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 int size; 337 338 if (timeo == MAX_SCHEDULE_TIMEOUT) { 339 tv.tv_sec = 0; 340 tv.tv_usec = 0; 341 } else { 342 tv.tv_sec = timeo / HZ; 343 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 344 } 345 346 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 347 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 348 *(struct old_timeval32 *)optval = tv32; 349 return sizeof(tv32); 350 } 351 352 if (old_timeval) { 353 struct __kernel_old_timeval old_tv; 354 old_tv.tv_sec = tv.tv_sec; 355 old_tv.tv_usec = tv.tv_usec; 356 *(struct __kernel_old_timeval *)optval = old_tv; 357 size = sizeof(old_tv); 358 } else { 359 *(struct __kernel_sock_timeval *)optval = tv; 360 size = sizeof(tv); 361 } 362 363 return size; 364 } 365 366 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval) 367 { 368 struct __kernel_sock_timeval tv; 369 370 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 371 struct old_timeval32 tv32; 372 373 if (optlen < sizeof(tv32)) 374 return -EINVAL; 375 376 if (copy_from_user(&tv32, optval, sizeof(tv32))) 377 return -EFAULT; 378 tv.tv_sec = tv32.tv_sec; 379 tv.tv_usec = tv32.tv_usec; 380 } else if (old_timeval) { 381 struct __kernel_old_timeval old_tv; 382 383 if (optlen < sizeof(old_tv)) 384 return -EINVAL; 385 if (copy_from_user(&old_tv, optval, sizeof(old_tv))) 386 return -EFAULT; 387 tv.tv_sec = old_tv.tv_sec; 388 tv.tv_usec = old_tv.tv_usec; 389 } else { 390 if (optlen < sizeof(tv)) 391 return -EINVAL; 392 if (copy_from_user(&tv, optval, sizeof(tv))) 393 return -EFAULT; 394 } 395 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 396 return -EDOM; 397 398 if (tv.tv_sec < 0) { 399 static int warned __read_mostly; 400 401 *timeo_p = 0; 402 if (warned < 10 && net_ratelimit()) { 403 warned++; 404 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 405 __func__, current->comm, task_pid_nr(current)); 406 } 407 return 0; 408 } 409 *timeo_p = MAX_SCHEDULE_TIMEOUT; 410 if (tv.tv_sec == 0 && tv.tv_usec == 0) 411 return 0; 412 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 413 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 414 return 0; 415 } 416 417 static void sock_warn_obsolete_bsdism(const char *name) 418 { 419 static int warned; 420 static char warncomm[TASK_COMM_LEN]; 421 if (strcmp(warncomm, current->comm) && warned < 5) { 422 strcpy(warncomm, current->comm); 423 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 424 warncomm, name); 425 warned++; 426 } 427 } 428 429 static bool sock_needs_netstamp(const struct sock *sk) 430 { 431 switch (sk->sk_family) { 432 case AF_UNSPEC: 433 case AF_UNIX: 434 return false; 435 default: 436 return true; 437 } 438 } 439 440 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 441 { 442 if (sk->sk_flags & flags) { 443 sk->sk_flags &= ~flags; 444 if (sock_needs_netstamp(sk) && 445 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 446 net_disable_timestamp(); 447 } 448 } 449 450 451 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 452 { 453 unsigned long flags; 454 struct sk_buff_head *list = &sk->sk_receive_queue; 455 456 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 457 atomic_inc(&sk->sk_drops); 458 trace_sock_rcvqueue_full(sk, skb); 459 return -ENOMEM; 460 } 461 462 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 463 atomic_inc(&sk->sk_drops); 464 return -ENOBUFS; 465 } 466 467 skb->dev = NULL; 468 skb_set_owner_r(skb, sk); 469 470 /* we escape from rcu protected region, make sure we dont leak 471 * a norefcounted dst 472 */ 473 skb_dst_force(skb); 474 475 spin_lock_irqsave(&list->lock, flags); 476 sock_skb_set_dropcount(sk, skb); 477 __skb_queue_tail(list, skb); 478 spin_unlock_irqrestore(&list->lock, flags); 479 480 if (!sock_flag(sk, SOCK_DEAD)) 481 sk->sk_data_ready(sk); 482 return 0; 483 } 484 EXPORT_SYMBOL(__sock_queue_rcv_skb); 485 486 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 487 { 488 int err; 489 490 err = sk_filter(sk, skb); 491 if (err) 492 return err; 493 494 return __sock_queue_rcv_skb(sk, skb); 495 } 496 EXPORT_SYMBOL(sock_queue_rcv_skb); 497 498 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 499 const int nested, unsigned int trim_cap, bool refcounted) 500 { 501 int rc = NET_RX_SUCCESS; 502 503 if (sk_filter_trim_cap(sk, skb, trim_cap)) 504 goto discard_and_relse; 505 506 skb->dev = NULL; 507 508 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 509 atomic_inc(&sk->sk_drops); 510 goto discard_and_relse; 511 } 512 if (nested) 513 bh_lock_sock_nested(sk); 514 else 515 bh_lock_sock(sk); 516 if (!sock_owned_by_user(sk)) { 517 /* 518 * trylock + unlock semantics: 519 */ 520 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 521 522 rc = sk_backlog_rcv(sk, skb); 523 524 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 525 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 526 bh_unlock_sock(sk); 527 atomic_inc(&sk->sk_drops); 528 goto discard_and_relse; 529 } 530 531 bh_unlock_sock(sk); 532 out: 533 if (refcounted) 534 sock_put(sk); 535 return rc; 536 discard_and_relse: 537 kfree_skb(skb); 538 goto out; 539 } 540 EXPORT_SYMBOL(__sk_receive_skb); 541 542 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 543 { 544 struct dst_entry *dst = __sk_dst_get(sk); 545 546 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 547 sk_tx_queue_clear(sk); 548 sk->sk_dst_pending_confirm = 0; 549 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 550 dst_release(dst); 551 return NULL; 552 } 553 554 return dst; 555 } 556 EXPORT_SYMBOL(__sk_dst_check); 557 558 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 559 { 560 struct dst_entry *dst = sk_dst_get(sk); 561 562 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 563 sk_dst_reset(sk); 564 dst_release(dst); 565 return NULL; 566 } 567 568 return dst; 569 } 570 EXPORT_SYMBOL(sk_dst_check); 571 572 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex) 573 { 574 int ret = -ENOPROTOOPT; 575 #ifdef CONFIG_NETDEVICES 576 struct net *net = sock_net(sk); 577 578 /* Sorry... */ 579 ret = -EPERM; 580 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 581 goto out; 582 583 ret = -EINVAL; 584 if (ifindex < 0) 585 goto out; 586 587 sk->sk_bound_dev_if = ifindex; 588 if (sk->sk_prot->rehash) 589 sk->sk_prot->rehash(sk); 590 sk_dst_reset(sk); 591 592 ret = 0; 593 594 out: 595 #endif 596 597 return ret; 598 } 599 600 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 601 int optlen) 602 { 603 int ret = -ENOPROTOOPT; 604 #ifdef CONFIG_NETDEVICES 605 struct net *net = sock_net(sk); 606 char devname[IFNAMSIZ]; 607 int index; 608 609 ret = -EINVAL; 610 if (optlen < 0) 611 goto out; 612 613 /* Bind this socket to a particular device like "eth0", 614 * as specified in the passed interface name. If the 615 * name is "" or the option length is zero the socket 616 * is not bound. 617 */ 618 if (optlen > IFNAMSIZ - 1) 619 optlen = IFNAMSIZ - 1; 620 memset(devname, 0, sizeof(devname)); 621 622 ret = -EFAULT; 623 if (copy_from_user(devname, optval, optlen)) 624 goto out; 625 626 index = 0; 627 if (devname[0] != '\0') { 628 struct net_device *dev; 629 630 rcu_read_lock(); 631 dev = dev_get_by_name_rcu(net, devname); 632 if (dev) 633 index = dev->ifindex; 634 rcu_read_unlock(); 635 ret = -ENODEV; 636 if (!dev) 637 goto out; 638 } 639 640 lock_sock(sk); 641 ret = sock_setbindtodevice_locked(sk, index); 642 release_sock(sk); 643 644 out: 645 #endif 646 647 return ret; 648 } 649 650 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 651 int __user *optlen, int len) 652 { 653 int ret = -ENOPROTOOPT; 654 #ifdef CONFIG_NETDEVICES 655 struct net *net = sock_net(sk); 656 char devname[IFNAMSIZ]; 657 658 if (sk->sk_bound_dev_if == 0) { 659 len = 0; 660 goto zero; 661 } 662 663 ret = -EINVAL; 664 if (len < IFNAMSIZ) 665 goto out; 666 667 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 668 if (ret) 669 goto out; 670 671 len = strlen(devname) + 1; 672 673 ret = -EFAULT; 674 if (copy_to_user(optval, devname, len)) 675 goto out; 676 677 zero: 678 ret = -EFAULT; 679 if (put_user(len, optlen)) 680 goto out; 681 682 ret = 0; 683 684 out: 685 #endif 686 687 return ret; 688 } 689 690 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 691 { 692 if (valbool) 693 sock_set_flag(sk, bit); 694 else 695 sock_reset_flag(sk, bit); 696 } 697 698 bool sk_mc_loop(struct sock *sk) 699 { 700 if (dev_recursion_level()) 701 return false; 702 if (!sk) 703 return true; 704 switch (sk->sk_family) { 705 case AF_INET: 706 return inet_sk(sk)->mc_loop; 707 #if IS_ENABLED(CONFIG_IPV6) 708 case AF_INET6: 709 return inet6_sk(sk)->mc_loop; 710 #endif 711 } 712 WARN_ON(1); 713 return true; 714 } 715 EXPORT_SYMBOL(sk_mc_loop); 716 717 /* 718 * This is meant for all protocols to use and covers goings on 719 * at the socket level. Everything here is generic. 720 */ 721 722 int sock_setsockopt(struct socket *sock, int level, int optname, 723 char __user *optval, unsigned int optlen) 724 { 725 struct sock_txtime sk_txtime; 726 struct sock *sk = sock->sk; 727 int val; 728 int valbool; 729 struct linger ling; 730 int ret = 0; 731 732 /* 733 * Options without arguments 734 */ 735 736 if (optname == SO_BINDTODEVICE) 737 return sock_setbindtodevice(sk, optval, optlen); 738 739 if (optlen < sizeof(int)) 740 return -EINVAL; 741 742 if (get_user(val, (int __user *)optval)) 743 return -EFAULT; 744 745 valbool = val ? 1 : 0; 746 747 lock_sock(sk); 748 749 switch (optname) { 750 case SO_DEBUG: 751 if (val && !capable(CAP_NET_ADMIN)) 752 ret = -EACCES; 753 else 754 sock_valbool_flag(sk, SOCK_DBG, valbool); 755 break; 756 case SO_REUSEADDR: 757 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 758 break; 759 case SO_REUSEPORT: 760 sk->sk_reuseport = valbool; 761 break; 762 case SO_TYPE: 763 case SO_PROTOCOL: 764 case SO_DOMAIN: 765 case SO_ERROR: 766 ret = -ENOPROTOOPT; 767 break; 768 case SO_DONTROUTE: 769 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 770 sk_dst_reset(sk); 771 break; 772 case SO_BROADCAST: 773 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 774 break; 775 case SO_SNDBUF: 776 /* Don't error on this BSD doesn't and if you think 777 * about it this is right. Otherwise apps have to 778 * play 'guess the biggest size' games. RCVBUF/SNDBUF 779 * are treated in BSD as hints 780 */ 781 val = min_t(u32, val, sysctl_wmem_max); 782 set_sndbuf: 783 /* Ensure val * 2 fits into an int, to prevent max_t() 784 * from treating it as a negative value. 785 */ 786 val = min_t(int, val, INT_MAX / 2); 787 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 788 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 789 /* Wake up sending tasks if we upped the value. */ 790 sk->sk_write_space(sk); 791 break; 792 793 case SO_SNDBUFFORCE: 794 if (!capable(CAP_NET_ADMIN)) { 795 ret = -EPERM; 796 break; 797 } 798 799 /* No negative values (to prevent underflow, as val will be 800 * multiplied by 2). 801 */ 802 if (val < 0) 803 val = 0; 804 goto set_sndbuf; 805 806 case SO_RCVBUF: 807 /* Don't error on this BSD doesn't and if you think 808 * about it this is right. Otherwise apps have to 809 * play 'guess the biggest size' games. RCVBUF/SNDBUF 810 * are treated in BSD as hints 811 */ 812 val = min_t(u32, val, sysctl_rmem_max); 813 set_rcvbuf: 814 /* Ensure val * 2 fits into an int, to prevent max_t() 815 * from treating it as a negative value. 816 */ 817 val = min_t(int, val, INT_MAX / 2); 818 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 819 /* 820 * We double it on the way in to account for 821 * "struct sk_buff" etc. overhead. Applications 822 * assume that the SO_RCVBUF setting they make will 823 * allow that much actual data to be received on that 824 * socket. 825 * 826 * Applications are unaware that "struct sk_buff" and 827 * other overheads allocate from the receive buffer 828 * during socket buffer allocation. 829 * 830 * And after considering the possible alternatives, 831 * returning the value we actually used in getsockopt 832 * is the most desirable behavior. 833 */ 834 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 835 break; 836 837 case SO_RCVBUFFORCE: 838 if (!capable(CAP_NET_ADMIN)) { 839 ret = -EPERM; 840 break; 841 } 842 843 /* No negative values (to prevent underflow, as val will be 844 * multiplied by 2). 845 */ 846 if (val < 0) 847 val = 0; 848 goto set_rcvbuf; 849 850 case SO_KEEPALIVE: 851 if (sk->sk_prot->keepalive) 852 sk->sk_prot->keepalive(sk, valbool); 853 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 854 break; 855 856 case SO_OOBINLINE: 857 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 858 break; 859 860 case SO_NO_CHECK: 861 sk->sk_no_check_tx = valbool; 862 break; 863 864 case SO_PRIORITY: 865 if ((val >= 0 && val <= 6) || 866 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 867 sk->sk_priority = val; 868 else 869 ret = -EPERM; 870 break; 871 872 case SO_LINGER: 873 if (optlen < sizeof(ling)) { 874 ret = -EINVAL; /* 1003.1g */ 875 break; 876 } 877 if (copy_from_user(&ling, optval, sizeof(ling))) { 878 ret = -EFAULT; 879 break; 880 } 881 if (!ling.l_onoff) 882 sock_reset_flag(sk, SOCK_LINGER); 883 else { 884 #if (BITS_PER_LONG == 32) 885 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 886 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 887 else 888 #endif 889 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 890 sock_set_flag(sk, SOCK_LINGER); 891 } 892 break; 893 894 case SO_BSDCOMPAT: 895 sock_warn_obsolete_bsdism("setsockopt"); 896 break; 897 898 case SO_PASSCRED: 899 if (valbool) 900 set_bit(SOCK_PASSCRED, &sock->flags); 901 else 902 clear_bit(SOCK_PASSCRED, &sock->flags); 903 break; 904 905 case SO_TIMESTAMP_OLD: 906 case SO_TIMESTAMP_NEW: 907 case SO_TIMESTAMPNS_OLD: 908 case SO_TIMESTAMPNS_NEW: 909 if (valbool) { 910 if (optname == SO_TIMESTAMP_NEW || optname == SO_TIMESTAMPNS_NEW) 911 sock_set_flag(sk, SOCK_TSTAMP_NEW); 912 else 913 sock_reset_flag(sk, SOCK_TSTAMP_NEW); 914 915 if (optname == SO_TIMESTAMP_OLD || optname == SO_TIMESTAMP_NEW) 916 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 917 else 918 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 919 sock_set_flag(sk, SOCK_RCVTSTAMP); 920 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 921 } else { 922 sock_reset_flag(sk, SOCK_RCVTSTAMP); 923 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 924 sock_reset_flag(sk, SOCK_TSTAMP_NEW); 925 } 926 break; 927 928 case SO_TIMESTAMPING_NEW: 929 sock_set_flag(sk, SOCK_TSTAMP_NEW); 930 /* fall through */ 931 case SO_TIMESTAMPING_OLD: 932 if (val & ~SOF_TIMESTAMPING_MASK) { 933 ret = -EINVAL; 934 break; 935 } 936 937 if (val & SOF_TIMESTAMPING_OPT_ID && 938 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 939 if (sk->sk_protocol == IPPROTO_TCP && 940 sk->sk_type == SOCK_STREAM) { 941 if ((1 << sk->sk_state) & 942 (TCPF_CLOSE | TCPF_LISTEN)) { 943 ret = -EINVAL; 944 break; 945 } 946 sk->sk_tskey = tcp_sk(sk)->snd_una; 947 } else { 948 sk->sk_tskey = 0; 949 } 950 } 951 952 if (val & SOF_TIMESTAMPING_OPT_STATS && 953 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) { 954 ret = -EINVAL; 955 break; 956 } 957 958 sk->sk_tsflags = val; 959 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 960 sock_enable_timestamp(sk, 961 SOCK_TIMESTAMPING_RX_SOFTWARE); 962 else { 963 if (optname == SO_TIMESTAMPING_NEW) 964 sock_reset_flag(sk, SOCK_TSTAMP_NEW); 965 966 sock_disable_timestamp(sk, 967 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 968 } 969 break; 970 971 case SO_RCVLOWAT: 972 if (val < 0) 973 val = INT_MAX; 974 if (sock->ops->set_rcvlowat) 975 ret = sock->ops->set_rcvlowat(sk, val); 976 else 977 sk->sk_rcvlowat = val ? : 1; 978 break; 979 980 case SO_RCVTIMEO_OLD: 981 case SO_RCVTIMEO_NEW: 982 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen, optname == SO_RCVTIMEO_OLD); 983 break; 984 985 case SO_SNDTIMEO_OLD: 986 case SO_SNDTIMEO_NEW: 987 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen, optname == SO_SNDTIMEO_OLD); 988 break; 989 990 case SO_ATTACH_FILTER: 991 ret = -EINVAL; 992 if (optlen == sizeof(struct sock_fprog)) { 993 struct sock_fprog fprog; 994 995 ret = -EFAULT; 996 if (copy_from_user(&fprog, optval, sizeof(fprog))) 997 break; 998 999 ret = sk_attach_filter(&fprog, sk); 1000 } 1001 break; 1002 1003 case SO_ATTACH_BPF: 1004 ret = -EINVAL; 1005 if (optlen == sizeof(u32)) { 1006 u32 ufd; 1007 1008 ret = -EFAULT; 1009 if (copy_from_user(&ufd, optval, sizeof(ufd))) 1010 break; 1011 1012 ret = sk_attach_bpf(ufd, sk); 1013 } 1014 break; 1015 1016 case SO_ATTACH_REUSEPORT_CBPF: 1017 ret = -EINVAL; 1018 if (optlen == sizeof(struct sock_fprog)) { 1019 struct sock_fprog fprog; 1020 1021 ret = -EFAULT; 1022 if (copy_from_user(&fprog, optval, sizeof(fprog))) 1023 break; 1024 1025 ret = sk_reuseport_attach_filter(&fprog, sk); 1026 } 1027 break; 1028 1029 case SO_ATTACH_REUSEPORT_EBPF: 1030 ret = -EINVAL; 1031 if (optlen == sizeof(u32)) { 1032 u32 ufd; 1033 1034 ret = -EFAULT; 1035 if (copy_from_user(&ufd, optval, sizeof(ufd))) 1036 break; 1037 1038 ret = sk_reuseport_attach_bpf(ufd, sk); 1039 } 1040 break; 1041 1042 case SO_DETACH_REUSEPORT_BPF: 1043 ret = reuseport_detach_prog(sk); 1044 break; 1045 1046 case SO_DETACH_FILTER: 1047 ret = sk_detach_filter(sk); 1048 break; 1049 1050 case SO_LOCK_FILTER: 1051 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 1052 ret = -EPERM; 1053 else 1054 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 1055 break; 1056 1057 case SO_PASSSEC: 1058 if (valbool) 1059 set_bit(SOCK_PASSSEC, &sock->flags); 1060 else 1061 clear_bit(SOCK_PASSSEC, &sock->flags); 1062 break; 1063 case SO_MARK: 1064 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1065 ret = -EPERM; 1066 } else if (val != sk->sk_mark) { 1067 sk->sk_mark = val; 1068 sk_dst_reset(sk); 1069 } 1070 break; 1071 1072 case SO_RXQ_OVFL: 1073 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 1074 break; 1075 1076 case SO_WIFI_STATUS: 1077 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 1078 break; 1079 1080 case SO_PEEK_OFF: 1081 if (sock->ops->set_peek_off) 1082 ret = sock->ops->set_peek_off(sk, val); 1083 else 1084 ret = -EOPNOTSUPP; 1085 break; 1086 1087 case SO_NOFCS: 1088 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 1089 break; 1090 1091 case SO_SELECT_ERR_QUEUE: 1092 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 1093 break; 1094 1095 #ifdef CONFIG_NET_RX_BUSY_POLL 1096 case SO_BUSY_POLL: 1097 /* allow unprivileged users to decrease the value */ 1098 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 1099 ret = -EPERM; 1100 else { 1101 if (val < 0) 1102 ret = -EINVAL; 1103 else 1104 sk->sk_ll_usec = val; 1105 } 1106 break; 1107 #endif 1108 1109 case SO_MAX_PACING_RATE: 1110 { 1111 unsigned long ulval = (val == ~0U) ? ~0UL : val; 1112 1113 if (sizeof(ulval) != sizeof(val) && 1114 optlen >= sizeof(ulval) && 1115 get_user(ulval, (unsigned long __user *)optval)) { 1116 ret = -EFAULT; 1117 break; 1118 } 1119 if (ulval != ~0UL) 1120 cmpxchg(&sk->sk_pacing_status, 1121 SK_PACING_NONE, 1122 SK_PACING_NEEDED); 1123 sk->sk_max_pacing_rate = ulval; 1124 sk->sk_pacing_rate = min(sk->sk_pacing_rate, ulval); 1125 break; 1126 } 1127 case SO_INCOMING_CPU: 1128 sk->sk_incoming_cpu = val; 1129 break; 1130 1131 case SO_CNX_ADVICE: 1132 if (val == 1) 1133 dst_negative_advice(sk); 1134 break; 1135 1136 case SO_ZEROCOPY: 1137 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1138 if (!((sk->sk_type == SOCK_STREAM && 1139 sk->sk_protocol == IPPROTO_TCP) || 1140 (sk->sk_type == SOCK_DGRAM && 1141 sk->sk_protocol == IPPROTO_UDP))) 1142 ret = -ENOTSUPP; 1143 } else if (sk->sk_family != PF_RDS) { 1144 ret = -ENOTSUPP; 1145 } 1146 if (!ret) { 1147 if (val < 0 || val > 1) 1148 ret = -EINVAL; 1149 else 1150 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1151 } 1152 break; 1153 1154 case SO_TXTIME: 1155 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1156 ret = -EPERM; 1157 } else if (optlen != sizeof(struct sock_txtime)) { 1158 ret = -EINVAL; 1159 } else if (copy_from_user(&sk_txtime, optval, 1160 sizeof(struct sock_txtime))) { 1161 ret = -EFAULT; 1162 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1163 ret = -EINVAL; 1164 } else { 1165 sock_valbool_flag(sk, SOCK_TXTIME, true); 1166 sk->sk_clockid = sk_txtime.clockid; 1167 sk->sk_txtime_deadline_mode = 1168 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1169 sk->sk_txtime_report_errors = 1170 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1171 } 1172 break; 1173 1174 case SO_BINDTOIFINDEX: 1175 ret = sock_setbindtodevice_locked(sk, val); 1176 break; 1177 1178 default: 1179 ret = -ENOPROTOOPT; 1180 break; 1181 } 1182 release_sock(sk); 1183 return ret; 1184 } 1185 EXPORT_SYMBOL(sock_setsockopt); 1186 1187 1188 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1189 struct ucred *ucred) 1190 { 1191 ucred->pid = pid_vnr(pid); 1192 ucred->uid = ucred->gid = -1; 1193 if (cred) { 1194 struct user_namespace *current_ns = current_user_ns(); 1195 1196 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1197 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1198 } 1199 } 1200 1201 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1202 { 1203 struct user_namespace *user_ns = current_user_ns(); 1204 int i; 1205 1206 for (i = 0; i < src->ngroups; i++) 1207 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1208 return -EFAULT; 1209 1210 return 0; 1211 } 1212 1213 int sock_getsockopt(struct socket *sock, int level, int optname, 1214 char __user *optval, int __user *optlen) 1215 { 1216 struct sock *sk = sock->sk; 1217 1218 union { 1219 int val; 1220 u64 val64; 1221 unsigned long ulval; 1222 struct linger ling; 1223 struct old_timeval32 tm32; 1224 struct __kernel_old_timeval tm; 1225 struct __kernel_sock_timeval stm; 1226 struct sock_txtime txtime; 1227 } v; 1228 1229 int lv = sizeof(int); 1230 int len; 1231 1232 if (get_user(len, optlen)) 1233 return -EFAULT; 1234 if (len < 0) 1235 return -EINVAL; 1236 1237 memset(&v, 0, sizeof(v)); 1238 1239 switch (optname) { 1240 case SO_DEBUG: 1241 v.val = sock_flag(sk, SOCK_DBG); 1242 break; 1243 1244 case SO_DONTROUTE: 1245 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1246 break; 1247 1248 case SO_BROADCAST: 1249 v.val = sock_flag(sk, SOCK_BROADCAST); 1250 break; 1251 1252 case SO_SNDBUF: 1253 v.val = sk->sk_sndbuf; 1254 break; 1255 1256 case SO_RCVBUF: 1257 v.val = sk->sk_rcvbuf; 1258 break; 1259 1260 case SO_REUSEADDR: 1261 v.val = sk->sk_reuse; 1262 break; 1263 1264 case SO_REUSEPORT: 1265 v.val = sk->sk_reuseport; 1266 break; 1267 1268 case SO_KEEPALIVE: 1269 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1270 break; 1271 1272 case SO_TYPE: 1273 v.val = sk->sk_type; 1274 break; 1275 1276 case SO_PROTOCOL: 1277 v.val = sk->sk_protocol; 1278 break; 1279 1280 case SO_DOMAIN: 1281 v.val = sk->sk_family; 1282 break; 1283 1284 case SO_ERROR: 1285 v.val = -sock_error(sk); 1286 if (v.val == 0) 1287 v.val = xchg(&sk->sk_err_soft, 0); 1288 break; 1289 1290 case SO_OOBINLINE: 1291 v.val = sock_flag(sk, SOCK_URGINLINE); 1292 break; 1293 1294 case SO_NO_CHECK: 1295 v.val = sk->sk_no_check_tx; 1296 break; 1297 1298 case SO_PRIORITY: 1299 v.val = sk->sk_priority; 1300 break; 1301 1302 case SO_LINGER: 1303 lv = sizeof(v.ling); 1304 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1305 v.ling.l_linger = sk->sk_lingertime / HZ; 1306 break; 1307 1308 case SO_BSDCOMPAT: 1309 sock_warn_obsolete_bsdism("getsockopt"); 1310 break; 1311 1312 case SO_TIMESTAMP_OLD: 1313 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1314 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1315 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1316 break; 1317 1318 case SO_TIMESTAMPNS_OLD: 1319 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1320 break; 1321 1322 case SO_TIMESTAMP_NEW: 1323 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1324 break; 1325 1326 case SO_TIMESTAMPNS_NEW: 1327 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1328 break; 1329 1330 case SO_TIMESTAMPING_OLD: 1331 v.val = sk->sk_tsflags; 1332 break; 1333 1334 case SO_RCVTIMEO_OLD: 1335 case SO_RCVTIMEO_NEW: 1336 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1337 break; 1338 1339 case SO_SNDTIMEO_OLD: 1340 case SO_SNDTIMEO_NEW: 1341 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1342 break; 1343 1344 case SO_RCVLOWAT: 1345 v.val = sk->sk_rcvlowat; 1346 break; 1347 1348 case SO_SNDLOWAT: 1349 v.val = 1; 1350 break; 1351 1352 case SO_PASSCRED: 1353 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1354 break; 1355 1356 case SO_PEERCRED: 1357 { 1358 struct ucred peercred; 1359 if (len > sizeof(peercred)) 1360 len = sizeof(peercred); 1361 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1362 if (copy_to_user(optval, &peercred, len)) 1363 return -EFAULT; 1364 goto lenout; 1365 } 1366 1367 case SO_PEERGROUPS: 1368 { 1369 int ret, n; 1370 1371 if (!sk->sk_peer_cred) 1372 return -ENODATA; 1373 1374 n = sk->sk_peer_cred->group_info->ngroups; 1375 if (len < n * sizeof(gid_t)) { 1376 len = n * sizeof(gid_t); 1377 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1378 } 1379 len = n * sizeof(gid_t); 1380 1381 ret = groups_to_user((gid_t __user *)optval, 1382 sk->sk_peer_cred->group_info); 1383 if (ret) 1384 return ret; 1385 goto lenout; 1386 } 1387 1388 case SO_PEERNAME: 1389 { 1390 char address[128]; 1391 1392 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1393 if (lv < 0) 1394 return -ENOTCONN; 1395 if (lv < len) 1396 return -EINVAL; 1397 if (copy_to_user(optval, address, len)) 1398 return -EFAULT; 1399 goto lenout; 1400 } 1401 1402 /* Dubious BSD thing... Probably nobody even uses it, but 1403 * the UNIX standard wants it for whatever reason... -DaveM 1404 */ 1405 case SO_ACCEPTCONN: 1406 v.val = sk->sk_state == TCP_LISTEN; 1407 break; 1408 1409 case SO_PASSSEC: 1410 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1411 break; 1412 1413 case SO_PEERSEC: 1414 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1415 1416 case SO_MARK: 1417 v.val = sk->sk_mark; 1418 break; 1419 1420 case SO_RXQ_OVFL: 1421 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1422 break; 1423 1424 case SO_WIFI_STATUS: 1425 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1426 break; 1427 1428 case SO_PEEK_OFF: 1429 if (!sock->ops->set_peek_off) 1430 return -EOPNOTSUPP; 1431 1432 v.val = sk->sk_peek_off; 1433 break; 1434 case SO_NOFCS: 1435 v.val = sock_flag(sk, SOCK_NOFCS); 1436 break; 1437 1438 case SO_BINDTODEVICE: 1439 return sock_getbindtodevice(sk, optval, optlen, len); 1440 1441 case SO_GET_FILTER: 1442 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1443 if (len < 0) 1444 return len; 1445 1446 goto lenout; 1447 1448 case SO_LOCK_FILTER: 1449 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1450 break; 1451 1452 case SO_BPF_EXTENSIONS: 1453 v.val = bpf_tell_extensions(); 1454 break; 1455 1456 case SO_SELECT_ERR_QUEUE: 1457 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1458 break; 1459 1460 #ifdef CONFIG_NET_RX_BUSY_POLL 1461 case SO_BUSY_POLL: 1462 v.val = sk->sk_ll_usec; 1463 break; 1464 #endif 1465 1466 case SO_MAX_PACING_RATE: 1467 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1468 lv = sizeof(v.ulval); 1469 v.ulval = sk->sk_max_pacing_rate; 1470 } else { 1471 /* 32bit version */ 1472 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1473 } 1474 break; 1475 1476 case SO_INCOMING_CPU: 1477 v.val = sk->sk_incoming_cpu; 1478 break; 1479 1480 case SO_MEMINFO: 1481 { 1482 u32 meminfo[SK_MEMINFO_VARS]; 1483 1484 sk_get_meminfo(sk, meminfo); 1485 1486 len = min_t(unsigned int, len, sizeof(meminfo)); 1487 if (copy_to_user(optval, &meminfo, len)) 1488 return -EFAULT; 1489 1490 goto lenout; 1491 } 1492 1493 #ifdef CONFIG_NET_RX_BUSY_POLL 1494 case SO_INCOMING_NAPI_ID: 1495 v.val = READ_ONCE(sk->sk_napi_id); 1496 1497 /* aggregate non-NAPI IDs down to 0 */ 1498 if (v.val < MIN_NAPI_ID) 1499 v.val = 0; 1500 1501 break; 1502 #endif 1503 1504 case SO_COOKIE: 1505 lv = sizeof(u64); 1506 if (len < lv) 1507 return -EINVAL; 1508 v.val64 = sock_gen_cookie(sk); 1509 break; 1510 1511 case SO_ZEROCOPY: 1512 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1513 break; 1514 1515 case SO_TXTIME: 1516 lv = sizeof(v.txtime); 1517 v.txtime.clockid = sk->sk_clockid; 1518 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1519 SOF_TXTIME_DEADLINE_MODE : 0; 1520 v.txtime.flags |= sk->sk_txtime_report_errors ? 1521 SOF_TXTIME_REPORT_ERRORS : 0; 1522 break; 1523 1524 case SO_BINDTOIFINDEX: 1525 v.val = sk->sk_bound_dev_if; 1526 break; 1527 1528 default: 1529 /* We implement the SO_SNDLOWAT etc to not be settable 1530 * (1003.1g 7). 1531 */ 1532 return -ENOPROTOOPT; 1533 } 1534 1535 if (len > lv) 1536 len = lv; 1537 if (copy_to_user(optval, &v, len)) 1538 return -EFAULT; 1539 lenout: 1540 if (put_user(len, optlen)) 1541 return -EFAULT; 1542 return 0; 1543 } 1544 1545 /* 1546 * Initialize an sk_lock. 1547 * 1548 * (We also register the sk_lock with the lock validator.) 1549 */ 1550 static inline void sock_lock_init(struct sock *sk) 1551 { 1552 if (sk->sk_kern_sock) 1553 sock_lock_init_class_and_name( 1554 sk, 1555 af_family_kern_slock_key_strings[sk->sk_family], 1556 af_family_kern_slock_keys + sk->sk_family, 1557 af_family_kern_key_strings[sk->sk_family], 1558 af_family_kern_keys + sk->sk_family); 1559 else 1560 sock_lock_init_class_and_name( 1561 sk, 1562 af_family_slock_key_strings[sk->sk_family], 1563 af_family_slock_keys + sk->sk_family, 1564 af_family_key_strings[sk->sk_family], 1565 af_family_keys + sk->sk_family); 1566 } 1567 1568 /* 1569 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1570 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1571 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1572 */ 1573 static void sock_copy(struct sock *nsk, const struct sock *osk) 1574 { 1575 #ifdef CONFIG_SECURITY_NETWORK 1576 void *sptr = nsk->sk_security; 1577 #endif 1578 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1579 1580 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1581 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1582 1583 #ifdef CONFIG_SECURITY_NETWORK 1584 nsk->sk_security = sptr; 1585 security_sk_clone(osk, nsk); 1586 #endif 1587 } 1588 1589 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1590 int family) 1591 { 1592 struct sock *sk; 1593 struct kmem_cache *slab; 1594 1595 slab = prot->slab; 1596 if (slab != NULL) { 1597 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1598 if (!sk) 1599 return sk; 1600 if (want_init_on_alloc(priority)) 1601 sk_prot_clear_nulls(sk, prot->obj_size); 1602 } else 1603 sk = kmalloc(prot->obj_size, priority); 1604 1605 if (sk != NULL) { 1606 if (security_sk_alloc(sk, family, priority)) 1607 goto out_free; 1608 1609 if (!try_module_get(prot->owner)) 1610 goto out_free_sec; 1611 sk_tx_queue_clear(sk); 1612 } 1613 1614 return sk; 1615 1616 out_free_sec: 1617 security_sk_free(sk); 1618 out_free: 1619 if (slab != NULL) 1620 kmem_cache_free(slab, sk); 1621 else 1622 kfree(sk); 1623 return NULL; 1624 } 1625 1626 static void sk_prot_free(struct proto *prot, struct sock *sk) 1627 { 1628 struct kmem_cache *slab; 1629 struct module *owner; 1630 1631 owner = prot->owner; 1632 slab = prot->slab; 1633 1634 cgroup_sk_free(&sk->sk_cgrp_data); 1635 mem_cgroup_sk_free(sk); 1636 security_sk_free(sk); 1637 if (slab != NULL) 1638 kmem_cache_free(slab, sk); 1639 else 1640 kfree(sk); 1641 module_put(owner); 1642 } 1643 1644 /** 1645 * sk_alloc - All socket objects are allocated here 1646 * @net: the applicable net namespace 1647 * @family: protocol family 1648 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1649 * @prot: struct proto associated with this new sock instance 1650 * @kern: is this to be a kernel socket? 1651 */ 1652 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1653 struct proto *prot, int kern) 1654 { 1655 struct sock *sk; 1656 1657 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1658 if (sk) { 1659 sk->sk_family = family; 1660 /* 1661 * See comment in struct sock definition to understand 1662 * why we need sk_prot_creator -acme 1663 */ 1664 sk->sk_prot = sk->sk_prot_creator = prot; 1665 sk->sk_kern_sock = kern; 1666 sock_lock_init(sk); 1667 sk->sk_net_refcnt = kern ? 0 : 1; 1668 if (likely(sk->sk_net_refcnt)) { 1669 get_net(net); 1670 sock_inuse_add(net, 1); 1671 } 1672 1673 sock_net_set(sk, net); 1674 refcount_set(&sk->sk_wmem_alloc, 1); 1675 1676 mem_cgroup_sk_alloc(sk); 1677 cgroup_sk_alloc(&sk->sk_cgrp_data); 1678 sock_update_classid(&sk->sk_cgrp_data); 1679 sock_update_netprioidx(&sk->sk_cgrp_data); 1680 } 1681 1682 return sk; 1683 } 1684 EXPORT_SYMBOL(sk_alloc); 1685 1686 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1687 * grace period. This is the case for UDP sockets and TCP listeners. 1688 */ 1689 static void __sk_destruct(struct rcu_head *head) 1690 { 1691 struct sock *sk = container_of(head, struct sock, sk_rcu); 1692 struct sk_filter *filter; 1693 1694 if (sk->sk_destruct) 1695 sk->sk_destruct(sk); 1696 1697 filter = rcu_dereference_check(sk->sk_filter, 1698 refcount_read(&sk->sk_wmem_alloc) == 0); 1699 if (filter) { 1700 sk_filter_uncharge(sk, filter); 1701 RCU_INIT_POINTER(sk->sk_filter, NULL); 1702 } 1703 1704 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1705 1706 #ifdef CONFIG_BPF_SYSCALL 1707 bpf_sk_storage_free(sk); 1708 #endif 1709 1710 if (atomic_read(&sk->sk_omem_alloc)) 1711 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1712 __func__, atomic_read(&sk->sk_omem_alloc)); 1713 1714 if (sk->sk_frag.page) { 1715 put_page(sk->sk_frag.page); 1716 sk->sk_frag.page = NULL; 1717 } 1718 1719 if (sk->sk_peer_cred) 1720 put_cred(sk->sk_peer_cred); 1721 put_pid(sk->sk_peer_pid); 1722 if (likely(sk->sk_net_refcnt)) 1723 put_net(sock_net(sk)); 1724 sk_prot_free(sk->sk_prot_creator, sk); 1725 } 1726 1727 void sk_destruct(struct sock *sk) 1728 { 1729 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 1730 1731 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 1732 reuseport_detach_sock(sk); 1733 use_call_rcu = true; 1734 } 1735 1736 if (use_call_rcu) 1737 call_rcu(&sk->sk_rcu, __sk_destruct); 1738 else 1739 __sk_destruct(&sk->sk_rcu); 1740 } 1741 1742 static void __sk_free(struct sock *sk) 1743 { 1744 if (likely(sk->sk_net_refcnt)) 1745 sock_inuse_add(sock_net(sk), -1); 1746 1747 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1748 sock_diag_broadcast_destroy(sk); 1749 else 1750 sk_destruct(sk); 1751 } 1752 1753 void sk_free(struct sock *sk) 1754 { 1755 /* 1756 * We subtract one from sk_wmem_alloc and can know if 1757 * some packets are still in some tx queue. 1758 * If not null, sock_wfree() will call __sk_free(sk) later 1759 */ 1760 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1761 __sk_free(sk); 1762 } 1763 EXPORT_SYMBOL(sk_free); 1764 1765 static void sk_init_common(struct sock *sk) 1766 { 1767 skb_queue_head_init(&sk->sk_receive_queue); 1768 skb_queue_head_init(&sk->sk_write_queue); 1769 skb_queue_head_init(&sk->sk_error_queue); 1770 1771 rwlock_init(&sk->sk_callback_lock); 1772 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1773 af_rlock_keys + sk->sk_family, 1774 af_family_rlock_key_strings[sk->sk_family]); 1775 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1776 af_wlock_keys + sk->sk_family, 1777 af_family_wlock_key_strings[sk->sk_family]); 1778 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1779 af_elock_keys + sk->sk_family, 1780 af_family_elock_key_strings[sk->sk_family]); 1781 lockdep_set_class_and_name(&sk->sk_callback_lock, 1782 af_callback_keys + sk->sk_family, 1783 af_family_clock_key_strings[sk->sk_family]); 1784 } 1785 1786 /** 1787 * sk_clone_lock - clone a socket, and lock its clone 1788 * @sk: the socket to clone 1789 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1790 * 1791 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1792 */ 1793 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1794 { 1795 struct sock *newsk; 1796 bool is_charged = true; 1797 1798 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1799 if (newsk != NULL) { 1800 struct sk_filter *filter; 1801 1802 sock_copy(newsk, sk); 1803 1804 newsk->sk_prot_creator = sk->sk_prot; 1805 1806 /* SANITY */ 1807 if (likely(newsk->sk_net_refcnt)) 1808 get_net(sock_net(newsk)); 1809 sk_node_init(&newsk->sk_node); 1810 sock_lock_init(newsk); 1811 bh_lock_sock(newsk); 1812 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1813 newsk->sk_backlog.len = 0; 1814 1815 atomic_set(&newsk->sk_rmem_alloc, 0); 1816 /* 1817 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1818 */ 1819 refcount_set(&newsk->sk_wmem_alloc, 1); 1820 atomic_set(&newsk->sk_omem_alloc, 0); 1821 sk_init_common(newsk); 1822 1823 newsk->sk_dst_cache = NULL; 1824 newsk->sk_dst_pending_confirm = 0; 1825 newsk->sk_wmem_queued = 0; 1826 newsk->sk_forward_alloc = 0; 1827 atomic_set(&newsk->sk_drops, 0); 1828 newsk->sk_send_head = NULL; 1829 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1830 atomic_set(&newsk->sk_zckey, 0); 1831 1832 sock_reset_flag(newsk, SOCK_DONE); 1833 mem_cgroup_sk_alloc(newsk); 1834 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1835 1836 rcu_read_lock(); 1837 filter = rcu_dereference(sk->sk_filter); 1838 if (filter != NULL) 1839 /* though it's an empty new sock, the charging may fail 1840 * if sysctl_optmem_max was changed between creation of 1841 * original socket and cloning 1842 */ 1843 is_charged = sk_filter_charge(newsk, filter); 1844 RCU_INIT_POINTER(newsk->sk_filter, filter); 1845 rcu_read_unlock(); 1846 1847 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1848 /* We need to make sure that we don't uncharge the new 1849 * socket if we couldn't charge it in the first place 1850 * as otherwise we uncharge the parent's filter. 1851 */ 1852 if (!is_charged) 1853 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1854 sk_free_unlock_clone(newsk); 1855 newsk = NULL; 1856 goto out; 1857 } 1858 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1859 1860 if (bpf_sk_storage_clone(sk, newsk)) { 1861 sk_free_unlock_clone(newsk); 1862 newsk = NULL; 1863 goto out; 1864 } 1865 1866 newsk->sk_err = 0; 1867 newsk->sk_err_soft = 0; 1868 newsk->sk_priority = 0; 1869 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1870 if (likely(newsk->sk_net_refcnt)) 1871 sock_inuse_add(sock_net(newsk), 1); 1872 1873 /* 1874 * Before updating sk_refcnt, we must commit prior changes to memory 1875 * (Documentation/RCU/rculist_nulls.txt for details) 1876 */ 1877 smp_wmb(); 1878 refcount_set(&newsk->sk_refcnt, 2); 1879 1880 /* 1881 * Increment the counter in the same struct proto as the master 1882 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1883 * is the same as sk->sk_prot->socks, as this field was copied 1884 * with memcpy). 1885 * 1886 * This _changes_ the previous behaviour, where 1887 * tcp_create_openreq_child always was incrementing the 1888 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1889 * to be taken into account in all callers. -acme 1890 */ 1891 sk_refcnt_debug_inc(newsk); 1892 sk_set_socket(newsk, NULL); 1893 RCU_INIT_POINTER(newsk->sk_wq, NULL); 1894 1895 if (newsk->sk_prot->sockets_allocated) 1896 sk_sockets_allocated_inc(newsk); 1897 1898 if (sock_needs_netstamp(sk) && 1899 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1900 net_enable_timestamp(); 1901 } 1902 out: 1903 return newsk; 1904 } 1905 EXPORT_SYMBOL_GPL(sk_clone_lock); 1906 1907 void sk_free_unlock_clone(struct sock *sk) 1908 { 1909 /* It is still raw copy of parent, so invalidate 1910 * destructor and make plain sk_free() */ 1911 sk->sk_destruct = NULL; 1912 bh_unlock_sock(sk); 1913 sk_free(sk); 1914 } 1915 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1916 1917 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1918 { 1919 u32 max_segs = 1; 1920 1921 sk_dst_set(sk, dst); 1922 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 1923 if (sk->sk_route_caps & NETIF_F_GSO) 1924 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1925 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1926 if (sk_can_gso(sk)) { 1927 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 1928 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1929 } else { 1930 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1931 sk->sk_gso_max_size = dst->dev->gso_max_size; 1932 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1933 } 1934 } 1935 sk->sk_gso_max_segs = max_segs; 1936 } 1937 EXPORT_SYMBOL_GPL(sk_setup_caps); 1938 1939 /* 1940 * Simple resource managers for sockets. 1941 */ 1942 1943 1944 /* 1945 * Write buffer destructor automatically called from kfree_skb. 1946 */ 1947 void sock_wfree(struct sk_buff *skb) 1948 { 1949 struct sock *sk = skb->sk; 1950 unsigned int len = skb->truesize; 1951 1952 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1953 /* 1954 * Keep a reference on sk_wmem_alloc, this will be released 1955 * after sk_write_space() call 1956 */ 1957 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 1958 sk->sk_write_space(sk); 1959 len = 1; 1960 } 1961 /* 1962 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1963 * could not do because of in-flight packets 1964 */ 1965 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 1966 __sk_free(sk); 1967 } 1968 EXPORT_SYMBOL(sock_wfree); 1969 1970 /* This variant of sock_wfree() is used by TCP, 1971 * since it sets SOCK_USE_WRITE_QUEUE. 1972 */ 1973 void __sock_wfree(struct sk_buff *skb) 1974 { 1975 struct sock *sk = skb->sk; 1976 1977 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1978 __sk_free(sk); 1979 } 1980 1981 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1982 { 1983 skb_orphan(skb); 1984 skb->sk = sk; 1985 #ifdef CONFIG_INET 1986 if (unlikely(!sk_fullsock(sk))) { 1987 skb->destructor = sock_edemux; 1988 sock_hold(sk); 1989 return; 1990 } 1991 #endif 1992 skb->destructor = sock_wfree; 1993 skb_set_hash_from_sk(skb, sk); 1994 /* 1995 * We used to take a refcount on sk, but following operation 1996 * is enough to guarantee sk_free() wont free this sock until 1997 * all in-flight packets are completed 1998 */ 1999 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2000 } 2001 EXPORT_SYMBOL(skb_set_owner_w); 2002 2003 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2004 { 2005 #ifdef CONFIG_TLS_DEVICE 2006 /* Drivers depend on in-order delivery for crypto offload, 2007 * partial orphan breaks out-of-order-OK logic. 2008 */ 2009 if (skb->decrypted) 2010 return false; 2011 #endif 2012 return (skb->destructor == sock_wfree || 2013 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2014 } 2015 2016 /* This helper is used by netem, as it can hold packets in its 2017 * delay queue. We want to allow the owner socket to send more 2018 * packets, as if they were already TX completed by a typical driver. 2019 * But we also want to keep skb->sk set because some packet schedulers 2020 * rely on it (sch_fq for example). 2021 */ 2022 void skb_orphan_partial(struct sk_buff *skb) 2023 { 2024 if (skb_is_tcp_pure_ack(skb)) 2025 return; 2026 2027 if (can_skb_orphan_partial(skb)) { 2028 struct sock *sk = skb->sk; 2029 2030 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 2031 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)); 2032 skb->destructor = sock_efree; 2033 } 2034 } else { 2035 skb_orphan(skb); 2036 } 2037 } 2038 EXPORT_SYMBOL(skb_orphan_partial); 2039 2040 /* 2041 * Read buffer destructor automatically called from kfree_skb. 2042 */ 2043 void sock_rfree(struct sk_buff *skb) 2044 { 2045 struct sock *sk = skb->sk; 2046 unsigned int len = skb->truesize; 2047 2048 atomic_sub(len, &sk->sk_rmem_alloc); 2049 sk_mem_uncharge(sk, len); 2050 } 2051 EXPORT_SYMBOL(sock_rfree); 2052 2053 /* 2054 * Buffer destructor for skbs that are not used directly in read or write 2055 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2056 */ 2057 void sock_efree(struct sk_buff *skb) 2058 { 2059 sock_put(skb->sk); 2060 } 2061 EXPORT_SYMBOL(sock_efree); 2062 2063 kuid_t sock_i_uid(struct sock *sk) 2064 { 2065 kuid_t uid; 2066 2067 read_lock_bh(&sk->sk_callback_lock); 2068 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2069 read_unlock_bh(&sk->sk_callback_lock); 2070 return uid; 2071 } 2072 EXPORT_SYMBOL(sock_i_uid); 2073 2074 unsigned long sock_i_ino(struct sock *sk) 2075 { 2076 unsigned long ino; 2077 2078 read_lock_bh(&sk->sk_callback_lock); 2079 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2080 read_unlock_bh(&sk->sk_callback_lock); 2081 return ino; 2082 } 2083 EXPORT_SYMBOL(sock_i_ino); 2084 2085 /* 2086 * Allocate a skb from the socket's send buffer. 2087 */ 2088 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2089 gfp_t priority) 2090 { 2091 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 2092 struct sk_buff *skb = alloc_skb(size, priority); 2093 if (skb) { 2094 skb_set_owner_w(skb, sk); 2095 return skb; 2096 } 2097 } 2098 return NULL; 2099 } 2100 EXPORT_SYMBOL(sock_wmalloc); 2101 2102 static void sock_ofree(struct sk_buff *skb) 2103 { 2104 struct sock *sk = skb->sk; 2105 2106 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2107 } 2108 2109 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2110 gfp_t priority) 2111 { 2112 struct sk_buff *skb; 2113 2114 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2115 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2116 sysctl_optmem_max) 2117 return NULL; 2118 2119 skb = alloc_skb(size, priority); 2120 if (!skb) 2121 return NULL; 2122 2123 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2124 skb->sk = sk; 2125 skb->destructor = sock_ofree; 2126 return skb; 2127 } 2128 2129 /* 2130 * Allocate a memory block from the socket's option memory buffer. 2131 */ 2132 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2133 { 2134 if ((unsigned int)size <= sysctl_optmem_max && 2135 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2136 void *mem; 2137 /* First do the add, to avoid the race if kmalloc 2138 * might sleep. 2139 */ 2140 atomic_add(size, &sk->sk_omem_alloc); 2141 mem = kmalloc(size, priority); 2142 if (mem) 2143 return mem; 2144 atomic_sub(size, &sk->sk_omem_alloc); 2145 } 2146 return NULL; 2147 } 2148 EXPORT_SYMBOL(sock_kmalloc); 2149 2150 /* Free an option memory block. Note, we actually want the inline 2151 * here as this allows gcc to detect the nullify and fold away the 2152 * condition entirely. 2153 */ 2154 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2155 const bool nullify) 2156 { 2157 if (WARN_ON_ONCE(!mem)) 2158 return; 2159 if (nullify) 2160 kzfree(mem); 2161 else 2162 kfree(mem); 2163 atomic_sub(size, &sk->sk_omem_alloc); 2164 } 2165 2166 void sock_kfree_s(struct sock *sk, void *mem, int size) 2167 { 2168 __sock_kfree_s(sk, mem, size, false); 2169 } 2170 EXPORT_SYMBOL(sock_kfree_s); 2171 2172 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2173 { 2174 __sock_kfree_s(sk, mem, size, true); 2175 } 2176 EXPORT_SYMBOL(sock_kzfree_s); 2177 2178 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2179 I think, these locks should be removed for datagram sockets. 2180 */ 2181 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2182 { 2183 DEFINE_WAIT(wait); 2184 2185 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2186 for (;;) { 2187 if (!timeo) 2188 break; 2189 if (signal_pending(current)) 2190 break; 2191 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2192 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2193 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 2194 break; 2195 if (sk->sk_shutdown & SEND_SHUTDOWN) 2196 break; 2197 if (sk->sk_err) 2198 break; 2199 timeo = schedule_timeout(timeo); 2200 } 2201 finish_wait(sk_sleep(sk), &wait); 2202 return timeo; 2203 } 2204 2205 2206 /* 2207 * Generic send/receive buffer handlers 2208 */ 2209 2210 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2211 unsigned long data_len, int noblock, 2212 int *errcode, int max_page_order) 2213 { 2214 struct sk_buff *skb; 2215 long timeo; 2216 int err; 2217 2218 timeo = sock_sndtimeo(sk, noblock); 2219 for (;;) { 2220 err = sock_error(sk); 2221 if (err != 0) 2222 goto failure; 2223 2224 err = -EPIPE; 2225 if (sk->sk_shutdown & SEND_SHUTDOWN) 2226 goto failure; 2227 2228 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 2229 break; 2230 2231 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2232 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2233 err = -EAGAIN; 2234 if (!timeo) 2235 goto failure; 2236 if (signal_pending(current)) 2237 goto interrupted; 2238 timeo = sock_wait_for_wmem(sk, timeo); 2239 } 2240 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2241 errcode, sk->sk_allocation); 2242 if (skb) 2243 skb_set_owner_w(skb, sk); 2244 return skb; 2245 2246 interrupted: 2247 err = sock_intr_errno(timeo); 2248 failure: 2249 *errcode = err; 2250 return NULL; 2251 } 2252 EXPORT_SYMBOL(sock_alloc_send_pskb); 2253 2254 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2255 int noblock, int *errcode) 2256 { 2257 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2258 } 2259 EXPORT_SYMBOL(sock_alloc_send_skb); 2260 2261 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2262 struct sockcm_cookie *sockc) 2263 { 2264 u32 tsflags; 2265 2266 switch (cmsg->cmsg_type) { 2267 case SO_MARK: 2268 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2269 return -EPERM; 2270 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2271 return -EINVAL; 2272 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2273 break; 2274 case SO_TIMESTAMPING_OLD: 2275 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2276 return -EINVAL; 2277 2278 tsflags = *(u32 *)CMSG_DATA(cmsg); 2279 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2280 return -EINVAL; 2281 2282 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2283 sockc->tsflags |= tsflags; 2284 break; 2285 case SCM_TXTIME: 2286 if (!sock_flag(sk, SOCK_TXTIME)) 2287 return -EINVAL; 2288 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2289 return -EINVAL; 2290 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2291 break; 2292 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2293 case SCM_RIGHTS: 2294 case SCM_CREDENTIALS: 2295 break; 2296 default: 2297 return -EINVAL; 2298 } 2299 return 0; 2300 } 2301 EXPORT_SYMBOL(__sock_cmsg_send); 2302 2303 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2304 struct sockcm_cookie *sockc) 2305 { 2306 struct cmsghdr *cmsg; 2307 int ret; 2308 2309 for_each_cmsghdr(cmsg, msg) { 2310 if (!CMSG_OK(msg, cmsg)) 2311 return -EINVAL; 2312 if (cmsg->cmsg_level != SOL_SOCKET) 2313 continue; 2314 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2315 if (ret) 2316 return ret; 2317 } 2318 return 0; 2319 } 2320 EXPORT_SYMBOL(sock_cmsg_send); 2321 2322 static void sk_enter_memory_pressure(struct sock *sk) 2323 { 2324 if (!sk->sk_prot->enter_memory_pressure) 2325 return; 2326 2327 sk->sk_prot->enter_memory_pressure(sk); 2328 } 2329 2330 static void sk_leave_memory_pressure(struct sock *sk) 2331 { 2332 if (sk->sk_prot->leave_memory_pressure) { 2333 sk->sk_prot->leave_memory_pressure(sk); 2334 } else { 2335 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2336 2337 if (memory_pressure && *memory_pressure) 2338 *memory_pressure = 0; 2339 } 2340 } 2341 2342 /* On 32bit arches, an skb frag is limited to 2^15 */ 2343 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2344 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2345 2346 /** 2347 * skb_page_frag_refill - check that a page_frag contains enough room 2348 * @sz: minimum size of the fragment we want to get 2349 * @pfrag: pointer to page_frag 2350 * @gfp: priority for memory allocation 2351 * 2352 * Note: While this allocator tries to use high order pages, there is 2353 * no guarantee that allocations succeed. Therefore, @sz MUST be 2354 * less or equal than PAGE_SIZE. 2355 */ 2356 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2357 { 2358 if (pfrag->page) { 2359 if (page_ref_count(pfrag->page) == 1) { 2360 pfrag->offset = 0; 2361 return true; 2362 } 2363 if (pfrag->offset + sz <= pfrag->size) 2364 return true; 2365 put_page(pfrag->page); 2366 } 2367 2368 pfrag->offset = 0; 2369 if (SKB_FRAG_PAGE_ORDER && 2370 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2371 /* Avoid direct reclaim but allow kswapd to wake */ 2372 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2373 __GFP_COMP | __GFP_NOWARN | 2374 __GFP_NORETRY, 2375 SKB_FRAG_PAGE_ORDER); 2376 if (likely(pfrag->page)) { 2377 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2378 return true; 2379 } 2380 } 2381 pfrag->page = alloc_page(gfp); 2382 if (likely(pfrag->page)) { 2383 pfrag->size = PAGE_SIZE; 2384 return true; 2385 } 2386 return false; 2387 } 2388 EXPORT_SYMBOL(skb_page_frag_refill); 2389 2390 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2391 { 2392 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2393 return true; 2394 2395 sk_enter_memory_pressure(sk); 2396 sk_stream_moderate_sndbuf(sk); 2397 return false; 2398 } 2399 EXPORT_SYMBOL(sk_page_frag_refill); 2400 2401 static void __lock_sock(struct sock *sk) 2402 __releases(&sk->sk_lock.slock) 2403 __acquires(&sk->sk_lock.slock) 2404 { 2405 DEFINE_WAIT(wait); 2406 2407 for (;;) { 2408 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2409 TASK_UNINTERRUPTIBLE); 2410 spin_unlock_bh(&sk->sk_lock.slock); 2411 schedule(); 2412 spin_lock_bh(&sk->sk_lock.slock); 2413 if (!sock_owned_by_user(sk)) 2414 break; 2415 } 2416 finish_wait(&sk->sk_lock.wq, &wait); 2417 } 2418 2419 void __release_sock(struct sock *sk) 2420 __releases(&sk->sk_lock.slock) 2421 __acquires(&sk->sk_lock.slock) 2422 { 2423 struct sk_buff *skb, *next; 2424 2425 while ((skb = sk->sk_backlog.head) != NULL) { 2426 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2427 2428 spin_unlock_bh(&sk->sk_lock.slock); 2429 2430 do { 2431 next = skb->next; 2432 prefetch(next); 2433 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2434 skb_mark_not_on_list(skb); 2435 sk_backlog_rcv(sk, skb); 2436 2437 cond_resched(); 2438 2439 skb = next; 2440 } while (skb != NULL); 2441 2442 spin_lock_bh(&sk->sk_lock.slock); 2443 } 2444 2445 /* 2446 * Doing the zeroing here guarantee we can not loop forever 2447 * while a wild producer attempts to flood us. 2448 */ 2449 sk->sk_backlog.len = 0; 2450 } 2451 2452 void __sk_flush_backlog(struct sock *sk) 2453 { 2454 spin_lock_bh(&sk->sk_lock.slock); 2455 __release_sock(sk); 2456 spin_unlock_bh(&sk->sk_lock.slock); 2457 } 2458 2459 /** 2460 * sk_wait_data - wait for data to arrive at sk_receive_queue 2461 * @sk: sock to wait on 2462 * @timeo: for how long 2463 * @skb: last skb seen on sk_receive_queue 2464 * 2465 * Now socket state including sk->sk_err is changed only under lock, 2466 * hence we may omit checks after joining wait queue. 2467 * We check receive queue before schedule() only as optimization; 2468 * it is very likely that release_sock() added new data. 2469 */ 2470 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2471 { 2472 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2473 int rc; 2474 2475 add_wait_queue(sk_sleep(sk), &wait); 2476 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2477 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2478 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2479 remove_wait_queue(sk_sleep(sk), &wait); 2480 return rc; 2481 } 2482 EXPORT_SYMBOL(sk_wait_data); 2483 2484 /** 2485 * __sk_mem_raise_allocated - increase memory_allocated 2486 * @sk: socket 2487 * @size: memory size to allocate 2488 * @amt: pages to allocate 2489 * @kind: allocation type 2490 * 2491 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2492 */ 2493 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2494 { 2495 struct proto *prot = sk->sk_prot; 2496 long allocated = sk_memory_allocated_add(sk, amt); 2497 bool charged = true; 2498 2499 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2500 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2501 goto suppress_allocation; 2502 2503 /* Under limit. */ 2504 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2505 sk_leave_memory_pressure(sk); 2506 return 1; 2507 } 2508 2509 /* Under pressure. */ 2510 if (allocated > sk_prot_mem_limits(sk, 1)) 2511 sk_enter_memory_pressure(sk); 2512 2513 /* Over hard limit. */ 2514 if (allocated > sk_prot_mem_limits(sk, 2)) 2515 goto suppress_allocation; 2516 2517 /* guarantee minimum buffer size under pressure */ 2518 if (kind == SK_MEM_RECV) { 2519 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2520 return 1; 2521 2522 } else { /* SK_MEM_SEND */ 2523 int wmem0 = sk_get_wmem0(sk, prot); 2524 2525 if (sk->sk_type == SOCK_STREAM) { 2526 if (sk->sk_wmem_queued < wmem0) 2527 return 1; 2528 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2529 return 1; 2530 } 2531 } 2532 2533 if (sk_has_memory_pressure(sk)) { 2534 u64 alloc; 2535 2536 if (!sk_under_memory_pressure(sk)) 2537 return 1; 2538 alloc = sk_sockets_allocated_read_positive(sk); 2539 if (sk_prot_mem_limits(sk, 2) > alloc * 2540 sk_mem_pages(sk->sk_wmem_queued + 2541 atomic_read(&sk->sk_rmem_alloc) + 2542 sk->sk_forward_alloc)) 2543 return 1; 2544 } 2545 2546 suppress_allocation: 2547 2548 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2549 sk_stream_moderate_sndbuf(sk); 2550 2551 /* Fail only if socket is _under_ its sndbuf. 2552 * In this case we cannot block, so that we have to fail. 2553 */ 2554 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2555 return 1; 2556 } 2557 2558 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2559 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2560 2561 sk_memory_allocated_sub(sk, amt); 2562 2563 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2564 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2565 2566 return 0; 2567 } 2568 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2569 2570 /** 2571 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2572 * @sk: socket 2573 * @size: memory size to allocate 2574 * @kind: allocation type 2575 * 2576 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2577 * rmem allocation. This function assumes that protocols which have 2578 * memory_pressure use sk_wmem_queued as write buffer accounting. 2579 */ 2580 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2581 { 2582 int ret, amt = sk_mem_pages(size); 2583 2584 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2585 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2586 if (!ret) 2587 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2588 return ret; 2589 } 2590 EXPORT_SYMBOL(__sk_mem_schedule); 2591 2592 /** 2593 * __sk_mem_reduce_allocated - reclaim memory_allocated 2594 * @sk: socket 2595 * @amount: number of quanta 2596 * 2597 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2598 */ 2599 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2600 { 2601 sk_memory_allocated_sub(sk, amount); 2602 2603 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2604 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2605 2606 if (sk_under_memory_pressure(sk) && 2607 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2608 sk_leave_memory_pressure(sk); 2609 } 2610 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2611 2612 /** 2613 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2614 * @sk: socket 2615 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2616 */ 2617 void __sk_mem_reclaim(struct sock *sk, int amount) 2618 { 2619 amount >>= SK_MEM_QUANTUM_SHIFT; 2620 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2621 __sk_mem_reduce_allocated(sk, amount); 2622 } 2623 EXPORT_SYMBOL(__sk_mem_reclaim); 2624 2625 int sk_set_peek_off(struct sock *sk, int val) 2626 { 2627 sk->sk_peek_off = val; 2628 return 0; 2629 } 2630 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2631 2632 /* 2633 * Set of default routines for initialising struct proto_ops when 2634 * the protocol does not support a particular function. In certain 2635 * cases where it makes no sense for a protocol to have a "do nothing" 2636 * function, some default processing is provided. 2637 */ 2638 2639 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2640 { 2641 return -EOPNOTSUPP; 2642 } 2643 EXPORT_SYMBOL(sock_no_bind); 2644 2645 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2646 int len, int flags) 2647 { 2648 return -EOPNOTSUPP; 2649 } 2650 EXPORT_SYMBOL(sock_no_connect); 2651 2652 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2653 { 2654 return -EOPNOTSUPP; 2655 } 2656 EXPORT_SYMBOL(sock_no_socketpair); 2657 2658 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2659 bool kern) 2660 { 2661 return -EOPNOTSUPP; 2662 } 2663 EXPORT_SYMBOL(sock_no_accept); 2664 2665 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2666 int peer) 2667 { 2668 return -EOPNOTSUPP; 2669 } 2670 EXPORT_SYMBOL(sock_no_getname); 2671 2672 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2673 { 2674 return -EOPNOTSUPP; 2675 } 2676 EXPORT_SYMBOL(sock_no_ioctl); 2677 2678 int sock_no_listen(struct socket *sock, int backlog) 2679 { 2680 return -EOPNOTSUPP; 2681 } 2682 EXPORT_SYMBOL(sock_no_listen); 2683 2684 int sock_no_shutdown(struct socket *sock, int how) 2685 { 2686 return -EOPNOTSUPP; 2687 } 2688 EXPORT_SYMBOL(sock_no_shutdown); 2689 2690 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2691 char __user *optval, unsigned int optlen) 2692 { 2693 return -EOPNOTSUPP; 2694 } 2695 EXPORT_SYMBOL(sock_no_setsockopt); 2696 2697 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2698 char __user *optval, int __user *optlen) 2699 { 2700 return -EOPNOTSUPP; 2701 } 2702 EXPORT_SYMBOL(sock_no_getsockopt); 2703 2704 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2705 { 2706 return -EOPNOTSUPP; 2707 } 2708 EXPORT_SYMBOL(sock_no_sendmsg); 2709 2710 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2711 { 2712 return -EOPNOTSUPP; 2713 } 2714 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2715 2716 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2717 int flags) 2718 { 2719 return -EOPNOTSUPP; 2720 } 2721 EXPORT_SYMBOL(sock_no_recvmsg); 2722 2723 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2724 { 2725 /* Mirror missing mmap method error code */ 2726 return -ENODEV; 2727 } 2728 EXPORT_SYMBOL(sock_no_mmap); 2729 2730 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2731 { 2732 ssize_t res; 2733 struct msghdr msg = {.msg_flags = flags}; 2734 struct kvec iov; 2735 char *kaddr = kmap(page); 2736 iov.iov_base = kaddr + offset; 2737 iov.iov_len = size; 2738 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2739 kunmap(page); 2740 return res; 2741 } 2742 EXPORT_SYMBOL(sock_no_sendpage); 2743 2744 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2745 int offset, size_t size, int flags) 2746 { 2747 ssize_t res; 2748 struct msghdr msg = {.msg_flags = flags}; 2749 struct kvec iov; 2750 char *kaddr = kmap(page); 2751 2752 iov.iov_base = kaddr + offset; 2753 iov.iov_len = size; 2754 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2755 kunmap(page); 2756 return res; 2757 } 2758 EXPORT_SYMBOL(sock_no_sendpage_locked); 2759 2760 /* 2761 * Default Socket Callbacks 2762 */ 2763 2764 static void sock_def_wakeup(struct sock *sk) 2765 { 2766 struct socket_wq *wq; 2767 2768 rcu_read_lock(); 2769 wq = rcu_dereference(sk->sk_wq); 2770 if (skwq_has_sleeper(wq)) 2771 wake_up_interruptible_all(&wq->wait); 2772 rcu_read_unlock(); 2773 } 2774 2775 static void sock_def_error_report(struct sock *sk) 2776 { 2777 struct socket_wq *wq; 2778 2779 rcu_read_lock(); 2780 wq = rcu_dereference(sk->sk_wq); 2781 if (skwq_has_sleeper(wq)) 2782 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 2783 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2784 rcu_read_unlock(); 2785 } 2786 2787 static void sock_def_readable(struct sock *sk) 2788 { 2789 struct socket_wq *wq; 2790 2791 rcu_read_lock(); 2792 wq = rcu_dereference(sk->sk_wq); 2793 if (skwq_has_sleeper(wq)) 2794 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 2795 EPOLLRDNORM | EPOLLRDBAND); 2796 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2797 rcu_read_unlock(); 2798 } 2799 2800 static void sock_def_write_space(struct sock *sk) 2801 { 2802 struct socket_wq *wq; 2803 2804 rcu_read_lock(); 2805 2806 /* Do not wake up a writer until he can make "significant" 2807 * progress. --DaveM 2808 */ 2809 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2810 wq = rcu_dereference(sk->sk_wq); 2811 if (skwq_has_sleeper(wq)) 2812 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 2813 EPOLLWRNORM | EPOLLWRBAND); 2814 2815 /* Should agree with poll, otherwise some programs break */ 2816 if (sock_writeable(sk)) 2817 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2818 } 2819 2820 rcu_read_unlock(); 2821 } 2822 2823 static void sock_def_destruct(struct sock *sk) 2824 { 2825 } 2826 2827 void sk_send_sigurg(struct sock *sk) 2828 { 2829 if (sk->sk_socket && sk->sk_socket->file) 2830 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2831 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2832 } 2833 EXPORT_SYMBOL(sk_send_sigurg); 2834 2835 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2836 unsigned long expires) 2837 { 2838 if (!mod_timer(timer, expires)) 2839 sock_hold(sk); 2840 } 2841 EXPORT_SYMBOL(sk_reset_timer); 2842 2843 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2844 { 2845 if (del_timer(timer)) 2846 __sock_put(sk); 2847 } 2848 EXPORT_SYMBOL(sk_stop_timer); 2849 2850 void sock_init_data(struct socket *sock, struct sock *sk) 2851 { 2852 sk_init_common(sk); 2853 sk->sk_send_head = NULL; 2854 2855 timer_setup(&sk->sk_timer, NULL, 0); 2856 2857 sk->sk_allocation = GFP_KERNEL; 2858 sk->sk_rcvbuf = sysctl_rmem_default; 2859 sk->sk_sndbuf = sysctl_wmem_default; 2860 sk->sk_state = TCP_CLOSE; 2861 sk_set_socket(sk, sock); 2862 2863 sock_set_flag(sk, SOCK_ZAPPED); 2864 2865 if (sock) { 2866 sk->sk_type = sock->type; 2867 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 2868 sock->sk = sk; 2869 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2870 } else { 2871 RCU_INIT_POINTER(sk->sk_wq, NULL); 2872 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2873 } 2874 2875 rwlock_init(&sk->sk_callback_lock); 2876 if (sk->sk_kern_sock) 2877 lockdep_set_class_and_name( 2878 &sk->sk_callback_lock, 2879 af_kern_callback_keys + sk->sk_family, 2880 af_family_kern_clock_key_strings[sk->sk_family]); 2881 else 2882 lockdep_set_class_and_name( 2883 &sk->sk_callback_lock, 2884 af_callback_keys + sk->sk_family, 2885 af_family_clock_key_strings[sk->sk_family]); 2886 2887 sk->sk_state_change = sock_def_wakeup; 2888 sk->sk_data_ready = sock_def_readable; 2889 sk->sk_write_space = sock_def_write_space; 2890 sk->sk_error_report = sock_def_error_report; 2891 sk->sk_destruct = sock_def_destruct; 2892 2893 sk->sk_frag.page = NULL; 2894 sk->sk_frag.offset = 0; 2895 sk->sk_peek_off = -1; 2896 2897 sk->sk_peer_pid = NULL; 2898 sk->sk_peer_cred = NULL; 2899 sk->sk_write_pending = 0; 2900 sk->sk_rcvlowat = 1; 2901 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2902 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2903 2904 sk->sk_stamp = SK_DEFAULT_STAMP; 2905 #if BITS_PER_LONG==32 2906 seqlock_init(&sk->sk_stamp_seq); 2907 #endif 2908 atomic_set(&sk->sk_zckey, 0); 2909 2910 #ifdef CONFIG_NET_RX_BUSY_POLL 2911 sk->sk_napi_id = 0; 2912 sk->sk_ll_usec = sysctl_net_busy_read; 2913 #endif 2914 2915 sk->sk_max_pacing_rate = ~0UL; 2916 sk->sk_pacing_rate = ~0UL; 2917 sk->sk_pacing_shift = 10; 2918 sk->sk_incoming_cpu = -1; 2919 2920 sk_rx_queue_clear(sk); 2921 /* 2922 * Before updating sk_refcnt, we must commit prior changes to memory 2923 * (Documentation/RCU/rculist_nulls.txt for details) 2924 */ 2925 smp_wmb(); 2926 refcount_set(&sk->sk_refcnt, 1); 2927 atomic_set(&sk->sk_drops, 0); 2928 } 2929 EXPORT_SYMBOL(sock_init_data); 2930 2931 void lock_sock_nested(struct sock *sk, int subclass) 2932 { 2933 might_sleep(); 2934 spin_lock_bh(&sk->sk_lock.slock); 2935 if (sk->sk_lock.owned) 2936 __lock_sock(sk); 2937 sk->sk_lock.owned = 1; 2938 spin_unlock(&sk->sk_lock.slock); 2939 /* 2940 * The sk_lock has mutex_lock() semantics here: 2941 */ 2942 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2943 local_bh_enable(); 2944 } 2945 EXPORT_SYMBOL(lock_sock_nested); 2946 2947 void release_sock(struct sock *sk) 2948 { 2949 spin_lock_bh(&sk->sk_lock.slock); 2950 if (sk->sk_backlog.tail) 2951 __release_sock(sk); 2952 2953 /* Warning : release_cb() might need to release sk ownership, 2954 * ie call sock_release_ownership(sk) before us. 2955 */ 2956 if (sk->sk_prot->release_cb) 2957 sk->sk_prot->release_cb(sk); 2958 2959 sock_release_ownership(sk); 2960 if (waitqueue_active(&sk->sk_lock.wq)) 2961 wake_up(&sk->sk_lock.wq); 2962 spin_unlock_bh(&sk->sk_lock.slock); 2963 } 2964 EXPORT_SYMBOL(release_sock); 2965 2966 /** 2967 * lock_sock_fast - fast version of lock_sock 2968 * @sk: socket 2969 * 2970 * This version should be used for very small section, where process wont block 2971 * return false if fast path is taken: 2972 * 2973 * sk_lock.slock locked, owned = 0, BH disabled 2974 * 2975 * return true if slow path is taken: 2976 * 2977 * sk_lock.slock unlocked, owned = 1, BH enabled 2978 */ 2979 bool lock_sock_fast(struct sock *sk) 2980 { 2981 might_sleep(); 2982 spin_lock_bh(&sk->sk_lock.slock); 2983 2984 if (!sk->sk_lock.owned) 2985 /* 2986 * Note : We must disable BH 2987 */ 2988 return false; 2989 2990 __lock_sock(sk); 2991 sk->sk_lock.owned = 1; 2992 spin_unlock(&sk->sk_lock.slock); 2993 /* 2994 * The sk_lock has mutex_lock() semantics here: 2995 */ 2996 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2997 local_bh_enable(); 2998 return true; 2999 } 3000 EXPORT_SYMBOL(lock_sock_fast); 3001 3002 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3003 bool timeval, bool time32) 3004 { 3005 struct sock *sk = sock->sk; 3006 struct timespec64 ts; 3007 3008 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3009 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3010 if (ts.tv_sec == -1) 3011 return -ENOENT; 3012 if (ts.tv_sec == 0) { 3013 ktime_t kt = ktime_get_real(); 3014 sock_write_timestamp(sk, kt);; 3015 ts = ktime_to_timespec64(kt); 3016 } 3017 3018 if (timeval) 3019 ts.tv_nsec /= 1000; 3020 3021 #ifdef CONFIG_COMPAT_32BIT_TIME 3022 if (time32) 3023 return put_old_timespec32(&ts, userstamp); 3024 #endif 3025 #ifdef CONFIG_SPARC64 3026 /* beware of padding in sparc64 timeval */ 3027 if (timeval && !in_compat_syscall()) { 3028 struct __kernel_old_timeval __user tv = { 3029 .tv_sec = ts.tv_sec, 3030 .tv_usec = ts.tv_nsec, 3031 }; 3032 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3033 return -EFAULT; 3034 return 0; 3035 } 3036 #endif 3037 return put_timespec64(&ts, userstamp); 3038 } 3039 EXPORT_SYMBOL(sock_gettstamp); 3040 3041 void sock_enable_timestamp(struct sock *sk, int flag) 3042 { 3043 if (!sock_flag(sk, flag)) { 3044 unsigned long previous_flags = sk->sk_flags; 3045 3046 sock_set_flag(sk, flag); 3047 /* 3048 * we just set one of the two flags which require net 3049 * time stamping, but time stamping might have been on 3050 * already because of the other one 3051 */ 3052 if (sock_needs_netstamp(sk) && 3053 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3054 net_enable_timestamp(); 3055 } 3056 } 3057 3058 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3059 int level, int type) 3060 { 3061 struct sock_exterr_skb *serr; 3062 struct sk_buff *skb; 3063 int copied, err; 3064 3065 err = -EAGAIN; 3066 skb = sock_dequeue_err_skb(sk); 3067 if (skb == NULL) 3068 goto out; 3069 3070 copied = skb->len; 3071 if (copied > len) { 3072 msg->msg_flags |= MSG_TRUNC; 3073 copied = len; 3074 } 3075 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3076 if (err) 3077 goto out_free_skb; 3078 3079 sock_recv_timestamp(msg, sk, skb); 3080 3081 serr = SKB_EXT_ERR(skb); 3082 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3083 3084 msg->msg_flags |= MSG_ERRQUEUE; 3085 err = copied; 3086 3087 out_free_skb: 3088 kfree_skb(skb); 3089 out: 3090 return err; 3091 } 3092 EXPORT_SYMBOL(sock_recv_errqueue); 3093 3094 /* 3095 * Get a socket option on an socket. 3096 * 3097 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3098 * asynchronous errors should be reported by getsockopt. We assume 3099 * this means if you specify SO_ERROR (otherwise whats the point of it). 3100 */ 3101 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3102 char __user *optval, int __user *optlen) 3103 { 3104 struct sock *sk = sock->sk; 3105 3106 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3107 } 3108 EXPORT_SYMBOL(sock_common_getsockopt); 3109 3110 #ifdef CONFIG_COMPAT 3111 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 3112 char __user *optval, int __user *optlen) 3113 { 3114 struct sock *sk = sock->sk; 3115 3116 if (sk->sk_prot->compat_getsockopt != NULL) 3117 return sk->sk_prot->compat_getsockopt(sk, level, optname, 3118 optval, optlen); 3119 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3120 } 3121 EXPORT_SYMBOL(compat_sock_common_getsockopt); 3122 #endif 3123 3124 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3125 int flags) 3126 { 3127 struct sock *sk = sock->sk; 3128 int addr_len = 0; 3129 int err; 3130 3131 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3132 flags & ~MSG_DONTWAIT, &addr_len); 3133 if (err >= 0) 3134 msg->msg_namelen = addr_len; 3135 return err; 3136 } 3137 EXPORT_SYMBOL(sock_common_recvmsg); 3138 3139 /* 3140 * Set socket options on an inet socket. 3141 */ 3142 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3143 char __user *optval, unsigned int optlen) 3144 { 3145 struct sock *sk = sock->sk; 3146 3147 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3148 } 3149 EXPORT_SYMBOL(sock_common_setsockopt); 3150 3151 #ifdef CONFIG_COMPAT 3152 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 3153 char __user *optval, unsigned int optlen) 3154 { 3155 struct sock *sk = sock->sk; 3156 3157 if (sk->sk_prot->compat_setsockopt != NULL) 3158 return sk->sk_prot->compat_setsockopt(sk, level, optname, 3159 optval, optlen); 3160 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3161 } 3162 EXPORT_SYMBOL(compat_sock_common_setsockopt); 3163 #endif 3164 3165 void sk_common_release(struct sock *sk) 3166 { 3167 if (sk->sk_prot->destroy) 3168 sk->sk_prot->destroy(sk); 3169 3170 /* 3171 * Observation: when sock_common_release is called, processes have 3172 * no access to socket. But net still has. 3173 * Step one, detach it from networking: 3174 * 3175 * A. Remove from hash tables. 3176 */ 3177 3178 sk->sk_prot->unhash(sk); 3179 3180 /* 3181 * In this point socket cannot receive new packets, but it is possible 3182 * that some packets are in flight because some CPU runs receiver and 3183 * did hash table lookup before we unhashed socket. They will achieve 3184 * receive queue and will be purged by socket destructor. 3185 * 3186 * Also we still have packets pending on receive queue and probably, 3187 * our own packets waiting in device queues. sock_destroy will drain 3188 * receive queue, but transmitted packets will delay socket destruction 3189 * until the last reference will be released. 3190 */ 3191 3192 sock_orphan(sk); 3193 3194 xfrm_sk_free_policy(sk); 3195 3196 sk_refcnt_debug_release(sk); 3197 3198 sock_put(sk); 3199 } 3200 EXPORT_SYMBOL(sk_common_release); 3201 3202 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3203 { 3204 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3205 3206 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3207 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf; 3208 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3209 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf; 3210 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3211 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued; 3212 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3213 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len; 3214 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3215 } 3216 3217 #ifdef CONFIG_PROC_FS 3218 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3219 struct prot_inuse { 3220 int val[PROTO_INUSE_NR]; 3221 }; 3222 3223 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3224 3225 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3226 { 3227 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3228 } 3229 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3230 3231 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3232 { 3233 int cpu, idx = prot->inuse_idx; 3234 int res = 0; 3235 3236 for_each_possible_cpu(cpu) 3237 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3238 3239 return res >= 0 ? res : 0; 3240 } 3241 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3242 3243 static void sock_inuse_add(struct net *net, int val) 3244 { 3245 this_cpu_add(*net->core.sock_inuse, val); 3246 } 3247 3248 int sock_inuse_get(struct net *net) 3249 { 3250 int cpu, res = 0; 3251 3252 for_each_possible_cpu(cpu) 3253 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3254 3255 return res; 3256 } 3257 3258 EXPORT_SYMBOL_GPL(sock_inuse_get); 3259 3260 static int __net_init sock_inuse_init_net(struct net *net) 3261 { 3262 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3263 if (net->core.prot_inuse == NULL) 3264 return -ENOMEM; 3265 3266 net->core.sock_inuse = alloc_percpu(int); 3267 if (net->core.sock_inuse == NULL) 3268 goto out; 3269 3270 return 0; 3271 3272 out: 3273 free_percpu(net->core.prot_inuse); 3274 return -ENOMEM; 3275 } 3276 3277 static void __net_exit sock_inuse_exit_net(struct net *net) 3278 { 3279 free_percpu(net->core.prot_inuse); 3280 free_percpu(net->core.sock_inuse); 3281 } 3282 3283 static struct pernet_operations net_inuse_ops = { 3284 .init = sock_inuse_init_net, 3285 .exit = sock_inuse_exit_net, 3286 }; 3287 3288 static __init int net_inuse_init(void) 3289 { 3290 if (register_pernet_subsys(&net_inuse_ops)) 3291 panic("Cannot initialize net inuse counters"); 3292 3293 return 0; 3294 } 3295 3296 core_initcall(net_inuse_init); 3297 3298 static int assign_proto_idx(struct proto *prot) 3299 { 3300 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3301 3302 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3303 pr_err("PROTO_INUSE_NR exhausted\n"); 3304 return -ENOSPC; 3305 } 3306 3307 set_bit(prot->inuse_idx, proto_inuse_idx); 3308 return 0; 3309 } 3310 3311 static void release_proto_idx(struct proto *prot) 3312 { 3313 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3314 clear_bit(prot->inuse_idx, proto_inuse_idx); 3315 } 3316 #else 3317 static inline int assign_proto_idx(struct proto *prot) 3318 { 3319 return 0; 3320 } 3321 3322 static inline void release_proto_idx(struct proto *prot) 3323 { 3324 } 3325 3326 static void sock_inuse_add(struct net *net, int val) 3327 { 3328 } 3329 #endif 3330 3331 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3332 { 3333 if (!rsk_prot) 3334 return; 3335 kfree(rsk_prot->slab_name); 3336 rsk_prot->slab_name = NULL; 3337 kmem_cache_destroy(rsk_prot->slab); 3338 rsk_prot->slab = NULL; 3339 } 3340 3341 static int req_prot_init(const struct proto *prot) 3342 { 3343 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3344 3345 if (!rsk_prot) 3346 return 0; 3347 3348 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3349 prot->name); 3350 if (!rsk_prot->slab_name) 3351 return -ENOMEM; 3352 3353 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3354 rsk_prot->obj_size, 0, 3355 SLAB_ACCOUNT | prot->slab_flags, 3356 NULL); 3357 3358 if (!rsk_prot->slab) { 3359 pr_crit("%s: Can't create request sock SLAB cache!\n", 3360 prot->name); 3361 return -ENOMEM; 3362 } 3363 return 0; 3364 } 3365 3366 int proto_register(struct proto *prot, int alloc_slab) 3367 { 3368 int ret = -ENOBUFS; 3369 3370 if (alloc_slab) { 3371 prot->slab = kmem_cache_create_usercopy(prot->name, 3372 prot->obj_size, 0, 3373 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3374 prot->slab_flags, 3375 prot->useroffset, prot->usersize, 3376 NULL); 3377 3378 if (prot->slab == NULL) { 3379 pr_crit("%s: Can't create sock SLAB cache!\n", 3380 prot->name); 3381 goto out; 3382 } 3383 3384 if (req_prot_init(prot)) 3385 goto out_free_request_sock_slab; 3386 3387 if (prot->twsk_prot != NULL) { 3388 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3389 3390 if (prot->twsk_prot->twsk_slab_name == NULL) 3391 goto out_free_request_sock_slab; 3392 3393 prot->twsk_prot->twsk_slab = 3394 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3395 prot->twsk_prot->twsk_obj_size, 3396 0, 3397 SLAB_ACCOUNT | 3398 prot->slab_flags, 3399 NULL); 3400 if (prot->twsk_prot->twsk_slab == NULL) 3401 goto out_free_timewait_sock_slab_name; 3402 } 3403 } 3404 3405 mutex_lock(&proto_list_mutex); 3406 ret = assign_proto_idx(prot); 3407 if (ret) { 3408 mutex_unlock(&proto_list_mutex); 3409 goto out_free_timewait_sock_slab_name; 3410 } 3411 list_add(&prot->node, &proto_list); 3412 mutex_unlock(&proto_list_mutex); 3413 return ret; 3414 3415 out_free_timewait_sock_slab_name: 3416 if (alloc_slab && prot->twsk_prot) 3417 kfree(prot->twsk_prot->twsk_slab_name); 3418 out_free_request_sock_slab: 3419 if (alloc_slab) { 3420 req_prot_cleanup(prot->rsk_prot); 3421 3422 kmem_cache_destroy(prot->slab); 3423 prot->slab = NULL; 3424 } 3425 out: 3426 return ret; 3427 } 3428 EXPORT_SYMBOL(proto_register); 3429 3430 void proto_unregister(struct proto *prot) 3431 { 3432 mutex_lock(&proto_list_mutex); 3433 release_proto_idx(prot); 3434 list_del(&prot->node); 3435 mutex_unlock(&proto_list_mutex); 3436 3437 kmem_cache_destroy(prot->slab); 3438 prot->slab = NULL; 3439 3440 req_prot_cleanup(prot->rsk_prot); 3441 3442 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3443 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3444 kfree(prot->twsk_prot->twsk_slab_name); 3445 prot->twsk_prot->twsk_slab = NULL; 3446 } 3447 } 3448 EXPORT_SYMBOL(proto_unregister); 3449 3450 int sock_load_diag_module(int family, int protocol) 3451 { 3452 if (!protocol) { 3453 if (!sock_is_registered(family)) 3454 return -ENOENT; 3455 3456 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3457 NETLINK_SOCK_DIAG, family); 3458 } 3459 3460 #ifdef CONFIG_INET 3461 if (family == AF_INET && 3462 protocol != IPPROTO_RAW && 3463 !rcu_access_pointer(inet_protos[protocol])) 3464 return -ENOENT; 3465 #endif 3466 3467 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3468 NETLINK_SOCK_DIAG, family, protocol); 3469 } 3470 EXPORT_SYMBOL(sock_load_diag_module); 3471 3472 #ifdef CONFIG_PROC_FS 3473 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3474 __acquires(proto_list_mutex) 3475 { 3476 mutex_lock(&proto_list_mutex); 3477 return seq_list_start_head(&proto_list, *pos); 3478 } 3479 3480 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3481 { 3482 return seq_list_next(v, &proto_list, pos); 3483 } 3484 3485 static void proto_seq_stop(struct seq_file *seq, void *v) 3486 __releases(proto_list_mutex) 3487 { 3488 mutex_unlock(&proto_list_mutex); 3489 } 3490 3491 static char proto_method_implemented(const void *method) 3492 { 3493 return method == NULL ? 'n' : 'y'; 3494 } 3495 static long sock_prot_memory_allocated(struct proto *proto) 3496 { 3497 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3498 } 3499 3500 static const char *sock_prot_memory_pressure(struct proto *proto) 3501 { 3502 return proto->memory_pressure != NULL ? 3503 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3504 } 3505 3506 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3507 { 3508 3509 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3510 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3511 proto->name, 3512 proto->obj_size, 3513 sock_prot_inuse_get(seq_file_net(seq), proto), 3514 sock_prot_memory_allocated(proto), 3515 sock_prot_memory_pressure(proto), 3516 proto->max_header, 3517 proto->slab == NULL ? "no" : "yes", 3518 module_name(proto->owner), 3519 proto_method_implemented(proto->close), 3520 proto_method_implemented(proto->connect), 3521 proto_method_implemented(proto->disconnect), 3522 proto_method_implemented(proto->accept), 3523 proto_method_implemented(proto->ioctl), 3524 proto_method_implemented(proto->init), 3525 proto_method_implemented(proto->destroy), 3526 proto_method_implemented(proto->shutdown), 3527 proto_method_implemented(proto->setsockopt), 3528 proto_method_implemented(proto->getsockopt), 3529 proto_method_implemented(proto->sendmsg), 3530 proto_method_implemented(proto->recvmsg), 3531 proto_method_implemented(proto->sendpage), 3532 proto_method_implemented(proto->bind), 3533 proto_method_implemented(proto->backlog_rcv), 3534 proto_method_implemented(proto->hash), 3535 proto_method_implemented(proto->unhash), 3536 proto_method_implemented(proto->get_port), 3537 proto_method_implemented(proto->enter_memory_pressure)); 3538 } 3539 3540 static int proto_seq_show(struct seq_file *seq, void *v) 3541 { 3542 if (v == &proto_list) 3543 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3544 "protocol", 3545 "size", 3546 "sockets", 3547 "memory", 3548 "press", 3549 "maxhdr", 3550 "slab", 3551 "module", 3552 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3553 else 3554 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3555 return 0; 3556 } 3557 3558 static const struct seq_operations proto_seq_ops = { 3559 .start = proto_seq_start, 3560 .next = proto_seq_next, 3561 .stop = proto_seq_stop, 3562 .show = proto_seq_show, 3563 }; 3564 3565 static __net_init int proto_init_net(struct net *net) 3566 { 3567 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3568 sizeof(struct seq_net_private))) 3569 return -ENOMEM; 3570 3571 return 0; 3572 } 3573 3574 static __net_exit void proto_exit_net(struct net *net) 3575 { 3576 remove_proc_entry("protocols", net->proc_net); 3577 } 3578 3579 3580 static __net_initdata struct pernet_operations proto_net_ops = { 3581 .init = proto_init_net, 3582 .exit = proto_exit_net, 3583 }; 3584 3585 static int __init proto_init(void) 3586 { 3587 return register_pernet_subsys(&proto_net_ops); 3588 } 3589 3590 subsys_initcall(proto_init); 3591 3592 #endif /* PROC_FS */ 3593 3594 #ifdef CONFIG_NET_RX_BUSY_POLL 3595 bool sk_busy_loop_end(void *p, unsigned long start_time) 3596 { 3597 struct sock *sk = p; 3598 3599 return !skb_queue_empty(&sk->sk_receive_queue) || 3600 sk_busy_loop_timeout(sk, start_time); 3601 } 3602 EXPORT_SYMBOL(sk_busy_loop_end); 3603 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3604