_OLD: 1243 case SO_TIMESTAMP_NEW: 1244 case SO_TIMESTAMPNS_OLD: 1245 case SO_TIMESTAMPNS_NEW: 1246 sock_set_timestamp(sk, optname, valbool); 1247 break; 1248 1249 case SO_TIMESTAMPING_NEW: 1250 case SO_TIMESTAMPING_OLD: 1251 if (optlen == sizeof(timestamping)) { 1252 if (copy_from_sockptr(×tamping, optval, 1253 sizeof(timestamping))) { 1254 ret = -EFAULT; 1255 break; 1256 } 1257 } else { 1258 memset(×tamping, 0, sizeof(timestamping)); 1259 timestamping.flags = val; 1260 } 1261 ret = sock_set_timestamping(sk, optname, timestamping); 1262 break; 1263 1264 case SO_RCVLOWAT: 1265 if (val < 0) 1266 val = INT_MAX; 1267 if (sock && sock->ops->set_rcvlowat) 1268 ret = sock->ops->set_rcvlowat(sk, val); 1269 else 1270 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1271 break; 1272 1273 case SO_RCVTIMEO_OLD: 1274 case SO_RCVTIMEO_NEW: 1275 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, 1276 optlen, optname == SO_RCVTIMEO_OLD); 1277 break; 1278 1279 case SO_SNDTIMEO_OLD: 1280 case SO_SNDTIMEO_NEW: 1281 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, 1282 optlen, optname == SO_SNDTIMEO_OLD); 1283 break; 1284 1285 case SO_ATTACH_FILTER: { 1286 struct sock_fprog fprog; 1287 1288 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1289 if (!ret) 1290 ret = sk_attach_filter(&fprog, sk); 1291 break; 1292 } 1293 case SO_ATTACH_BPF: 1294 ret = -EINVAL; 1295 if (optlen == sizeof(u32)) { 1296 u32 ufd; 1297 1298 ret = -EFAULT; 1299 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1300 break; 1301 1302 ret = sk_attach_bpf(ufd, sk); 1303 } 1304 break; 1305 1306 case SO_ATTACH_REUSEPORT_CBPF: { 1307 struct sock_fprog fprog; 1308 1309 ret = copy_bpf_fprog_from_user(&fprog, optval, optlen); 1310 if (!ret) 1311 ret = sk_reuseport_attach_filter(&fprog, sk); 1312 break; 1313 } 1314 case SO_ATTACH_REUSEPORT_EBPF: 1315 ret = -EINVAL; 1316 if (optlen == sizeof(u32)) { 1317 u32 ufd; 1318 1319 ret = -EFAULT; 1320 if (copy_from_sockptr(&ufd, optval, sizeof(ufd))) 1321 break; 1322 1323 ret = sk_reuseport_attach_bpf(ufd, sk); 1324 } 1325 break; 1326 1327 case SO_DETACH_REUSEPORT_BPF: 1328 ret = reuseport_detach_prog(sk); 1329 break; 1330 1331 case SO_DETACH_FILTER: 1332 ret = sk_detach_filter(sk); 1333 break; 1334 1335 case SO_LOCK_FILTER: 1336 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1337 ret = -EPERM; 1338 else 1339 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1340 break; 1341 1342 case SO_PASSSEC: 1343 if (valbool) 1344 set_bit(SOCK_PASSSEC, &sock->flags); 1345 else 1346 clear_bit(SOCK_PASSSEC, &sock->flags); 1347 break; 1348 case SO_MARK: 1349 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1350 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1351 ret = -EPERM; 1352 break; 1353 } 1354 1355 __sock_set_mark(sk, val); 1356 break; 1357 case SO_RCVMARK: 1358 if (!sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 1359 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1360 ret = -EPERM; 1361 break; 1362 } 1363 1364 sock_valbool_flag(sk, SOCK_RCVMARK, valbool); 1365 break; 1366 1367 case SO_RXQ_OVFL: 1368 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1369 break; 1370 1371 case SO_WIFI_STATUS: 1372 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1373 break; 1374 1375 case SO_PEEK_OFF: 1376 if (sock->ops->set_peek_off) 1377 ret = sock->ops->set_peek_off(sk, val); 1378 else 1379 ret = -EOPNOTSUPP; 1380 break; 1381 1382 case SO_NOFCS: 1383 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1384 break; 1385 1386 case SO_SELECT_ERR_QUEUE: 1387 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1388 break; 1389 1390 #ifdef CONFIG_NET_RX_BUSY_POLL 1391 case SO_BUSY_POLL: 1392 /* allow unprivileged users to decrease the value */ 1393 if ((val > sk->sk_ll_usec) && !sockopt_capable(CAP_NET_ADMIN)) 1394 ret = -EPERM; 1395 else { 1396 if (val < 0) 1397 ret = -EINVAL; 1398 else 1399 WRITE_ONCE(sk->sk_ll_usec, val); 1400 } 1401 break; 1402 case SO_PREFER_BUSY_POLL: 1403 if (valbool && !sockopt_capable(CAP_NET_ADMIN)) 1404 ret = -EPERM; 1405 else 1406 WRITE_ONCE(sk->sk_prefer_busy_poll, valbool); 1407 break; 1408 case SO_BUSY_POLL_BUDGET: 1409 if (val > READ_ONCE(sk->sk_busy_poll_budget) && !sockopt_capable(CAP_NET_ADMIN)) { 1410 ret = -EPERM; 1411 } else { 1412 if (val < 0 || val > U16_MAX) 1413 ret = -EINVAL; 1414 else 1415 WRITE_ONCE(sk->sk_busy_poll_budget, val); 1416 } 1417 break; 1418 #endif 1419 1420 case SO_MAX_PACING_RATE: 1421 { 1422 unsigned long ulval = (val == ~0U) ? ~0UL : (unsigned int)val; 1423 1424 if (sizeof(ulval) != sizeof(val) && 1425 optlen >= sizeof(ulval) && 1426 copy_from_sockptr(&ulval, optval, sizeof(ulval))) { 1427 ret = -EFAULT; 1428 break; 1429 } 1430 if (ulval != ~0UL) 1431 cmpxchg(&sk->sk_pacing_status, 1432 SK_PACING_NONE, 1433 SK_PACING_NEEDED); 1434 sk->sk_max_pacing_rate = ulval; 1435 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1436 break; 1437 } 1438 case SO_INCOMING_CPU: 1439 WRITE_ONCE(sk->sk_incoming_cpu, val); 1440 break; 1441 1442 case SO_CNX_ADVICE: 1443 if (val == 1) 1444 dst_negative_advice(sk); 1445 break; 1446 1447 case SO_ZEROCOPY: 1448 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1449 if (!(sk_is_tcp(sk) || 1450 (sk->sk_type == SOCK_DGRAM && 1451 sk->sk_protocol == IPPROTO_UDP))) 1452 ret = -EOPNOTSUPP; 1453 } else if (sk->sk_family != PF_RDS) { 1454 ret = -EOPNOTSUPP; 1455 } 1456 if (!ret) { 1457 if (val < 0 || val > 1) 1458 ret = -EINVAL; 1459 else 1460 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1461 } 1462 break; 1463 1464 case SO_TXTIME: 1465 if (optlen != sizeof(struct sock_txtime)) { 1466 ret = -EINVAL; 1467 break; 1468 } else if (copy_from_sockptr(&sk_txtime, optval, 1469 sizeof(struct sock_txtime))) { 1470 ret = -EFAULT; 1471 break; 1472 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1473 ret = -EINVAL; 1474 break; 1475 } 1476 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1477 * scheduler has enough safe guards. 1478 */ 1479 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1480 !sockopt_ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1481 ret = -EPERM; 1482 break; 1483 } 1484 sock_valbool_flag(sk, SOCK_TXTIME, true); 1485 sk->sk_clockid = sk_txtime.clockid; 1486 sk->sk_txtime_deadline_mode = 1487 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1488 sk->sk_txtime_report_errors = 1489 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1490 break; 1491 1492 case SO_BINDTOIFINDEX: 1493 ret = sock_bindtoindex_locked(sk, val); 1494 break; 1495 1496 case SO_BUF_LOCK: 1497 if (val & ~SOCK_BUF_LOCK_MASK) { 1498 ret = -EINVAL; 1499 break; 1500 } 1501 sk->sk_userlocks = val | (sk->sk_userlocks & 1502 ~SOCK_BUF_LOCK_MASK); 1503 break; 1504 1505 case SO_RESERVE_MEM: 1506 { 1507 int delta; 1508 1509 if (val < 0) { 1510 ret = -EINVAL; 1511 break; 1512 } 1513 1514 delta = val - sk->sk_reserved_mem; 1515 if (delta < 0) 1516 sock_release_reserved_memory(sk, -delta); 1517 else 1518 ret = sock_reserve_memory(sk, delta); 1519 break; 1520 } 1521 1522 case SO_TXREHASH: 1523 if (val < -1 || val > 1) { 1524 ret = -EINVAL; 1525 break; 1526 } 1527 /* Paired with READ_ONCE() in tcp_rtx_synack() */ 1528 WRITE_ONCE(sk->sk_txrehash, (u8)val); 1529 break; 1530 1531 default: 1532 ret = -ENOPROTOOPT; 1533 break; 1534 } 1535 sockopt_release_sock(sk); 1536 return ret; 1537 } 1538 1539 int sock_setsockopt(struct socket *sock, int level, int optname, 1540 sockptr_t optval, unsigned int optlen) 1541 { 1542 return sk_setsockopt(sock->sk, level, optname, 1543 optval, optlen); 1544 } 1545 EXPORT_SYMBOL(sock_setsockopt); 1546 1547 static const struct cred *sk_get_peer_cred(struct sock *sk) 1548 { 1549 const struct cred *cred; 1550 1551 spin_lock(&sk->sk_peer_lock); 1552 cred = get_cred(sk->sk_peer_cred); 1553 spin_unlock(&sk->sk_peer_lock); 1554 1555 return cred; 1556 } 1557 1558 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1559 struct ucred *ucred) 1560 { 1561 ucred->pid = pid_vnr(pid); 1562 ucred->uid = ucred->gid = -1; 1563 if (cred) { 1564 struct user_namespace *current_ns = current_user_ns(); 1565 1566 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1567 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1568 } 1569 } 1570 1571 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1572 { 1573 struct user_namespace *user_ns = current_user_ns(); 1574 int i; 1575 1576 for (i = 0; i < src->ngroups; i++) 1577 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1578 return -EFAULT; 1579 1580 return 0; 1581 } 1582 1583 int sock_getsockopt(struct socket *sock, int level, int optname, 1584 char __user *optval, int __user *optlen) 1585 { 1586 struct sock *sk = sock->sk; 1587 1588 union { 1589 int val; 1590 u64 val64; 1591 unsigned long ulval; 1592 struct linger ling; 1593 struct old_timeval32 tm32; 1594 struct __kernel_old_timeval tm; 1595 struct __kernel_sock_timeval stm; 1596 struct sock_txtime txtime; 1597 struct so_timestamping timestamping; 1598 } v; 1599 1600 int lv = sizeof(int); 1601 int len; 1602 1603 if (get_user(len, optlen)) 1604 return -EFAULT; 1605 if (len < 0) 1606 return -EINVAL; 1607 1608 memset(&v, 0, sizeof(v)); 1609 1610 switch (optname) { 1611 case SO_DEBUG: 1612 v.val = sock_flag(sk, SOCK_DBG); 1613 break; 1614 1615 case SO_DONTROUTE: 1616 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1617 break; 1618 1619 case SO_BROADCAST: 1620 v.val = sock_flag(sk, SOCK_BROADCAST); 1621 break; 1622 1623 case SO_SNDBUF: 1624 v.val = sk->sk_sndbuf; 1625 break; 1626 1627 case SO_RCVBUF: 1628 v.val = sk->sk_rcvbuf; 1629 break; 1630 1631 case SO_REUSEADDR: 1632 v.val = sk->sk_reuse; 1633 break; 1634 1635 case SO_REUSEPORT: 1636 v.val = sk->sk_reuseport; 1637 break; 1638 1639 case SO_KEEPALIVE: 1640 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1641 break; 1642 1643 case SO_TYPE: 1644 v.val = sk->sk_type; 1645 break; 1646 1647 case SO_PROTOCOL: 1648 v.val = sk->sk_protocol; 1649 break; 1650 1651 case SO_DOMAIN: 1652 v.val = sk->sk_family; 1653 break; 1654 1655 case SO_ERROR: 1656 v.val = -sock_error(sk); 1657 if (v.val == 0) 1658 v.val = xchg(&sk->sk_err_soft, 0); 1659 break; 1660 1661 case SO_OOBINLINE: 1662 v.val = sock_flag(sk, SOCK_URGINLINE); 1663 break; 1664 1665 case SO_NO_CHECK: 1666 v.val = sk->sk_no_check_tx; 1667 break; 1668 1669 case SO_PRIORITY: 1670 v.val = sk->sk_priority; 1671 break; 1672 1673 case SO_LINGER: 1674 lv = sizeof(v.ling); 1675 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1676 v.ling.l_linger = sk->sk_lingertime / HZ; 1677 break; 1678 1679 case SO_BSDCOMPAT: 1680 break; 1681 1682 case SO_TIMESTAMP_OLD: 1683 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1684 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1685 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1686 break; 1687 1688 case SO_TIMESTAMPNS_OLD: 1689 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1690 break; 1691 1692 case SO_TIMESTAMP_NEW: 1693 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1694 break; 1695 1696 case SO_TIMESTAMPNS_NEW: 1697 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1698 break; 1699 1700 case SO_TIMESTAMPING_OLD: 1701 lv = sizeof(v.timestamping); 1702 v.timestamping.flags = sk->sk_tsflags; 1703 v.timestamping.bind_phc = sk->sk_bind_phc; 1704 break; 1705 1706 case SO_RCVTIMEO_OLD: 1707 case SO_RCVTIMEO_NEW: 1708 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1709 break; 1710 1711 case SO_SNDTIMEO_OLD: 1712 case SO_SNDTIMEO_NEW: 1713 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1714 break; 1715 1716 case SO_RCVLOWAT: 1717 v.val = sk->sk_rcvlowat; 1718 break; 1719 1720 case SO_SNDLOWAT: 1721 v.val = 1; 1722 break; 1723 1724 case SO_PASSCRED: 1725 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1726 break; 1727 1728 case SO_PEERCRED: 1729 { 1730 struct ucred peercred; 1731 if (len > sizeof(peercred)) 1732 len = sizeof(peercred); 1733 1734 spin_lock(&sk->sk_peer_lock); 1735 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1736 spin_unlock(&sk->sk_peer_lock); 1737 1738 if (copy_to_user(optval, &peercred, len)) 1739 return -EFAULT; 1740 goto lenout; 1741 } 1742 1743 case SO_PEERGROUPS: 1744 { 1745 const struct cred *cred; 1746 int ret, n; 1747 1748 cred = sk_get_peer_cred(sk); 1749 if (!cred) 1750 return -ENODATA; 1751 1752 n = cred->group_info->ngroups; 1753 if (len < n * sizeof(gid_t)) { 1754 len = n * sizeof(gid_t); 1755 put_cred(cred); 1756 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1757 } 1758 len = n * sizeof(gid_t); 1759 1760 ret = groups_to_user((gid_t __user *)optval, cred->group_info); 1761 put_cred(cred); 1762 if (ret) 1763 return ret; 1764 goto lenout; 1765 } 1766 1767 case SO_PEERNAME: 1768 { 1769 char address[128]; 1770 1771 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1772 if (lv < 0) 1773 return -ENOTCONN; 1774 if (lv < len) 1775 return -EINVAL; 1776 if (copy_to_user(optval, address, len)) 1777 return -EFAULT; 1778 goto lenout; 1779 } 1780 1781 /* Dubious BSD thing... Probably nobody even uses it, but 1782 * the UNIX standard wants it for whatever reason... -DaveM 1783 */ 1784 case SO_ACCEPTCONN: 1785 v.val = sk->sk_state == TCP_LISTEN; 1786 break; 1787 1788 case SO_PASSSEC: 1789 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1790 break; 1791 1792 case SO_PEERSEC: 1793 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1794 1795 case SO_MARK: 1796 v.val = sk->sk_mark; 1797 break; 1798 1799 case SO_RCVMARK: 1800 v.val = sock_flag(sk, SOCK_RCVMARK); 1801 break; 1802 1803 case SO_RXQ_OVFL: 1804 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1805 break; 1806 1807 case SO_WIFI_STATUS: 1808 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1809 break; 1810 1811 case SO_PEEK_OFF: 1812 if (!sock->ops->set_peek_off) 1813 return -EOPNOTSUPP; 1814 1815 v.val = sk->sk_peek_off; 1816 break; 1817 case SO_NOFCS: 1818 v.val = sock_flag(sk, SOCK_NOFCS); 1819 break; 1820 1821 case SO_BINDTODEVICE: 1822 return sock_getbindtodevice(sk, optval, optlen, len); 1823 1824 case SO_GET_FILTER: 1825 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1826 if (len < 0) 1827 return len; 1828 1829 goto lenout; 1830 1831 case SO_LOCK_FILTER: 1832 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1833 break; 1834 1835 case SO_BPF_EXTENSIONS: 1836 v.val = bpf_tell_extensions(); 1837 break; 1838 1839 case SO_SELECT_ERR_QUEUE: 1840 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1841 break; 1842 1843 #ifdef CONFIG_NET_RX_BUSY_POLL 1844 case SO_BUSY_POLL: 1845 v.val = sk->sk_ll_usec; 1846 break; 1847 case SO_PREFER_BUSY_POLL: 1848 v.val = READ_ONCE(sk->sk_prefer_busy_poll); 1849 break; 1850 #endif 1851 1852 case SO_MAX_PACING_RATE: 1853 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1854 lv = sizeof(v.ulval); 1855 v.ulval = sk->sk_max_pacing_rate; 1856 } else { 1857 /* 32bit version */ 1858 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1859 } 1860 break; 1861 1862 case SO_INCOMING_CPU: 1863 v.val = READ_ONCE(sk->sk_incoming_cpu); 1864 break; 1865 1866 case SO_MEMINFO: 1867 { 1868 u32 meminfo[SK_MEMINFO_VARS]; 1869 1870 sk_get_meminfo(sk, meminfo); 1871 1872 len = min_t(unsigned int, len, sizeof(meminfo)); 1873 if (copy_to_user(optval, &meminfo, len)) 1874 return -EFAULT; 1875 1876 goto lenout; 1877 } 1878 1879 #ifdef CONFIG_NET_RX_BUSY_POLL 1880 case SO_INCOMING_NAPI_ID: 1881 v.val = READ_ONCE(sk->sk_napi_id); 1882 1883 /* aggregate non-NAPI IDs down to 0 */ 1884 if (v.val < MIN_NAPI_ID) 1885 v.val = 0; 1886 1887 break; 1888 #endif 1889 1890 case SO_COOKIE: 1891 lv = sizeof(u64); 1892 if (len < lv) 1893 return -EINVAL; 1894 v.val64 = sock_gen_cookie(sk); 1895 break; 1896 1897 case SO_ZEROCOPY: 1898 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1899 break; 1900 1901 case SO_TXTIME: 1902 lv = sizeof(v.txtime); 1903 v.txtime.clockid = sk->sk_clockid; 1904 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1905 SOF_TXTIME_DEADLINE_MODE : 0; 1906 v.txtime.flags |= sk->sk_txtime_report_errors ? 1907 SOF_TXTIME_REPORT_ERRORS : 0; 1908 break; 1909 1910 case SO_BINDTOIFINDEX: 1911 v.val = READ_ONCE(sk->sk_bound_dev_if); 1912 break; 1913 1914 case SO_NETNS_COOKIE: 1915 lv = sizeof(u64); 1916 if (len != lv) 1917 return -EINVAL; 1918 v.val64 = sock_net(sk)->net_cookie; 1919 break; 1920 1921 case SO_BUF_LOCK: 1922 v.val = sk->sk_userlocks & SOCK_BUF_LOCK_MASK; 1923 break; 1924 1925 case SO_RESERVE_MEM: 1926 v.val = sk->sk_reserved_mem; 1927 break; 1928 1929 case SO_TXREHASH: 1930 v.val = sk->sk_txrehash; 1931 break; 1932 1933 default: 1934 /* We implement the SO_SNDLOWAT etc to not be settable 1935 * (1003.1g 7). 1936 */ 1937 return -ENOPROTOOPT; 1938 } 1939 1940 if (len > lv) 1941 len = lv; 1942 if (copy_to_user(optval, &v, len)) 1943 return -EFAULT; 1944 lenout: 1945 if (put_user(len, optlen)) 1946 return -EFAULT; 1947 return 0; 1948 } 1949 1950 /* 1951 * Initialize an sk_lock. 1952 * 1953 * (We also register the sk_lock with the lock validator.) 1954 */ 1955 static inline void sock_lock_init(struct sock *sk) 1956 { 1957 if (sk->sk_kern_sock) 1958 sock_lock_init_class_and_name( 1959 sk, 1960 af_family_kern_slock_key_strings[sk->sk_family], 1961 af_family_kern_slock_keys + sk->sk_family, 1962 af_family_kern_key_strings[sk->sk_family], 1963 af_family_kern_keys + sk->sk_family); 1964 else 1965 sock_lock_init_class_and_name( 1966 sk, 1967 af_family_slock_key_strings[sk->sk_family], 1968 af_family_slock_keys + sk->sk_family, 1969 af_family_key_strings[sk->sk_family], 1970 af_family_keys + sk->sk_family); 1971 } 1972 1973 /* 1974 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1975 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1976 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1977 */ 1978 static void sock_copy(struct sock *nsk, const struct sock *osk) 1979 { 1980 const struct proto *prot = READ_ONCE(osk->sk_prot); 1981 #ifdef CONFIG_SECURITY_NETWORK 1982 void *sptr = nsk->sk_security; 1983 #endif 1984 1985 /* If we move sk_tx_queue_mapping out of the private section, 1986 * we must check if sk_tx_queue_clear() is called after 1987 * sock_copy() in sk_clone_lock(). 1988 */ 1989 BUILD_BUG_ON(offsetof(struct sock, sk_tx_queue_mapping) < 1990 offsetof(struct sock, sk_dontcopy_begin) || 1991 offsetof(struct sock, sk_tx_queue_mapping) >= 1992 offsetof(struct sock, sk_dontcopy_end)); 1993 1994 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1995 1996 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1997 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1998 1999 #ifdef CONFIG_SECURITY_NETWORK 2000 nsk->sk_security = sptr; 2001 security_sk_clone(osk, nsk); 2002 #endif 2003 } 2004 2005 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 2006 int family) 2007 { 2008 struct sock *sk; 2009 struct kmem_cache *slab; 2010 2011 slab = prot->slab; 2012 if (slab != NULL) { 2013 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 2014 if (!sk) 2015 return sk; 2016 if (want_init_on_alloc(priority)) 2017 sk_prot_clear_nulls(sk, prot->obj_size); 2018 } else 2019 sk = kmalloc(prot->obj_size, priority); 2020 2021 if (sk != NULL) { 2022 if (security_sk_alloc(sk, family, priority)) 2023 goto out_free; 2024 2025 if (!try_module_get(prot->owner)) 2026 goto out_free_sec; 2027 } 2028 2029 return sk; 2030 2031 out_free_sec: 2032 security_sk_free(sk); 2033 out_free: 2034 if (slab != NULL) 2035 kmem_cache_free(slab, sk); 2036 else 2037 kfree(sk); 2038 return NULL; 2039 } 2040 2041 static void sk_prot_free(struct proto *prot, struct sock *sk) 2042 { 2043 struct kmem_cache *slab; 2044 struct module *owner; 2045 2046 owner = prot->owner; 2047 slab = prot->slab; 2048 2049 cgroup_sk_free(&sk->sk_cgrp_data); 2050 mem_cgroup_sk_free(sk); 2051 security_sk_free(sk); 2052 if (slab != NULL) 2053 kmem_cache_free(slab, sk); 2054 else 2055 kfree(sk); 2056 module_put(owner); 2057 } 2058 2059 /** 2060 * sk_alloc - All socket objects are allocated here 2061 * @net: the applicable net namespace 2062 * @family: protocol family 2063 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2064 * @prot: struct proto associated with this new sock instance 2065 * @kern: is this to be a kernel socket? 2066 */ 2067 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 2068 struct proto *prot, int kern) 2069 { 2070 struct sock *sk; 2071 2072 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 2073 if (sk) { 2074 sk->sk_family = family; 2075 /* 2076 * See comment in struct sock definition to understand 2077 * why we need sk_prot_creator -acme 2078 */ 2079 sk->sk_prot = sk->sk_prot_creator = prot; 2080 sk->sk_kern_sock = kern; 2081 sock_lock_init(sk); 2082 sk->sk_net_refcnt = kern ? 0 : 1; 2083 if (likely(sk->sk_net_refcnt)) { 2084 get_net_track(net, &sk->ns_tracker, priority); 2085 sock_inuse_add(net, 1); 2086 } 2087 2088 sock_net_set(sk, net); 2089 refcount_set(&sk->sk_wmem_alloc, 1); 2090 2091 mem_cgroup_sk_alloc(sk); 2092 cgroup_sk_alloc(&sk->sk_cgrp_data); 2093 sock_update_classid(&sk->sk_cgrp_data); 2094 sock_update_netprioidx(&sk->sk_cgrp_data); 2095 sk_tx_queue_clear(sk); 2096 } 2097 2098 return sk; 2099 } 2100 EXPORT_SYMBOL(sk_alloc); 2101 2102 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 2103 * grace period. This is the case for UDP sockets and TCP listeners. 2104 */ 2105 static void __sk_destruct(struct rcu_head *head) 2106 { 2107 struct sock *sk = container_of(head, struct sock, sk_rcu); 2108 struct sk_filter *filter; 2109 2110 if (sk->sk_destruct) 2111 sk->sk_destruct(sk); 2112 2113 filter = rcu_dereference_check(sk->sk_filter, 2114 refcount_read(&sk->sk_wmem_alloc) == 0); 2115 if (filter) { 2116 sk_filter_uncharge(sk, filter); 2117 RCU_INIT_POINTER(sk->sk_filter, NULL); 2118 } 2119 2120 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 2121 2122 #ifdef CONFIG_BPF_SYSCALL 2123 bpf_sk_storage_free(sk); 2124 #endif 2125 2126 if (atomic_read(&sk->sk_omem_alloc)) 2127 pr_debug("%s: optmem leakage (%d bytes) detected\n", 2128 __func__, atomic_read(&sk->sk_omem_alloc)); 2129 2130 if (sk->sk_frag.page) { 2131 put_page(sk->sk_frag.page); 2132 sk->sk_frag.page = NULL; 2133 } 2134 2135 /* We do not need to acquire sk->sk_peer_lock, we are the last user. */ 2136 put_cred(sk->sk_peer_cred); 2137 put_pid(sk->sk_peer_pid); 2138 2139 if (likely(sk->sk_net_refcnt)) 2140 put_net_track(sock_net(sk), &sk->ns_tracker); 2141 sk_prot_free(sk->sk_prot_creator, sk); 2142 } 2143 2144 void sk_destruct(struct sock *sk) 2145 { 2146 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 2147 2148 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 2149 reuseport_detach_sock(sk); 2150 use_call_rcu = true; 2151 } 2152 2153 if (use_call_rcu) 2154 call_rcu(&sk->sk_rcu, __sk_destruct); 2155 else 2156 __sk_destruct(&sk->sk_rcu); 2157 } 2158 2159 static void __sk_free(struct sock *sk) 2160 { 2161 if (likely(sk->sk_net_refcnt)) 2162 sock_inuse_add(sock_net(sk), -1); 2163 2164 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 2165 sock_diag_broadcast_destroy(sk); 2166 else 2167 sk_destruct(sk); 2168 } 2169 2170 void sk_free(struct sock *sk) 2171 { 2172 /* 2173 * We subtract one from sk_wmem_alloc and can know if 2174 * some packets are still in some tx queue. 2175 * If not null, sock_wfree() will call __sk_free(sk) later 2176 */ 2177 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 2178 __sk_free(sk); 2179 } 2180 EXPORT_SYMBOL(sk_free); 2181 2182 static void sk_init_common(struct sock *sk) 2183 { 2184 skb_queue_head_init(&sk->sk_receive_queue); 2185 skb_queue_head_init(&sk->sk_write_queue); 2186 skb_queue_head_init(&sk->sk_error_queue); 2187 2188 rwlock_init(&sk->sk_callback_lock); 2189 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 2190 af_rlock_keys + sk->sk_family, 2191 af_family_rlock_key_strings[sk->sk_family]); 2192 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 2193 af_wlock_keys + sk->sk_family, 2194 af_family_wlock_key_strings[sk->sk_family]); 2195 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 2196 af_elock_keys + sk->sk_family, 2197 af_family_elock_key_strings[sk->sk_family]); 2198 lockdep_set_class_and_name(&sk->sk_callback_lock, 2199 af_callback_keys + sk->sk_family, 2200 af_family_clock_key_strings[sk->sk_family]); 2201 } 2202 2203 /** 2204 * sk_clone_lock - clone a socket, and lock its clone 2205 * @sk: the socket to clone 2206 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 2207 * 2208 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 2209 */ 2210 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 2211 { 2212 struct proto *prot = READ_ONCE(sk->sk_prot); 2213 struct sk_filter *filter; 2214 bool is_charged = true; 2215 struct sock *newsk; 2216 2217 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 2218 if (!newsk) 2219 goto out; 2220 2221 sock_copy(newsk, sk); 2222 2223 newsk->sk_prot_creator = prot; 2224 2225 /* SANITY */ 2226 if (likely(newsk->sk_net_refcnt)) { 2227 get_net_track(sock_net(newsk), &newsk->ns_tracker, priority); 2228 sock_inuse_add(sock_net(newsk), 1); 2229 } 2230 sk_node_init(&newsk->sk_node); 2231 sock_lock_init(newsk); 2232 bh_lock_sock(newsk); 2233 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 2234 newsk->sk_backlog.len = 0; 2235 2236 atomic_set(&newsk->sk_rmem_alloc, 0); 2237 2238 /* sk_wmem_alloc set to one (see sk_free() and sock_wfree()) */ 2239 refcount_set(&newsk->sk_wmem_alloc, 1); 2240 2241 atomic_set(&newsk->sk_omem_alloc, 0); 2242 sk_init_common(newsk); 2243 2244 newsk->sk_dst_cache = NULL; 2245 newsk->sk_dst_pending_confirm = 0; 2246 newsk->sk_wmem_queued = 0; 2247 newsk->sk_forward_alloc = 0; 2248 newsk->sk_reserved_mem = 0; 2249 atomic_set(&newsk->sk_drops, 0); 2250 newsk->sk_send_head = NULL; 2251 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 2252 atomic_set(&newsk->sk_zckey, 0); 2253 2254 sock_reset_flag(newsk, SOCK_DONE); 2255 2256 /* sk->sk_memcg will be populated at accept() time */ 2257 newsk->sk_memcg = NULL; 2258 2259 cgroup_sk_clone(&newsk->sk_cgrp_data); 2260 2261 rcu_read_lock(); 2262 filter = rcu_dereference(sk->sk_filter); 2263 if (filter != NULL) 2264 /* though it's an empty new sock, the charging may fail 2265 * if sysctl_optmem_max was changed between creation of 2266 * original socket and cloning 2267 */ 2268 is_charged = sk_filter_charge(newsk, filter); 2269 RCU_INIT_POINTER(newsk->sk_filter, filter); 2270 rcu_read_unlock(); 2271 2272 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 2273 /* We need to make sure that we don't uncharge the new 2274 * socket if we couldn't charge it in the first place 2275 * as otherwise we uncharge the parent's filter. 2276 */ 2277 if (!is_charged) 2278 RCU_INIT_POINTER(newsk->sk_filter, NULL); 2279 sk_free_unlock_clone(newsk); 2280 newsk = NULL; 2281 goto out; 2282 } 2283 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 2284 2285 if (bpf_sk_storage_clone(sk, newsk)) { 2286 sk_free_unlock_clone(newsk); 2287 newsk = NULL; 2288 goto out; 2289 } 2290 2291 /* Clear sk_user_data if parent had the pointer tagged 2292 * as not suitable for copying when cloning. 2293 */ 2294 if (sk_user_data_is_nocopy(newsk)) 2295 newsk->sk_user_data = NULL; 2296 2297 newsk->sk_err = 0; 2298 newsk->sk_err_soft = 0; 2299 newsk->sk_priority = 0; 2300 newsk->sk_incoming_cpu = raw_smp_processor_id(); 2301 2302 /* Before updating sk_refcnt, we must commit prior changes to memory 2303 * (Documentation/RCU/rculist_nulls.rst for details) 2304 */ 2305 smp_wmb(); 2306 refcount_set(&newsk->sk_refcnt, 2); 2307 2308 /* Increment the counter in the same struct proto as the master 2309 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 2310 * is the same as sk->sk_prot->socks, as this field was copied 2311 * with memcpy). 2312 * 2313 * This _changes_ the previous behaviour, where 2314 * tcp_create_openreq_child always was incrementing the 2315 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 2316 * to be taken into account in all callers. -acme 2317 */ 2318 sk_refcnt_debug_inc(newsk); 2319 sk_set_socket(newsk, NULL); 2320 sk_tx_queue_clear(newsk); 2321 RCU_INIT_POINTER(newsk->sk_wq, NULL); 2322 2323 if (newsk->sk_prot->sockets_allocated) 2324 sk_sockets_allocated_inc(newsk); 2325 2326 if (sock_needs_netstamp(sk) && newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2327 net_enable_timestamp(); 2328 out: 2329 return newsk; 2330 } 2331 EXPORT_SYMBOL_GPL(sk_clone_lock); 2332 2333 void sk_free_unlock_clone(struct sock *sk) 2334 { 2335 /* It is still raw copy of parent, so invalidate 2336 * destructor and make plain sk_free() */ 2337 sk->sk_destruct = NULL; 2338 bh_unlock_sock(sk); 2339 sk_free(sk); 2340 } 2341 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2342 2343 static void sk_trim_gso_size(struct sock *sk) 2344 { 2345 if (sk->sk_gso_max_size <= GSO_LEGACY_MAX_SIZE) 2346 return; 2347 #if IS_ENABLED(CONFIG_IPV6) 2348 if (sk->sk_family == AF_INET6 && 2349 sk_is_tcp(sk) && 2350 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr)) 2351 return; 2352 #endif 2353 sk->sk_gso_max_size = GSO_LEGACY_MAX_SIZE; 2354 } 2355 2356 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2357 { 2358 u32 max_segs = 1; 2359 2360 sk_dst_set(sk, dst); 2361 sk->sk_route_caps = dst->dev->features; 2362 if (sk_is_tcp(sk)) 2363 sk->sk_route_caps |= NETIF_F_GSO; 2364 if (sk->sk_route_caps & NETIF_F_GSO) 2365 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2366 if (unlikely(sk->sk_gso_disabled)) 2367 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2368 if (sk_can_gso(sk)) { 2369 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2370 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2371 } else { 2372 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2373 /* pairs with the WRITE_ONCE() in netif_set_gso_max_size() */ 2374 sk->sk_gso_max_size = READ_ONCE(dst->dev->gso_max_size); 2375 sk_trim_gso_size(sk); 2376 sk->sk_gso_max_size -= (MAX_TCP_HEADER + 1); 2377 /* pairs with the WRITE_ONCE() in netif_set_gso_max_segs() */ 2378 max_segs = max_t(u32, READ_ONCE(dst->dev->gso_max_segs), 1); 2379 } 2380 } 2381 sk->sk_gso_max_segs = max_segs; 2382 } 2383 EXPORT_SYMBOL_GPL(sk_setup_caps); 2384 2385 /* 2386 * Simple resource managers for sockets. 2387 */ 2388 2389 2390 /* 2391 * Write buffer destructor automatically called from kfree_skb. 2392 */ 2393 void sock_wfree(struct sk_buff *skb) 2394 { 2395 struct sock *sk = skb->sk; 2396 unsigned int len = skb->truesize; 2397 bool free; 2398 2399 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2400 if (sock_flag(sk, SOCK_RCU_FREE) && 2401 sk->sk_write_space == sock_def_write_space) { 2402 rcu_read_lock(); 2403 free = refcount_sub_and_test(len, &sk->sk_wmem_alloc); 2404 sock_def_write_space_wfree(sk); 2405 rcu_read_unlock(); 2406 if (unlikely(free)) 2407 __sk_free(sk); 2408 return; 2409 } 2410 2411 /* 2412 * Keep a reference on sk_wmem_alloc, this will be released 2413 * after sk_write_space() call 2414 */ 2415 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2416 sk->sk_write_space(sk); 2417 len = 1; 2418 } 2419 /* 2420 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2421 * could not do because of in-flight packets 2422 */ 2423 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2424 __sk_free(sk); 2425 } 2426 EXPORT_SYMBOL(sock_wfree); 2427 2428 /* This variant of sock_wfree() is used by TCP, 2429 * since it sets SOCK_USE_WRITE_QUEUE. 2430 */ 2431 void __sock_wfree(struct sk_buff *skb) 2432 { 2433 struct sock *sk = skb->sk; 2434 2435 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2436 __sk_free(sk); 2437 } 2438 2439 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2440 { 2441 skb_orphan(skb); 2442 skb->sk = sk; 2443 #ifdef CONFIG_INET 2444 if (unlikely(!sk_fullsock(sk))) { 2445 skb->destructor = sock_edemux; 2446 sock_hold(sk); 2447 return; 2448 } 2449 #endif 2450 skb->destructor = sock_wfree; 2451 skb_set_hash_from_sk(skb, sk); 2452 /* 2453 * We used to take a refcount on sk, but following operation 2454 * is enough to guarantee sk_free() wont free this sock until 2455 * all in-flight packets are completed 2456 */ 2457 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2458 } 2459 EXPORT_SYMBOL(skb_set_owner_w); 2460 2461 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2462 { 2463 #ifdef CONFIG_TLS_DEVICE 2464 /* Drivers depend on in-order delivery for crypto offload, 2465 * partial orphan breaks out-of-order-OK logic. 2466 */ 2467 if (skb->decrypted) 2468 return false; 2469 #endif 2470 return (skb->destructor == sock_wfree || 2471 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2472 } 2473 2474 /* This helper is used by netem, as it can hold packets in its 2475 * delay queue. We want to allow the owner socket to send more 2476 * packets, as if they were already TX completed by a typical driver. 2477 * But we also want to keep skb->sk set because some packet schedulers 2478 * rely on it (sch_fq for example). 2479 */ 2480 void skb_orphan_partial(struct sk_buff *skb) 2481 { 2482 if (skb_is_tcp_pure_ack(skb)) 2483 return; 2484 2485 if (can_skb_orphan_partial(skb) && skb_set_owner_sk_safe(skb, skb->sk)) 2486 return; 2487 2488 skb_orphan(skb); 2489 } 2490 EXPORT_SYMBOL(skb_orphan_partial); 2491 2492 /* 2493 * Read buffer destructor automatically called from kfree_skb. 2494 */ 2495 void sock_rfree(struct sk_buff *skb) 2496 { 2497 struct sock *sk = skb->sk; 2498 unsigned int len = skb->truesize; 2499 2500 atomic_sub(len, &sk->sk_rmem_alloc); 2501 sk_mem_uncharge(sk, len); 2502 } 2503 EXPORT_SYMBOL(sock_rfree); 2504 2505 /* 2506 * Buffer destructor for skbs that are not used directly in read or write 2507 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2508 */ 2509 void sock_efree(struct sk_buff *skb) 2510 { 2511 sock_put(skb->sk); 2512 } 2513 EXPORT_SYMBOL(sock_efree); 2514 2515 /* Buffer destructor for prefetch/receive path where reference count may 2516 * not be held, e.g. for listen sockets. 2517 */ 2518 #ifdef CONFIG_INET 2519 void sock_pfree(struct sk_buff *skb) 2520 { 2521 if (sk_is_refcounted(skb->sk)) 2522 sock_gen_put(skb->sk); 2523 } 2524 EXPORT_SYMBOL(sock_pfree); 2525 #endif /* CONFIG_INET */ 2526 2527 kuid_t sock_i_uid(struct sock *sk) 2528 { 2529 kuid_t uid; 2530 2531 read_lock_bh(&sk->sk_callback_lock); 2532 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2533 read_unlock_bh(&sk->sk_callback_lock); 2534 return uid; 2535 } 2536 EXPORT_SYMBOL(sock_i_uid); 2537 2538 unsigned long sock_i_ino(struct sock *sk) 2539 { 2540 unsigned long ino; 2541 2542 read_lock_bh(&sk->sk_callback_lock); 2543 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2544 read_unlock_bh(&sk->sk_callback_lock); 2545 return ino; 2546 } 2547 EXPORT_SYMBOL(sock_i_ino); 2548 2549 /* 2550 * Allocate a skb from the socket's send buffer. 2551 */ 2552 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2553 gfp_t priority) 2554 { 2555 if (force || 2556 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2557 struct sk_buff *skb = alloc_skb(size, priority); 2558 2559 if (skb) { 2560 skb_set_owner_w(skb, sk); 2561 return skb; 2562 } 2563 } 2564 return NULL; 2565 } 2566 EXPORT_SYMBOL(sock_wmalloc); 2567 2568 static void sock_ofree(struct sk_buff *skb) 2569 { 2570 struct sock *sk = skb->sk; 2571 2572 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2573 } 2574 2575 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2576 gfp_t priority) 2577 { 2578 struct sk_buff *skb; 2579 2580 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2581 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2582 sysctl_optmem_max) 2583 return NULL; 2584 2585 skb = alloc_skb(size, priority); 2586 if (!skb) 2587 return NULL; 2588 2589 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2590 skb->sk = sk; 2591 skb->destructor = sock_ofree; 2592 return skb; 2593 } 2594 2595 /* 2596 * Allocate a memory block from the socket's option memory buffer. 2597 */ 2598 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2599 { 2600 if ((unsigned int)size <= sysctl_optmem_max && 2601 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2602 void *mem; 2603 /* First do the add, to avoid the race if kmalloc 2604 * might sleep. 2605 */ 2606 atomic_add(size, &sk->sk_omem_alloc); 2607 mem = kmalloc(size, priority); 2608 if (mem) 2609 return mem; 2610 atomic_sub(size, &sk->sk_omem_alloc); 2611 } 2612 return NULL; 2613 } 2614 EXPORT_SYMBOL(sock_kmalloc); 2615 2616 /* Free an option memory block. Note, we actually want the inline 2617 * here as this allows gcc to detect the nullify and fold away the 2618 * condition entirely. 2619 */ 2620 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2621 const bool nullify) 2622 { 2623 if (WARN_ON_ONCE(!mem)) 2624 return; 2625 if (nullify) 2626 kfree_sensitive(mem); 2627 else 2628 kfree(mem); 2629 atomic_sub(size, &sk->sk_omem_alloc); 2630 } 2631 2632 void sock_kfree_s(struct sock *sk, void *mem, int size) 2633 { 2634 __sock_kfree_s(sk, mem, size, false); 2635 } 2636 EXPORT_SYMBOL(sock_kfree_s); 2637 2638 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2639 { 2640 __sock_kfree_s(sk, mem, size, true); 2641 } 2642 EXPORT_SYMBOL(sock_kzfree_s); 2643 2644 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2645 I think, these locks should be removed for datagram sockets. 2646 */ 2647 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2648 { 2649 DEFINE_WAIT(wait); 2650 2651 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2652 for (;;) { 2653 if (!timeo) 2654 break; 2655 if (signal_pending(current)) 2656 break; 2657 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2658 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2659 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2660 break; 2661 if (sk->sk_shutdown & SEND_SHUTDOWN) 2662 break; 2663 if (sk->sk_err) 2664 break; 2665 timeo = schedule_timeout(timeo); 2666 } 2667 finish_wait(sk_sleep(sk), &wait); 2668 return timeo; 2669 } 2670 2671 2672 /* 2673 * Generic send/receive buffer handlers 2674 */ 2675 2676 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2677 unsigned long data_len, int noblock, 2678 int *errcode, int max_page_order) 2679 { 2680 struct sk_buff *skb; 2681 long timeo; 2682 int err; 2683 2684 timeo = sock_sndtimeo(sk, noblock); 2685 for (;;) { 2686 err = sock_error(sk); 2687 if (err != 0) 2688 goto failure; 2689 2690 err = -EPIPE; 2691 if (sk->sk_shutdown & SEND_SHUTDOWN) 2692 goto failure; 2693 2694 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2695 break; 2696 2697 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2698 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2699 err = -EAGAIN; 2700 if (!timeo) 2701 goto failure; 2702 if (signal_pending(current)) 2703 goto interrupted; 2704 timeo = sock_wait_for_wmem(sk, timeo); 2705 } 2706 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2707 errcode, sk->sk_allocation); 2708 if (skb) 2709 skb_set_owner_w(skb, sk); 2710 return skb; 2711 2712 interrupted: 2713 err = sock_intr_errno(timeo); 2714 failure: 2715 *errcode = err; 2716 return NULL; 2717 } 2718 EXPORT_SYMBOL(sock_alloc_send_pskb); 2719 2720 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2721 struct sockcm_cookie *sockc) 2722 { 2723 u32 tsflags; 2724 2725 switch (cmsg->cmsg_type) { 2726 case SO_MARK: 2727 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_RAW) && 2728 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2729 return -EPERM; 2730 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2731 return -EINVAL; 2732 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2733 break; 2734 case SO_TIMESTAMPING_OLD: 2735 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2736 return -EINVAL; 2737 2738 tsflags = *(u32 *)CMSG_DATA(cmsg); 2739 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2740 return -EINVAL; 2741 2742 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2743 sockc->tsflags |= tsflags; 2744 break; 2745 case SCM_TXTIME: 2746 if (!sock_flag(sk, SOCK_TXTIME)) 2747 return -EINVAL; 2748 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2749 return -EINVAL; 2750 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2751 break; 2752 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2753 case SCM_RIGHTS: 2754 case SCM_CREDENTIALS: 2755 break; 2756 default: 2757 return -EINVAL; 2758 } 2759 return 0; 2760 } 2761 EXPORT_SYMBOL(__sock_cmsg_send); 2762 2763 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2764 struct sockcm_cookie *sockc) 2765 { 2766 struct cmsghdr *cmsg; 2767 int ret; 2768 2769 for_each_cmsghdr(cmsg, msg) { 2770 if (!CMSG_OK(msg, cmsg)) 2771 return -EINVAL; 2772 if (cmsg->cmsg_level != SOL_SOCKET) 2773 continue; 2774 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2775 if (ret) 2776 return ret; 2777 } 2778 return 0; 2779 } 2780 EXPORT_SYMBOL(sock_cmsg_send); 2781 2782 static void sk_enter_memory_pressure(struct sock *sk) 2783 { 2784 if (!sk->sk_prot->enter_memory_pressure) 2785 return; 2786 2787 sk->sk_prot->enter_memory_pressure(sk); 2788 } 2789 2790 static void sk_leave_memory_pressure(struct sock *sk) 2791 { 2792 if (sk->sk_prot->leave_memory_pressure) { 2793 sk->sk_prot->leave_memory_pressure(sk); 2794 } else { 2795 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2796 2797 if (memory_pressure && READ_ONCE(*memory_pressure)) 2798 WRITE_ONCE(*memory_pressure, 0); 2799 } 2800 } 2801 2802 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2803 2804 /** 2805 * skb_page_frag_refill - check that a page_frag contains enough room 2806 * @sz: minimum size of the fragment we want to get 2807 * @pfrag: pointer to page_frag 2808 * @gfp: priority for memory allocation 2809 * 2810 * Note: While this allocator tries to use high order pages, there is 2811 * no guarantee that allocations succeed. Therefore, @sz MUST be 2812 * less or equal than PAGE_SIZE. 2813 */ 2814 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2815 { 2816 if (pfrag->page) { 2817 if (page_ref_count(pfrag->page) == 1) { 2818 pfrag->offset = 0; 2819 return true; 2820 } 2821 if (pfrag->offset + sz <= pfrag->size) 2822 return true; 2823 put_page(pfrag->page); 2824 } 2825 2826 pfrag->offset = 0; 2827 if (SKB_FRAG_PAGE_ORDER && 2828 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2829 /* Avoid direct reclaim but allow kswapd to wake */ 2830 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2831 __GFP_COMP | __GFP_NOWARN | 2832 __GFP_NORETRY, 2833 SKB_FRAG_PAGE_ORDER); 2834 if (likely(pfrag->page)) { 2835 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2836 return true; 2837 } 2838 } 2839 pfrag->page = alloc_page(gfp); 2840 if (likely(pfrag->page)) { 2841 pfrag->size = PAGE_SIZE; 2842 return true; 2843 } 2844 return false; 2845 } 2846 EXPORT_SYMBOL(skb_page_frag_refill); 2847 2848 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2849 { 2850 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2851 return true; 2852 2853 sk_enter_memory_pressure(sk); 2854 sk_stream_moderate_sndbuf(sk); 2855 return false; 2856 } 2857 EXPORT_SYMBOL(sk_page_frag_refill); 2858 2859 void __lock_sock(struct sock *sk) 2860 __releases(&sk->sk_lock.slock) 2861 __acquires(&sk->sk_lock.slock) 2862 { 2863 DEFINE_WAIT(wait); 2864 2865 for (;;) { 2866 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2867 TASK_UNINTERRUPTIBLE); 2868 spin_unlock_bh(&sk->sk_lock.slock); 2869 schedule(); 2870 spin_lock_bh(&sk->sk_lock.slock); 2871 if (!sock_owned_by_user(sk)) 2872 break; 2873 } 2874 finish_wait(&sk->sk_lock.wq, &wait); 2875 } 2876 2877 void __release_sock(struct sock *sk) 2878 __releases(&sk->sk_lock.slock) 2879 __acquires(&sk->sk_lock.slock) 2880 { 2881 struct sk_buff *skb, *next; 2882 2883 while ((skb = sk->sk_backlog.head) != NULL) { 2884 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2885 2886 spin_unlock_bh(&sk->sk_lock.slock); 2887 2888 do { 2889 next = skb->next; 2890 prefetch(next); 2891 DEBUG_NET_WARN_ON_ONCE(skb_dst_is_noref(skb)); 2892 skb_mark_not_on_list(skb); 2893 sk_backlog_rcv(sk, skb); 2894 2895 cond_resched(); 2896 2897 skb = next; 2898 } while (skb != NULL); 2899 2900 spin_lock_bh(&sk->sk_lock.slock); 2901 } 2902 2903 /* 2904 * Doing the zeroing here guarantee we can not loop forever 2905 * while a wild producer attempts to flood us. 2906 */ 2907 sk->sk_backlog.len = 0; 2908 } 2909 2910 void __sk_flush_backlog(struct sock *sk) 2911 { 2912 spin_lock_bh(&sk->sk_lock.slock); 2913 __release_sock(sk); 2914 spin_unlock_bh(&sk->sk_lock.slock); 2915 } 2916 EXPORT_SYMBOL_GPL(__sk_flush_backlog); 2917 2918 /** 2919 * sk_wait_data - wait for data to arrive at sk_receive_queue 2920 * @sk: sock to wait on 2921 * @timeo: for how long 2922 * @skb: last skb seen on sk_receive_queue 2923 * 2924 * Now socket state including sk->sk_err is changed only under lock, 2925 * hence we may omit checks after joining wait queue. 2926 * We check receive queue before schedule() only as optimization; 2927 * it is very likely that release_sock() added new data. 2928 */ 2929 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2930 { 2931 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2932 int rc; 2933 2934 add_wait_queue(sk_sleep(sk), &wait); 2935 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2936 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2937 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2938 remove_wait_queue(sk_sleep(sk), &wait); 2939 return rc; 2940 } 2941 EXPORT_SYMBOL(sk_wait_data); 2942 2943 /** 2944 * __sk_mem_raise_allocated - increase memory_allocated 2945 * @sk: socket 2946 * @size: memory size to allocate 2947 * @amt: pages to allocate 2948 * @kind: allocation type 2949 * 2950 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2951 */ 2952 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2953 { 2954 bool memcg_charge = mem_cgroup_sockets_enabled && sk->sk_memcg; 2955 struct proto *prot = sk->sk_prot; 2956 bool charged = true; 2957 long allocated; 2958 2959 sk_memory_allocated_add(sk, amt); 2960 allocated = sk_memory_allocated(sk); 2961 if (memcg_charge && 2962 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt, 2963 gfp_memcg_charge()))) 2964 goto suppress_allocation; 2965 2966 /* Under limit. */ 2967 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2968 sk_leave_memory_pressure(sk); 2969 return 1; 2970 } 2971 2972 /* Under pressure. */ 2973 if (allocated > sk_prot_mem_limits(sk, 1)) 2974 sk_enter_memory_pressure(sk); 2975 2976 /* Over hard limit. */ 2977 if (allocated > sk_prot_mem_limits(sk, 2)) 2978 goto suppress_allocation; 2979 2980 /* guarantee minimum buffer size under pressure */ 2981 if (kind == SK_MEM_RECV) { 2982 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2983 return 1; 2984 2985 } else { /* SK_MEM_SEND */ 2986 int wmem0 = sk_get_wmem0(sk, prot); 2987 2988 if (sk->sk_type == SOCK_STREAM) { 2989 if (sk->sk_wmem_queued < wmem0) 2990 return 1; 2991 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2992 return 1; 2993 } 2994 } 2995 2996 if (sk_has_memory_pressure(sk)) { 2997 u64 alloc; 2998 2999 if (!sk_under_memory_pressure(sk)) 3000 return 1; 3001 alloc = sk_sockets_allocated_read_positive(sk); 3002 if (sk_prot_mem_limits(sk, 2) > alloc * 3003 sk_mem_pages(sk->sk_wmem_queued + 3004 atomic_read(&sk->sk_rmem_alloc) + 3005 sk->sk_forward_alloc)) 3006 return 1; 3007 } 3008 3009 suppress_allocation: 3010 3011 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 3012 sk_stream_moderate_sndbuf(sk); 3013 3014 /* Fail only if socket is _under_ its sndbuf. 3015 * In this case we cannot block, so that we have to fail. 3016 */ 3017 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) { 3018 /* Force charge with __GFP_NOFAIL */ 3019 if (memcg_charge && !charged) { 3020 mem_cgroup_charge_skmem(sk->sk_memcg, amt, 3021 gfp_memcg_charge() | __GFP_NOFAIL); 3022 } 3023 return 1; 3024 } 3025 } 3026 3027 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 3028 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 3029 3030 sk_memory_allocated_sub(sk, amt); 3031 3032 if (memcg_charge && charged) 3033 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 3034 3035 return 0; 3036 } 3037 3038 /** 3039 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 3040 * @sk: socket 3041 * @size: memory size to allocate 3042 * @kind: allocation type 3043 * 3044 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 3045 * rmem allocation. This function assumes that protocols which have 3046 * memory_pressure use sk_wmem_queued as write buffer accounting. 3047 */ 3048 int __sk_mem_schedule(struct sock *sk, int size, int kind) 3049 { 3050 int ret, amt = sk_mem_pages(size); 3051 3052 sk->sk_forward_alloc += amt << PAGE_SHIFT; 3053 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 3054 if (!ret) 3055 sk->sk_forward_alloc -= amt << PAGE_SHIFT; 3056 return ret; 3057 } 3058 EXPORT_SYMBOL(__sk_mem_schedule); 3059 3060 /** 3061 * __sk_mem_reduce_allocated - reclaim memory_allocated 3062 * @sk: socket 3063 * @amount: number of quanta 3064 * 3065 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 3066 */ 3067 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 3068 { 3069 sk_memory_allocated_sub(sk, amount); 3070 3071 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 3072 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 3073 3074 if (sk_under_memory_pressure(sk) && 3075 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 3076 sk_leave_memory_pressure(sk); 3077 } 3078 3079 /** 3080 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 3081 * @sk: socket 3082 * @amount: number of bytes (rounded down to a PAGE_SIZE multiple) 3083 */ 3084 void __sk_mem_reclaim(struct sock *sk, int amount) 3085 { 3086 amount >>= PAGE_SHIFT; 3087 sk->sk_forward_alloc -= amount << PAGE_SHIFT; 3088 __sk_mem_reduce_allocated(sk, amount); 3089 } 3090 EXPORT_SYMBOL(__sk_mem_reclaim); 3091 3092 int sk_set_peek_off(struct sock *sk, int val) 3093 { 3094 sk->sk_peek_off = val; 3095 return 0; 3096 } 3097 EXPORT_SYMBOL_GPL(sk_set_peek_off); 3098 3099 /* 3100 * Set of default routines for initialising struct proto_ops when 3101 * the protocol does not support a particular function. In certain 3102 * cases where it makes no sense for a protocol to have a "do nothing" 3103 * function, some default processing is provided. 3104 */ 3105 3106 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 3107 { 3108 return -EOPNOTSUPP; 3109 } 3110 EXPORT_SYMBOL(sock_no_bind); 3111 3112 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 3113 int len, int flags) 3114 { 3115 return -EOPNOTSUPP; 3116 } 3117 EXPORT_SYMBOL(sock_no_connect); 3118 3119 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 3120 { 3121 return -EOPNOTSUPP; 3122 } 3123 EXPORT_SYMBOL(sock_no_socketpair); 3124 3125 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 3126 bool kern) 3127 { 3128 return -EOPNOTSUPP; 3129 } 3130 EXPORT_SYMBOL(sock_no_accept); 3131 3132 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 3133 int peer) 3134 { 3135 return -EOPNOTSUPP; 3136 } 3137 EXPORT_SYMBOL(sock_no_getname); 3138 3139 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 3140 { 3141 return -EOPNOTSUPP; 3142 } 3143 EXPORT_SYMBOL(sock_no_ioctl); 3144 3145 int sock_no_listen(struct socket *sock, int backlog) 3146 { 3147 return -EOPNOTSUPP; 3148 } 3149 EXPORT_SYMBOL(sock_no_listen); 3150 3151 int sock_no_shutdown(struct socket *sock, int how) 3152 { 3153 return -EOPNOTSUPP; 3154 } 3155 EXPORT_SYMBOL(sock_no_shutdown); 3156 3157 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 3158 { 3159 return -EOPNOTSUPP; 3160 } 3161 EXPORT_SYMBOL(sock_no_sendmsg); 3162 3163 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 3164 { 3165 return -EOPNOTSUPP; 3166 } 3167 EXPORT_SYMBOL(sock_no_sendmsg_locked); 3168 3169 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 3170 int flags) 3171 { 3172 return -EOPNOTSUPP; 3173 } 3174 EXPORT_SYMBOL(sock_no_recvmsg); 3175 3176 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 3177 { 3178 /* Mirror missing mmap method error code */ 3179 return -ENODEV; 3180 } 3181 EXPORT_SYMBOL(sock_no_mmap); 3182 3183 /* 3184 * When a file is received (via SCM_RIGHTS, etc), we must bump the 3185 * various sock-based usage counts. 3186 */ 3187 void __receive_sock(struct file *file) 3188 { 3189 struct socket *sock; 3190 3191 sock = sock_from_file(file); 3192 if (sock) { 3193 sock_update_netprioidx(&sock->sk->sk_cgrp_data); 3194 sock_update_classid(&sock->sk->sk_cgrp_data); 3195 } 3196 } 3197 3198 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 3199 { 3200 ssize_t res; 3201 struct msghdr msg = {.msg_flags = flags}; 3202 struct kvec iov; 3203 char *kaddr = kmap(page); 3204 iov.iov_base = kaddr + offset; 3205 iov.iov_len = size; 3206 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 3207 kunmap(page); 3208 return res; 3209 } 3210 EXPORT_SYMBOL(sock_no_sendpage); 3211 3212 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 3213 int offset, size_t size, int flags) 3214 { 3215 ssize_t res; 3216 struct msghdr msg = {.msg_flags = flags}; 3217 struct kvec iov; 3218 char *kaddr = kmap(page); 3219 3220 iov.iov_base = kaddr + offset; 3221 iov.iov_len = size; 3222 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 3223 kunmap(page); 3224 return res; 3225 } 3226 EXPORT_SYMBOL(sock_no_sendpage_locked); 3227 3228 /* 3229 * Default Socket Callbacks 3230 */ 3231 3232 static void sock_def_wakeup(struct sock *sk) 3233 { 3234 struct socket_wq *wq; 3235 3236 rcu_read_lock(); 3237 wq = rcu_dereference(sk->sk_wq); 3238 if (skwq_has_sleeper(wq)) 3239 wake_up_interruptible_all(&wq->wait); 3240 rcu_read_unlock(); 3241 } 3242 3243 static void sock_def_error_report(struct sock *sk) 3244 { 3245 struct socket_wq *wq; 3246 3247 rcu_read_lock(); 3248 wq = rcu_dereference(sk->sk_wq); 3249 if (skwq_has_sleeper(wq)) 3250 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 3251 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 3252 rcu_read_unlock(); 3253 } 3254 3255 void sock_def_readable(struct sock *sk) 3256 { 3257 struct socket_wq *wq; 3258 3259 rcu_read_lock(); 3260 wq = rcu_dereference(sk->sk_wq); 3261 if (skwq_has_sleeper(wq)) 3262 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 3263 EPOLLRDNORM | EPOLLRDBAND); 3264 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 3265 rcu_read_unlock(); 3266 } 3267 3268 static void sock_def_write_space(struct sock *sk) 3269 { 3270 struct socket_wq *wq; 3271 3272 rcu_read_lock(); 3273 3274 /* Do not wake up a writer until he can make "significant" 3275 * progress. --DaveM 3276 */ 3277 if (sock_writeable(sk)) { 3278 wq = rcu_dereference(sk->sk_wq); 3279 if (skwq_has_sleeper(wq)) 3280 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3281 EPOLLWRNORM | EPOLLWRBAND); 3282 3283 /* Should agree with poll, otherwise some programs break */ 3284 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3285 } 3286 3287 rcu_read_unlock(); 3288 } 3289 3290 /* An optimised version of sock_def_write_space(), should only be called 3291 * for SOCK_RCU_FREE sockets under RCU read section and after putting 3292 * ->sk_wmem_alloc. 3293 */ 3294 static void sock_def_write_space_wfree(struct sock *sk) 3295 { 3296 /* Do not wake up a writer until he can make "significant" 3297 * progress. --DaveM 3298 */ 3299 if (sock_writeable(sk)) { 3300 struct socket_wq *wq = rcu_dereference(sk->sk_wq); 3301 3302 /* rely on refcount_sub from sock_wfree() */ 3303 smp_mb__after_atomic(); 3304 if (wq && waitqueue_active(&wq->wait)) 3305 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 3306 EPOLLWRNORM | EPOLLWRBAND); 3307 3308 /* Should agree with poll, otherwise some programs break */ 3309 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 3310 } 3311 } 3312 3313 static void sock_def_destruct(struct sock *sk) 3314 { 3315 } 3316 3317 void sk_send_sigurg(struct sock *sk) 3318 { 3319 if (sk->sk_socket && sk->sk_socket->file) 3320 if (send_sigurg(&sk->sk_socket->file->f_owner)) 3321 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 3322 } 3323 EXPORT_SYMBOL(sk_send_sigurg); 3324 3325 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 3326 unsigned long expires) 3327 { 3328 if (!mod_timer(timer, expires)) 3329 sock_hold(sk); 3330 } 3331 EXPORT_SYMBOL(sk_reset_timer); 3332 3333 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 3334 { 3335 if (del_timer(timer)) 3336 __sock_put(sk); 3337 } 3338 EXPORT_SYMBOL(sk_stop_timer); 3339 3340 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer) 3341 { 3342 if (del_timer_sync(timer)) 3343 __sock_put(sk); 3344 } 3345 EXPORT_SYMBOL(sk_stop_timer_sync); 3346 3347 void sock_init_data(struct socket *sock, struct sock *sk) 3348 { 3349 sk_init_common(sk); 3350 sk->sk_send_head = NULL; 3351 3352 timer_setup(&sk->sk_timer, NULL, 0); 3353 3354 sk->sk_allocation = GFP_KERNEL; 3355 sk->sk_rcvbuf = sysctl_rmem_default; 3356 sk->sk_sndbuf = sysctl_wmem_default; 3357 sk->sk_state = TCP_CLOSE; 3358 sk_set_socket(sk, sock); 3359 3360 sock_set_flag(sk, SOCK_ZAPPED); 3361 3362 if (sock) { 3363 sk->sk_type = sock->type; 3364 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 3365 sock->sk = sk; 3366 sk->sk_uid = SOCK_INODE(sock)->i_uid; 3367 } else { 3368 RCU_INIT_POINTER(sk->sk_wq, NULL); 3369 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 3370 } 3371 3372 rwlock_init(&sk->sk_callback_lock); 3373 if (sk->sk_kern_sock) 3374 lockdep_set_class_and_name( 3375 &sk->sk_callback_lock, 3376 af_kern_callback_keys + sk->sk_family, 3377 af_family_kern_clock_key_strings[sk->sk_family]); 3378 else 3379 lockdep_set_class_and_name( 3380 &sk->sk_callback_lock, 3381 af_callback_keys + sk->sk_family, 3382 af_family_clock_key_strings[sk->sk_family]); 3383 3384 sk->sk_state_change = sock_def_wakeup; 3385 sk->sk_data_ready = sock_def_readable; 3386 sk->sk_write_space = sock_def_write_space; 3387 sk->sk_error_report = sock_def_error_report; 3388 sk->sk_destruct = sock_def_destruct; 3389 3390 sk->sk_frag.page = NULL; 3391 sk->sk_frag.offset = 0; 3392 sk->sk_peek_off = -1; 3393 3394 sk->sk_peer_pid = NULL; 3395 sk->sk_peer_cred = NULL; 3396 spin_lock_init(&sk->sk_peer_lock); 3397 3398 sk->sk_write_pending = 0; 3399 sk->sk_rcvlowat = 1; 3400 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3401 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3402 3403 sk->sk_stamp = SK_DEFAULT_STAMP; 3404 #if BITS_PER_LONG==32 3405 seqlock_init(&sk->sk_stamp_seq); 3406 #endif 3407 atomic_set(&sk->sk_zckey, 0); 3408 3409 #ifdef CONFIG_NET_RX_BUSY_POLL 3410 sk->sk_napi_id = 0; 3411 sk->sk_ll_usec = sysctl_net_busy_read; 3412 #endif 3413 3414 sk->sk_max_pacing_rate = ~0UL; 3415 sk->sk_pacing_rate = ~0UL; 3416 WRITE_ONCE(sk->sk_pacing_shift, 10); 3417 sk->sk_incoming_cpu = -1; 3418 sk->sk_txrehash = SOCK_TXREHASH_DEFAULT; 3419 3420 sk_rx_queue_clear(sk); 3421 /* 3422 * Before updating sk_refcnt, we must commit prior changes to memory 3423 * (Documentation/RCU/rculist_nulls.rst for details) 3424 */ 3425 smp_wmb(); 3426 refcount_set(&sk->sk_refcnt, 1); 3427 atomic_set(&sk->sk_drops, 0); 3428 } 3429 EXPORT_SYMBOL(sock_init_data); 3430 3431 void lock_sock_nested(struct sock *sk, int subclass) 3432 { 3433 /* The sk_lock has mutex_lock() semantics here. */ 3434 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3435 3436 might_sleep(); 3437 spin_lock_bh(&sk->sk_lock.slock); 3438 if (sock_owned_by_user_nocheck(sk)) 3439 __lock_sock(sk); 3440 sk->sk_lock.owned = 1; 3441 spin_unlock_bh(&sk->sk_lock.slock); 3442 } 3443 EXPORT_SYMBOL(lock_sock_nested); 3444 3445 void release_sock(struct sock *sk) 3446 { 3447 spin_lock_bh(&sk->sk_lock.slock); 3448 if (sk->sk_backlog.tail) 3449 __release_sock(sk); 3450 3451 /* Warning : release_cb() might need to release sk ownership, 3452 * ie call sock_release_ownership(sk) before us. 3453 */ 3454 if (sk->sk_prot->release_cb) 3455 sk->sk_prot->release_cb(sk); 3456 3457 sock_release_ownership(sk); 3458 if (waitqueue_active(&sk->sk_lock.wq)) 3459 wake_up(&sk->sk_lock.wq); 3460 spin_unlock_bh(&sk->sk_lock.slock); 3461 } 3462 EXPORT_SYMBOL(release_sock); 3463 3464 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock) 3465 { 3466 might_sleep(); 3467 spin_lock_bh(&sk->sk_lock.slock); 3468 3469 if (!sock_owned_by_user_nocheck(sk)) { 3470 /* 3471 * Fast path return with bottom halves disabled and 3472 * sock::sk_lock.slock held. 3473 * 3474 * The 'mutex' is not contended and holding 3475 * sock::sk_lock.slock prevents all other lockers to 3476 * proceed so the corresponding unlock_sock_fast() can 3477 * avoid the slow path of release_sock() completely and 3478 * just release slock. 3479 * 3480 * From a semantical POV this is equivalent to 'acquiring' 3481 * the 'mutex', hence the corresponding lockdep 3482 * mutex_release() has to happen in the fast path of 3483 * unlock_sock_fast(). 3484 */ 3485 return false; 3486 } 3487 3488 __lock_sock(sk); 3489 sk->sk_lock.owned = 1; 3490 __acquire(&sk->sk_lock.slock); 3491 spin_unlock_bh(&sk->sk_lock.slock); 3492 return true; 3493 } 3494 EXPORT_SYMBOL(__lock_sock_fast); 3495 3496 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3497 bool timeval, bool time32) 3498 { 3499 struct sock *sk = sock->sk; 3500 struct timespec64 ts; 3501 3502 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3503 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3504 if (ts.tv_sec == -1) 3505 return -ENOENT; 3506 if (ts.tv_sec == 0) { 3507 ktime_t kt = ktime_get_real(); 3508 sock_write_timestamp(sk, kt); 3509 ts = ktime_to_timespec64(kt); 3510 } 3511 3512 if (timeval) 3513 ts.tv_nsec /= 1000; 3514 3515 #ifdef CONFIG_COMPAT_32BIT_TIME 3516 if (time32) 3517 return put_old_timespec32(&ts, userstamp); 3518 #endif 3519 #ifdef CONFIG_SPARC64 3520 /* beware of padding in sparc64 timeval */ 3521 if (timeval && !in_compat_syscall()) { 3522 struct __kernel_old_timeval __user tv = { 3523 .tv_sec = ts.tv_sec, 3524 .tv_usec = ts.tv_nsec, 3525 }; 3526 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3527 return -EFAULT; 3528 return 0; 3529 } 3530 #endif 3531 return put_timespec64(&ts, userstamp); 3532 } 3533 EXPORT_SYMBOL(sock_gettstamp); 3534 3535 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3536 { 3537 if (!sock_flag(sk, flag)) { 3538 unsigned long previous_flags = sk->sk_flags; 3539 3540 sock_set_flag(sk, flag); 3541 /* 3542 * we just set one of the two flags which require net 3543 * time stamping, but time stamping might have been on 3544 * already because of the other one 3545 */ 3546 if (sock_needs_netstamp(sk) && 3547 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3548 net_enable_timestamp(); 3549 } 3550 } 3551 3552 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3553 int level, int type) 3554 { 3555 struct sock_exterr_skb *serr; 3556 struct sk_buff *skb; 3557 int copied, err; 3558 3559 err = -EAGAIN; 3560 skb = sock_dequeue_err_skb(sk); 3561 if (skb == NULL) 3562 goto out; 3563 3564 copied = skb->len; 3565 if (copied > len) { 3566 msg->msg_flags |= MSG_TRUNC; 3567 copied = len; 3568 } 3569 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3570 if (err) 3571 goto out_free_skb; 3572 3573 sock_recv_timestamp(msg, sk, skb); 3574 3575 serr = SKB_EXT_ERR(skb); 3576 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3577 3578 msg->msg_flags |= MSG_ERRQUEUE; 3579 err = copied; 3580 3581 out_free_skb: 3582 kfree_skb(skb); 3583 out: 3584 return err; 3585 } 3586 EXPORT_SYMBOL(sock_recv_errqueue); 3587 3588 /* 3589 * Get a socket option on an socket. 3590 * 3591 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3592 * asynchronous errors should be reported by getsockopt. We assume 3593 * this means if you specify SO_ERROR (otherwise whats the point of it). 3594 */ 3595 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3596 char __user *optval, int __user *optlen) 3597 { 3598 struct sock *sk = sock->sk; 3599 3600 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3601 } 3602 EXPORT_SYMBOL(sock_common_getsockopt); 3603 3604 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3605 int flags) 3606 { 3607 struct sock *sk = sock->sk; 3608 int addr_len = 0; 3609 int err; 3610 3611 err = sk->sk_prot->recvmsg(sk, msg, size, flags, &addr_len); 3612 if (err >= 0) 3613 msg->msg_namelen = addr_len; 3614 return err; 3615 } 3616 EXPORT_SYMBOL(sock_common_recvmsg); 3617 3618 /* 3619 * Set socket options on an inet socket. 3620 */ 3621 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3622 sockptr_t optval, unsigned int optlen) 3623 { 3624 struct sock *sk = sock->sk; 3625 3626 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3627 } 3628 EXPORT_SYMBOL(sock_common_setsockopt); 3629 3630 void sk_common_release(struct sock *sk) 3631 { 3632 if (sk->sk_prot->destroy) 3633 sk->sk_prot->destroy(sk); 3634 3635 /* 3636 * Observation: when sk_common_release is called, processes have 3637 * no access to socket. But net still has. 3638 * Step one, detach it from networking: 3639 * 3640 * A. Remove from hash tables. 3641 */ 3642 3643 sk->sk_prot->unhash(sk); 3644 3645 /* 3646 * In this point socket cannot receive new packets, but it is possible 3647 * that some packets are in flight because some CPU runs receiver and 3648 * did hash table lookup before we unhashed socket. They will achieve 3649 * receive queue and will be purged by socket destructor. 3650 * 3651 * Also we still have packets pending on receive queue and probably, 3652 * our own packets waiting in device queues. sock_destroy will drain 3653 * receive queue, but transmitted packets will delay socket destruction 3654 * until the last reference will be released. 3655 */ 3656 3657 sock_orphan(sk); 3658 3659 xfrm_sk_free_policy(sk); 3660 3661 sk_refcnt_debug_release(sk); 3662 3663 sock_put(sk); 3664 } 3665 EXPORT_SYMBOL(sk_common_release); 3666 3667 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3668 { 3669 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3670 3671 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3672 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3673 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3674 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3675 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3676 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3677 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3678 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3679 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3680 } 3681 3682 #ifdef CONFIG_PROC_FS 3683 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3684 3685 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3686 { 3687 int cpu, idx = prot->inuse_idx; 3688 int res = 0; 3689 3690 for_each_possible_cpu(cpu) 3691 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3692 3693 return res >= 0 ? res : 0; 3694 } 3695 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3696 3697 int sock_inuse_get(struct net *net) 3698 { 3699 int cpu, res = 0; 3700 3701 for_each_possible_cpu(cpu) 3702 res += per_cpu_ptr(net->core.prot_inuse, cpu)->all; 3703 3704 return res; 3705 } 3706 3707 EXPORT_SYMBOL_GPL(sock_inuse_get); 3708 3709 static int __net_init sock_inuse_init_net(struct net *net) 3710 { 3711 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3712 if (net->core.prot_inuse == NULL) 3713 return -ENOMEM; 3714 return 0; 3715 } 3716 3717 static void __net_exit sock_inuse_exit_net(struct net *net) 3718 { 3719 free_percpu(net->core.prot_inuse); 3720 } 3721 3722 static struct pernet_operations net_inuse_ops = { 3723 .init = sock_inuse_init_net, 3724 .exit = sock_inuse_exit_net, 3725 }; 3726 3727 static __init int net_inuse_init(void) 3728 { 3729 if (register_pernet_subsys(&net_inuse_ops)) 3730 panic("Cannot initialize net inuse counters"); 3731 3732 return 0; 3733 } 3734 3735 core_initcall(net_inuse_init); 3736 3737 static int assign_proto_idx(struct proto *prot) 3738 { 3739 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3740 3741 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3742 pr_err("PROTO_INUSE_NR exhausted\n"); 3743 return -ENOSPC; 3744 } 3745 3746 set_bit(prot->inuse_idx, proto_inuse_idx); 3747 return 0; 3748 } 3749 3750 static void release_proto_idx(struct proto *prot) 3751 { 3752 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3753 clear_bit(prot->inuse_idx, proto_inuse_idx); 3754 } 3755 #else 3756 static inline int assign_proto_idx(struct proto *prot) 3757 { 3758 return 0; 3759 } 3760 3761 static inline void release_proto_idx(struct proto *prot) 3762 { 3763 } 3764 3765 #endif 3766 3767 static void tw_prot_cleanup(struct timewait_sock_ops *twsk_prot) 3768 { 3769 if (!twsk_prot) 3770 return; 3771 kfree(twsk_prot->twsk_slab_name); 3772 twsk_prot->twsk_slab_name = NULL; 3773 kmem_cache_destroy(twsk_prot->twsk_slab); 3774 twsk_prot->twsk_slab = NULL; 3775 } 3776 3777 static int tw_prot_init(const struct proto *prot) 3778 { 3779 struct timewait_sock_ops *twsk_prot = prot->twsk_prot; 3780 3781 if (!twsk_prot) 3782 return 0; 3783 3784 twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", 3785 prot->name); 3786 if (!twsk_prot->twsk_slab_name) 3787 return -ENOMEM; 3788 3789 twsk_prot->twsk_slab = 3790 kmem_cache_create(twsk_prot->twsk_slab_name, 3791 twsk_prot->twsk_obj_size, 0, 3792 SLAB_ACCOUNT | prot->slab_flags, 3793 NULL); 3794 if (!twsk_prot->twsk_slab) { 3795 pr_crit("%s: Can't create timewait sock SLAB cache!\n", 3796 prot->name); 3797 return -ENOMEM; 3798 } 3799 3800 return 0; 3801 } 3802 3803 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3804 { 3805 if (!rsk_prot) 3806 return; 3807 kfree(rsk_prot->slab_name); 3808 rsk_prot->slab_name = NULL; 3809 kmem_cache_destroy(rsk_prot->slab); 3810 rsk_prot->slab = NULL; 3811 } 3812 3813 static int req_prot_init(const struct proto *prot) 3814 { 3815 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3816 3817 if (!rsk_prot) 3818 return 0; 3819 3820 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3821 prot->name); 3822 if (!rsk_prot->slab_name) 3823 return -ENOMEM; 3824 3825 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3826 rsk_prot->obj_size, 0, 3827 SLAB_ACCOUNT | prot->slab_flags, 3828 NULL); 3829 3830 if (!rsk_prot->slab) { 3831 pr_crit("%s: Can't create request sock SLAB cache!\n", 3832 prot->name); 3833 return -ENOMEM; 3834 } 3835 return 0; 3836 } 3837 3838 int proto_register(struct proto *prot, int alloc_slab) 3839 { 3840 int ret = -ENOBUFS; 3841 3842 if (prot->memory_allocated && !prot->sysctl_mem) { 3843 pr_err("%s: missing sysctl_mem\n", prot->name); 3844 return -EINVAL; 3845 } 3846 if (prot->memory_allocated && !prot->per_cpu_fw_alloc) { 3847 pr_err("%s: missing per_cpu_fw_alloc\n", prot->name); 3848 return -EINVAL; 3849 } 3850 if (alloc_slab) { 3851 prot->slab = kmem_cache_create_usercopy(prot->name, 3852 prot->obj_size, 0, 3853 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3854 prot->slab_flags, 3855 prot->useroffset, prot->usersize, 3856 NULL); 3857 3858 if (prot->slab == NULL) { 3859 pr_crit("%s: Can't create sock SLAB cache!\n", 3860 prot->name); 3861 goto out; 3862 } 3863 3864 if (req_prot_init(prot)) 3865 goto out_free_request_sock_slab; 3866 3867 if (tw_prot_init(prot)) 3868 goto out_free_timewait_sock_slab; 3869 } 3870 3871 mutex_lock(&proto_list_mutex); 3872 ret = assign_proto_idx(prot); 3873 if (ret) { 3874 mutex_unlock(&proto_list_mutex); 3875 goto out_free_timewait_sock_slab; 3876 } 3877 list_add(&prot->node, &proto_list); 3878 mutex_unlock(&proto_list_mutex); 3879 return ret; 3880 3881 out_free_timewait_sock_slab: 3882 if (alloc_slab) 3883 tw_prot_cleanup(prot->twsk_prot); 3884 out_free_request_sock_slab: 3885 if (alloc_slab) { 3886 req_prot_cleanup(prot->rsk_prot); 3887 3888 kmem_cache_destroy(prot->slab); 3889 prot->slab = NULL; 3890 } 3891 out: 3892 return ret; 3893 } 3894 EXPORT_SYMBOL(proto_register); 3895 3896 void proto_unregister(struct proto *prot) 3897 { 3898 mutex_lock(&proto_list_mutex); 3899 release_proto_idx(prot); 3900 list_del(&prot->node); 3901 mutex_unlock(&proto_list_mutex); 3902 3903 kmem_cache_destroy(prot->slab); 3904 prot->slab = NULL; 3905 3906 req_prot_cleanup(prot->rsk_prot); 3907 tw_prot_cleanup(prot->twsk_prot); 3908 } 3909 EXPORT_SYMBOL(proto_unregister); 3910 3911 int sock_load_diag_module(int family, int protocol) 3912 { 3913 if (!protocol) { 3914 if (!sock_is_registered(family)) 3915 return -ENOENT; 3916 3917 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3918 NETLINK_SOCK_DIAG, family); 3919 } 3920 3921 #ifdef CONFIG_INET 3922 if (family == AF_INET && 3923 protocol != IPPROTO_RAW && 3924 protocol < MAX_INET_PROTOS && 3925 !rcu_access_pointer(inet_protos[protocol])) 3926 return -ENOENT; 3927 #endif 3928 3929 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3930 NETLINK_SOCK_DIAG, family, protocol); 3931 } 3932 EXPORT_SYMBOL(sock_load_diag_module); 3933 3934 #ifdef CONFIG_PROC_FS 3935 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3936 __acquires(proto_list_mutex) 3937 { 3938 mutex_lock(&proto_list_mutex); 3939 return seq_list_start_head(&proto_list, *pos); 3940 } 3941 3942 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3943 { 3944 return seq_list_next(v, &proto_list, pos); 3945 } 3946 3947 static void proto_seq_stop(struct seq_file *seq, void *v) 3948 __releases(proto_list_mutex) 3949 { 3950 mutex_unlock(&proto_list_mutex); 3951 } 3952 3953 static char proto_method_implemented(const void *method) 3954 { 3955 return method == NULL ? 'n' : 'y'; 3956 } 3957 static long sock_prot_memory_allocated(struct proto *proto) 3958 { 3959 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3960 } 3961 3962 static const char *sock_prot_memory_pressure(struct proto *proto) 3963 { 3964 return proto->memory_pressure != NULL ? 3965 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3966 } 3967 3968 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3969 { 3970 3971 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3972 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3973 proto->name, 3974 proto->obj_size, 3975 sock_prot_inuse_get(seq_file_net(seq), proto), 3976 sock_prot_memory_allocated(proto), 3977 sock_prot_memory_pressure(proto), 3978 proto->max_header, 3979 proto->slab == NULL ? "no" : "yes", 3980 module_name(proto->owner), 3981 proto_method_implemented(proto->close), 3982 proto_method_implemented(proto->connect), 3983 proto_method_implemented(proto->disconnect), 3984 proto_method_implemented(proto->accept), 3985 proto_method_implemented(proto->ioctl), 3986 proto_method_implemented(proto->init), 3987 proto_method_implemented(proto->destroy), 3988 proto_method_implemented(proto->shutdown), 3989 proto_method_implemented(proto->setsockopt), 3990 proto_method_implemented(proto->getsockopt), 3991 proto_method_implemented(proto->sendmsg), 3992 proto_method_implemented(proto->recvmsg), 3993 proto_method_implemented(proto->sendpage), 3994 proto_method_implemented(proto->bind), 3995 proto_method_implemented(proto->backlog_rcv), 3996 proto_method_implemented(proto->hash), 3997 proto_method_implemented(proto->unhash), 3998 proto_method_implemented(proto->get_port), 3999 proto_method_implemented(proto->enter_memory_pressure)); 4000 } 4001 4002 static int proto_seq_show(struct seq_file *seq, void *v) 4003 { 4004 if (v == &proto_list) 4005 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 4006 "protocol", 4007 "size", 4008 "sockets", 4009 "memory", 4010 "press", 4011 "maxhdr", 4012 "slab", 4013 "module", 4014 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 4015 else 4016 proto_seq_printf(seq, list_entry(v, struct proto, node)); 4017 return 0; 4018 } 4019 4020 static const struct seq_operations proto_seq_ops = { 4021 .start = proto_seq_start, 4022 .next = proto_seq_next, 4023 .stop = proto_seq_stop, 4024 .show = proto_seq_show, 4025 }; 4026 4027 static __net_init int proto_init_net(struct net *net) 4028 { 4029 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 4030 sizeof(struct seq_net_private))) 4031 return -ENOMEM; 4032 4033 return 0; 4034 } 4035 4036 static __net_exit void proto_exit_net(struct net *net) 4037 { 4038 remove_proc_entry("protocols", net->proc_net); 4039 } 4040 4041 4042 static __net_initdata struct pernet_operations proto_net_ops = { 4043 .init = proto_init_net, 4044 .exit = proto_exit_net, 4045 }; 4046 4047 static int __init proto_init(void) 4048 { 4049 return register_pernet_subsys(&proto_net_ops); 4050 } 4051 4052 subsys_initcall(proto_init); 4053 4054 #endif /* PROC_FS */ 4055 4056 #ifdef CONFIG_NET_RX_BUSY_POLL 4057 bool sk_busy_loop_end(void *p, unsigned long start_time) 4058 { 4059 struct sock *sk = p; 4060 4061 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 4062 sk_busy_loop_timeout(sk, start_time); 4063 } 4064 EXPORT_SYMBOL(sk_busy_loop_end); 4065 #endif /* CONFIG_NET_RX_BUSY_POLL */ 4066 4067 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 4068 { 4069 if (!sk->sk_prot->bind_add) 4070 return -EOPNOTSUPP; 4071 return sk->sk_prot->bind_add(sk, addr, addr_len); 4072 } 4073 EXPORT_SYMBOL(sock_bind_add); 4074