1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Generic socket support routines. Memory allocators, socket lock/release 8 * handler for protocols to use and generic option handler. 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Florian La Roche, <flla@stud.uni-sb.de> 13 * Alan Cox, <A.Cox@swansea.ac.uk> 14 * 15 * Fixes: 16 * Alan Cox : Numerous verify_area() problems 17 * Alan Cox : Connecting on a connecting socket 18 * now returns an error for tcp. 19 * Alan Cox : sock->protocol is set correctly. 20 * and is not sometimes left as 0. 21 * Alan Cox : connect handles icmp errors on a 22 * connect properly. Unfortunately there 23 * is a restart syscall nasty there. I 24 * can't match BSD without hacking the C 25 * library. Ideas urgently sought! 26 * Alan Cox : Disallow bind() to addresses that are 27 * not ours - especially broadcast ones!! 28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost) 29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets, 30 * instead they leave that for the DESTROY timer. 31 * Alan Cox : Clean up error flag in accept 32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer 33 * was buggy. Put a remove_sock() in the handler 34 * for memory when we hit 0. Also altered the timer 35 * code. The ACK stuff can wait and needs major 36 * TCP layer surgery. 37 * Alan Cox : Fixed TCP ack bug, removed remove sock 38 * and fixed timer/inet_bh race. 39 * Alan Cox : Added zapped flag for TCP 40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code 41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb 42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources 43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing. 44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so... 45 * Rick Sladkey : Relaxed UDP rules for matching packets. 46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support 47 * Pauline Middelink : identd support 48 * Alan Cox : Fixed connect() taking signals I think. 49 * Alan Cox : SO_LINGER supported 50 * Alan Cox : Error reporting fixes 51 * Anonymous : inet_create tidied up (sk->reuse setting) 52 * Alan Cox : inet sockets don't set sk->type! 53 * Alan Cox : Split socket option code 54 * Alan Cox : Callbacks 55 * Alan Cox : Nagle flag for Charles & Johannes stuff 56 * Alex : Removed restriction on inet fioctl 57 * Alan Cox : Splitting INET from NET core 58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt() 59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code 60 * Alan Cox : Split IP from generic code 61 * Alan Cox : New kfree_skbmem() 62 * Alan Cox : Make SO_DEBUG superuser only. 63 * Alan Cox : Allow anyone to clear SO_DEBUG 64 * (compatibility fix) 65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput. 66 * Alan Cox : Allocator for a socket is settable. 67 * Alan Cox : SO_ERROR includes soft errors. 68 * Alan Cox : Allow NULL arguments on some SO_ opts 69 * Alan Cox : Generic socket allocation to make hooks 70 * easier (suggested by Craig Metz). 71 * Michael Pall : SO_ERROR returns positive errno again 72 * Steve Whitehouse: Added default destructor to free 73 * protocol private data. 74 * Steve Whitehouse: Added various other default routines 75 * common to several socket families. 76 * Chris Evans : Call suser() check last on F_SETOWN 77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER. 78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s() 79 * Andi Kleen : Fix write_space callback 80 * Chris Evans : Security fixes - signedness again 81 * Arnaldo C. Melo : cleanups, use skb_queue_purge 82 * 83 * To Fix: 84 */ 85 86 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 87 88 #include <asm/unaligned.h> 89 #include <linux/capability.h> 90 #include <linux/errno.h> 91 #include <linux/errqueue.h> 92 #include <linux/types.h> 93 #include <linux/socket.h> 94 #include <linux/in.h> 95 #include <linux/kernel.h> 96 #include <linux/module.h> 97 #include <linux/proc_fs.h> 98 #include <linux/seq_file.h> 99 #include <linux/sched.h> 100 #include <linux/sched/mm.h> 101 #include <linux/timer.h> 102 #include <linux/string.h> 103 #include <linux/sockios.h> 104 #include <linux/net.h> 105 #include <linux/mm.h> 106 #include <linux/slab.h> 107 #include <linux/interrupt.h> 108 #include <linux/poll.h> 109 #include <linux/tcp.h> 110 #include <linux/init.h> 111 #include <linux/highmem.h> 112 #include <linux/user_namespace.h> 113 #include <linux/static_key.h> 114 #include <linux/memcontrol.h> 115 #include <linux/prefetch.h> 116 117 #include <linux/uaccess.h> 118 119 #include <linux/netdevice.h> 120 #include <net/protocol.h> 121 #include <linux/skbuff.h> 122 #include <net/net_namespace.h> 123 #include <net/request_sock.h> 124 #include <net/sock.h> 125 #include <linux/net_tstamp.h> 126 #include <net/xfrm.h> 127 #include <linux/ipsec.h> 128 #include <net/cls_cgroup.h> 129 #include <net/netprio_cgroup.h> 130 #include <linux/sock_diag.h> 131 132 #include <linux/filter.h> 133 #include <net/sock_reuseport.h> 134 #include <net/bpf_sk_storage.h> 135 136 #include <trace/events/sock.h> 137 138 #include <net/tcp.h> 139 #include <net/busy_poll.h> 140 141 static DEFINE_MUTEX(proto_list_mutex); 142 static LIST_HEAD(proto_list); 143 144 static void sock_inuse_add(struct net *net, int val); 145 146 /** 147 * sk_ns_capable - General socket capability test 148 * @sk: Socket to use a capability on or through 149 * @user_ns: The user namespace of the capability to use 150 * @cap: The capability to use 151 * 152 * Test to see if the opener of the socket had when the socket was 153 * created and the current process has the capability @cap in the user 154 * namespace @user_ns. 155 */ 156 bool sk_ns_capable(const struct sock *sk, 157 struct user_namespace *user_ns, int cap) 158 { 159 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 160 ns_capable(user_ns, cap); 161 } 162 EXPORT_SYMBOL(sk_ns_capable); 163 164 /** 165 * sk_capable - Socket global capability test 166 * @sk: Socket to use a capability on or through 167 * @cap: The global capability to use 168 * 169 * Test to see if the opener of the socket had when the socket was 170 * created and the current process has the capability @cap in all user 171 * namespaces. 172 */ 173 bool sk_capable(const struct sock *sk, int cap) 174 { 175 return sk_ns_capable(sk, &init_user_ns, cap); 176 } 177 EXPORT_SYMBOL(sk_capable); 178 179 /** 180 * sk_net_capable - Network namespace socket capability test 181 * @sk: Socket to use a capability on or through 182 * @cap: The capability to use 183 * 184 * Test to see if the opener of the socket had when the socket was created 185 * and the current process has the capability @cap over the network namespace 186 * the socket is a member of. 187 */ 188 bool sk_net_capable(const struct sock *sk, int cap) 189 { 190 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 191 } 192 EXPORT_SYMBOL(sk_net_capable); 193 194 /* 195 * Each address family might have different locking rules, so we have 196 * one slock key per address family and separate keys for internal and 197 * userspace sockets. 198 */ 199 static struct lock_class_key af_family_keys[AF_MAX]; 200 static struct lock_class_key af_family_kern_keys[AF_MAX]; 201 static struct lock_class_key af_family_slock_keys[AF_MAX]; 202 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 203 204 /* 205 * Make lock validator output more readable. (we pre-construct these 206 * strings build-time, so that runtime initialization of socket 207 * locks is fast): 208 */ 209 210 #define _sock_locks(x) \ 211 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 212 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 213 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 214 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 215 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 216 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 217 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 218 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 219 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 220 x "27" , x "28" , x "AF_CAN" , \ 221 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 222 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 223 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 224 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 225 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 226 x "AF_MAX" 227 228 static const char *const af_family_key_strings[AF_MAX+1] = { 229 _sock_locks("sk_lock-") 230 }; 231 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 232 _sock_locks("slock-") 233 }; 234 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 235 _sock_locks("clock-") 236 }; 237 238 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 239 _sock_locks("k-sk_lock-") 240 }; 241 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 242 _sock_locks("k-slock-") 243 }; 244 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 245 _sock_locks("k-clock-") 246 }; 247 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 248 _sock_locks("rlock-") 249 }; 250 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 251 _sock_locks("wlock-") 252 }; 253 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 254 _sock_locks("elock-") 255 }; 256 257 /* 258 * sk_callback_lock and sk queues locking rules are per-address-family, 259 * so split the lock classes by using a per-AF key: 260 */ 261 static struct lock_class_key af_callback_keys[AF_MAX]; 262 static struct lock_class_key af_rlock_keys[AF_MAX]; 263 static struct lock_class_key af_wlock_keys[AF_MAX]; 264 static struct lock_class_key af_elock_keys[AF_MAX]; 265 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 266 267 /* Run time adjustable parameters. */ 268 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 269 EXPORT_SYMBOL(sysctl_wmem_max); 270 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 271 EXPORT_SYMBOL(sysctl_rmem_max); 272 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 273 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 274 275 /* Maximal space eaten by iovec or ancillary data plus some space */ 276 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 277 EXPORT_SYMBOL(sysctl_optmem_max); 278 279 int sysctl_tstamp_allow_data __read_mostly = 1; 280 281 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 282 EXPORT_SYMBOL_GPL(memalloc_socks_key); 283 284 /** 285 * sk_set_memalloc - sets %SOCK_MEMALLOC 286 * @sk: socket to set it on 287 * 288 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 289 * It's the responsibility of the admin to adjust min_free_kbytes 290 * to meet the requirements 291 */ 292 void sk_set_memalloc(struct sock *sk) 293 { 294 sock_set_flag(sk, SOCK_MEMALLOC); 295 sk->sk_allocation |= __GFP_MEMALLOC; 296 static_branch_inc(&memalloc_socks_key); 297 } 298 EXPORT_SYMBOL_GPL(sk_set_memalloc); 299 300 void sk_clear_memalloc(struct sock *sk) 301 { 302 sock_reset_flag(sk, SOCK_MEMALLOC); 303 sk->sk_allocation &= ~__GFP_MEMALLOC; 304 static_branch_dec(&memalloc_socks_key); 305 306 /* 307 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 308 * progress of swapping. SOCK_MEMALLOC may be cleared while 309 * it has rmem allocations due to the last swapfile being deactivated 310 * but there is a risk that the socket is unusable due to exceeding 311 * the rmem limits. Reclaim the reserves and obey rmem limits again. 312 */ 313 sk_mem_reclaim(sk); 314 } 315 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 316 317 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 318 { 319 int ret; 320 unsigned int noreclaim_flag; 321 322 /* these should have been dropped before queueing */ 323 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 324 325 noreclaim_flag = memalloc_noreclaim_save(); 326 ret = sk->sk_backlog_rcv(sk, skb); 327 memalloc_noreclaim_restore(noreclaim_flag); 328 329 return ret; 330 } 331 EXPORT_SYMBOL(__sk_backlog_rcv); 332 333 static int sock_get_timeout(long timeo, void *optval, bool old_timeval) 334 { 335 struct __kernel_sock_timeval tv; 336 337 if (timeo == MAX_SCHEDULE_TIMEOUT) { 338 tv.tv_sec = 0; 339 tv.tv_usec = 0; 340 } else { 341 tv.tv_sec = timeo / HZ; 342 tv.tv_usec = ((timeo % HZ) * USEC_PER_SEC) / HZ; 343 } 344 345 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 346 struct old_timeval32 tv32 = { tv.tv_sec, tv.tv_usec }; 347 *(struct old_timeval32 *)optval = tv32; 348 return sizeof(tv32); 349 } 350 351 if (old_timeval) { 352 struct __kernel_old_timeval old_tv; 353 old_tv.tv_sec = tv.tv_sec; 354 old_tv.tv_usec = tv.tv_usec; 355 *(struct __kernel_old_timeval *)optval = old_tv; 356 return sizeof(old_tv); 357 } 358 359 *(struct __kernel_sock_timeval *)optval = tv; 360 return sizeof(tv); 361 } 362 363 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen, bool old_timeval) 364 { 365 struct __kernel_sock_timeval tv; 366 367 if (old_timeval && in_compat_syscall() && !COMPAT_USE_64BIT_TIME) { 368 struct old_timeval32 tv32; 369 370 if (optlen < sizeof(tv32)) 371 return -EINVAL; 372 373 if (copy_from_user(&tv32, optval, sizeof(tv32))) 374 return -EFAULT; 375 tv.tv_sec = tv32.tv_sec; 376 tv.tv_usec = tv32.tv_usec; 377 } else if (old_timeval) { 378 struct __kernel_old_timeval old_tv; 379 380 if (optlen < sizeof(old_tv)) 381 return -EINVAL; 382 if (copy_from_user(&old_tv, optval, sizeof(old_tv))) 383 return -EFAULT; 384 tv.tv_sec = old_tv.tv_sec; 385 tv.tv_usec = old_tv.tv_usec; 386 } else { 387 if (optlen < sizeof(tv)) 388 return -EINVAL; 389 if (copy_from_user(&tv, optval, sizeof(tv))) 390 return -EFAULT; 391 } 392 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 393 return -EDOM; 394 395 if (tv.tv_sec < 0) { 396 static int warned __read_mostly; 397 398 *timeo_p = 0; 399 if (warned < 10 && net_ratelimit()) { 400 warned++; 401 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 402 __func__, current->comm, task_pid_nr(current)); 403 } 404 return 0; 405 } 406 *timeo_p = MAX_SCHEDULE_TIMEOUT; 407 if (tv.tv_sec == 0 && tv.tv_usec == 0) 408 return 0; 409 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT / HZ - 1)) 410 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP((unsigned long)tv.tv_usec, USEC_PER_SEC / HZ); 411 return 0; 412 } 413 414 static void sock_warn_obsolete_bsdism(const char *name) 415 { 416 static int warned; 417 static char warncomm[TASK_COMM_LEN]; 418 if (strcmp(warncomm, current->comm) && warned < 5) { 419 strcpy(warncomm, current->comm); 420 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 421 warncomm, name); 422 warned++; 423 } 424 } 425 426 static bool sock_needs_netstamp(const struct sock *sk) 427 { 428 switch (sk->sk_family) { 429 case AF_UNSPEC: 430 case AF_UNIX: 431 return false; 432 default: 433 return true; 434 } 435 } 436 437 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 438 { 439 if (sk->sk_flags & flags) { 440 sk->sk_flags &= ~flags; 441 if (sock_needs_netstamp(sk) && 442 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 443 net_disable_timestamp(); 444 } 445 } 446 447 448 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 449 { 450 unsigned long flags; 451 struct sk_buff_head *list = &sk->sk_receive_queue; 452 453 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 454 atomic_inc(&sk->sk_drops); 455 trace_sock_rcvqueue_full(sk, skb); 456 return -ENOMEM; 457 } 458 459 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 460 atomic_inc(&sk->sk_drops); 461 return -ENOBUFS; 462 } 463 464 skb->dev = NULL; 465 skb_set_owner_r(skb, sk); 466 467 /* we escape from rcu protected region, make sure we dont leak 468 * a norefcounted dst 469 */ 470 skb_dst_force(skb); 471 472 spin_lock_irqsave(&list->lock, flags); 473 sock_skb_set_dropcount(sk, skb); 474 __skb_queue_tail(list, skb); 475 spin_unlock_irqrestore(&list->lock, flags); 476 477 if (!sock_flag(sk, SOCK_DEAD)) 478 sk->sk_data_ready(sk); 479 return 0; 480 } 481 EXPORT_SYMBOL(__sock_queue_rcv_skb); 482 483 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 484 { 485 int err; 486 487 err = sk_filter(sk, skb); 488 if (err) 489 return err; 490 491 return __sock_queue_rcv_skb(sk, skb); 492 } 493 EXPORT_SYMBOL(sock_queue_rcv_skb); 494 495 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 496 const int nested, unsigned int trim_cap, bool refcounted) 497 { 498 int rc = NET_RX_SUCCESS; 499 500 if (sk_filter_trim_cap(sk, skb, trim_cap)) 501 goto discard_and_relse; 502 503 skb->dev = NULL; 504 505 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 506 atomic_inc(&sk->sk_drops); 507 goto discard_and_relse; 508 } 509 if (nested) 510 bh_lock_sock_nested(sk); 511 else 512 bh_lock_sock(sk); 513 if (!sock_owned_by_user(sk)) { 514 /* 515 * trylock + unlock semantics: 516 */ 517 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 518 519 rc = sk_backlog_rcv(sk, skb); 520 521 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 522 } else if (sk_add_backlog(sk, skb, READ_ONCE(sk->sk_rcvbuf))) { 523 bh_unlock_sock(sk); 524 atomic_inc(&sk->sk_drops); 525 goto discard_and_relse; 526 } 527 528 bh_unlock_sock(sk); 529 out: 530 if (refcounted) 531 sock_put(sk); 532 return rc; 533 discard_and_relse: 534 kfree_skb(skb); 535 goto out; 536 } 537 EXPORT_SYMBOL(__sk_receive_skb); 538 539 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 540 { 541 struct dst_entry *dst = __sk_dst_get(sk); 542 543 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 544 sk_tx_queue_clear(sk); 545 sk->sk_dst_pending_confirm = 0; 546 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 547 dst_release(dst); 548 return NULL; 549 } 550 551 return dst; 552 } 553 EXPORT_SYMBOL(__sk_dst_check); 554 555 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 556 { 557 struct dst_entry *dst = sk_dst_get(sk); 558 559 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 560 sk_dst_reset(sk); 561 dst_release(dst); 562 return NULL; 563 } 564 565 return dst; 566 } 567 EXPORT_SYMBOL(sk_dst_check); 568 569 static int sock_setbindtodevice_locked(struct sock *sk, int ifindex) 570 { 571 int ret = -ENOPROTOOPT; 572 #ifdef CONFIG_NETDEVICES 573 struct net *net = sock_net(sk); 574 575 /* Sorry... */ 576 ret = -EPERM; 577 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 578 goto out; 579 580 ret = -EINVAL; 581 if (ifindex < 0) 582 goto out; 583 584 sk->sk_bound_dev_if = ifindex; 585 if (sk->sk_prot->rehash) 586 sk->sk_prot->rehash(sk); 587 sk_dst_reset(sk); 588 589 ret = 0; 590 591 out: 592 #endif 593 594 return ret; 595 } 596 597 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 598 int optlen) 599 { 600 int ret = -ENOPROTOOPT; 601 #ifdef CONFIG_NETDEVICES 602 struct net *net = sock_net(sk); 603 char devname[IFNAMSIZ]; 604 int index; 605 606 ret = -EINVAL; 607 if (optlen < 0) 608 goto out; 609 610 /* Bind this socket to a particular device like "eth0", 611 * as specified in the passed interface name. If the 612 * name is "" or the option length is zero the socket 613 * is not bound. 614 */ 615 if (optlen > IFNAMSIZ - 1) 616 optlen = IFNAMSIZ - 1; 617 memset(devname, 0, sizeof(devname)); 618 619 ret = -EFAULT; 620 if (copy_from_user(devname, optval, optlen)) 621 goto out; 622 623 index = 0; 624 if (devname[0] != '\0') { 625 struct net_device *dev; 626 627 rcu_read_lock(); 628 dev = dev_get_by_name_rcu(net, devname); 629 if (dev) 630 index = dev->ifindex; 631 rcu_read_unlock(); 632 ret = -ENODEV; 633 if (!dev) 634 goto out; 635 } 636 637 lock_sock(sk); 638 ret = sock_setbindtodevice_locked(sk, index); 639 release_sock(sk); 640 641 out: 642 #endif 643 644 return ret; 645 } 646 647 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 648 int __user *optlen, int len) 649 { 650 int ret = -ENOPROTOOPT; 651 #ifdef CONFIG_NETDEVICES 652 struct net *net = sock_net(sk); 653 char devname[IFNAMSIZ]; 654 655 if (sk->sk_bound_dev_if == 0) { 656 len = 0; 657 goto zero; 658 } 659 660 ret = -EINVAL; 661 if (len < IFNAMSIZ) 662 goto out; 663 664 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 665 if (ret) 666 goto out; 667 668 len = strlen(devname) + 1; 669 670 ret = -EFAULT; 671 if (copy_to_user(optval, devname, len)) 672 goto out; 673 674 zero: 675 ret = -EFAULT; 676 if (put_user(len, optlen)) 677 goto out; 678 679 ret = 0; 680 681 out: 682 #endif 683 684 return ret; 685 } 686 687 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 688 int valbool) 689 { 690 if (valbool) 691 sock_set_flag(sk, bit); 692 else 693 sock_reset_flag(sk, bit); 694 } 695 696 bool sk_mc_loop(struct sock *sk) 697 { 698 if (dev_recursion_level()) 699 return false; 700 if (!sk) 701 return true; 702 switch (sk->sk_family) { 703 case AF_INET: 704 return inet_sk(sk)->mc_loop; 705 #if IS_ENABLED(CONFIG_IPV6) 706 case AF_INET6: 707 return inet6_sk(sk)->mc_loop; 708 #endif 709 } 710 WARN_ON(1); 711 return true; 712 } 713 EXPORT_SYMBOL(sk_mc_loop); 714 715 /* 716 * This is meant for all protocols to use and covers goings on 717 * at the socket level. Everything here is generic. 718 */ 719 720 int sock_setsockopt(struct socket *sock, int level, int optname, 721 char __user *optval, unsigned int optlen) 722 { 723 struct sock_txtime sk_txtime; 724 struct sock *sk = sock->sk; 725 int val; 726 int valbool; 727 struct linger ling; 728 int ret = 0; 729 730 /* 731 * Options without arguments 732 */ 733 734 if (optname == SO_BINDTODEVICE) 735 return sock_setbindtodevice(sk, optval, optlen); 736 737 if (optlen < sizeof(int)) 738 return -EINVAL; 739 740 if (get_user(val, (int __user *)optval)) 741 return -EFAULT; 742 743 valbool = val ? 1 : 0; 744 745 lock_sock(sk); 746 747 switch (optname) { 748 case SO_DEBUG: 749 if (val && !capable(CAP_NET_ADMIN)) 750 ret = -EACCES; 751 else 752 sock_valbool_flag(sk, SOCK_DBG, valbool); 753 break; 754 case SO_REUSEADDR: 755 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 756 break; 757 case SO_REUSEPORT: 758 sk->sk_reuseport = valbool; 759 break; 760 case SO_TYPE: 761 case SO_PROTOCOL: 762 case SO_DOMAIN: 763 case SO_ERROR: 764 ret = -ENOPROTOOPT; 765 break; 766 case SO_DONTROUTE: 767 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 768 sk_dst_reset(sk); 769 break; 770 case SO_BROADCAST: 771 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 772 break; 773 case SO_SNDBUF: 774 /* Don't error on this BSD doesn't and if you think 775 * about it this is right. Otherwise apps have to 776 * play 'guess the biggest size' games. RCVBUF/SNDBUF 777 * are treated in BSD as hints 778 */ 779 val = min_t(u32, val, sysctl_wmem_max); 780 set_sndbuf: 781 /* Ensure val * 2 fits into an int, to prevent max_t() 782 * from treating it as a negative value. 783 */ 784 val = min_t(int, val, INT_MAX / 2); 785 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 786 WRITE_ONCE(sk->sk_sndbuf, 787 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 788 /* Wake up sending tasks if we upped the value. */ 789 sk->sk_write_space(sk); 790 break; 791 792 case SO_SNDBUFFORCE: 793 if (!capable(CAP_NET_ADMIN)) { 794 ret = -EPERM; 795 break; 796 } 797 798 /* No negative values (to prevent underflow, as val will be 799 * multiplied by 2). 800 */ 801 if (val < 0) 802 val = 0; 803 goto set_sndbuf; 804 805 case SO_RCVBUF: 806 /* Don't error on this BSD doesn't and if you think 807 * about it this is right. Otherwise apps have to 808 * play 'guess the biggest size' games. RCVBUF/SNDBUF 809 * are treated in BSD as hints 810 */ 811 val = min_t(u32, val, sysctl_rmem_max); 812 set_rcvbuf: 813 /* Ensure val * 2 fits into an int, to prevent max_t() 814 * from treating it as a negative value. 815 */ 816 val = min_t(int, val, INT_MAX / 2); 817 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 818 /* 819 * We double it on the way in to account for 820 * "struct sk_buff" etc. overhead. Applications 821 * assume that the SO_RCVBUF setting they make will 822 * allow that much actual data to be received on that 823 * socket. 824 * 825 * Applications are unaware that "struct sk_buff" and 826 * other overheads allocate from the receive buffer 827 * during socket buffer allocation. 828 * 829 * And after considering the possible alternatives, 830 * returning the value we actually used in getsockopt 831 * is the most desirable behavior. 832 */ 833 WRITE_ONCE(sk->sk_rcvbuf, 834 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 WRITE_ONCE(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 WRITE_ONCE(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 = READ_ONCE(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 1834 /* sk->sk_memcg will be populated at accept() time */ 1835 newsk->sk_memcg = NULL; 1836 1837 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1838 1839 rcu_read_lock(); 1840 filter = rcu_dereference(sk->sk_filter); 1841 if (filter != NULL) 1842 /* though it's an empty new sock, the charging may fail 1843 * if sysctl_optmem_max was changed between creation of 1844 * original socket and cloning 1845 */ 1846 is_charged = sk_filter_charge(newsk, filter); 1847 RCU_INIT_POINTER(newsk->sk_filter, filter); 1848 rcu_read_unlock(); 1849 1850 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1851 /* We need to make sure that we don't uncharge the new 1852 * socket if we couldn't charge it in the first place 1853 * as otherwise we uncharge the parent's filter. 1854 */ 1855 if (!is_charged) 1856 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1857 sk_free_unlock_clone(newsk); 1858 newsk = NULL; 1859 goto out; 1860 } 1861 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1862 1863 if (bpf_sk_storage_clone(sk, newsk)) { 1864 sk_free_unlock_clone(newsk); 1865 newsk = NULL; 1866 goto out; 1867 } 1868 1869 newsk->sk_err = 0; 1870 newsk->sk_err_soft = 0; 1871 newsk->sk_priority = 0; 1872 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1873 if (likely(newsk->sk_net_refcnt)) 1874 sock_inuse_add(sock_net(newsk), 1); 1875 1876 /* 1877 * Before updating sk_refcnt, we must commit prior changes to memory 1878 * (Documentation/RCU/rculist_nulls.txt for details) 1879 */ 1880 smp_wmb(); 1881 refcount_set(&newsk->sk_refcnt, 2); 1882 1883 /* 1884 * Increment the counter in the same struct proto as the master 1885 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1886 * is the same as sk->sk_prot->socks, as this field was copied 1887 * with memcpy). 1888 * 1889 * This _changes_ the previous behaviour, where 1890 * tcp_create_openreq_child always was incrementing the 1891 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1892 * to be taken into account in all callers. -acme 1893 */ 1894 sk_refcnt_debug_inc(newsk); 1895 sk_set_socket(newsk, NULL); 1896 RCU_INIT_POINTER(newsk->sk_wq, NULL); 1897 1898 if (newsk->sk_prot->sockets_allocated) 1899 sk_sockets_allocated_inc(newsk); 1900 1901 if (sock_needs_netstamp(sk) && 1902 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1903 net_enable_timestamp(); 1904 } 1905 out: 1906 return newsk; 1907 } 1908 EXPORT_SYMBOL_GPL(sk_clone_lock); 1909 1910 void sk_free_unlock_clone(struct sock *sk) 1911 { 1912 /* It is still raw copy of parent, so invalidate 1913 * destructor and make plain sk_free() */ 1914 sk->sk_destruct = NULL; 1915 bh_unlock_sock(sk); 1916 sk_free(sk); 1917 } 1918 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1919 1920 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1921 { 1922 u32 max_segs = 1; 1923 1924 sk_dst_set(sk, dst); 1925 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 1926 if (sk->sk_route_caps & NETIF_F_GSO) 1927 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1928 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1929 if (sk_can_gso(sk)) { 1930 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 1931 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1932 } else { 1933 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1934 sk->sk_gso_max_size = dst->dev->gso_max_size; 1935 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1936 } 1937 } 1938 sk->sk_gso_max_segs = max_segs; 1939 } 1940 EXPORT_SYMBOL_GPL(sk_setup_caps); 1941 1942 /* 1943 * Simple resource managers for sockets. 1944 */ 1945 1946 1947 /* 1948 * Write buffer destructor automatically called from kfree_skb. 1949 */ 1950 void sock_wfree(struct sk_buff *skb) 1951 { 1952 struct sock *sk = skb->sk; 1953 unsigned int len = skb->truesize; 1954 1955 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1956 /* 1957 * Keep a reference on sk_wmem_alloc, this will be released 1958 * after sk_write_space() call 1959 */ 1960 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 1961 sk->sk_write_space(sk); 1962 len = 1; 1963 } 1964 /* 1965 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1966 * could not do because of in-flight packets 1967 */ 1968 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 1969 __sk_free(sk); 1970 } 1971 EXPORT_SYMBOL(sock_wfree); 1972 1973 /* This variant of sock_wfree() is used by TCP, 1974 * since it sets SOCK_USE_WRITE_QUEUE. 1975 */ 1976 void __sock_wfree(struct sk_buff *skb) 1977 { 1978 struct sock *sk = skb->sk; 1979 1980 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1981 __sk_free(sk); 1982 } 1983 1984 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1985 { 1986 skb_orphan(skb); 1987 skb->sk = sk; 1988 #ifdef CONFIG_INET 1989 if (unlikely(!sk_fullsock(sk))) { 1990 skb->destructor = sock_edemux; 1991 sock_hold(sk); 1992 return; 1993 } 1994 #endif 1995 skb->destructor = sock_wfree; 1996 skb_set_hash_from_sk(skb, sk); 1997 /* 1998 * We used to take a refcount on sk, but following operation 1999 * is enough to guarantee sk_free() wont free this sock until 2000 * all in-flight packets are completed 2001 */ 2002 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2003 } 2004 EXPORT_SYMBOL(skb_set_owner_w); 2005 2006 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2007 { 2008 #ifdef CONFIG_TLS_DEVICE 2009 /* Drivers depend on in-order delivery for crypto offload, 2010 * partial orphan breaks out-of-order-OK logic. 2011 */ 2012 if (skb->decrypted) 2013 return false; 2014 #endif 2015 return (skb->destructor == sock_wfree || 2016 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2017 } 2018 2019 /* This helper is used by netem, as it can hold packets in its 2020 * delay queue. We want to allow the owner socket to send more 2021 * packets, as if they were already TX completed by a typical driver. 2022 * But we also want to keep skb->sk set because some packet schedulers 2023 * rely on it (sch_fq for example). 2024 */ 2025 void skb_orphan_partial(struct sk_buff *skb) 2026 { 2027 if (skb_is_tcp_pure_ack(skb)) 2028 return; 2029 2030 if (can_skb_orphan_partial(skb)) { 2031 struct sock *sk = skb->sk; 2032 2033 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 2034 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)); 2035 skb->destructor = sock_efree; 2036 } 2037 } else { 2038 skb_orphan(skb); 2039 } 2040 } 2041 EXPORT_SYMBOL(skb_orphan_partial); 2042 2043 /* 2044 * Read buffer destructor automatically called from kfree_skb. 2045 */ 2046 void sock_rfree(struct sk_buff *skb) 2047 { 2048 struct sock *sk = skb->sk; 2049 unsigned int len = skb->truesize; 2050 2051 atomic_sub(len, &sk->sk_rmem_alloc); 2052 sk_mem_uncharge(sk, len); 2053 } 2054 EXPORT_SYMBOL(sock_rfree); 2055 2056 /* 2057 * Buffer destructor for skbs that are not used directly in read or write 2058 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2059 */ 2060 void sock_efree(struct sk_buff *skb) 2061 { 2062 sock_put(skb->sk); 2063 } 2064 EXPORT_SYMBOL(sock_efree); 2065 2066 kuid_t sock_i_uid(struct sock *sk) 2067 { 2068 kuid_t uid; 2069 2070 read_lock_bh(&sk->sk_callback_lock); 2071 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2072 read_unlock_bh(&sk->sk_callback_lock); 2073 return uid; 2074 } 2075 EXPORT_SYMBOL(sock_i_uid); 2076 2077 unsigned long sock_i_ino(struct sock *sk) 2078 { 2079 unsigned long ino; 2080 2081 read_lock_bh(&sk->sk_callback_lock); 2082 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2083 read_unlock_bh(&sk->sk_callback_lock); 2084 return ino; 2085 } 2086 EXPORT_SYMBOL(sock_i_ino); 2087 2088 /* 2089 * Allocate a skb from the socket's send buffer. 2090 */ 2091 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2092 gfp_t priority) 2093 { 2094 if (force || 2095 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2096 struct sk_buff *skb = alloc_skb(size, priority); 2097 2098 if (skb) { 2099 skb_set_owner_w(skb, sk); 2100 return skb; 2101 } 2102 } 2103 return NULL; 2104 } 2105 EXPORT_SYMBOL(sock_wmalloc); 2106 2107 static void sock_ofree(struct sk_buff *skb) 2108 { 2109 struct sock *sk = skb->sk; 2110 2111 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2112 } 2113 2114 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2115 gfp_t priority) 2116 { 2117 struct sk_buff *skb; 2118 2119 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2120 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2121 sysctl_optmem_max) 2122 return NULL; 2123 2124 skb = alloc_skb(size, priority); 2125 if (!skb) 2126 return NULL; 2127 2128 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2129 skb->sk = sk; 2130 skb->destructor = sock_ofree; 2131 return skb; 2132 } 2133 2134 /* 2135 * Allocate a memory block from the socket's option memory buffer. 2136 */ 2137 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2138 { 2139 if ((unsigned int)size <= sysctl_optmem_max && 2140 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2141 void *mem; 2142 /* First do the add, to avoid the race if kmalloc 2143 * might sleep. 2144 */ 2145 atomic_add(size, &sk->sk_omem_alloc); 2146 mem = kmalloc(size, priority); 2147 if (mem) 2148 return mem; 2149 atomic_sub(size, &sk->sk_omem_alloc); 2150 } 2151 return NULL; 2152 } 2153 EXPORT_SYMBOL(sock_kmalloc); 2154 2155 /* Free an option memory block. Note, we actually want the inline 2156 * here as this allows gcc to detect the nullify and fold away the 2157 * condition entirely. 2158 */ 2159 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2160 const bool nullify) 2161 { 2162 if (WARN_ON_ONCE(!mem)) 2163 return; 2164 if (nullify) 2165 kzfree(mem); 2166 else 2167 kfree(mem); 2168 atomic_sub(size, &sk->sk_omem_alloc); 2169 } 2170 2171 void sock_kfree_s(struct sock *sk, void *mem, int size) 2172 { 2173 __sock_kfree_s(sk, mem, size, false); 2174 } 2175 EXPORT_SYMBOL(sock_kfree_s); 2176 2177 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2178 { 2179 __sock_kfree_s(sk, mem, size, true); 2180 } 2181 EXPORT_SYMBOL(sock_kzfree_s); 2182 2183 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2184 I think, these locks should be removed for datagram sockets. 2185 */ 2186 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2187 { 2188 DEFINE_WAIT(wait); 2189 2190 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2191 for (;;) { 2192 if (!timeo) 2193 break; 2194 if (signal_pending(current)) 2195 break; 2196 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2197 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2198 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2199 break; 2200 if (sk->sk_shutdown & SEND_SHUTDOWN) 2201 break; 2202 if (sk->sk_err) 2203 break; 2204 timeo = schedule_timeout(timeo); 2205 } 2206 finish_wait(sk_sleep(sk), &wait); 2207 return timeo; 2208 } 2209 2210 2211 /* 2212 * Generic send/receive buffer handlers 2213 */ 2214 2215 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2216 unsigned long data_len, int noblock, 2217 int *errcode, int max_page_order) 2218 { 2219 struct sk_buff *skb; 2220 long timeo; 2221 int err; 2222 2223 timeo = sock_sndtimeo(sk, noblock); 2224 for (;;) { 2225 err = sock_error(sk); 2226 if (err != 0) 2227 goto failure; 2228 2229 err = -EPIPE; 2230 if (sk->sk_shutdown & SEND_SHUTDOWN) 2231 goto failure; 2232 2233 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2234 break; 2235 2236 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2237 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2238 err = -EAGAIN; 2239 if (!timeo) 2240 goto failure; 2241 if (signal_pending(current)) 2242 goto interrupted; 2243 timeo = sock_wait_for_wmem(sk, timeo); 2244 } 2245 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2246 errcode, sk->sk_allocation); 2247 if (skb) 2248 skb_set_owner_w(skb, sk); 2249 return skb; 2250 2251 interrupted: 2252 err = sock_intr_errno(timeo); 2253 failure: 2254 *errcode = err; 2255 return NULL; 2256 } 2257 EXPORT_SYMBOL(sock_alloc_send_pskb); 2258 2259 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2260 int noblock, int *errcode) 2261 { 2262 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2263 } 2264 EXPORT_SYMBOL(sock_alloc_send_skb); 2265 2266 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2267 struct sockcm_cookie *sockc) 2268 { 2269 u32 tsflags; 2270 2271 switch (cmsg->cmsg_type) { 2272 case SO_MARK: 2273 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2274 return -EPERM; 2275 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2276 return -EINVAL; 2277 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2278 break; 2279 case SO_TIMESTAMPING_OLD: 2280 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2281 return -EINVAL; 2282 2283 tsflags = *(u32 *)CMSG_DATA(cmsg); 2284 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2285 return -EINVAL; 2286 2287 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2288 sockc->tsflags |= tsflags; 2289 break; 2290 case SCM_TXTIME: 2291 if (!sock_flag(sk, SOCK_TXTIME)) 2292 return -EINVAL; 2293 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2294 return -EINVAL; 2295 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2296 break; 2297 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2298 case SCM_RIGHTS: 2299 case SCM_CREDENTIALS: 2300 break; 2301 default: 2302 return -EINVAL; 2303 } 2304 return 0; 2305 } 2306 EXPORT_SYMBOL(__sock_cmsg_send); 2307 2308 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2309 struct sockcm_cookie *sockc) 2310 { 2311 struct cmsghdr *cmsg; 2312 int ret; 2313 2314 for_each_cmsghdr(cmsg, msg) { 2315 if (!CMSG_OK(msg, cmsg)) 2316 return -EINVAL; 2317 if (cmsg->cmsg_level != SOL_SOCKET) 2318 continue; 2319 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2320 if (ret) 2321 return ret; 2322 } 2323 return 0; 2324 } 2325 EXPORT_SYMBOL(sock_cmsg_send); 2326 2327 static void sk_enter_memory_pressure(struct sock *sk) 2328 { 2329 if (!sk->sk_prot->enter_memory_pressure) 2330 return; 2331 2332 sk->sk_prot->enter_memory_pressure(sk); 2333 } 2334 2335 static void sk_leave_memory_pressure(struct sock *sk) 2336 { 2337 if (sk->sk_prot->leave_memory_pressure) { 2338 sk->sk_prot->leave_memory_pressure(sk); 2339 } else { 2340 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2341 2342 if (memory_pressure && READ_ONCE(*memory_pressure)) 2343 WRITE_ONCE(*memory_pressure, 0); 2344 } 2345 } 2346 2347 /* On 32bit arches, an skb frag is limited to 2^15 */ 2348 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2349 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2350 2351 /** 2352 * skb_page_frag_refill - check that a page_frag contains enough room 2353 * @sz: minimum size of the fragment we want to get 2354 * @pfrag: pointer to page_frag 2355 * @gfp: priority for memory allocation 2356 * 2357 * Note: While this allocator tries to use high order pages, there is 2358 * no guarantee that allocations succeed. Therefore, @sz MUST be 2359 * less or equal than PAGE_SIZE. 2360 */ 2361 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2362 { 2363 if (pfrag->page) { 2364 if (page_ref_count(pfrag->page) == 1) { 2365 pfrag->offset = 0; 2366 return true; 2367 } 2368 if (pfrag->offset + sz <= pfrag->size) 2369 return true; 2370 put_page(pfrag->page); 2371 } 2372 2373 pfrag->offset = 0; 2374 if (SKB_FRAG_PAGE_ORDER && 2375 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2376 /* Avoid direct reclaim but allow kswapd to wake */ 2377 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2378 __GFP_COMP | __GFP_NOWARN | 2379 __GFP_NORETRY, 2380 SKB_FRAG_PAGE_ORDER); 2381 if (likely(pfrag->page)) { 2382 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2383 return true; 2384 } 2385 } 2386 pfrag->page = alloc_page(gfp); 2387 if (likely(pfrag->page)) { 2388 pfrag->size = PAGE_SIZE; 2389 return true; 2390 } 2391 return false; 2392 } 2393 EXPORT_SYMBOL(skb_page_frag_refill); 2394 2395 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2396 { 2397 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2398 return true; 2399 2400 sk_enter_memory_pressure(sk); 2401 sk_stream_moderate_sndbuf(sk); 2402 return false; 2403 } 2404 EXPORT_SYMBOL(sk_page_frag_refill); 2405 2406 static void __lock_sock(struct sock *sk) 2407 __releases(&sk->sk_lock.slock) 2408 __acquires(&sk->sk_lock.slock) 2409 { 2410 DEFINE_WAIT(wait); 2411 2412 for (;;) { 2413 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2414 TASK_UNINTERRUPTIBLE); 2415 spin_unlock_bh(&sk->sk_lock.slock); 2416 schedule(); 2417 spin_lock_bh(&sk->sk_lock.slock); 2418 if (!sock_owned_by_user(sk)) 2419 break; 2420 } 2421 finish_wait(&sk->sk_lock.wq, &wait); 2422 } 2423 2424 void __release_sock(struct sock *sk) 2425 __releases(&sk->sk_lock.slock) 2426 __acquires(&sk->sk_lock.slock) 2427 { 2428 struct sk_buff *skb, *next; 2429 2430 while ((skb = sk->sk_backlog.head) != NULL) { 2431 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2432 2433 spin_unlock_bh(&sk->sk_lock.slock); 2434 2435 do { 2436 next = skb->next; 2437 prefetch(next); 2438 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2439 skb_mark_not_on_list(skb); 2440 sk_backlog_rcv(sk, skb); 2441 2442 cond_resched(); 2443 2444 skb = next; 2445 } while (skb != NULL); 2446 2447 spin_lock_bh(&sk->sk_lock.slock); 2448 } 2449 2450 /* 2451 * Doing the zeroing here guarantee we can not loop forever 2452 * while a wild producer attempts to flood us. 2453 */ 2454 sk->sk_backlog.len = 0; 2455 } 2456 2457 void __sk_flush_backlog(struct sock *sk) 2458 { 2459 spin_lock_bh(&sk->sk_lock.slock); 2460 __release_sock(sk); 2461 spin_unlock_bh(&sk->sk_lock.slock); 2462 } 2463 2464 /** 2465 * sk_wait_data - wait for data to arrive at sk_receive_queue 2466 * @sk: sock to wait on 2467 * @timeo: for how long 2468 * @skb: last skb seen on sk_receive_queue 2469 * 2470 * Now socket state including sk->sk_err is changed only under lock, 2471 * hence we may omit checks after joining wait queue. 2472 * We check receive queue before schedule() only as optimization; 2473 * it is very likely that release_sock() added new data. 2474 */ 2475 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2476 { 2477 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2478 int rc; 2479 2480 add_wait_queue(sk_sleep(sk), &wait); 2481 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2482 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2483 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2484 remove_wait_queue(sk_sleep(sk), &wait); 2485 return rc; 2486 } 2487 EXPORT_SYMBOL(sk_wait_data); 2488 2489 /** 2490 * __sk_mem_raise_allocated - increase memory_allocated 2491 * @sk: socket 2492 * @size: memory size to allocate 2493 * @amt: pages to allocate 2494 * @kind: allocation type 2495 * 2496 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2497 */ 2498 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2499 { 2500 struct proto *prot = sk->sk_prot; 2501 long allocated = sk_memory_allocated_add(sk, amt); 2502 bool charged = true; 2503 2504 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2505 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2506 goto suppress_allocation; 2507 2508 /* Under limit. */ 2509 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2510 sk_leave_memory_pressure(sk); 2511 return 1; 2512 } 2513 2514 /* Under pressure. */ 2515 if (allocated > sk_prot_mem_limits(sk, 1)) 2516 sk_enter_memory_pressure(sk); 2517 2518 /* Over hard limit. */ 2519 if (allocated > sk_prot_mem_limits(sk, 2)) 2520 goto suppress_allocation; 2521 2522 /* guarantee minimum buffer size under pressure */ 2523 if (kind == SK_MEM_RECV) { 2524 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2525 return 1; 2526 2527 } else { /* SK_MEM_SEND */ 2528 int wmem0 = sk_get_wmem0(sk, prot); 2529 2530 if (sk->sk_type == SOCK_STREAM) { 2531 if (sk->sk_wmem_queued < wmem0) 2532 return 1; 2533 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2534 return 1; 2535 } 2536 } 2537 2538 if (sk_has_memory_pressure(sk)) { 2539 u64 alloc; 2540 2541 if (!sk_under_memory_pressure(sk)) 2542 return 1; 2543 alloc = sk_sockets_allocated_read_positive(sk); 2544 if (sk_prot_mem_limits(sk, 2) > alloc * 2545 sk_mem_pages(sk->sk_wmem_queued + 2546 atomic_read(&sk->sk_rmem_alloc) + 2547 sk->sk_forward_alloc)) 2548 return 1; 2549 } 2550 2551 suppress_allocation: 2552 2553 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2554 sk_stream_moderate_sndbuf(sk); 2555 2556 /* Fail only if socket is _under_ its sndbuf. 2557 * In this case we cannot block, so that we have to fail. 2558 */ 2559 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2560 return 1; 2561 } 2562 2563 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2564 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2565 2566 sk_memory_allocated_sub(sk, amt); 2567 2568 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2569 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2570 2571 return 0; 2572 } 2573 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2574 2575 /** 2576 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2577 * @sk: socket 2578 * @size: memory size to allocate 2579 * @kind: allocation type 2580 * 2581 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2582 * rmem allocation. This function assumes that protocols which have 2583 * memory_pressure use sk_wmem_queued as write buffer accounting. 2584 */ 2585 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2586 { 2587 int ret, amt = sk_mem_pages(size); 2588 2589 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2590 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2591 if (!ret) 2592 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2593 return ret; 2594 } 2595 EXPORT_SYMBOL(__sk_mem_schedule); 2596 2597 /** 2598 * __sk_mem_reduce_allocated - reclaim memory_allocated 2599 * @sk: socket 2600 * @amount: number of quanta 2601 * 2602 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2603 */ 2604 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2605 { 2606 sk_memory_allocated_sub(sk, amount); 2607 2608 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2609 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2610 2611 if (sk_under_memory_pressure(sk) && 2612 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2613 sk_leave_memory_pressure(sk); 2614 } 2615 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2616 2617 /** 2618 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2619 * @sk: socket 2620 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2621 */ 2622 void __sk_mem_reclaim(struct sock *sk, int amount) 2623 { 2624 amount >>= SK_MEM_QUANTUM_SHIFT; 2625 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2626 __sk_mem_reduce_allocated(sk, amount); 2627 } 2628 EXPORT_SYMBOL(__sk_mem_reclaim); 2629 2630 int sk_set_peek_off(struct sock *sk, int val) 2631 { 2632 sk->sk_peek_off = val; 2633 return 0; 2634 } 2635 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2636 2637 /* 2638 * Set of default routines for initialising struct proto_ops when 2639 * the protocol does not support a particular function. In certain 2640 * cases where it makes no sense for a protocol to have a "do nothing" 2641 * function, some default processing is provided. 2642 */ 2643 2644 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2645 { 2646 return -EOPNOTSUPP; 2647 } 2648 EXPORT_SYMBOL(sock_no_bind); 2649 2650 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2651 int len, int flags) 2652 { 2653 return -EOPNOTSUPP; 2654 } 2655 EXPORT_SYMBOL(sock_no_connect); 2656 2657 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2658 { 2659 return -EOPNOTSUPP; 2660 } 2661 EXPORT_SYMBOL(sock_no_socketpair); 2662 2663 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2664 bool kern) 2665 { 2666 return -EOPNOTSUPP; 2667 } 2668 EXPORT_SYMBOL(sock_no_accept); 2669 2670 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2671 int peer) 2672 { 2673 return -EOPNOTSUPP; 2674 } 2675 EXPORT_SYMBOL(sock_no_getname); 2676 2677 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2678 { 2679 return -EOPNOTSUPP; 2680 } 2681 EXPORT_SYMBOL(sock_no_ioctl); 2682 2683 int sock_no_listen(struct socket *sock, int backlog) 2684 { 2685 return -EOPNOTSUPP; 2686 } 2687 EXPORT_SYMBOL(sock_no_listen); 2688 2689 int sock_no_shutdown(struct socket *sock, int how) 2690 { 2691 return -EOPNOTSUPP; 2692 } 2693 EXPORT_SYMBOL(sock_no_shutdown); 2694 2695 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2696 char __user *optval, unsigned int optlen) 2697 { 2698 return -EOPNOTSUPP; 2699 } 2700 EXPORT_SYMBOL(sock_no_setsockopt); 2701 2702 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2703 char __user *optval, int __user *optlen) 2704 { 2705 return -EOPNOTSUPP; 2706 } 2707 EXPORT_SYMBOL(sock_no_getsockopt); 2708 2709 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2710 { 2711 return -EOPNOTSUPP; 2712 } 2713 EXPORT_SYMBOL(sock_no_sendmsg); 2714 2715 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2716 { 2717 return -EOPNOTSUPP; 2718 } 2719 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2720 2721 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2722 int flags) 2723 { 2724 return -EOPNOTSUPP; 2725 } 2726 EXPORT_SYMBOL(sock_no_recvmsg); 2727 2728 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2729 { 2730 /* Mirror missing mmap method error code */ 2731 return -ENODEV; 2732 } 2733 EXPORT_SYMBOL(sock_no_mmap); 2734 2735 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2736 { 2737 ssize_t res; 2738 struct msghdr msg = {.msg_flags = flags}; 2739 struct kvec iov; 2740 char *kaddr = kmap(page); 2741 iov.iov_base = kaddr + offset; 2742 iov.iov_len = size; 2743 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2744 kunmap(page); 2745 return res; 2746 } 2747 EXPORT_SYMBOL(sock_no_sendpage); 2748 2749 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2750 int offset, size_t size, int flags) 2751 { 2752 ssize_t res; 2753 struct msghdr msg = {.msg_flags = flags}; 2754 struct kvec iov; 2755 char *kaddr = kmap(page); 2756 2757 iov.iov_base = kaddr + offset; 2758 iov.iov_len = size; 2759 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2760 kunmap(page); 2761 return res; 2762 } 2763 EXPORT_SYMBOL(sock_no_sendpage_locked); 2764 2765 /* 2766 * Default Socket Callbacks 2767 */ 2768 2769 static void sock_def_wakeup(struct sock *sk) 2770 { 2771 struct socket_wq *wq; 2772 2773 rcu_read_lock(); 2774 wq = rcu_dereference(sk->sk_wq); 2775 if (skwq_has_sleeper(wq)) 2776 wake_up_interruptible_all(&wq->wait); 2777 rcu_read_unlock(); 2778 } 2779 2780 static void sock_def_error_report(struct sock *sk) 2781 { 2782 struct socket_wq *wq; 2783 2784 rcu_read_lock(); 2785 wq = rcu_dereference(sk->sk_wq); 2786 if (skwq_has_sleeper(wq)) 2787 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 2788 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2789 rcu_read_unlock(); 2790 } 2791 2792 void sock_def_readable(struct sock *sk) 2793 { 2794 struct socket_wq *wq; 2795 2796 rcu_read_lock(); 2797 wq = rcu_dereference(sk->sk_wq); 2798 if (skwq_has_sleeper(wq)) 2799 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 2800 EPOLLRDNORM | EPOLLRDBAND); 2801 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2802 rcu_read_unlock(); 2803 } 2804 2805 static void sock_def_write_space(struct sock *sk) 2806 { 2807 struct socket_wq *wq; 2808 2809 rcu_read_lock(); 2810 2811 /* Do not wake up a writer until he can make "significant" 2812 * progress. --DaveM 2813 */ 2814 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 2815 wq = rcu_dereference(sk->sk_wq); 2816 if (skwq_has_sleeper(wq)) 2817 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 2818 EPOLLWRNORM | EPOLLWRBAND); 2819 2820 /* Should agree with poll, otherwise some programs break */ 2821 if (sock_writeable(sk)) 2822 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2823 } 2824 2825 rcu_read_unlock(); 2826 } 2827 2828 static void sock_def_destruct(struct sock *sk) 2829 { 2830 } 2831 2832 void sk_send_sigurg(struct sock *sk) 2833 { 2834 if (sk->sk_socket && sk->sk_socket->file) 2835 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2836 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2837 } 2838 EXPORT_SYMBOL(sk_send_sigurg); 2839 2840 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2841 unsigned long expires) 2842 { 2843 if (!mod_timer(timer, expires)) 2844 sock_hold(sk); 2845 } 2846 EXPORT_SYMBOL(sk_reset_timer); 2847 2848 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2849 { 2850 if (del_timer(timer)) 2851 __sock_put(sk); 2852 } 2853 EXPORT_SYMBOL(sk_stop_timer); 2854 2855 void sock_init_data(struct socket *sock, struct sock *sk) 2856 { 2857 sk_init_common(sk); 2858 sk->sk_send_head = NULL; 2859 2860 timer_setup(&sk->sk_timer, NULL, 0); 2861 2862 sk->sk_allocation = GFP_KERNEL; 2863 sk->sk_rcvbuf = sysctl_rmem_default; 2864 sk->sk_sndbuf = sysctl_wmem_default; 2865 sk->sk_state = TCP_CLOSE; 2866 sk_set_socket(sk, sock); 2867 2868 sock_set_flag(sk, SOCK_ZAPPED); 2869 2870 if (sock) { 2871 sk->sk_type = sock->type; 2872 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 2873 sock->sk = sk; 2874 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2875 } else { 2876 RCU_INIT_POINTER(sk->sk_wq, NULL); 2877 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2878 } 2879 2880 rwlock_init(&sk->sk_callback_lock); 2881 if (sk->sk_kern_sock) 2882 lockdep_set_class_and_name( 2883 &sk->sk_callback_lock, 2884 af_kern_callback_keys + sk->sk_family, 2885 af_family_kern_clock_key_strings[sk->sk_family]); 2886 else 2887 lockdep_set_class_and_name( 2888 &sk->sk_callback_lock, 2889 af_callback_keys + sk->sk_family, 2890 af_family_clock_key_strings[sk->sk_family]); 2891 2892 sk->sk_state_change = sock_def_wakeup; 2893 sk->sk_data_ready = sock_def_readable; 2894 sk->sk_write_space = sock_def_write_space; 2895 sk->sk_error_report = sock_def_error_report; 2896 sk->sk_destruct = sock_def_destruct; 2897 2898 sk->sk_frag.page = NULL; 2899 sk->sk_frag.offset = 0; 2900 sk->sk_peek_off = -1; 2901 2902 sk->sk_peer_pid = NULL; 2903 sk->sk_peer_cred = NULL; 2904 sk->sk_write_pending = 0; 2905 sk->sk_rcvlowat = 1; 2906 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2907 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2908 2909 sk->sk_stamp = SK_DEFAULT_STAMP; 2910 #if BITS_PER_LONG==32 2911 seqlock_init(&sk->sk_stamp_seq); 2912 #endif 2913 atomic_set(&sk->sk_zckey, 0); 2914 2915 #ifdef CONFIG_NET_RX_BUSY_POLL 2916 sk->sk_napi_id = 0; 2917 sk->sk_ll_usec = sysctl_net_busy_read; 2918 #endif 2919 2920 sk->sk_max_pacing_rate = ~0UL; 2921 sk->sk_pacing_rate = ~0UL; 2922 WRITE_ONCE(sk->sk_pacing_shift, 10); 2923 sk->sk_incoming_cpu = -1; 2924 2925 sk_rx_queue_clear(sk); 2926 /* 2927 * Before updating sk_refcnt, we must commit prior changes to memory 2928 * (Documentation/RCU/rculist_nulls.txt for details) 2929 */ 2930 smp_wmb(); 2931 refcount_set(&sk->sk_refcnt, 1); 2932 atomic_set(&sk->sk_drops, 0); 2933 } 2934 EXPORT_SYMBOL(sock_init_data); 2935 2936 void lock_sock_nested(struct sock *sk, int subclass) 2937 { 2938 might_sleep(); 2939 spin_lock_bh(&sk->sk_lock.slock); 2940 if (sk->sk_lock.owned) 2941 __lock_sock(sk); 2942 sk->sk_lock.owned = 1; 2943 spin_unlock(&sk->sk_lock.slock); 2944 /* 2945 * The sk_lock has mutex_lock() semantics here: 2946 */ 2947 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2948 local_bh_enable(); 2949 } 2950 EXPORT_SYMBOL(lock_sock_nested); 2951 2952 void release_sock(struct sock *sk) 2953 { 2954 spin_lock_bh(&sk->sk_lock.slock); 2955 if (sk->sk_backlog.tail) 2956 __release_sock(sk); 2957 2958 /* Warning : release_cb() might need to release sk ownership, 2959 * ie call sock_release_ownership(sk) before us. 2960 */ 2961 if (sk->sk_prot->release_cb) 2962 sk->sk_prot->release_cb(sk); 2963 2964 sock_release_ownership(sk); 2965 if (waitqueue_active(&sk->sk_lock.wq)) 2966 wake_up(&sk->sk_lock.wq); 2967 spin_unlock_bh(&sk->sk_lock.slock); 2968 } 2969 EXPORT_SYMBOL(release_sock); 2970 2971 /** 2972 * lock_sock_fast - fast version of lock_sock 2973 * @sk: socket 2974 * 2975 * This version should be used for very small section, where process wont block 2976 * return false if fast path is taken: 2977 * 2978 * sk_lock.slock locked, owned = 0, BH disabled 2979 * 2980 * return true if slow path is taken: 2981 * 2982 * sk_lock.slock unlocked, owned = 1, BH enabled 2983 */ 2984 bool lock_sock_fast(struct sock *sk) 2985 { 2986 might_sleep(); 2987 spin_lock_bh(&sk->sk_lock.slock); 2988 2989 if (!sk->sk_lock.owned) 2990 /* 2991 * Note : We must disable BH 2992 */ 2993 return false; 2994 2995 __lock_sock(sk); 2996 sk->sk_lock.owned = 1; 2997 spin_unlock(&sk->sk_lock.slock); 2998 /* 2999 * The sk_lock has mutex_lock() semantics here: 3000 */ 3001 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 3002 local_bh_enable(); 3003 return true; 3004 } 3005 EXPORT_SYMBOL(lock_sock_fast); 3006 3007 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3008 bool timeval, bool time32) 3009 { 3010 struct sock *sk = sock->sk; 3011 struct timespec64 ts; 3012 3013 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3014 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3015 if (ts.tv_sec == -1) 3016 return -ENOENT; 3017 if (ts.tv_sec == 0) { 3018 ktime_t kt = ktime_get_real(); 3019 sock_write_timestamp(sk, kt); 3020 ts = ktime_to_timespec64(kt); 3021 } 3022 3023 if (timeval) 3024 ts.tv_nsec /= 1000; 3025 3026 #ifdef CONFIG_COMPAT_32BIT_TIME 3027 if (time32) 3028 return put_old_timespec32(&ts, userstamp); 3029 #endif 3030 #ifdef CONFIG_SPARC64 3031 /* beware of padding in sparc64 timeval */ 3032 if (timeval && !in_compat_syscall()) { 3033 struct __kernel_old_timeval __user tv = { 3034 .tv_sec = ts.tv_sec, 3035 .tv_usec = ts.tv_nsec, 3036 }; 3037 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3038 return -EFAULT; 3039 return 0; 3040 } 3041 #endif 3042 return put_timespec64(&ts, userstamp); 3043 } 3044 EXPORT_SYMBOL(sock_gettstamp); 3045 3046 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3047 { 3048 if (!sock_flag(sk, flag)) { 3049 unsigned long previous_flags = sk->sk_flags; 3050 3051 sock_set_flag(sk, flag); 3052 /* 3053 * we just set one of the two flags which require net 3054 * time stamping, but time stamping might have been on 3055 * already because of the other one 3056 */ 3057 if (sock_needs_netstamp(sk) && 3058 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3059 net_enable_timestamp(); 3060 } 3061 } 3062 3063 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3064 int level, int type) 3065 { 3066 struct sock_exterr_skb *serr; 3067 struct sk_buff *skb; 3068 int copied, err; 3069 3070 err = -EAGAIN; 3071 skb = sock_dequeue_err_skb(sk); 3072 if (skb == NULL) 3073 goto out; 3074 3075 copied = skb->len; 3076 if (copied > len) { 3077 msg->msg_flags |= MSG_TRUNC; 3078 copied = len; 3079 } 3080 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3081 if (err) 3082 goto out_free_skb; 3083 3084 sock_recv_timestamp(msg, sk, skb); 3085 3086 serr = SKB_EXT_ERR(skb); 3087 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3088 3089 msg->msg_flags |= MSG_ERRQUEUE; 3090 err = copied; 3091 3092 out_free_skb: 3093 kfree_skb(skb); 3094 out: 3095 return err; 3096 } 3097 EXPORT_SYMBOL(sock_recv_errqueue); 3098 3099 /* 3100 * Get a socket option on an socket. 3101 * 3102 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3103 * asynchronous errors should be reported by getsockopt. We assume 3104 * this means if you specify SO_ERROR (otherwise whats the point of it). 3105 */ 3106 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3107 char __user *optval, int __user *optlen) 3108 { 3109 struct sock *sk = sock->sk; 3110 3111 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3112 } 3113 EXPORT_SYMBOL(sock_common_getsockopt); 3114 3115 #ifdef CONFIG_COMPAT 3116 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 3117 char __user *optval, int __user *optlen) 3118 { 3119 struct sock *sk = sock->sk; 3120 3121 if (sk->sk_prot->compat_getsockopt != NULL) 3122 return sk->sk_prot->compat_getsockopt(sk, level, optname, 3123 optval, optlen); 3124 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3125 } 3126 EXPORT_SYMBOL(compat_sock_common_getsockopt); 3127 #endif 3128 3129 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3130 int flags) 3131 { 3132 struct sock *sk = sock->sk; 3133 int addr_len = 0; 3134 int err; 3135 3136 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3137 flags & ~MSG_DONTWAIT, &addr_len); 3138 if (err >= 0) 3139 msg->msg_namelen = addr_len; 3140 return err; 3141 } 3142 EXPORT_SYMBOL(sock_common_recvmsg); 3143 3144 /* 3145 * Set socket options on an inet socket. 3146 */ 3147 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3148 char __user *optval, unsigned int optlen) 3149 { 3150 struct sock *sk = sock->sk; 3151 3152 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3153 } 3154 EXPORT_SYMBOL(sock_common_setsockopt); 3155 3156 #ifdef CONFIG_COMPAT 3157 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 3158 char __user *optval, unsigned int optlen) 3159 { 3160 struct sock *sk = sock->sk; 3161 3162 if (sk->sk_prot->compat_setsockopt != NULL) 3163 return sk->sk_prot->compat_setsockopt(sk, level, optname, 3164 optval, optlen); 3165 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3166 } 3167 EXPORT_SYMBOL(compat_sock_common_setsockopt); 3168 #endif 3169 3170 void sk_common_release(struct sock *sk) 3171 { 3172 if (sk->sk_prot->destroy) 3173 sk->sk_prot->destroy(sk); 3174 3175 /* 3176 * Observation: when sock_common_release is called, processes have 3177 * no access to socket. But net still has. 3178 * Step one, detach it from networking: 3179 * 3180 * A. Remove from hash tables. 3181 */ 3182 3183 sk->sk_prot->unhash(sk); 3184 3185 /* 3186 * In this point socket cannot receive new packets, but it is possible 3187 * that some packets are in flight because some CPU runs receiver and 3188 * did hash table lookup before we unhashed socket. They will achieve 3189 * receive queue and will be purged by socket destructor. 3190 * 3191 * Also we still have packets pending on receive queue and probably, 3192 * our own packets waiting in device queues. sock_destroy will drain 3193 * receive queue, but transmitted packets will delay socket destruction 3194 * until the last reference will be released. 3195 */ 3196 3197 sock_orphan(sk); 3198 3199 xfrm_sk_free_policy(sk); 3200 3201 sk_refcnt_debug_release(sk); 3202 3203 sock_put(sk); 3204 } 3205 EXPORT_SYMBOL(sk_common_release); 3206 3207 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3208 { 3209 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3210 3211 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3212 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3213 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3214 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3215 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3216 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3217 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3218 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3219 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3220 } 3221 3222 #ifdef CONFIG_PROC_FS 3223 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3224 struct prot_inuse { 3225 int val[PROTO_INUSE_NR]; 3226 }; 3227 3228 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3229 3230 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3231 { 3232 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3233 } 3234 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3235 3236 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3237 { 3238 int cpu, idx = prot->inuse_idx; 3239 int res = 0; 3240 3241 for_each_possible_cpu(cpu) 3242 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3243 3244 return res >= 0 ? res : 0; 3245 } 3246 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3247 3248 static void sock_inuse_add(struct net *net, int val) 3249 { 3250 this_cpu_add(*net->core.sock_inuse, val); 3251 } 3252 3253 int sock_inuse_get(struct net *net) 3254 { 3255 int cpu, res = 0; 3256 3257 for_each_possible_cpu(cpu) 3258 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3259 3260 return res; 3261 } 3262 3263 EXPORT_SYMBOL_GPL(sock_inuse_get); 3264 3265 static int __net_init sock_inuse_init_net(struct net *net) 3266 { 3267 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3268 if (net->core.prot_inuse == NULL) 3269 return -ENOMEM; 3270 3271 net->core.sock_inuse = alloc_percpu(int); 3272 if (net->core.sock_inuse == NULL) 3273 goto out; 3274 3275 return 0; 3276 3277 out: 3278 free_percpu(net->core.prot_inuse); 3279 return -ENOMEM; 3280 } 3281 3282 static void __net_exit sock_inuse_exit_net(struct net *net) 3283 { 3284 free_percpu(net->core.prot_inuse); 3285 free_percpu(net->core.sock_inuse); 3286 } 3287 3288 static struct pernet_operations net_inuse_ops = { 3289 .init = sock_inuse_init_net, 3290 .exit = sock_inuse_exit_net, 3291 }; 3292 3293 static __init int net_inuse_init(void) 3294 { 3295 if (register_pernet_subsys(&net_inuse_ops)) 3296 panic("Cannot initialize net inuse counters"); 3297 3298 return 0; 3299 } 3300 3301 core_initcall(net_inuse_init); 3302 3303 static int assign_proto_idx(struct proto *prot) 3304 { 3305 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3306 3307 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3308 pr_err("PROTO_INUSE_NR exhausted\n"); 3309 return -ENOSPC; 3310 } 3311 3312 set_bit(prot->inuse_idx, proto_inuse_idx); 3313 return 0; 3314 } 3315 3316 static void release_proto_idx(struct proto *prot) 3317 { 3318 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3319 clear_bit(prot->inuse_idx, proto_inuse_idx); 3320 } 3321 #else 3322 static inline int assign_proto_idx(struct proto *prot) 3323 { 3324 return 0; 3325 } 3326 3327 static inline void release_proto_idx(struct proto *prot) 3328 { 3329 } 3330 3331 static void sock_inuse_add(struct net *net, int val) 3332 { 3333 } 3334 #endif 3335 3336 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3337 { 3338 if (!rsk_prot) 3339 return; 3340 kfree(rsk_prot->slab_name); 3341 rsk_prot->slab_name = NULL; 3342 kmem_cache_destroy(rsk_prot->slab); 3343 rsk_prot->slab = NULL; 3344 } 3345 3346 static int req_prot_init(const struct proto *prot) 3347 { 3348 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3349 3350 if (!rsk_prot) 3351 return 0; 3352 3353 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3354 prot->name); 3355 if (!rsk_prot->slab_name) 3356 return -ENOMEM; 3357 3358 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3359 rsk_prot->obj_size, 0, 3360 SLAB_ACCOUNT | prot->slab_flags, 3361 NULL); 3362 3363 if (!rsk_prot->slab) { 3364 pr_crit("%s: Can't create request sock SLAB cache!\n", 3365 prot->name); 3366 return -ENOMEM; 3367 } 3368 return 0; 3369 } 3370 3371 int proto_register(struct proto *prot, int alloc_slab) 3372 { 3373 int ret = -ENOBUFS; 3374 3375 if (alloc_slab) { 3376 prot->slab = kmem_cache_create_usercopy(prot->name, 3377 prot->obj_size, 0, 3378 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3379 prot->slab_flags, 3380 prot->useroffset, prot->usersize, 3381 NULL); 3382 3383 if (prot->slab == NULL) { 3384 pr_crit("%s: Can't create sock SLAB cache!\n", 3385 prot->name); 3386 goto out; 3387 } 3388 3389 if (req_prot_init(prot)) 3390 goto out_free_request_sock_slab; 3391 3392 if (prot->twsk_prot != NULL) { 3393 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3394 3395 if (prot->twsk_prot->twsk_slab_name == NULL) 3396 goto out_free_request_sock_slab; 3397 3398 prot->twsk_prot->twsk_slab = 3399 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3400 prot->twsk_prot->twsk_obj_size, 3401 0, 3402 SLAB_ACCOUNT | 3403 prot->slab_flags, 3404 NULL); 3405 if (prot->twsk_prot->twsk_slab == NULL) 3406 goto out_free_timewait_sock_slab_name; 3407 } 3408 } 3409 3410 mutex_lock(&proto_list_mutex); 3411 ret = assign_proto_idx(prot); 3412 if (ret) { 3413 mutex_unlock(&proto_list_mutex); 3414 goto out_free_timewait_sock_slab_name; 3415 } 3416 list_add(&prot->node, &proto_list); 3417 mutex_unlock(&proto_list_mutex); 3418 return ret; 3419 3420 out_free_timewait_sock_slab_name: 3421 if (alloc_slab && prot->twsk_prot) 3422 kfree(prot->twsk_prot->twsk_slab_name); 3423 out_free_request_sock_slab: 3424 if (alloc_slab) { 3425 req_prot_cleanup(prot->rsk_prot); 3426 3427 kmem_cache_destroy(prot->slab); 3428 prot->slab = NULL; 3429 } 3430 out: 3431 return ret; 3432 } 3433 EXPORT_SYMBOL(proto_register); 3434 3435 void proto_unregister(struct proto *prot) 3436 { 3437 mutex_lock(&proto_list_mutex); 3438 release_proto_idx(prot); 3439 list_del(&prot->node); 3440 mutex_unlock(&proto_list_mutex); 3441 3442 kmem_cache_destroy(prot->slab); 3443 prot->slab = NULL; 3444 3445 req_prot_cleanup(prot->rsk_prot); 3446 3447 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3448 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3449 kfree(prot->twsk_prot->twsk_slab_name); 3450 prot->twsk_prot->twsk_slab = NULL; 3451 } 3452 } 3453 EXPORT_SYMBOL(proto_unregister); 3454 3455 int sock_load_diag_module(int family, int protocol) 3456 { 3457 if (!protocol) { 3458 if (!sock_is_registered(family)) 3459 return -ENOENT; 3460 3461 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3462 NETLINK_SOCK_DIAG, family); 3463 } 3464 3465 #ifdef CONFIG_INET 3466 if (family == AF_INET && 3467 protocol != IPPROTO_RAW && 3468 !rcu_access_pointer(inet_protos[protocol])) 3469 return -ENOENT; 3470 #endif 3471 3472 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3473 NETLINK_SOCK_DIAG, family, protocol); 3474 } 3475 EXPORT_SYMBOL(sock_load_diag_module); 3476 3477 #ifdef CONFIG_PROC_FS 3478 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3479 __acquires(proto_list_mutex) 3480 { 3481 mutex_lock(&proto_list_mutex); 3482 return seq_list_start_head(&proto_list, *pos); 3483 } 3484 3485 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3486 { 3487 return seq_list_next(v, &proto_list, pos); 3488 } 3489 3490 static void proto_seq_stop(struct seq_file *seq, void *v) 3491 __releases(proto_list_mutex) 3492 { 3493 mutex_unlock(&proto_list_mutex); 3494 } 3495 3496 static char proto_method_implemented(const void *method) 3497 { 3498 return method == NULL ? 'n' : 'y'; 3499 } 3500 static long sock_prot_memory_allocated(struct proto *proto) 3501 { 3502 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3503 } 3504 3505 static const char *sock_prot_memory_pressure(struct proto *proto) 3506 { 3507 return proto->memory_pressure != NULL ? 3508 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3509 } 3510 3511 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3512 { 3513 3514 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3515 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3516 proto->name, 3517 proto->obj_size, 3518 sock_prot_inuse_get(seq_file_net(seq), proto), 3519 sock_prot_memory_allocated(proto), 3520 sock_prot_memory_pressure(proto), 3521 proto->max_header, 3522 proto->slab == NULL ? "no" : "yes", 3523 module_name(proto->owner), 3524 proto_method_implemented(proto->close), 3525 proto_method_implemented(proto->connect), 3526 proto_method_implemented(proto->disconnect), 3527 proto_method_implemented(proto->accept), 3528 proto_method_implemented(proto->ioctl), 3529 proto_method_implemented(proto->init), 3530 proto_method_implemented(proto->destroy), 3531 proto_method_implemented(proto->shutdown), 3532 proto_method_implemented(proto->setsockopt), 3533 proto_method_implemented(proto->getsockopt), 3534 proto_method_implemented(proto->sendmsg), 3535 proto_method_implemented(proto->recvmsg), 3536 proto_method_implemented(proto->sendpage), 3537 proto_method_implemented(proto->bind), 3538 proto_method_implemented(proto->backlog_rcv), 3539 proto_method_implemented(proto->hash), 3540 proto_method_implemented(proto->unhash), 3541 proto_method_implemented(proto->get_port), 3542 proto_method_implemented(proto->enter_memory_pressure)); 3543 } 3544 3545 static int proto_seq_show(struct seq_file *seq, void *v) 3546 { 3547 if (v == &proto_list) 3548 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3549 "protocol", 3550 "size", 3551 "sockets", 3552 "memory", 3553 "press", 3554 "maxhdr", 3555 "slab", 3556 "module", 3557 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3558 else 3559 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3560 return 0; 3561 } 3562 3563 static const struct seq_operations proto_seq_ops = { 3564 .start = proto_seq_start, 3565 .next = proto_seq_next, 3566 .stop = proto_seq_stop, 3567 .show = proto_seq_show, 3568 }; 3569 3570 static __net_init int proto_init_net(struct net *net) 3571 { 3572 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3573 sizeof(struct seq_net_private))) 3574 return -ENOMEM; 3575 3576 return 0; 3577 } 3578 3579 static __net_exit void proto_exit_net(struct net *net) 3580 { 3581 remove_proc_entry("protocols", net->proc_net); 3582 } 3583 3584 3585 static __net_initdata struct pernet_operations proto_net_ops = { 3586 .init = proto_init_net, 3587 .exit = proto_exit_net, 3588 }; 3589 3590 static int __init proto_init(void) 3591 { 3592 return register_pernet_subsys(&proto_net_ops); 3593 } 3594 3595 subsys_initcall(proto_init); 3596 3597 #endif /* PROC_FS */ 3598 3599 #ifdef CONFIG_NET_RX_BUSY_POLL 3600 bool sk_busy_loop_end(void *p, unsigned long start_time) 3601 { 3602 struct sock *sk = p; 3603 3604 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3605 sk_busy_loop_timeout(sk, start_time); 3606 } 3607 EXPORT_SYMBOL(sk_busy_loop_end); 3608 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3609