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> 116 117 #include <linux/uaccess.h> 118 119 #include <linux/netdevice.h> 120 #include <net/protocol.h> 121 #include <linux/skbuff.h> 122 #include <net/net_namespace.h> 123 #include <net/request_sock.h> 124 #include <net/sock.h> 125 #include <linux/net_tstamp.h> 126 #include <net/xfrm.h> 127 #include <linux/ipsec.h> 128 #include <net/cls_cgroup.h> 129 #include <net/netprio_cgroup.h> 130 #include <linux/sock_diag.h> 131 132 #include <linux/filter.h> 133 #include <net/sock_reuseport.h> 134 #include <net/bpf_sk_storage.h> 135 136 #include <trace/events/sock.h> 137 138 #include <net/tcp.h> 139 #include <net/busy_poll.h> 140 141 static DEFINE_MUTEX(proto_list_mutex); 142 static LIST_HEAD(proto_list); 143 144 static void sock_inuse_add(struct net *net, int val); 145 146 /** 147 * sk_ns_capable - General socket capability test 148 * @sk: Socket to use a capability on or through 149 * @user_ns: The user namespace of the capability to use 150 * @cap: The capability to use 151 * 152 * Test to see if the opener of the socket had when the socket was 153 * created and the current process has the capability @cap in the user 154 * namespace @user_ns. 155 */ 156 bool sk_ns_capable(const struct sock *sk, 157 struct user_namespace *user_ns, int cap) 158 { 159 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 160 ns_capable(user_ns, cap); 161 } 162 EXPORT_SYMBOL(sk_ns_capable); 163 164 /** 165 * sk_capable - Socket global capability test 166 * @sk: Socket to use a capability on or through 167 * @cap: The global capability to use 168 * 169 * Test to see if the opener of the socket had when the socket was 170 * created and the current process has the capability @cap in all user 171 * namespaces. 172 */ 173 bool sk_capable(const struct sock *sk, int cap) 174 { 175 return sk_ns_capable(sk, &init_user_ns, cap); 176 } 177 EXPORT_SYMBOL(sk_capable); 178 179 /** 180 * sk_net_capable - Network namespace socket capability test 181 * @sk: Socket to use a capability on or through 182 * @cap: The capability to use 183 * 184 * Test to see if the opener of the socket had when the socket was created 185 * and the current process has the capability @cap over the network namespace 186 * the socket is a member of. 187 */ 188 bool sk_net_capable(const struct sock *sk, int cap) 189 { 190 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 191 } 192 EXPORT_SYMBOL(sk_net_capable); 193 194 /* 195 * Each address family might have different locking rules, so we have 196 * one slock key per address family and separate keys for internal and 197 * userspace sockets. 198 */ 199 static struct lock_class_key af_family_keys[AF_MAX]; 200 static struct lock_class_key af_family_kern_keys[AF_MAX]; 201 static struct lock_class_key af_family_slock_keys[AF_MAX]; 202 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 203 204 /* 205 * Make lock validator output more readable. (we pre-construct these 206 * strings build-time, so that runtime initialization of socket 207 * locks is fast): 208 */ 209 210 #define _sock_locks(x) \ 211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 220 x "27" , x "28" , x "AF_CAN" , \ 221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 226 x "AF_MAX" 227 228 static const char *const af_family_key_strings[AF_MAX+1] = { 229 _sock_locks("sk_lock-") 230 }; 231 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 232 _sock_locks("slock-") 233 }; 234 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 235 _sock_locks("clock-") 236 }; 237 238 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 239 _sock_locks("k-sk_lock-") 240 }; 241 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 242 _sock_locks("k-slock-") 243 }; 244 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 245 _sock_locks("k-clock-") 246 }; 247 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 248 _sock_locks("rlock-") 249 }; 250 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 251 _sock_locks("wlock-") 252 }; 253 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 254 _sock_locks("elock-") 255 }; 256 257 /* 258 * sk_callback_lock and sk queues locking rules are per-address-family, 259 * so split the lock classes by using a per-AF key: 260 */ 261 static struct lock_class_key af_callback_keys[AF_MAX]; 262 static struct lock_class_key af_rlock_keys[AF_MAX]; 263 static struct lock_class_key af_wlock_keys[AF_MAX]; 264 static struct lock_class_key af_elock_keys[AF_MAX]; 265 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 266 267 /* Run time adjustable parameters. */ 268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 269 EXPORT_SYMBOL(sysctl_wmem_max); 270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 271 EXPORT_SYMBOL(sysctl_rmem_max); 272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 274 275 /* Maximal space eaten by iovec or ancillary data plus some space */ 276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 277 EXPORT_SYMBOL(sysctl_optmem_max); 278 279 int sysctl_tstamp_allow_data __read_mostly = 1; 280 281 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 282 EXPORT_SYMBOL_GPL(memalloc_socks_key); 283 284 /** 285 * sk_set_memalloc - sets %SOCK_MEMALLOC 286 * @sk: socket to set it on 287 * 288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 289 * It's the responsibility of the admin to adjust min_free_kbytes 290 * to meet the requirements 291 */ 292 void sk_set_memalloc(struct sock *sk) 293 { 294 sock_set_flag(sk, SOCK_MEMALLOC); 295 sk->sk_allocation |= __GFP_MEMALLOC; 296 static_branch_inc(&memalloc_socks_key); 297 } 298 EXPORT_SYMBOL_GPL(sk_set_memalloc); 299 300 void sk_clear_memalloc(struct sock *sk) 301 { 302 sock_reset_flag(sk, SOCK_MEMALLOC); 303 sk->sk_allocation &= ~__GFP_MEMALLOC; 304 static_branch_dec(&memalloc_socks_key); 305 306 /* 307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 308 * progress of swapping. SOCK_MEMALLOC may be cleared while 309 * it has rmem allocations due to the last swapfile being deactivated 310 * but there is a risk that the socket is unusable due to exceeding 311 * the rmem limits. Reclaim the reserves and obey rmem limits again. 312 */ 313 sk_mem_reclaim(sk); 314 } 315 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 316 317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 318 { 319 int ret; 320 unsigned int noreclaim_flag; 321 322 /* these should have been dropped before queueing */ 323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 324 325 noreclaim_flag = memalloc_noreclaim_save(); 326 ret = sk->sk_backlog_rcv(sk, skb); 327 memalloc_noreclaim_restore(noreclaim_flag); 328 329 return ret; 330 } 331 EXPORT_SYMBOL(__sk_backlog_rcv); 332 333 static int sock_get_timeout(long timeo, void *optval, bool old_timeval) 334 { 335 struct __kernel_sock_timeval tv; 336 337 if (timeo == MAX_SCHEDULE_TIMEOUT) { 338 tv.tv_sec = 0; 339 tv.tv_usec = 0; 340 } else { 341 tv.tv_sec = timeo / HZ; 342 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 343 } 344 345 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 346 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 347 *(struct old_timeval32 *)optval = tv32; 348 return sizeof(tv32); 349 } 350 351 if (old_timeval) { 352 struct __kernel_old_timeval old_tv; 353 old_tv.tv_sec = tv.tv_sec; 354 old_tv.tv_usec = tv.tv_usec; 355 *(struct __kernel_old_timeval *)optval = old_tv; 356 return sizeof(old_tv); 357 } 358 359 *(struct __kernel_sock_timeval *)optval = tv; 360 return sizeof(tv); 361 } 362 363 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval) 364 { 365 struct __kernel_sock_timeval tv; 366 367 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 368 struct old_timeval32 tv32; 369 370 if (optlen < sizeof(tv32)) 371 return -EINVAL; 372 373 if (copy_from_user(&tv32, optval, sizeof(tv32))) 374 return -EFAULT; 375 tv.tv_sec = tv32.tv_sec; 376 tv.tv_usec = tv32.tv_usec; 377 } else if (old_timeval) { 378 struct __kernel_old_timeval old_tv; 379 380 if (optlen < sizeof(old_tv)) 381 return -EINVAL; 382 if (copy_from_user(&old_tv, optval, sizeof(old_tv))) 383 return -EFAULT; 384 tv.tv_sec = old_tv.tv_sec; 385 tv.tv_usec = old_tv.tv_usec; 386 } else { 387 if (optlen < sizeof(tv)) 388 return -EINVAL; 389 if (copy_from_user(&tv, optval, sizeof(tv))) 390 return -EFAULT; 391 } 392 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 393 return -EDOM; 394 395 if (tv.tv_sec < 0) { 396 static int warned __read_mostly; 397 398 *timeo_p = 0; 399 if (warned < 10 && net_ratelimit()) { 400 warned++; 401 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 402 __func__, current->comm, task_pid_nr(current)); 403 } 404 return 0; 405 } 406 *timeo_p = MAX_SCHEDULE_TIMEOUT; 407 if (tv.tv_sec == 0 && tv.tv_usec == 0) 408 return 0; 409 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 410 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 411 return 0; 412 } 413 414 static void sock_warn_obsolete_bsdism(const char *name) 415 { 416 static int warned; 417 static char warncomm[TASK_COMM_LEN]; 418 if (strcmp(warncomm, current->comm) && warned < 5) { 419 strcpy(warncomm, current->comm); 420 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 421 warncomm, name); 422 warned++; 423 } 424 } 425 426 static bool sock_needs_netstamp(const struct sock *sk) 427 { 428 switch (sk->sk_family) { 429 case AF_UNSPEC: 430 case AF_UNIX: 431 return false; 432 default: 433 return true; 434 } 435 } 436 437 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 438 { 439 if (sk->sk_flags & flags) { 440 sk->sk_flags &= ~flags; 441 if (sock_needs_netstamp(sk) && 442 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 443 net_disable_timestamp(); 444 } 445 } 446 447 448 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 449 { 450 unsigned long flags; 451 struct sk_buff_head *list = &sk->sk_receive_queue; 452 453 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 454 atomic_inc(&sk->sk_drops); 455 trace_sock_rcvqueue_full(sk, skb); 456 return -ENOMEM; 457 } 458 459 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 460 atomic_inc(&sk->sk_drops); 461 return -ENOBUFS; 462 } 463 464 skb->dev = NULL; 465 skb_set_owner_r(skb, sk); 466 467 /* we escape from rcu protected region, make sure we dont leak 468 * a norefcounted dst 469 */ 470 skb_dst_force(skb); 471 472 spin_lock_irqsave(&list->lock, flags); 473 sock_skb_set_dropcount(sk, skb); 474 __skb_queue_tail(list, skb); 475 spin_unlock_irqrestore(&list->lock, flags); 476 477 if (!sock_flag(sk, SOCK_DEAD)) 478 sk->sk_data_ready(sk); 479 return 0; 480 } 481 EXPORT_SYMBOL(__sock_queue_rcv_skb); 482 483 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 484 { 485 int err; 486 487 err = sk_filter(sk, skb); 488 if (err) 489 return err; 490 491 return __sock_queue_rcv_skb(sk, skb); 492 } 493 EXPORT_SYMBOL(sock_queue_rcv_skb); 494 495 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 496 const int nested, unsigned int trim_cap, bool refcounted) 497 { 498 int rc = NET_RX_SUCCESS; 499 500 if (sk_filter_trim_cap(sk, skb, trim_cap)) 501 goto discard_and_relse; 502 503 skb->dev = NULL; 504 505 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 506 atomic_inc(&sk->sk_drops); 507 goto discard_and_relse; 508 } 509 if (nested) 510 bh_lock_sock_nested(sk); 511 else 512 bh_lock_sock(sk); 513 if (!sock_owned_by_user(sk)) { 514 /* 515 * trylock + unlock semantics: 516 */ 517 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 518 519 rc = sk_backlog_rcv(sk, skb); 520 521 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 522 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 523 bh_unlock_sock(sk); 524 atomic_inc(&sk->sk_drops); 525 goto discard_and_relse; 526 } 527 528 bh_unlock_sock(sk); 529 out: 530 if (refcounted) 531 sock_put(sk); 532 return rc; 533 discard_and_relse: 534 kfree_skb(skb); 535 goto out; 536 } 537 EXPORT_SYMBOL(__sk_receive_skb); 538 539 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 540 { 541 struct dst_entry *dst = __sk_dst_get(sk); 542 543 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 544 sk_tx_queue_clear(sk); 545 sk->sk_dst_pending_confirm = 0; 546 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 547 dst_release(dst); 548 return NULL; 549 } 550 551 return dst; 552 } 553 EXPORT_SYMBOL(__sk_dst_check); 554 555 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 556 { 557 struct dst_entry *dst = sk_dst_get(sk); 558 559 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 560 sk_dst_reset(sk); 561 dst_release(dst); 562 return NULL; 563 } 564 565 return dst; 566 } 567 EXPORT_SYMBOL(sk_dst_check); 568 569 static int sock_bindtoindex_locked(struct sock *sk, int ifindex) 570 { 571 int ret = -ENOPROTOOPT; 572 #ifdef CONFIG_NETDEVICES 573 struct net *net = sock_net(sk); 574 575 /* Sorry... */ 576 ret = -EPERM; 577 if (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 578 goto out; 579 580 ret = -EINVAL; 581 if (ifindex < 0) 582 goto out; 583 584 sk->sk_bound_dev_if = ifindex; 585 if (sk->sk_prot->rehash) 586 sk->sk_prot->rehash(sk); 587 sk_dst_reset(sk); 588 589 ret = 0; 590 591 out: 592 #endif 593 594 return ret; 595 } 596 597 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk) 598 { 599 int ret; 600 601 if (lock_sk) 602 lock_sock(sk); 603 ret = sock_bindtoindex_locked(sk, ifindex); 604 if (lock_sk) 605 release_sock(sk); 606 607 return ret; 608 } 609 EXPORT_SYMBOL(sock_bindtoindex); 610 611 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 612 int optlen) 613 { 614 int ret = -ENOPROTOOPT; 615 #ifdef CONFIG_NETDEVICES 616 struct net *net = sock_net(sk); 617 char devname[IFNAMSIZ]; 618 int index; 619 620 ret = -EINVAL; 621 if (optlen < 0) 622 goto out; 623 624 /* Bind this socket to a particular device like "eth0", 625 * as specified in the passed interface name. If the 626 * name is "" or the option length is zero the socket 627 * is not bound. 628 */ 629 if (optlen > IFNAMSIZ - 1) 630 optlen = IFNAMSIZ - 1; 631 memset(devname, 0, sizeof(devname)); 632 633 ret = -EFAULT; 634 if (copy_from_user(devname, optval, optlen)) 635 goto out; 636 637 index = 0; 638 if (devname[0] != '\0') { 639 struct net_device *dev; 640 641 rcu_read_lock(); 642 dev = dev_get_by_name_rcu(net, devname); 643 if (dev) 644 index = dev->ifindex; 645 rcu_read_unlock(); 646 ret = -ENODEV; 647 if (!dev) 648 goto out; 649 } 650 651 return sock_bindtoindex(sk, index, true); 652 out: 653 #endif 654 655 return ret; 656 } 657 658 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 659 int __user *optlen, int len) 660 { 661 int ret = -ENOPROTOOPT; 662 #ifdef CONFIG_NETDEVICES 663 struct net *net = sock_net(sk); 664 char devname[IFNAMSIZ]; 665 666 if (sk->sk_bound_dev_if == 0) { 667 len = 0; 668 goto zero; 669 } 670 671 ret = -EINVAL; 672 if (len < IFNAMSIZ) 673 goto out; 674 675 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 676 if (ret) 677 goto out; 678 679 len = strlen(devname) + 1; 680 681 ret = -EFAULT; 682 if (copy_to_user(optval, devname, len)) 683 goto out; 684 685 zero: 686 ret = -EFAULT; 687 if (put_user(len, optlen)) 688 goto out; 689 690 ret = 0; 691 692 out: 693 #endif 694 695 return ret; 696 } 697 698 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 699 int valbool) 700 { 701 if (valbool) 702 sock_set_flag(sk, bit); 703 else 704 sock_reset_flag(sk, bit); 705 } 706 707 bool sk_mc_loop(struct sock *sk) 708 { 709 if (dev_recursion_level()) 710 return false; 711 if (!sk) 712 return true; 713 switch (sk->sk_family) { 714 case AF_INET: 715 return inet_sk(sk)->mc_loop; 716 #if IS_ENABLED(CONFIG_IPV6) 717 case AF_INET6: 718 return inet6_sk(sk)->mc_loop; 719 #endif 720 } 721 WARN_ON(1); 722 return true; 723 } 724 EXPORT_SYMBOL(sk_mc_loop); 725 726 void sock_set_reuseaddr(struct sock *sk) 727 { 728 lock_sock(sk); 729 sk->sk_reuse = SK_CAN_REUSE; 730 release_sock(sk); 731 } 732 EXPORT_SYMBOL(sock_set_reuseaddr); 733 734 void sock_set_reuseport(struct sock *sk) 735 { 736 lock_sock(sk); 737 sk->sk_reuseport = true; 738 release_sock(sk); 739 } 740 EXPORT_SYMBOL(sock_set_reuseport); 741 742 void sock_no_linger(struct sock *sk) 743 { 744 lock_sock(sk); 745 sk->sk_lingertime = 0; 746 sock_set_flag(sk, SOCK_LINGER); 747 release_sock(sk); 748 } 749 EXPORT_SYMBOL(sock_no_linger); 750 751 void sock_set_priority(struct sock *sk, u32 priority) 752 { 753 lock_sock(sk); 754 sk->sk_priority = priority; 755 release_sock(sk); 756 } 757 EXPORT_SYMBOL(sock_set_priority); 758 759 void sock_set_sndtimeo(struct sock *sk, s64 secs) 760 { 761 lock_sock(sk); 762 if (secs && secs < MAX_SCHEDULE_TIMEOUT / HZ - 1) 763 sk->sk_sndtimeo = secs * HZ; 764 else 765 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 766 release_sock(sk); 767 } 768 EXPORT_SYMBOL(sock_set_sndtimeo); 769 770 static void __sock_set_timestamps(struct sock *sk, bool val, bool new, bool ns) 771 { 772 if (val) { 773 sock_valbool_flag(sk, SOCK_TSTAMP_NEW, new); 774 sock_valbool_flag(sk, SOCK_RCVTSTAMPNS, ns); 775 sock_set_flag(sk, SOCK_RCVTSTAMP); 776 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 777 } else { 778 sock_reset_flag(sk, SOCK_RCVTSTAMP); 779 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 780 sock_reset_flag(sk, SOCK_TSTAMP_NEW); 781 } 782 } 783 784 void sock_enable_timestamps(struct sock *sk) 785 { 786 lock_sock(sk); 787 __sock_set_timestamps(sk, true, false, true); 788 release_sock(sk); 789 } 790 EXPORT_SYMBOL(sock_enable_timestamps); 791 792 void sock_set_keepalive(struct sock *sk) 793 { 794 lock_sock(sk); 795 if (sk->sk_prot->keepalive) 796 sk->sk_prot->keepalive(sk, true); 797 sock_valbool_flag(sk, SOCK_KEEPOPEN, true); 798 release_sock(sk); 799 } 800 EXPORT_SYMBOL(sock_set_keepalive); 801 802 static void __sock_set_rcvbuf(struct sock *sk, int val) 803 { 804 /* Ensure val * 2 fits into an int, to prevent max_t() from treating it 805 * as a negative value. 806 */ 807 val = min_t(int, val, INT_MAX / 2); 808 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 809 810 /* We double it on the way in to account for "struct sk_buff" etc. 811 * overhead. Applications assume that the SO_RCVBUF setting they make 812 * will allow that much actual data to be received on that socket. 813 * 814 * Applications are unaware that "struct sk_buff" and other overheads 815 * allocate from the receive buffer during socket buffer allocation. 816 * 817 * And after considering the possible alternatives, returning the value 818 * we actually used in getsockopt is the most desirable behavior. 819 */ 820 WRITE_ONCE(sk->sk_rcvbuf, max_t(int, val * 2, SOCK_MIN_RCVBUF)); 821 } 822 823 void sock_set_rcvbuf(struct sock *sk, int val) 824 { 825 lock_sock(sk); 826 __sock_set_rcvbuf(sk, val); 827 release_sock(sk); 828 } 829 EXPORT_SYMBOL(sock_set_rcvbuf); 830 831 /* 832 * This is meant for all protocols to use and covers goings on 833 * at the socket level. Everything here is generic. 834 */ 835 836 int sock_setsockopt(struct socket *sock, int level, int optname, 837 char __user *optval, unsigned int optlen) 838 { 839 struct sock_txtime sk_txtime; 840 struct sock *sk = sock->sk; 841 int val; 842 int valbool; 843 struct linger ling; 844 int ret = 0; 845 846 /* 847 * Options without arguments 848 */ 849 850 if (optname == SO_BINDTODEVICE) 851 return sock_setbindtodevice(sk, optval, optlen); 852 853 if (optlen < sizeof(int)) 854 return -EINVAL; 855 856 if (get_user(val, (int __user *)optval)) 857 return -EFAULT; 858 859 valbool = val ? 1 : 0; 860 861 lock_sock(sk); 862 863 switch (optname) { 864 case SO_DEBUG: 865 if (val && !capable(CAP_NET_ADMIN)) 866 ret = -EACCES; 867 else 868 sock_valbool_flag(sk, SOCK_DBG, valbool); 869 break; 870 case SO_REUSEADDR: 871 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 872 break; 873 case SO_REUSEPORT: 874 sk->sk_reuseport = valbool; 875 break; 876 case SO_TYPE: 877 case SO_PROTOCOL: 878 case SO_DOMAIN: 879 case SO_ERROR: 880 ret = -ENOPROTOOPT; 881 break; 882 case SO_DONTROUTE: 883 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 884 sk_dst_reset(sk); 885 break; 886 case SO_BROADCAST: 887 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 888 break; 889 case SO_SNDBUF: 890 /* Don't error on this BSD doesn't and if you think 891 * about it this is right. Otherwise apps have to 892 * play 'guess the biggest size' games. RCVBUF/SNDBUF 893 * are treated in BSD as hints 894 */ 895 val = min_t(u32, val, sysctl_wmem_max); 896 set_sndbuf: 897 /* Ensure val * 2 fits into an int, to prevent max_t() 898 * from treating it as a negative value. 899 */ 900 val = min_t(int, val, INT_MAX / 2); 901 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 902 WRITE_ONCE(sk->sk_sndbuf, 903 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 904 /* Wake up sending tasks if we upped the value. */ 905 sk->sk_write_space(sk); 906 break; 907 908 case SO_SNDBUFFORCE: 909 if (!capable(CAP_NET_ADMIN)) { 910 ret = -EPERM; 911 break; 912 } 913 914 /* No negative values (to prevent underflow, as val will be 915 * multiplied by 2). 916 */ 917 if (val < 0) 918 val = 0; 919 goto set_sndbuf; 920 921 case SO_RCVBUF: 922 /* Don't error on this BSD doesn't and if you think 923 * about it this is right. Otherwise apps have to 924 * play 'guess the biggest size' games. RCVBUF/SNDBUF 925 * are treated in BSD as hints 926 */ 927 __sock_set_rcvbuf(sk, min_t(u32, val, sysctl_rmem_max)); 928 break; 929 930 case SO_RCVBUFFORCE: 931 if (!capable(CAP_NET_ADMIN)) { 932 ret = -EPERM; 933 break; 934 } 935 936 /* No negative values (to prevent underflow, as val will be 937 * multiplied by 2). 938 */ 939 __sock_set_rcvbuf(sk, max(val, 0)); 940 break; 941 942 case SO_KEEPALIVE: 943 if (sk->sk_prot->keepalive) 944 sk->sk_prot->keepalive(sk, valbool); 945 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 946 break; 947 948 case SO_OOBINLINE: 949 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 950 break; 951 952 case SO_NO_CHECK: 953 sk->sk_no_check_tx = valbool; 954 break; 955 956 case SO_PRIORITY: 957 if ((val >= 0 && val <= 6) || 958 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 959 sk->sk_priority = val; 960 else 961 ret = -EPERM; 962 break; 963 964 case SO_LINGER: 965 if (optlen < sizeof(ling)) { 966 ret = -EINVAL; /* 1003.1g */ 967 break; 968 } 969 if (copy_from_user(&ling, optval, sizeof(ling))) { 970 ret = -EFAULT; 971 break; 972 } 973 if (!ling.l_onoff) 974 sock_reset_flag(sk, SOCK_LINGER); 975 else { 976 #if (BITS_PER_LONG == 32) 977 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 978 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 979 else 980 #endif 981 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 982 sock_set_flag(sk, SOCK_LINGER); 983 } 984 break; 985 986 case SO_BSDCOMPAT: 987 sock_warn_obsolete_bsdism("setsockopt"); 988 break; 989 990 case SO_PASSCRED: 991 if (valbool) 992 set_bit(SOCK_PASSCRED, &sock->flags); 993 else 994 clear_bit(SOCK_PASSCRED, &sock->flags); 995 break; 996 997 case SO_TIMESTAMP_OLD: 998 __sock_set_timestamps(sk, valbool, false, false); 999 break; 1000 case SO_TIMESTAMP_NEW: 1001 __sock_set_timestamps(sk, valbool, true, false); 1002 break; 1003 case SO_TIMESTAMPNS_OLD: 1004 __sock_set_timestamps(sk, valbool, false, true); 1005 break; 1006 case SO_TIMESTAMPNS_NEW: 1007 __sock_set_timestamps(sk, valbool, true, true); 1008 break; 1009 case SO_TIMESTAMPING_NEW: 1010 sock_set_flag(sk, SOCK_TSTAMP_NEW); 1011 /* fall through */ 1012 case SO_TIMESTAMPING_OLD: 1013 if (val & ~SOF_TIMESTAMPING_MASK) { 1014 ret = -EINVAL; 1015 break; 1016 } 1017 1018 if (val & SOF_TIMESTAMPING_OPT_ID && 1019 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 1020 if (sk->sk_protocol == IPPROTO_TCP && 1021 sk->sk_type == SOCK_STREAM) { 1022 if ((1 << sk->sk_state) & 1023 (TCPF_CLOSE | TCPF_LISTEN)) { 1024 ret = -EINVAL; 1025 break; 1026 } 1027 sk->sk_tskey = tcp_sk(sk)->snd_una; 1028 } else { 1029 sk->sk_tskey = 0; 1030 } 1031 } 1032 1033 if (val & SOF_TIMESTAMPING_OPT_STATS && 1034 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) { 1035 ret = -EINVAL; 1036 break; 1037 } 1038 1039 sk->sk_tsflags = val; 1040 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 1041 sock_enable_timestamp(sk, 1042 SOCK_TIMESTAMPING_RX_SOFTWARE); 1043 else { 1044 if (optname == SO_TIMESTAMPING_NEW) 1045 sock_reset_flag(sk, SOCK_TSTAMP_NEW); 1046 1047 sock_disable_timestamp(sk, 1048 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 1049 } 1050 break; 1051 1052 case SO_RCVLOWAT: 1053 if (val < 0) 1054 val = INT_MAX; 1055 if (sock->ops->set_rcvlowat) 1056 ret = sock->ops->set_rcvlowat(sk, val); 1057 else 1058 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 1059 break; 1060 1061 case SO_RCVTIMEO_OLD: 1062 case SO_RCVTIMEO_NEW: 1063 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD); 1064 break; 1065 1066 case SO_SNDTIMEO_OLD: 1067 case SO_SNDTIMEO_NEW: 1068 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD); 1069 break; 1070 1071 case SO_ATTACH_FILTER: 1072 ret = -EINVAL; 1073 if (optlen == sizeof(struct sock_fprog)) { 1074 struct sock_fprog fprog; 1075 1076 ret = -EFAULT; 1077 if (copy_from_user(&fprog, optval, sizeof(fprog))) 1078 break; 1079 1080 ret = sk_attach_filter(&fprog, sk); 1081 } 1082 break; 1083 1084 case SO_ATTACH_BPF: 1085 ret = -EINVAL; 1086 if (optlen == sizeof(u32)) { 1087 u32 ufd; 1088 1089 ret = -EFAULT; 1090 if (copy_from_user(&ufd, optval, sizeof(ufd))) 1091 break; 1092 1093 ret = sk_attach_bpf(ufd, sk); 1094 } 1095 break; 1096 1097 case SO_ATTACH_REUSEPORT_CBPF: 1098 ret = -EINVAL; 1099 if (optlen == sizeof(struct sock_fprog)) { 1100 struct sock_fprog fprog; 1101 1102 ret = -EFAULT; 1103 if (copy_from_user(&fprog, optval, sizeof(fprog))) 1104 break; 1105 1106 ret = sk_reuseport_attach_filter(&fprog, sk); 1107 } 1108 break; 1109 1110 case SO_ATTACH_REUSEPORT_EBPF: 1111 ret = -EINVAL; 1112 if (optlen == sizeof(u32)) { 1113 u32 ufd; 1114 1115 ret = -EFAULT; 1116 if (copy_from_user(&ufd, optval, sizeof(ufd))) 1117 break; 1118 1119 ret = sk_reuseport_attach_bpf(ufd, sk); 1120 } 1121 break; 1122 1123 case SO_DETACH_REUSEPORT_BPF: 1124 ret = reuseport_detach_prog(sk); 1125 break; 1126 1127 case SO_DETACH_FILTER: 1128 ret = sk_detach_filter(sk); 1129 break; 1130 1131 case SO_LOCK_FILTER: 1132 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1133 ret = -EPERM; 1134 else 1135 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1136 break; 1137 1138 case SO_PASSSEC: 1139 if (valbool) 1140 set_bit(SOCK_PASSSEC, &sock->flags); 1141 else 1142 clear_bit(SOCK_PASSSEC, &sock->flags); 1143 break; 1144 case SO_MARK: 1145 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1146 ret = -EPERM; 1147 } else if (val != sk->sk_mark) { 1148 sk->sk_mark = val; 1149 sk_dst_reset(sk); 1150 } 1151 break; 1152 1153 case SO_RXQ_OVFL: 1154 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1155 break; 1156 1157 case SO_WIFI_STATUS: 1158 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1159 break; 1160 1161 case SO_PEEK_OFF: 1162 if (sock->ops->set_peek_off) 1163 ret = sock->ops->set_peek_off(sk, val); 1164 else 1165 ret = -EOPNOTSUPP; 1166 break; 1167 1168 case SO_NOFCS: 1169 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1170 break; 1171 1172 case SO_SELECT_ERR_QUEUE: 1173 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1174 break; 1175 1176 #ifdef CONFIG_NET_RX_BUSY_POLL 1177 case SO_BUSY_POLL: 1178 /* allow unprivileged users to decrease the value */ 1179 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1180 ret = -EPERM; 1181 else { 1182 if (val < 0) 1183 ret = -EINVAL; 1184 else 1185 sk->sk_ll_usec = val; 1186 } 1187 break; 1188 #endif 1189 1190 case SO_MAX_PACING_RATE: 1191 { 1192 unsigned long ulval = (val == ~0U) ? ~0UL : val; 1193 1194 if (sizeof(ulval) != sizeof(val) && 1195 optlen >= sizeof(ulval) && 1196 get_user(ulval, (unsigned long __user *)optval)) { 1197 ret = -EFAULT; 1198 break; 1199 } 1200 if (ulval != ~0UL) 1201 cmpxchg(&sk->sk_pacing_status, 1202 SK_PACING_NONE, 1203 SK_PACING_NEEDED); 1204 sk->sk_max_pacing_rate = ulval; 1205 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1206 break; 1207 } 1208 case SO_INCOMING_CPU: 1209 WRITE_ONCE(sk->sk_incoming_cpu, val); 1210 break; 1211 1212 case SO_CNX_ADVICE: 1213 if (val == 1) 1214 dst_negative_advice(sk); 1215 break; 1216 1217 case SO_ZEROCOPY: 1218 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1219 if (!((sk->sk_type == SOCK_STREAM && 1220 sk->sk_protocol == IPPROTO_TCP) || 1221 (sk->sk_type == SOCK_DGRAM && 1222 sk->sk_protocol == IPPROTO_UDP))) 1223 ret = -ENOTSUPP; 1224 } else if (sk->sk_family != PF_RDS) { 1225 ret = -ENOTSUPP; 1226 } 1227 if (!ret) { 1228 if (val < 0 || val > 1) 1229 ret = -EINVAL; 1230 else 1231 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1232 } 1233 break; 1234 1235 case SO_TXTIME: 1236 if (optlen != sizeof(struct sock_txtime)) { 1237 ret = -EINVAL; 1238 break; 1239 } else if (copy_from_user(&sk_txtime, optval, 1240 sizeof(struct sock_txtime))) { 1241 ret = -EFAULT; 1242 break; 1243 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1244 ret = -EINVAL; 1245 break; 1246 } 1247 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1248 * scheduler has enough safe guards. 1249 */ 1250 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1251 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1252 ret = -EPERM; 1253 break; 1254 } 1255 sock_valbool_flag(sk, SOCK_TXTIME, true); 1256 sk->sk_clockid = sk_txtime.clockid; 1257 sk->sk_txtime_deadline_mode = 1258 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1259 sk->sk_txtime_report_errors = 1260 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1261 break; 1262 1263 case SO_BINDTOIFINDEX: 1264 ret = sock_bindtoindex_locked(sk, val); 1265 break; 1266 1267 default: 1268 ret = -ENOPROTOOPT; 1269 break; 1270 } 1271 release_sock(sk); 1272 return ret; 1273 } 1274 EXPORT_SYMBOL(sock_setsockopt); 1275 1276 1277 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1278 struct ucred *ucred) 1279 { 1280 ucred->pid = pid_vnr(pid); 1281 ucred->uid = ucred->gid = -1; 1282 if (cred) { 1283 struct user_namespace *current_ns = current_user_ns(); 1284 1285 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1286 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1287 } 1288 } 1289 1290 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1291 { 1292 struct user_namespace *user_ns = current_user_ns(); 1293 int i; 1294 1295 for (i = 0; i < src->ngroups; i++) 1296 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1297 return -EFAULT; 1298 1299 return 0; 1300 } 1301 1302 int sock_getsockopt(struct socket *sock, int level, int optname, 1303 char __user *optval, int __user *optlen) 1304 { 1305 struct sock *sk = sock->sk; 1306 1307 union { 1308 int val; 1309 u64 val64; 1310 unsigned long ulval; 1311 struct linger ling; 1312 struct old_timeval32 tm32; 1313 struct __kernel_old_timeval tm; 1314 struct __kernel_sock_timeval stm; 1315 struct sock_txtime txtime; 1316 } v; 1317 1318 int lv = sizeof(int); 1319 int len; 1320 1321 if (get_user(len, optlen)) 1322 return -EFAULT; 1323 if (len < 0) 1324 return -EINVAL; 1325 1326 memset(&v, 0, sizeof(v)); 1327 1328 switch (optname) { 1329 case SO_DEBUG: 1330 v.val = sock_flag(sk, SOCK_DBG); 1331 break; 1332 1333 case SO_DONTROUTE: 1334 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1335 break; 1336 1337 case SO_BROADCAST: 1338 v.val = sock_flag(sk, SOCK_BROADCAST); 1339 break; 1340 1341 case SO_SNDBUF: 1342 v.val = sk->sk_sndbuf; 1343 break; 1344 1345 case SO_RCVBUF: 1346 v.val = sk->sk_rcvbuf; 1347 break; 1348 1349 case SO_REUSEADDR: 1350 v.val = sk->sk_reuse; 1351 break; 1352 1353 case SO_REUSEPORT: 1354 v.val = sk->sk_reuseport; 1355 break; 1356 1357 case SO_KEEPALIVE: 1358 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1359 break; 1360 1361 case SO_TYPE: 1362 v.val = sk->sk_type; 1363 break; 1364 1365 case SO_PROTOCOL: 1366 v.val = sk->sk_protocol; 1367 break; 1368 1369 case SO_DOMAIN: 1370 v.val = sk->sk_family; 1371 break; 1372 1373 case SO_ERROR: 1374 v.val = -sock_error(sk); 1375 if (v.val == 0) 1376 v.val = xchg(&sk->sk_err_soft, 0); 1377 break; 1378 1379 case SO_OOBINLINE: 1380 v.val = sock_flag(sk, SOCK_URGINLINE); 1381 break; 1382 1383 case SO_NO_CHECK: 1384 v.val = sk->sk_no_check_tx; 1385 break; 1386 1387 case SO_PRIORITY: 1388 v.val = sk->sk_priority; 1389 break; 1390 1391 case SO_LINGER: 1392 lv = sizeof(v.ling); 1393 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1394 v.ling.l_linger = sk->sk_lingertime / HZ; 1395 break; 1396 1397 case SO_BSDCOMPAT: 1398 sock_warn_obsolete_bsdism("getsockopt"); 1399 break; 1400 1401 case SO_TIMESTAMP_OLD: 1402 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1403 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1404 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1405 break; 1406 1407 case SO_TIMESTAMPNS_OLD: 1408 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1409 break; 1410 1411 case SO_TIMESTAMP_NEW: 1412 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1413 break; 1414 1415 case SO_TIMESTAMPNS_NEW: 1416 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1417 break; 1418 1419 case SO_TIMESTAMPING_OLD: 1420 v.val = sk->sk_tsflags; 1421 break; 1422 1423 case SO_RCVTIMEO_OLD: 1424 case SO_RCVTIMEO_NEW: 1425 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1426 break; 1427 1428 case SO_SNDTIMEO_OLD: 1429 case SO_SNDTIMEO_NEW: 1430 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1431 break; 1432 1433 case SO_RCVLOWAT: 1434 v.val = sk->sk_rcvlowat; 1435 break; 1436 1437 case SO_SNDLOWAT: 1438 v.val = 1; 1439 break; 1440 1441 case SO_PASSCRED: 1442 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1443 break; 1444 1445 case SO_PEERCRED: 1446 { 1447 struct ucred peercred; 1448 if (len > sizeof(peercred)) 1449 len = sizeof(peercred); 1450 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1451 if (copy_to_user(optval, &peercred, len)) 1452 return -EFAULT; 1453 goto lenout; 1454 } 1455 1456 case SO_PEERGROUPS: 1457 { 1458 int ret, n; 1459 1460 if (!sk->sk_peer_cred) 1461 return -ENODATA; 1462 1463 n = sk->sk_peer_cred->group_info->ngroups; 1464 if (len < n * sizeof(gid_t)) { 1465 len = n * sizeof(gid_t); 1466 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1467 } 1468 len = n * sizeof(gid_t); 1469 1470 ret = groups_to_user((gid_t __user *)optval, 1471 sk->sk_peer_cred->group_info); 1472 if (ret) 1473 return ret; 1474 goto lenout; 1475 } 1476 1477 case SO_PEERNAME: 1478 { 1479 char address[128]; 1480 1481 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1482 if (lv < 0) 1483 return -ENOTCONN; 1484 if (lv < len) 1485 return -EINVAL; 1486 if (copy_to_user(optval, address, len)) 1487 return -EFAULT; 1488 goto lenout; 1489 } 1490 1491 /* Dubious BSD thing... Probably nobody even uses it, but 1492 * the UNIX standard wants it for whatever reason... -DaveM 1493 */ 1494 case SO_ACCEPTCONN: 1495 v.val = sk->sk_state == TCP_LISTEN; 1496 break; 1497 1498 case SO_PASSSEC: 1499 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1500 break; 1501 1502 case SO_PEERSEC: 1503 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1504 1505 case SO_MARK: 1506 v.val = sk->sk_mark; 1507 break; 1508 1509 case SO_RXQ_OVFL: 1510 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1511 break; 1512 1513 case SO_WIFI_STATUS: 1514 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1515 break; 1516 1517 case SO_PEEK_OFF: 1518 if (!sock->ops->set_peek_off) 1519 return -EOPNOTSUPP; 1520 1521 v.val = sk->sk_peek_off; 1522 break; 1523 case SO_NOFCS: 1524 v.val = sock_flag(sk, SOCK_NOFCS); 1525 break; 1526 1527 case SO_BINDTODEVICE: 1528 return sock_getbindtodevice(sk, optval, optlen, len); 1529 1530 case SO_GET_FILTER: 1531 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1532 if (len < 0) 1533 return len; 1534 1535 goto lenout; 1536 1537 case SO_LOCK_FILTER: 1538 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1539 break; 1540 1541 case SO_BPF_EXTENSIONS: 1542 v.val = bpf_tell_extensions(); 1543 break; 1544 1545 case SO_SELECT_ERR_QUEUE: 1546 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1547 break; 1548 1549 #ifdef CONFIG_NET_RX_BUSY_POLL 1550 case SO_BUSY_POLL: 1551 v.val = sk->sk_ll_usec; 1552 break; 1553 #endif 1554 1555 case SO_MAX_PACING_RATE: 1556 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1557 lv = sizeof(v.ulval); 1558 v.ulval = sk->sk_max_pacing_rate; 1559 } else { 1560 /* 32bit version */ 1561 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1562 } 1563 break; 1564 1565 case SO_INCOMING_CPU: 1566 v.val = READ_ONCE(sk->sk_incoming_cpu); 1567 break; 1568 1569 case SO_MEMINFO: 1570 { 1571 u32 meminfo[SK_MEMINFO_VARS]; 1572 1573 sk_get_meminfo(sk, meminfo); 1574 1575 len = min_t(unsigned int, len, sizeof(meminfo)); 1576 if (copy_to_user(optval, &meminfo, len)) 1577 return -EFAULT; 1578 1579 goto lenout; 1580 } 1581 1582 #ifdef CONFIG_NET_RX_BUSY_POLL 1583 case SO_INCOMING_NAPI_ID: 1584 v.val = READ_ONCE(sk->sk_napi_id); 1585 1586 /* aggregate non-NAPI IDs down to 0 */ 1587 if (v.val < MIN_NAPI_ID) 1588 v.val = 0; 1589 1590 break; 1591 #endif 1592 1593 case SO_COOKIE: 1594 lv = sizeof(u64); 1595 if (len < lv) 1596 return -EINVAL; 1597 v.val64 = sock_gen_cookie(sk); 1598 break; 1599 1600 case SO_ZEROCOPY: 1601 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1602 break; 1603 1604 case SO_TXTIME: 1605 lv = sizeof(v.txtime); 1606 v.txtime.clockid = sk->sk_clockid; 1607 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1608 SOF_TXTIME_DEADLINE_MODE : 0; 1609 v.txtime.flags |= sk->sk_txtime_report_errors ? 1610 SOF_TXTIME_REPORT_ERRORS : 0; 1611 break; 1612 1613 case SO_BINDTOIFINDEX: 1614 v.val = sk->sk_bound_dev_if; 1615 break; 1616 1617 default: 1618 /* We implement the SO_SNDLOWAT etc to not be settable 1619 * (1003.1g 7). 1620 */ 1621 return -ENOPROTOOPT; 1622 } 1623 1624 if (len > lv) 1625 len = lv; 1626 if (copy_to_user(optval, &v, len)) 1627 return -EFAULT; 1628 lenout: 1629 if (put_user(len, optlen)) 1630 return -EFAULT; 1631 return 0; 1632 } 1633 1634 /* 1635 * Initialize an sk_lock. 1636 * 1637 * (We also register the sk_lock with the lock validator.) 1638 */ 1639 static inline void sock_lock_init(struct sock *sk) 1640 { 1641 if (sk->sk_kern_sock) 1642 sock_lock_init_class_and_name( 1643 sk, 1644 af_family_kern_slock_key_strings[sk->sk_family], 1645 af_family_kern_slock_keys + sk->sk_family, 1646 af_family_kern_key_strings[sk->sk_family], 1647 af_family_kern_keys + sk->sk_family); 1648 else 1649 sock_lock_init_class_and_name( 1650 sk, 1651 af_family_slock_key_strings[sk->sk_family], 1652 af_family_slock_keys + sk->sk_family, 1653 af_family_key_strings[sk->sk_family], 1654 af_family_keys + sk->sk_family); 1655 } 1656 1657 /* 1658 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1659 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1660 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1661 */ 1662 static void sock_copy(struct sock *nsk, const struct sock *osk) 1663 { 1664 const struct proto *prot = READ_ONCE(osk->sk_prot); 1665 #ifdef CONFIG_SECURITY_NETWORK 1666 void *sptr = nsk->sk_security; 1667 #endif 1668 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1669 1670 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1671 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1672 1673 #ifdef CONFIG_SECURITY_NETWORK 1674 nsk->sk_security = sptr; 1675 security_sk_clone(osk, nsk); 1676 #endif 1677 } 1678 1679 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1680 int family) 1681 { 1682 struct sock *sk; 1683 struct kmem_cache *slab; 1684 1685 slab = prot->slab; 1686 if (slab != NULL) { 1687 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1688 if (!sk) 1689 return sk; 1690 if (want_init_on_alloc(priority)) 1691 sk_prot_clear_nulls(sk, prot->obj_size); 1692 } else 1693 sk = kmalloc(prot->obj_size, priority); 1694 1695 if (sk != NULL) { 1696 if (security_sk_alloc(sk, family, priority)) 1697 goto out_free; 1698 1699 if (!try_module_get(prot->owner)) 1700 goto out_free_sec; 1701 sk_tx_queue_clear(sk); 1702 } 1703 1704 return sk; 1705 1706 out_free_sec: 1707 security_sk_free(sk); 1708 out_free: 1709 if (slab != NULL) 1710 kmem_cache_free(slab, sk); 1711 else 1712 kfree(sk); 1713 return NULL; 1714 } 1715 1716 static void sk_prot_free(struct proto *prot, struct sock *sk) 1717 { 1718 struct kmem_cache *slab; 1719 struct module *owner; 1720 1721 owner = prot->owner; 1722 slab = prot->slab; 1723 1724 cgroup_sk_free(&sk->sk_cgrp_data); 1725 mem_cgroup_sk_free(sk); 1726 security_sk_free(sk); 1727 if (slab != NULL) 1728 kmem_cache_free(slab, sk); 1729 else 1730 kfree(sk); 1731 module_put(owner); 1732 } 1733 1734 /** 1735 * sk_alloc - All socket objects are allocated here 1736 * @net: the applicable net namespace 1737 * @family: protocol family 1738 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1739 * @prot: struct proto associated with this new sock instance 1740 * @kern: is this to be a kernel socket? 1741 */ 1742 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1743 struct proto *prot, int kern) 1744 { 1745 struct sock *sk; 1746 1747 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1748 if (sk) { 1749 sk->sk_family = family; 1750 /* 1751 * See comment in struct sock definition to understand 1752 * why we need sk_prot_creator -acme 1753 */ 1754 sk->sk_prot = sk->sk_prot_creator = prot; 1755 sk->sk_kern_sock = kern; 1756 sock_lock_init(sk); 1757 sk->sk_net_refcnt = kern ? 0 : 1; 1758 if (likely(sk->sk_net_refcnt)) { 1759 get_net(net); 1760 sock_inuse_add(net, 1); 1761 } 1762 1763 sock_net_set(sk, net); 1764 refcount_set(&sk->sk_wmem_alloc, 1); 1765 1766 mem_cgroup_sk_alloc(sk); 1767 cgroup_sk_alloc(&sk->sk_cgrp_data); 1768 sock_update_classid(&sk->sk_cgrp_data); 1769 sock_update_netprioidx(&sk->sk_cgrp_data); 1770 } 1771 1772 return sk; 1773 } 1774 EXPORT_SYMBOL(sk_alloc); 1775 1776 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1777 * grace period. This is the case for UDP sockets and TCP listeners. 1778 */ 1779 static void __sk_destruct(struct rcu_head *head) 1780 { 1781 struct sock *sk = container_of(head, struct sock, sk_rcu); 1782 struct sk_filter *filter; 1783 1784 if (sk->sk_destruct) 1785 sk->sk_destruct(sk); 1786 1787 filter = rcu_dereference_check(sk->sk_filter, 1788 refcount_read(&sk->sk_wmem_alloc) == 0); 1789 if (filter) { 1790 sk_filter_uncharge(sk, filter); 1791 RCU_INIT_POINTER(sk->sk_filter, NULL); 1792 } 1793 1794 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1795 1796 #ifdef CONFIG_BPF_SYSCALL 1797 bpf_sk_storage_free(sk); 1798 #endif 1799 1800 if (atomic_read(&sk->sk_omem_alloc)) 1801 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1802 __func__, atomic_read(&sk->sk_omem_alloc)); 1803 1804 if (sk->sk_frag.page) { 1805 put_page(sk->sk_frag.page); 1806 sk->sk_frag.page = NULL; 1807 } 1808 1809 if (sk->sk_peer_cred) 1810 put_cred(sk->sk_peer_cred); 1811 put_pid(sk->sk_peer_pid); 1812 if (likely(sk->sk_net_refcnt)) 1813 put_net(sock_net(sk)); 1814 sk_prot_free(sk->sk_prot_creator, sk); 1815 } 1816 1817 void sk_destruct(struct sock *sk) 1818 { 1819 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 1820 1821 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 1822 reuseport_detach_sock(sk); 1823 use_call_rcu = true; 1824 } 1825 1826 if (use_call_rcu) 1827 call_rcu(&sk->sk_rcu, __sk_destruct); 1828 else 1829 __sk_destruct(&sk->sk_rcu); 1830 } 1831 1832 static void __sk_free(struct sock *sk) 1833 { 1834 if (likely(sk->sk_net_refcnt)) 1835 sock_inuse_add(sock_net(sk), -1); 1836 1837 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1838 sock_diag_broadcast_destroy(sk); 1839 else 1840 sk_destruct(sk); 1841 } 1842 1843 void sk_free(struct sock *sk) 1844 { 1845 /* 1846 * We subtract one from sk_wmem_alloc and can know if 1847 * some packets are still in some tx queue. 1848 * If not null, sock_wfree() will call __sk_free(sk) later 1849 */ 1850 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1851 __sk_free(sk); 1852 } 1853 EXPORT_SYMBOL(sk_free); 1854 1855 static void sk_init_common(struct sock *sk) 1856 { 1857 skb_queue_head_init(&sk->sk_receive_queue); 1858 skb_queue_head_init(&sk->sk_write_queue); 1859 skb_queue_head_init(&sk->sk_error_queue); 1860 1861 rwlock_init(&sk->sk_callback_lock); 1862 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1863 af_rlock_keys + sk->sk_family, 1864 af_family_rlock_key_strings[sk->sk_family]); 1865 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1866 af_wlock_keys + sk->sk_family, 1867 af_family_wlock_key_strings[sk->sk_family]); 1868 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1869 af_elock_keys + sk->sk_family, 1870 af_family_elock_key_strings[sk->sk_family]); 1871 lockdep_set_class_and_name(&sk->sk_callback_lock, 1872 af_callback_keys + sk->sk_family, 1873 af_family_clock_key_strings[sk->sk_family]); 1874 } 1875 1876 /** 1877 * sk_clone_lock - clone a socket, and lock its clone 1878 * @sk: the socket to clone 1879 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1880 * 1881 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1882 */ 1883 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1884 { 1885 struct proto *prot = READ_ONCE(sk->sk_prot); 1886 struct sock *newsk; 1887 bool is_charged = true; 1888 1889 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 1890 if (newsk != NULL) { 1891 struct sk_filter *filter; 1892 1893 sock_copy(newsk, sk); 1894 1895 newsk->sk_prot_creator = prot; 1896 1897 /* SANITY */ 1898 if (likely(newsk->sk_net_refcnt)) 1899 get_net(sock_net(newsk)); 1900 sk_node_init(&newsk->sk_node); 1901 sock_lock_init(newsk); 1902 bh_lock_sock(newsk); 1903 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1904 newsk->sk_backlog.len = 0; 1905 1906 atomic_set(&newsk->sk_rmem_alloc, 0); 1907 /* 1908 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1909 */ 1910 refcount_set(&newsk->sk_wmem_alloc, 1); 1911 atomic_set(&newsk->sk_omem_alloc, 0); 1912 sk_init_common(newsk); 1913 1914 newsk->sk_dst_cache = NULL; 1915 newsk->sk_dst_pending_confirm = 0; 1916 newsk->sk_wmem_queued = 0; 1917 newsk->sk_forward_alloc = 0; 1918 atomic_set(&newsk->sk_drops, 0); 1919 newsk->sk_send_head = NULL; 1920 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1921 atomic_set(&newsk->sk_zckey, 0); 1922 1923 sock_reset_flag(newsk, SOCK_DONE); 1924 1925 /* sk->sk_memcg will be populated at accept() time */ 1926 newsk->sk_memcg = NULL; 1927 1928 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1929 1930 rcu_read_lock(); 1931 filter = rcu_dereference(sk->sk_filter); 1932 if (filter != NULL) 1933 /* though it's an empty new sock, the charging may fail 1934 * if sysctl_optmem_max was changed between creation of 1935 * original socket and cloning 1936 */ 1937 is_charged = sk_filter_charge(newsk, filter); 1938 RCU_INIT_POINTER(newsk->sk_filter, filter); 1939 rcu_read_unlock(); 1940 1941 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1942 /* We need to make sure that we don't uncharge the new 1943 * socket if we couldn't charge it in the first place 1944 * as otherwise we uncharge the parent's filter. 1945 */ 1946 if (!is_charged) 1947 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1948 sk_free_unlock_clone(newsk); 1949 newsk = NULL; 1950 goto out; 1951 } 1952 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1953 1954 if (bpf_sk_storage_clone(sk, newsk)) { 1955 sk_free_unlock_clone(newsk); 1956 newsk = NULL; 1957 goto out; 1958 } 1959 1960 /* Clear sk_user_data if parent had the pointer tagged 1961 * as not suitable for copying when cloning. 1962 */ 1963 if (sk_user_data_is_nocopy(newsk)) 1964 newsk->sk_user_data = NULL; 1965 1966 newsk->sk_err = 0; 1967 newsk->sk_err_soft = 0; 1968 newsk->sk_priority = 0; 1969 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1970 if (likely(newsk->sk_net_refcnt)) 1971 sock_inuse_add(sock_net(newsk), 1); 1972 1973 /* 1974 * Before updating sk_refcnt, we must commit prior changes to memory 1975 * (Documentation/RCU/rculist_nulls.txt for details) 1976 */ 1977 smp_wmb(); 1978 refcount_set(&newsk->sk_refcnt, 2); 1979 1980 /* 1981 * Increment the counter in the same struct proto as the master 1982 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1983 * is the same as sk->sk_prot->socks, as this field was copied 1984 * with memcpy). 1985 * 1986 * This _changes_ the previous behaviour, where 1987 * tcp_create_openreq_child always was incrementing the 1988 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1989 * to be taken into account in all callers. -acme 1990 */ 1991 sk_refcnt_debug_inc(newsk); 1992 sk_set_socket(newsk, NULL); 1993 RCU_INIT_POINTER(newsk->sk_wq, NULL); 1994 1995 if (newsk->sk_prot->sockets_allocated) 1996 sk_sockets_allocated_inc(newsk); 1997 1998 if (sock_needs_netstamp(sk) && 1999 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 2000 net_enable_timestamp(); 2001 } 2002 out: 2003 return newsk; 2004 } 2005 EXPORT_SYMBOL_GPL(sk_clone_lock); 2006 2007 void sk_free_unlock_clone(struct sock *sk) 2008 { 2009 /* It is still raw copy of parent, so invalidate 2010 * destructor and make plain sk_free() */ 2011 sk->sk_destruct = NULL; 2012 bh_unlock_sock(sk); 2013 sk_free(sk); 2014 } 2015 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 2016 2017 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 2018 { 2019 u32 max_segs = 1; 2020 2021 sk_dst_set(sk, dst); 2022 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 2023 if (sk->sk_route_caps & NETIF_F_GSO) 2024 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 2025 sk->sk_route_caps &= ~sk->sk_route_nocaps; 2026 if (sk_can_gso(sk)) { 2027 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 2028 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2029 } else { 2030 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 2031 sk->sk_gso_max_size = dst->dev->gso_max_size; 2032 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 2033 } 2034 } 2035 sk->sk_gso_max_segs = max_segs; 2036 } 2037 EXPORT_SYMBOL_GPL(sk_setup_caps); 2038 2039 /* 2040 * Simple resource managers for sockets. 2041 */ 2042 2043 2044 /* 2045 * Write buffer destructor automatically called from kfree_skb. 2046 */ 2047 void sock_wfree(struct sk_buff *skb) 2048 { 2049 struct sock *sk = skb->sk; 2050 unsigned int len = skb->truesize; 2051 2052 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 2053 /* 2054 * Keep a reference on sk_wmem_alloc, this will be released 2055 * after sk_write_space() call 2056 */ 2057 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 2058 sk->sk_write_space(sk); 2059 len = 1; 2060 } 2061 /* 2062 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 2063 * could not do because of in-flight packets 2064 */ 2065 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 2066 __sk_free(sk); 2067 } 2068 EXPORT_SYMBOL(sock_wfree); 2069 2070 /* This variant of sock_wfree() is used by TCP, 2071 * since it sets SOCK_USE_WRITE_QUEUE. 2072 */ 2073 void __sock_wfree(struct sk_buff *skb) 2074 { 2075 struct sock *sk = skb->sk; 2076 2077 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 2078 __sk_free(sk); 2079 } 2080 2081 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2082 { 2083 skb_orphan(skb); 2084 skb->sk = sk; 2085 #ifdef CONFIG_INET 2086 if (unlikely(!sk_fullsock(sk))) { 2087 skb->destructor = sock_edemux; 2088 sock_hold(sk); 2089 return; 2090 } 2091 #endif 2092 skb->destructor = sock_wfree; 2093 skb_set_hash_from_sk(skb, sk); 2094 /* 2095 * We used to take a refcount on sk, but following operation 2096 * is enough to guarantee sk_free() wont free this sock until 2097 * all in-flight packets are completed 2098 */ 2099 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2100 } 2101 EXPORT_SYMBOL(skb_set_owner_w); 2102 2103 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2104 { 2105 #ifdef CONFIG_TLS_DEVICE 2106 /* Drivers depend on in-order delivery for crypto offload, 2107 * partial orphan breaks out-of-order-OK logic. 2108 */ 2109 if (skb->decrypted) 2110 return false; 2111 #endif 2112 return (skb->destructor == sock_wfree || 2113 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2114 } 2115 2116 /* This helper is used by netem, as it can hold packets in its 2117 * delay queue. We want to allow the owner socket to send more 2118 * packets, as if they were already TX completed by a typical driver. 2119 * But we also want to keep skb->sk set because some packet schedulers 2120 * rely on it (sch_fq for example). 2121 */ 2122 void skb_orphan_partial(struct sk_buff *skb) 2123 { 2124 if (skb_is_tcp_pure_ack(skb)) 2125 return; 2126 2127 if (can_skb_orphan_partial(skb)) { 2128 struct sock *sk = skb->sk; 2129 2130 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 2131 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)); 2132 skb->destructor = sock_efree; 2133 } 2134 } else { 2135 skb_orphan(skb); 2136 } 2137 } 2138 EXPORT_SYMBOL(skb_orphan_partial); 2139 2140 /* 2141 * Read buffer destructor automatically called from kfree_skb. 2142 */ 2143 void sock_rfree(struct sk_buff *skb) 2144 { 2145 struct sock *sk = skb->sk; 2146 unsigned int len = skb->truesize; 2147 2148 atomic_sub(len, &sk->sk_rmem_alloc); 2149 sk_mem_uncharge(sk, len); 2150 } 2151 EXPORT_SYMBOL(sock_rfree); 2152 2153 /* 2154 * Buffer destructor for skbs that are not used directly in read or write 2155 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2156 */ 2157 void sock_efree(struct sk_buff *skb) 2158 { 2159 sock_put(skb->sk); 2160 } 2161 EXPORT_SYMBOL(sock_efree); 2162 2163 /* Buffer destructor for prefetch/receive path where reference count may 2164 * not be held, e.g. for listen sockets. 2165 */ 2166 #ifdef CONFIG_INET 2167 void sock_pfree(struct sk_buff *skb) 2168 { 2169 if (sk_is_refcounted(skb->sk)) 2170 sock_gen_put(skb->sk); 2171 } 2172 EXPORT_SYMBOL(sock_pfree); 2173 #endif /* CONFIG_INET */ 2174 2175 kuid_t sock_i_uid(struct sock *sk) 2176 { 2177 kuid_t uid; 2178 2179 read_lock_bh(&sk->sk_callback_lock); 2180 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2181 read_unlock_bh(&sk->sk_callback_lock); 2182 return uid; 2183 } 2184 EXPORT_SYMBOL(sock_i_uid); 2185 2186 unsigned long sock_i_ino(struct sock *sk) 2187 { 2188 unsigned long ino; 2189 2190 read_lock_bh(&sk->sk_callback_lock); 2191 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2192 read_unlock_bh(&sk->sk_callback_lock); 2193 return ino; 2194 } 2195 EXPORT_SYMBOL(sock_i_ino); 2196 2197 /* 2198 * Allocate a skb from the socket's send buffer. 2199 */ 2200 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2201 gfp_t priority) 2202 { 2203 if (force || 2204 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2205 struct sk_buff *skb = alloc_skb(size, priority); 2206 2207 if (skb) { 2208 skb_set_owner_w(skb, sk); 2209 return skb; 2210 } 2211 } 2212 return NULL; 2213 } 2214 EXPORT_SYMBOL(sock_wmalloc); 2215 2216 static void sock_ofree(struct sk_buff *skb) 2217 { 2218 struct sock *sk = skb->sk; 2219 2220 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2221 } 2222 2223 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2224 gfp_t priority) 2225 { 2226 struct sk_buff *skb; 2227 2228 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2229 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2230 sysctl_optmem_max) 2231 return NULL; 2232 2233 skb = alloc_skb(size, priority); 2234 if (!skb) 2235 return NULL; 2236 2237 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2238 skb->sk = sk; 2239 skb->destructor = sock_ofree; 2240 return skb; 2241 } 2242 2243 /* 2244 * Allocate a memory block from the socket's option memory buffer. 2245 */ 2246 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2247 { 2248 if ((unsigned int)size <= sysctl_optmem_max && 2249 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2250 void *mem; 2251 /* First do the add, to avoid the race if kmalloc 2252 * might sleep. 2253 */ 2254 atomic_add(size, &sk->sk_omem_alloc); 2255 mem = kmalloc(size, priority); 2256 if (mem) 2257 return mem; 2258 atomic_sub(size, &sk->sk_omem_alloc); 2259 } 2260 return NULL; 2261 } 2262 EXPORT_SYMBOL(sock_kmalloc); 2263 2264 /* Free an option memory block. Note, we actually want the inline 2265 * here as this allows gcc to detect the nullify and fold away the 2266 * condition entirely. 2267 */ 2268 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2269 const bool nullify) 2270 { 2271 if (WARN_ON_ONCE(!mem)) 2272 return; 2273 if (nullify) 2274 kzfree(mem); 2275 else 2276 kfree(mem); 2277 atomic_sub(size, &sk->sk_omem_alloc); 2278 } 2279 2280 void sock_kfree_s(struct sock *sk, void *mem, int size) 2281 { 2282 __sock_kfree_s(sk, mem, size, false); 2283 } 2284 EXPORT_SYMBOL(sock_kfree_s); 2285 2286 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2287 { 2288 __sock_kfree_s(sk, mem, size, true); 2289 } 2290 EXPORT_SYMBOL(sock_kzfree_s); 2291 2292 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2293 I think, these locks should be removed for datagram sockets. 2294 */ 2295 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2296 { 2297 DEFINE_WAIT(wait); 2298 2299 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2300 for (;;) { 2301 if (!timeo) 2302 break; 2303 if (signal_pending(current)) 2304 break; 2305 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2306 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2307 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2308 break; 2309 if (sk->sk_shutdown & SEND_SHUTDOWN) 2310 break; 2311 if (sk->sk_err) 2312 break; 2313 timeo = schedule_timeout(timeo); 2314 } 2315 finish_wait(sk_sleep(sk), &wait); 2316 return timeo; 2317 } 2318 2319 2320 /* 2321 * Generic send/receive buffer handlers 2322 */ 2323 2324 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2325 unsigned long data_len, int noblock, 2326 int *errcode, int max_page_order) 2327 { 2328 struct sk_buff *skb; 2329 long timeo; 2330 int err; 2331 2332 timeo = sock_sndtimeo(sk, noblock); 2333 for (;;) { 2334 err = sock_error(sk); 2335 if (err != 0) 2336 goto failure; 2337 2338 err = -EPIPE; 2339 if (sk->sk_shutdown & SEND_SHUTDOWN) 2340 goto failure; 2341 2342 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2343 break; 2344 2345 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2346 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2347 err = -EAGAIN; 2348 if (!timeo) 2349 goto failure; 2350 if (signal_pending(current)) 2351 goto interrupted; 2352 timeo = sock_wait_for_wmem(sk, timeo); 2353 } 2354 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2355 errcode, sk->sk_allocation); 2356 if (skb) 2357 skb_set_owner_w(skb, sk); 2358 return skb; 2359 2360 interrupted: 2361 err = sock_intr_errno(timeo); 2362 failure: 2363 *errcode = err; 2364 return NULL; 2365 } 2366 EXPORT_SYMBOL(sock_alloc_send_pskb); 2367 2368 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2369 int noblock, int *errcode) 2370 { 2371 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2372 } 2373 EXPORT_SYMBOL(sock_alloc_send_skb); 2374 2375 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2376 struct sockcm_cookie *sockc) 2377 { 2378 u32 tsflags; 2379 2380 switch (cmsg->cmsg_type) { 2381 case SO_MARK: 2382 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2383 return -EPERM; 2384 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2385 return -EINVAL; 2386 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2387 break; 2388 case SO_TIMESTAMPING_OLD: 2389 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2390 return -EINVAL; 2391 2392 tsflags = *(u32 *)CMSG_DATA(cmsg); 2393 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2394 return -EINVAL; 2395 2396 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2397 sockc->tsflags |= tsflags; 2398 break; 2399 case SCM_TXTIME: 2400 if (!sock_flag(sk, SOCK_TXTIME)) 2401 return -EINVAL; 2402 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2403 return -EINVAL; 2404 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2405 break; 2406 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2407 case SCM_RIGHTS: 2408 case SCM_CREDENTIALS: 2409 break; 2410 default: 2411 return -EINVAL; 2412 } 2413 return 0; 2414 } 2415 EXPORT_SYMBOL(__sock_cmsg_send); 2416 2417 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2418 struct sockcm_cookie *sockc) 2419 { 2420 struct cmsghdr *cmsg; 2421 int ret; 2422 2423 for_each_cmsghdr(cmsg, msg) { 2424 if (!CMSG_OK(msg, cmsg)) 2425 return -EINVAL; 2426 if (cmsg->cmsg_level != SOL_SOCKET) 2427 continue; 2428 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2429 if (ret) 2430 return ret; 2431 } 2432 return 0; 2433 } 2434 EXPORT_SYMBOL(sock_cmsg_send); 2435 2436 static void sk_enter_memory_pressure(struct sock *sk) 2437 { 2438 if (!sk->sk_prot->enter_memory_pressure) 2439 return; 2440 2441 sk->sk_prot->enter_memory_pressure(sk); 2442 } 2443 2444 static void sk_leave_memory_pressure(struct sock *sk) 2445 { 2446 if (sk->sk_prot->leave_memory_pressure) { 2447 sk->sk_prot->leave_memory_pressure(sk); 2448 } else { 2449 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2450 2451 if (memory_pressure && READ_ONCE(*memory_pressure)) 2452 WRITE_ONCE(*memory_pressure, 0); 2453 } 2454 } 2455 2456 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2457 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2458 2459 /** 2460 * skb_page_frag_refill - check that a page_frag contains enough room 2461 * @sz: minimum size of the fragment we want to get 2462 * @pfrag: pointer to page_frag 2463 * @gfp: priority for memory allocation 2464 * 2465 * Note: While this allocator tries to use high order pages, there is 2466 * no guarantee that allocations succeed. Therefore, @sz MUST be 2467 * less or equal than PAGE_SIZE. 2468 */ 2469 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2470 { 2471 if (pfrag->page) { 2472 if (page_ref_count(pfrag->page) == 1) { 2473 pfrag->offset = 0; 2474 return true; 2475 } 2476 if (pfrag->offset + sz <= pfrag->size) 2477 return true; 2478 put_page(pfrag->page); 2479 } 2480 2481 pfrag->offset = 0; 2482 if (SKB_FRAG_PAGE_ORDER && 2483 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2484 /* Avoid direct reclaim but allow kswapd to wake */ 2485 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2486 __GFP_COMP | __GFP_NOWARN | 2487 __GFP_NORETRY, 2488 SKB_FRAG_PAGE_ORDER); 2489 if (likely(pfrag->page)) { 2490 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2491 return true; 2492 } 2493 } 2494 pfrag->page = alloc_page(gfp); 2495 if (likely(pfrag->page)) { 2496 pfrag->size = PAGE_SIZE; 2497 return true; 2498 } 2499 return false; 2500 } 2501 EXPORT_SYMBOL(skb_page_frag_refill); 2502 2503 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2504 { 2505 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2506 return true; 2507 2508 sk_enter_memory_pressure(sk); 2509 sk_stream_moderate_sndbuf(sk); 2510 return false; 2511 } 2512 EXPORT_SYMBOL(sk_page_frag_refill); 2513 2514 static void __lock_sock(struct sock *sk) 2515 __releases(&sk->sk_lock.slock) 2516 __acquires(&sk->sk_lock.slock) 2517 { 2518 DEFINE_WAIT(wait); 2519 2520 for (;;) { 2521 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2522 TASK_UNINTERRUPTIBLE); 2523 spin_unlock_bh(&sk->sk_lock.slock); 2524 schedule(); 2525 spin_lock_bh(&sk->sk_lock.slock); 2526 if (!sock_owned_by_user(sk)) 2527 break; 2528 } 2529 finish_wait(&sk->sk_lock.wq, &wait); 2530 } 2531 2532 void __release_sock(struct sock *sk) 2533 __releases(&sk->sk_lock.slock) 2534 __acquires(&sk->sk_lock.slock) 2535 { 2536 struct sk_buff *skb, *next; 2537 2538 while ((skb = sk->sk_backlog.head) != NULL) { 2539 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2540 2541 spin_unlock_bh(&sk->sk_lock.slock); 2542 2543 do { 2544 next = skb->next; 2545 prefetch(next); 2546 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2547 skb_mark_not_on_list(skb); 2548 sk_backlog_rcv(sk, skb); 2549 2550 cond_resched(); 2551 2552 skb = next; 2553 } while (skb != NULL); 2554 2555 spin_lock_bh(&sk->sk_lock.slock); 2556 } 2557 2558 /* 2559 * Doing the zeroing here guarantee we can not loop forever 2560 * while a wild producer attempts to flood us. 2561 */ 2562 sk->sk_backlog.len = 0; 2563 } 2564 2565 void __sk_flush_backlog(struct sock *sk) 2566 { 2567 spin_lock_bh(&sk->sk_lock.slock); 2568 __release_sock(sk); 2569 spin_unlock_bh(&sk->sk_lock.slock); 2570 } 2571 2572 /** 2573 * sk_wait_data - wait for data to arrive at sk_receive_queue 2574 * @sk: sock to wait on 2575 * @timeo: for how long 2576 * @skb: last skb seen on sk_receive_queue 2577 * 2578 * Now socket state including sk->sk_err is changed only under lock, 2579 * hence we may omit checks after joining wait queue. 2580 * We check receive queue before schedule() only as optimization; 2581 * it is very likely that release_sock() added new data. 2582 */ 2583 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2584 { 2585 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2586 int rc; 2587 2588 add_wait_queue(sk_sleep(sk), &wait); 2589 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2590 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2591 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2592 remove_wait_queue(sk_sleep(sk), &wait); 2593 return rc; 2594 } 2595 EXPORT_SYMBOL(sk_wait_data); 2596 2597 /** 2598 * __sk_mem_raise_allocated - increase memory_allocated 2599 * @sk: socket 2600 * @size: memory size to allocate 2601 * @amt: pages to allocate 2602 * @kind: allocation type 2603 * 2604 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2605 */ 2606 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2607 { 2608 struct proto *prot = sk->sk_prot; 2609 long allocated = sk_memory_allocated_add(sk, amt); 2610 bool charged = true; 2611 2612 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2613 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2614 goto suppress_allocation; 2615 2616 /* Under limit. */ 2617 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2618 sk_leave_memory_pressure(sk); 2619 return 1; 2620 } 2621 2622 /* Under pressure. */ 2623 if (allocated > sk_prot_mem_limits(sk, 1)) 2624 sk_enter_memory_pressure(sk); 2625 2626 /* Over hard limit. */ 2627 if (allocated > sk_prot_mem_limits(sk, 2)) 2628 goto suppress_allocation; 2629 2630 /* guarantee minimum buffer size under pressure */ 2631 if (kind == SK_MEM_RECV) { 2632 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2633 return 1; 2634 2635 } else { /* SK_MEM_SEND */ 2636 int wmem0 = sk_get_wmem0(sk, prot); 2637 2638 if (sk->sk_type == SOCK_STREAM) { 2639 if (sk->sk_wmem_queued < wmem0) 2640 return 1; 2641 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2642 return 1; 2643 } 2644 } 2645 2646 if (sk_has_memory_pressure(sk)) { 2647 u64 alloc; 2648 2649 if (!sk_under_memory_pressure(sk)) 2650 return 1; 2651 alloc = sk_sockets_allocated_read_positive(sk); 2652 if (sk_prot_mem_limits(sk, 2) > alloc * 2653 sk_mem_pages(sk->sk_wmem_queued + 2654 atomic_read(&sk->sk_rmem_alloc) + 2655 sk->sk_forward_alloc)) 2656 return 1; 2657 } 2658 2659 suppress_allocation: 2660 2661 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2662 sk_stream_moderate_sndbuf(sk); 2663 2664 /* Fail only if socket is _under_ its sndbuf. 2665 * In this case we cannot block, so that we have to fail. 2666 */ 2667 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2668 return 1; 2669 } 2670 2671 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2672 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2673 2674 sk_memory_allocated_sub(sk, amt); 2675 2676 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2677 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2678 2679 return 0; 2680 } 2681 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2682 2683 /** 2684 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2685 * @sk: socket 2686 * @size: memory size to allocate 2687 * @kind: allocation type 2688 * 2689 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2690 * rmem allocation. This function assumes that protocols which have 2691 * memory_pressure use sk_wmem_queued as write buffer accounting. 2692 */ 2693 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2694 { 2695 int ret, amt = sk_mem_pages(size); 2696 2697 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2698 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2699 if (!ret) 2700 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2701 return ret; 2702 } 2703 EXPORT_SYMBOL(__sk_mem_schedule); 2704 2705 /** 2706 * __sk_mem_reduce_allocated - reclaim memory_allocated 2707 * @sk: socket 2708 * @amount: number of quanta 2709 * 2710 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2711 */ 2712 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2713 { 2714 sk_memory_allocated_sub(sk, amount); 2715 2716 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2717 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2718 2719 if (sk_under_memory_pressure(sk) && 2720 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2721 sk_leave_memory_pressure(sk); 2722 } 2723 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2724 2725 /** 2726 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2727 * @sk: socket 2728 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2729 */ 2730 void __sk_mem_reclaim(struct sock *sk, int amount) 2731 { 2732 amount >>= SK_MEM_QUANTUM_SHIFT; 2733 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2734 __sk_mem_reduce_allocated(sk, amount); 2735 } 2736 EXPORT_SYMBOL(__sk_mem_reclaim); 2737 2738 int sk_set_peek_off(struct sock *sk, int val) 2739 { 2740 sk->sk_peek_off = val; 2741 return 0; 2742 } 2743 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2744 2745 /* 2746 * Set of default routines for initialising struct proto_ops when 2747 * the protocol does not support a particular function. In certain 2748 * cases where it makes no sense for a protocol to have a "do nothing" 2749 * function, some default processing is provided. 2750 */ 2751 2752 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2753 { 2754 return -EOPNOTSUPP; 2755 } 2756 EXPORT_SYMBOL(sock_no_bind); 2757 2758 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2759 int len, int flags) 2760 { 2761 return -EOPNOTSUPP; 2762 } 2763 EXPORT_SYMBOL(sock_no_connect); 2764 2765 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2766 { 2767 return -EOPNOTSUPP; 2768 } 2769 EXPORT_SYMBOL(sock_no_socketpair); 2770 2771 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2772 bool kern) 2773 { 2774 return -EOPNOTSUPP; 2775 } 2776 EXPORT_SYMBOL(sock_no_accept); 2777 2778 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2779 int peer) 2780 { 2781 return -EOPNOTSUPP; 2782 } 2783 EXPORT_SYMBOL(sock_no_getname); 2784 2785 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2786 { 2787 return -EOPNOTSUPP; 2788 } 2789 EXPORT_SYMBOL(sock_no_ioctl); 2790 2791 int sock_no_listen(struct socket *sock, int backlog) 2792 { 2793 return -EOPNOTSUPP; 2794 } 2795 EXPORT_SYMBOL(sock_no_listen); 2796 2797 int sock_no_shutdown(struct socket *sock, int how) 2798 { 2799 return -EOPNOTSUPP; 2800 } 2801 EXPORT_SYMBOL(sock_no_shutdown); 2802 2803 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2804 char __user *optval, unsigned int optlen) 2805 { 2806 return -EOPNOTSUPP; 2807 } 2808 EXPORT_SYMBOL(sock_no_setsockopt); 2809 2810 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2811 char __user *optval, int __user *optlen) 2812 { 2813 return -EOPNOTSUPP; 2814 } 2815 EXPORT_SYMBOL(sock_no_getsockopt); 2816 2817 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2818 { 2819 return -EOPNOTSUPP; 2820 } 2821 EXPORT_SYMBOL(sock_no_sendmsg); 2822 2823 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2824 { 2825 return -EOPNOTSUPP; 2826 } 2827 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2828 2829 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2830 int flags) 2831 { 2832 return -EOPNOTSUPP; 2833 } 2834 EXPORT_SYMBOL(sock_no_recvmsg); 2835 2836 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2837 { 2838 /* Mirror missing mmap method error code */ 2839 return -ENODEV; 2840 } 2841 EXPORT_SYMBOL(sock_no_mmap); 2842 2843 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2844 { 2845 ssize_t res; 2846 struct msghdr msg = {.msg_flags = flags}; 2847 struct kvec iov; 2848 char *kaddr = kmap(page); 2849 iov.iov_base = kaddr + offset; 2850 iov.iov_len = size; 2851 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2852 kunmap(page); 2853 return res; 2854 } 2855 EXPORT_SYMBOL(sock_no_sendpage); 2856 2857 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2858 int offset, size_t size, int flags) 2859 { 2860 ssize_t res; 2861 struct msghdr msg = {.msg_flags = flags}; 2862 struct kvec iov; 2863 char *kaddr = kmap(page); 2864 2865 iov.iov_base = kaddr + offset; 2866 iov.iov_len = size; 2867 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2868 kunmap(page); 2869 return res; 2870 } 2871 EXPORT_SYMBOL(sock_no_sendpage_locked); 2872 2873 /* 2874 * Default Socket Callbacks 2875 */ 2876 2877 static void sock_def_wakeup(struct sock *sk) 2878 { 2879 struct socket_wq *wq; 2880 2881 rcu_read_lock(); 2882 wq = rcu_dereference(sk->sk_wq); 2883 if (skwq_has_sleeper(wq)) 2884 wake_up_interruptible_all(&wq->wait); 2885 rcu_read_unlock(); 2886 } 2887 2888 static void sock_def_error_report(struct sock *sk) 2889 { 2890 struct socket_wq *wq; 2891 2892 rcu_read_lock(); 2893 wq = rcu_dereference(sk->sk_wq); 2894 if (skwq_has_sleeper(wq)) 2895 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 2896 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2897 rcu_read_unlock(); 2898 } 2899 2900 void sock_def_readable(struct sock *sk) 2901 { 2902 struct socket_wq *wq; 2903 2904 rcu_read_lock(); 2905 wq = rcu_dereference(sk->sk_wq); 2906 if (skwq_has_sleeper(wq)) 2907 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 2908 EPOLLRDNORM | EPOLLRDBAND); 2909 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2910 rcu_read_unlock(); 2911 } 2912 2913 static void sock_def_write_space(struct sock *sk) 2914 { 2915 struct socket_wq *wq; 2916 2917 rcu_read_lock(); 2918 2919 /* Do not wake up a writer until he can make "significant" 2920 * progress. --DaveM 2921 */ 2922 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 2923 wq = rcu_dereference(sk->sk_wq); 2924 if (skwq_has_sleeper(wq)) 2925 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 2926 EPOLLWRNORM | EPOLLWRBAND); 2927 2928 /* Should agree with poll, otherwise some programs break */ 2929 if (sock_writeable(sk)) 2930 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2931 } 2932 2933 rcu_read_unlock(); 2934 } 2935 2936 static void sock_def_destruct(struct sock *sk) 2937 { 2938 } 2939 2940 void sk_send_sigurg(struct sock *sk) 2941 { 2942 if (sk->sk_socket && sk->sk_socket->file) 2943 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2944 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2945 } 2946 EXPORT_SYMBOL(sk_send_sigurg); 2947 2948 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2949 unsigned long expires) 2950 { 2951 if (!mod_timer(timer, expires)) 2952 sock_hold(sk); 2953 } 2954 EXPORT_SYMBOL(sk_reset_timer); 2955 2956 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2957 { 2958 if (del_timer(timer)) 2959 __sock_put(sk); 2960 } 2961 EXPORT_SYMBOL(sk_stop_timer); 2962 2963 void sock_init_data(struct socket *sock, struct sock *sk) 2964 { 2965 sk_init_common(sk); 2966 sk->sk_send_head = NULL; 2967 2968 timer_setup(&sk->sk_timer, NULL, 0); 2969 2970 sk->sk_allocation = GFP_KERNEL; 2971 sk->sk_rcvbuf = sysctl_rmem_default; 2972 sk->sk_sndbuf = sysctl_wmem_default; 2973 sk->sk_state = TCP_CLOSE; 2974 sk_set_socket(sk, sock); 2975 2976 sock_set_flag(sk, SOCK_ZAPPED); 2977 2978 if (sock) { 2979 sk->sk_type = sock->type; 2980 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 2981 sock->sk = sk; 2982 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2983 } else { 2984 RCU_INIT_POINTER(sk->sk_wq, NULL); 2985 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2986 } 2987 2988 rwlock_init(&sk->sk_callback_lock); 2989 if (sk->sk_kern_sock) 2990 lockdep_set_class_and_name( 2991 &sk->sk_callback_lock, 2992 af_kern_callback_keys + sk->sk_family, 2993 af_family_kern_clock_key_strings[sk->sk_family]); 2994 else 2995 lockdep_set_class_and_name( 2996 &sk->sk_callback_lock, 2997 af_callback_keys + sk->sk_family, 2998 af_family_clock_key_strings[sk->sk_family]); 2999 3000 sk->sk_state_change = sock_def_wakeup; 3001 sk->sk_data_ready = sock_def_readable; 3002 sk->sk_write_space = sock_def_write_space; 3003 sk->sk_error_report = sock_def_error_report; 3004 sk->sk_destruct = sock_def_destruct; 3005 3006 sk->sk_frag.page = NULL; 3007 sk->sk_frag.offset = 0; 3008 sk->sk_peek_off = -1; 3009 3010 sk->sk_peer_pid = NULL; 3011 sk->sk_peer_cred = NULL; 3012 sk->sk_write_pending = 0; 3013 sk->sk_rcvlowat = 1; 3014 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 3015 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 3016 3017 sk->sk_stamp = SK_DEFAULT_STAMP; 3018 #if BITS_PER_LONG==32 3019 seqlock_init(&sk->sk_stamp_seq); 3020 #endif 3021 atomic_set(&sk->sk_zckey, 0); 3022 3023 #ifdef CONFIG_NET_RX_BUSY_POLL 3024 sk->sk_napi_id = 0; 3025 sk->sk_ll_usec = sysctl_net_busy_read; 3026 #endif 3027 3028 sk->sk_max_pacing_rate = ~0UL; 3029 sk->sk_pacing_rate = ~0UL; 3030 WRITE_ONCE(sk->sk_pacing_shift, 10); 3031 sk->sk_incoming_cpu = -1; 3032 3033 sk_rx_queue_clear(sk); 3034 /* 3035 * Before updating sk_refcnt, we must commit prior changes to memory 3036 * (Documentation/RCU/rculist_nulls.txt for details) 3037 */ 3038 smp_wmb(); 3039 refcount_set(&sk->sk_refcnt, 1); 3040 atomic_set(&sk->sk_drops, 0); 3041 } 3042 EXPORT_SYMBOL(sock_init_data); 3043 3044 void lock_sock_nested(struct sock *sk, int subclass) 3045 { 3046 might_sleep(); 3047 spin_lock_bh(&sk->sk_lock.slock); 3048 if (sk->sk_lock.owned) 3049 __lock_sock(sk); 3050 sk->sk_lock.owned = 1; 3051 spin_unlock(&sk->sk_lock.slock); 3052 /* 3053 * The sk_lock has mutex_lock() semantics here: 3054 */ 3055 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 3056 local_bh_enable(); 3057 } 3058 EXPORT_SYMBOL(lock_sock_nested); 3059 3060 void release_sock(struct sock *sk) 3061 { 3062 spin_lock_bh(&sk->sk_lock.slock); 3063 if (sk->sk_backlog.tail) 3064 __release_sock(sk); 3065 3066 /* Warning : release_cb() might need to release sk ownership, 3067 * ie call sock_release_ownership(sk) before us. 3068 */ 3069 if (sk->sk_prot->release_cb) 3070 sk->sk_prot->release_cb(sk); 3071 3072 sock_release_ownership(sk); 3073 if (waitqueue_active(&sk->sk_lock.wq)) 3074 wake_up(&sk->sk_lock.wq); 3075 spin_unlock_bh(&sk->sk_lock.slock); 3076 } 3077 EXPORT_SYMBOL(release_sock); 3078 3079 /** 3080 * lock_sock_fast - fast version of lock_sock 3081 * @sk: socket 3082 * 3083 * This version should be used for very small section, where process wont block 3084 * return false if fast path is taken: 3085 * 3086 * sk_lock.slock locked, owned = 0, BH disabled 3087 * 3088 * return true if slow path is taken: 3089 * 3090 * sk_lock.slock unlocked, owned = 1, BH enabled 3091 */ 3092 bool lock_sock_fast(struct sock *sk) 3093 { 3094 might_sleep(); 3095 spin_lock_bh(&sk->sk_lock.slock); 3096 3097 if (!sk->sk_lock.owned) 3098 /* 3099 * Note : We must disable BH 3100 */ 3101 return false; 3102 3103 __lock_sock(sk); 3104 sk->sk_lock.owned = 1; 3105 spin_unlock(&sk->sk_lock.slock); 3106 /* 3107 * The sk_lock has mutex_lock() semantics here: 3108 */ 3109 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 3110 local_bh_enable(); 3111 return true; 3112 } 3113 EXPORT_SYMBOL(lock_sock_fast); 3114 3115 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3116 bool timeval, bool time32) 3117 { 3118 struct sock *sk = sock->sk; 3119 struct timespec64 ts; 3120 3121 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3122 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3123 if (ts.tv_sec == -1) 3124 return -ENOENT; 3125 if (ts.tv_sec == 0) { 3126 ktime_t kt = ktime_get_real(); 3127 sock_write_timestamp(sk, kt); 3128 ts = ktime_to_timespec64(kt); 3129 } 3130 3131 if (timeval) 3132 ts.tv_nsec /= 1000; 3133 3134 #ifdef CONFIG_COMPAT_32BIT_TIME 3135 if (time32) 3136 return put_old_timespec32(&ts, userstamp); 3137 #endif 3138 #ifdef CONFIG_SPARC64 3139 /* beware of padding in sparc64 timeval */ 3140 if (timeval && !in_compat_syscall()) { 3141 struct __kernel_old_timeval __user tv = { 3142 .tv_sec = ts.tv_sec, 3143 .tv_usec = ts.tv_nsec, 3144 }; 3145 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3146 return -EFAULT; 3147 return 0; 3148 } 3149 #endif 3150 return put_timespec64(&ts, userstamp); 3151 } 3152 EXPORT_SYMBOL(sock_gettstamp); 3153 3154 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3155 { 3156 if (!sock_flag(sk, flag)) { 3157 unsigned long previous_flags = sk->sk_flags; 3158 3159 sock_set_flag(sk, flag); 3160 /* 3161 * we just set one of the two flags which require net 3162 * time stamping, but time stamping might have been on 3163 * already because of the other one 3164 */ 3165 if (sock_needs_netstamp(sk) && 3166 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3167 net_enable_timestamp(); 3168 } 3169 } 3170 3171 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3172 int level, int type) 3173 { 3174 struct sock_exterr_skb *serr; 3175 struct sk_buff *skb; 3176 int copied, err; 3177 3178 err = -EAGAIN; 3179 skb = sock_dequeue_err_skb(sk); 3180 if (skb == NULL) 3181 goto out; 3182 3183 copied = skb->len; 3184 if (copied > len) { 3185 msg->msg_flags |= MSG_TRUNC; 3186 copied = len; 3187 } 3188 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3189 if (err) 3190 goto out_free_skb; 3191 3192 sock_recv_timestamp(msg, sk, skb); 3193 3194 serr = SKB_EXT_ERR(skb); 3195 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3196 3197 msg->msg_flags |= MSG_ERRQUEUE; 3198 err = copied; 3199 3200 out_free_skb: 3201 kfree_skb(skb); 3202 out: 3203 return err; 3204 } 3205 EXPORT_SYMBOL(sock_recv_errqueue); 3206 3207 /* 3208 * Get a socket option on an socket. 3209 * 3210 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3211 * asynchronous errors should be reported by getsockopt. We assume 3212 * this means if you specify SO_ERROR (otherwise whats the point of it). 3213 */ 3214 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3215 char __user *optval, int __user *optlen) 3216 { 3217 struct sock *sk = sock->sk; 3218 3219 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3220 } 3221 EXPORT_SYMBOL(sock_common_getsockopt); 3222 3223 #ifdef CONFIG_COMPAT 3224 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 3225 char __user *optval, int __user *optlen) 3226 { 3227 struct sock *sk = sock->sk; 3228 3229 if (sk->sk_prot->compat_getsockopt != NULL) 3230 return sk->sk_prot->compat_getsockopt(sk, level, optname, 3231 optval, optlen); 3232 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3233 } 3234 EXPORT_SYMBOL(compat_sock_common_getsockopt); 3235 #endif 3236 3237 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3238 int flags) 3239 { 3240 struct sock *sk = sock->sk; 3241 int addr_len = 0; 3242 int err; 3243 3244 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3245 flags & ~MSG_DONTWAIT, &addr_len); 3246 if (err >= 0) 3247 msg->msg_namelen = addr_len; 3248 return err; 3249 } 3250 EXPORT_SYMBOL(sock_common_recvmsg); 3251 3252 /* 3253 * Set socket options on an inet socket. 3254 */ 3255 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3256 char __user *optval, unsigned int optlen) 3257 { 3258 struct sock *sk = sock->sk; 3259 3260 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3261 } 3262 EXPORT_SYMBOL(sock_common_setsockopt); 3263 3264 #ifdef CONFIG_COMPAT 3265 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 3266 char __user *optval, unsigned int optlen) 3267 { 3268 struct sock *sk = sock->sk; 3269 3270 if (sk->sk_prot->compat_setsockopt != NULL) 3271 return sk->sk_prot->compat_setsockopt(sk, level, optname, 3272 optval, optlen); 3273 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3274 } 3275 EXPORT_SYMBOL(compat_sock_common_setsockopt); 3276 #endif 3277 3278 void sk_common_release(struct sock *sk) 3279 { 3280 if (sk->sk_prot->destroy) 3281 sk->sk_prot->destroy(sk); 3282 3283 /* 3284 * Observation: when sock_common_release is called, processes have 3285 * no access to socket. But net still has. 3286 * Step one, detach it from networking: 3287 * 3288 * A. Remove from hash tables. 3289 */ 3290 3291 sk->sk_prot->unhash(sk); 3292 3293 /* 3294 * In this point socket cannot receive new packets, but it is possible 3295 * that some packets are in flight because some CPU runs receiver and 3296 * did hash table lookup before we unhashed socket. They will achieve 3297 * receive queue and will be purged by socket destructor. 3298 * 3299 * Also we still have packets pending on receive queue and probably, 3300 * our own packets waiting in device queues. sock_destroy will drain 3301 * receive queue, but transmitted packets will delay socket destruction 3302 * until the last reference will be released. 3303 */ 3304 3305 sock_orphan(sk); 3306 3307 xfrm_sk_free_policy(sk); 3308 3309 sk_refcnt_debug_release(sk); 3310 3311 sock_put(sk); 3312 } 3313 EXPORT_SYMBOL(sk_common_release); 3314 3315 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3316 { 3317 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3318 3319 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3320 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3321 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3322 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3323 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3324 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3325 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3326 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3327 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3328 } 3329 3330 #ifdef CONFIG_PROC_FS 3331 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3332 struct prot_inuse { 3333 int val[PROTO_INUSE_NR]; 3334 }; 3335 3336 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3337 3338 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3339 { 3340 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3341 } 3342 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3343 3344 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3345 { 3346 int cpu, idx = prot->inuse_idx; 3347 int res = 0; 3348 3349 for_each_possible_cpu(cpu) 3350 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3351 3352 return res >= 0 ? res : 0; 3353 } 3354 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3355 3356 static void sock_inuse_add(struct net *net, int val) 3357 { 3358 this_cpu_add(*net->core.sock_inuse, val); 3359 } 3360 3361 int sock_inuse_get(struct net *net) 3362 { 3363 int cpu, res = 0; 3364 3365 for_each_possible_cpu(cpu) 3366 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3367 3368 return res; 3369 } 3370 3371 EXPORT_SYMBOL_GPL(sock_inuse_get); 3372 3373 static int __net_init sock_inuse_init_net(struct net *net) 3374 { 3375 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3376 if (net->core.prot_inuse == NULL) 3377 return -ENOMEM; 3378 3379 net->core.sock_inuse = alloc_percpu(int); 3380 if (net->core.sock_inuse == NULL) 3381 goto out; 3382 3383 return 0; 3384 3385 out: 3386 free_percpu(net->core.prot_inuse); 3387 return -ENOMEM; 3388 } 3389 3390 static void __net_exit sock_inuse_exit_net(struct net *net) 3391 { 3392 free_percpu(net->core.prot_inuse); 3393 free_percpu(net->core.sock_inuse); 3394 } 3395 3396 static struct pernet_operations net_inuse_ops = { 3397 .init = sock_inuse_init_net, 3398 .exit = sock_inuse_exit_net, 3399 }; 3400 3401 static __init int net_inuse_init(void) 3402 { 3403 if (register_pernet_subsys(&net_inuse_ops)) 3404 panic("Cannot initialize net inuse counters"); 3405 3406 return 0; 3407 } 3408 3409 core_initcall(net_inuse_init); 3410 3411 static int assign_proto_idx(struct proto *prot) 3412 { 3413 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3414 3415 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3416 pr_err("PROTO_INUSE_NR exhausted\n"); 3417 return -ENOSPC; 3418 } 3419 3420 set_bit(prot->inuse_idx, proto_inuse_idx); 3421 return 0; 3422 } 3423 3424 static void release_proto_idx(struct proto *prot) 3425 { 3426 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3427 clear_bit(prot->inuse_idx, proto_inuse_idx); 3428 } 3429 #else 3430 static inline int assign_proto_idx(struct proto *prot) 3431 { 3432 return 0; 3433 } 3434 3435 static inline void release_proto_idx(struct proto *prot) 3436 { 3437 } 3438 3439 static void sock_inuse_add(struct net *net, int val) 3440 { 3441 } 3442 #endif 3443 3444 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3445 { 3446 if (!rsk_prot) 3447 return; 3448 kfree(rsk_prot->slab_name); 3449 rsk_prot->slab_name = NULL; 3450 kmem_cache_destroy(rsk_prot->slab); 3451 rsk_prot->slab = NULL; 3452 } 3453 3454 static int req_prot_init(const struct proto *prot) 3455 { 3456 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3457 3458 if (!rsk_prot) 3459 return 0; 3460 3461 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3462 prot->name); 3463 if (!rsk_prot->slab_name) 3464 return -ENOMEM; 3465 3466 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3467 rsk_prot->obj_size, 0, 3468 SLAB_ACCOUNT | prot->slab_flags, 3469 NULL); 3470 3471 if (!rsk_prot->slab) { 3472 pr_crit("%s: Can't create request sock SLAB cache!\n", 3473 prot->name); 3474 return -ENOMEM; 3475 } 3476 return 0; 3477 } 3478 3479 int proto_register(struct proto *prot, int alloc_slab) 3480 { 3481 int ret = -ENOBUFS; 3482 3483 if (alloc_slab) { 3484 prot->slab = kmem_cache_create_usercopy(prot->name, 3485 prot->obj_size, 0, 3486 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3487 prot->slab_flags, 3488 prot->useroffset, prot->usersize, 3489 NULL); 3490 3491 if (prot->slab == NULL) { 3492 pr_crit("%s: Can't create sock SLAB cache!\n", 3493 prot->name); 3494 goto out; 3495 } 3496 3497 if (req_prot_init(prot)) 3498 goto out_free_request_sock_slab; 3499 3500 if (prot->twsk_prot != NULL) { 3501 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3502 3503 if (prot->twsk_prot->twsk_slab_name == NULL) 3504 goto out_free_request_sock_slab; 3505 3506 prot->twsk_prot->twsk_slab = 3507 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3508 prot->twsk_prot->twsk_obj_size, 3509 0, 3510 SLAB_ACCOUNT | 3511 prot->slab_flags, 3512 NULL); 3513 if (prot->twsk_prot->twsk_slab == NULL) 3514 goto out_free_timewait_sock_slab_name; 3515 } 3516 } 3517 3518 mutex_lock(&proto_list_mutex); 3519 ret = assign_proto_idx(prot); 3520 if (ret) { 3521 mutex_unlock(&proto_list_mutex); 3522 goto out_free_timewait_sock_slab_name; 3523 } 3524 list_add(&prot->node, &proto_list); 3525 mutex_unlock(&proto_list_mutex); 3526 return ret; 3527 3528 out_free_timewait_sock_slab_name: 3529 if (alloc_slab && prot->twsk_prot) 3530 kfree(prot->twsk_prot->twsk_slab_name); 3531 out_free_request_sock_slab: 3532 if (alloc_slab) { 3533 req_prot_cleanup(prot->rsk_prot); 3534 3535 kmem_cache_destroy(prot->slab); 3536 prot->slab = NULL; 3537 } 3538 out: 3539 return ret; 3540 } 3541 EXPORT_SYMBOL(proto_register); 3542 3543 void proto_unregister(struct proto *prot) 3544 { 3545 mutex_lock(&proto_list_mutex); 3546 release_proto_idx(prot); 3547 list_del(&prot->node); 3548 mutex_unlock(&proto_list_mutex); 3549 3550 kmem_cache_destroy(prot->slab); 3551 prot->slab = NULL; 3552 3553 req_prot_cleanup(prot->rsk_prot); 3554 3555 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3556 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3557 kfree(prot->twsk_prot->twsk_slab_name); 3558 prot->twsk_prot->twsk_slab = NULL; 3559 } 3560 } 3561 EXPORT_SYMBOL(proto_unregister); 3562 3563 int sock_load_diag_module(int family, int protocol) 3564 { 3565 if (!protocol) { 3566 if (!sock_is_registered(family)) 3567 return -ENOENT; 3568 3569 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3570 NETLINK_SOCK_DIAG, family); 3571 } 3572 3573 #ifdef CONFIG_INET 3574 if (family == AF_INET && 3575 protocol != IPPROTO_RAW && 3576 !rcu_access_pointer(inet_protos[protocol])) 3577 return -ENOENT; 3578 #endif 3579 3580 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3581 NETLINK_SOCK_DIAG, family, protocol); 3582 } 3583 EXPORT_SYMBOL(sock_load_diag_module); 3584 3585 #ifdef CONFIG_PROC_FS 3586 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3587 __acquires(proto_list_mutex) 3588 { 3589 mutex_lock(&proto_list_mutex); 3590 return seq_list_start_head(&proto_list, *pos); 3591 } 3592 3593 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3594 { 3595 return seq_list_next(v, &proto_list, pos); 3596 } 3597 3598 static void proto_seq_stop(struct seq_file *seq, void *v) 3599 __releases(proto_list_mutex) 3600 { 3601 mutex_unlock(&proto_list_mutex); 3602 } 3603 3604 static char proto_method_implemented(const void *method) 3605 { 3606 return method == NULL ? 'n' : 'y'; 3607 } 3608 static long sock_prot_memory_allocated(struct proto *proto) 3609 { 3610 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3611 } 3612 3613 static const char *sock_prot_memory_pressure(struct proto *proto) 3614 { 3615 return proto->memory_pressure != NULL ? 3616 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3617 } 3618 3619 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3620 { 3621 3622 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3623 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3624 proto->name, 3625 proto->obj_size, 3626 sock_prot_inuse_get(seq_file_net(seq), proto), 3627 sock_prot_memory_allocated(proto), 3628 sock_prot_memory_pressure(proto), 3629 proto->max_header, 3630 proto->slab == NULL ? "no" : "yes", 3631 module_name(proto->owner), 3632 proto_method_implemented(proto->close), 3633 proto_method_implemented(proto->connect), 3634 proto_method_implemented(proto->disconnect), 3635 proto_method_implemented(proto->accept), 3636 proto_method_implemented(proto->ioctl), 3637 proto_method_implemented(proto->init), 3638 proto_method_implemented(proto->destroy), 3639 proto_method_implemented(proto->shutdown), 3640 proto_method_implemented(proto->setsockopt), 3641 proto_method_implemented(proto->getsockopt), 3642 proto_method_implemented(proto->sendmsg), 3643 proto_method_implemented(proto->recvmsg), 3644 proto_method_implemented(proto->sendpage), 3645 proto_method_implemented(proto->bind), 3646 proto_method_implemented(proto->backlog_rcv), 3647 proto_method_implemented(proto->hash), 3648 proto_method_implemented(proto->unhash), 3649 proto_method_implemented(proto->get_port), 3650 proto_method_implemented(proto->enter_memory_pressure)); 3651 } 3652 3653 static int proto_seq_show(struct seq_file *seq, void *v) 3654 { 3655 if (v == &proto_list) 3656 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3657 "protocol", 3658 "size", 3659 "sockets", 3660 "memory", 3661 "press", 3662 "maxhdr", 3663 "slab", 3664 "module", 3665 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3666 else 3667 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3668 return 0; 3669 } 3670 3671 static const struct seq_operations proto_seq_ops = { 3672 .start = proto_seq_start, 3673 .next = proto_seq_next, 3674 .stop = proto_seq_stop, 3675 .show = proto_seq_show, 3676 }; 3677 3678 static __net_init int proto_init_net(struct net *net) 3679 { 3680 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3681 sizeof(struct seq_net_private))) 3682 return -ENOMEM; 3683 3684 return 0; 3685 } 3686 3687 static __net_exit void proto_exit_net(struct net *net) 3688 { 3689 remove_proc_entry("protocols", net->proc_net); 3690 } 3691 3692 3693 static __net_initdata struct pernet_operations proto_net_ops = { 3694 .init = proto_init_net, 3695 .exit = proto_exit_net, 3696 }; 3697 3698 static int __init proto_init(void) 3699 { 3700 return register_pernet_subsys(&proto_net_ops); 3701 } 3702 3703 subsys_initcall(proto_init); 3704 3705 #endif /* PROC_FS */ 3706 3707 #ifdef CONFIG_NET_RX_BUSY_POLL 3708 bool sk_busy_loop_end(void *p, unsigned long start_time) 3709 { 3710 struct sock *sk = p; 3711 3712 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3713 sk_busy_loop_timeout(sk, start_time); 3714 } 3715 EXPORT_SYMBOL(sk_busy_loop_end); 3716 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3717 3718 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len) 3719 { 3720 if (!sk->sk_prot->bind_add) 3721 return -EOPNOTSUPP; 3722 return sk->sk_prot->bind_add(sk, addr, addr_len); 3723 } 3724 EXPORT_SYMBOL(sock_bind_add); 3725