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 const struct proto *prot = READ_ONCE(osk->sk_prot); 1576 #ifdef CONFIG_SECURITY_NETWORK 1577 void *sptr = nsk->sk_security; 1578 #endif 1579 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1580 1581 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1582 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1583 1584 #ifdef CONFIG_SECURITY_NETWORK 1585 nsk->sk_security = sptr; 1586 security_sk_clone(osk, nsk); 1587 #endif 1588 } 1589 1590 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1591 int family) 1592 { 1593 struct sock *sk; 1594 struct kmem_cache *slab; 1595 1596 slab = prot->slab; 1597 if (slab != NULL) { 1598 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1599 if (!sk) 1600 return sk; 1601 if (want_init_on_alloc(priority)) 1602 sk_prot_clear_nulls(sk, prot->obj_size); 1603 } else 1604 sk = kmalloc(prot->obj_size, priority); 1605 1606 if (sk != NULL) { 1607 if (security_sk_alloc(sk, family, priority)) 1608 goto out_free; 1609 1610 if (!try_module_get(prot->owner)) 1611 goto out_free_sec; 1612 sk_tx_queue_clear(sk); 1613 } 1614 1615 return sk; 1616 1617 out_free_sec: 1618 security_sk_free(sk); 1619 out_free: 1620 if (slab != NULL) 1621 kmem_cache_free(slab, sk); 1622 else 1623 kfree(sk); 1624 return NULL; 1625 } 1626 1627 static void sk_prot_free(struct proto *prot, struct sock *sk) 1628 { 1629 struct kmem_cache *slab; 1630 struct module *owner; 1631 1632 owner = prot->owner; 1633 slab = prot->slab; 1634 1635 cgroup_sk_free(&sk->sk_cgrp_data); 1636 mem_cgroup_sk_free(sk); 1637 security_sk_free(sk); 1638 if (slab != NULL) 1639 kmem_cache_free(slab, sk); 1640 else 1641 kfree(sk); 1642 module_put(owner); 1643 } 1644 1645 /** 1646 * sk_alloc - All socket objects are allocated here 1647 * @net: the applicable net namespace 1648 * @family: protocol family 1649 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1650 * @prot: struct proto associated with this new sock instance 1651 * @kern: is this to be a kernel socket? 1652 */ 1653 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1654 struct proto *prot, int kern) 1655 { 1656 struct sock *sk; 1657 1658 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1659 if (sk) { 1660 sk->sk_family = family; 1661 /* 1662 * See comment in struct sock definition to understand 1663 * why we need sk_prot_creator -acme 1664 */ 1665 sk->sk_prot = sk->sk_prot_creator = prot; 1666 sk->sk_kern_sock = kern; 1667 sock_lock_init(sk); 1668 sk->sk_net_refcnt = kern ? 0 : 1; 1669 if (likely(sk->sk_net_refcnt)) { 1670 get_net(net); 1671 sock_inuse_add(net, 1); 1672 } 1673 1674 sock_net_set(sk, net); 1675 refcount_set(&sk->sk_wmem_alloc, 1); 1676 1677 mem_cgroup_sk_alloc(sk); 1678 cgroup_sk_alloc(&sk->sk_cgrp_data); 1679 sock_update_classid(&sk->sk_cgrp_data); 1680 sock_update_netprioidx(&sk->sk_cgrp_data); 1681 } 1682 1683 return sk; 1684 } 1685 EXPORT_SYMBOL(sk_alloc); 1686 1687 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1688 * grace period. This is the case for UDP sockets and TCP listeners. 1689 */ 1690 static void __sk_destruct(struct rcu_head *head) 1691 { 1692 struct sock *sk = container_of(head, struct sock, sk_rcu); 1693 struct sk_filter *filter; 1694 1695 if (sk->sk_destruct) 1696 sk->sk_destruct(sk); 1697 1698 filter = rcu_dereference_check(sk->sk_filter, 1699 refcount_read(&sk->sk_wmem_alloc) == 0); 1700 if (filter) { 1701 sk_filter_uncharge(sk, filter); 1702 RCU_INIT_POINTER(sk->sk_filter, NULL); 1703 } 1704 1705 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1706 1707 #ifdef CONFIG_BPF_SYSCALL 1708 bpf_sk_storage_free(sk); 1709 #endif 1710 1711 if (atomic_read(&sk->sk_omem_alloc)) 1712 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1713 __func__, atomic_read(&sk->sk_omem_alloc)); 1714 1715 if (sk->sk_frag.page) { 1716 put_page(sk->sk_frag.page); 1717 sk->sk_frag.page = NULL; 1718 } 1719 1720 if (sk->sk_peer_cred) 1721 put_cred(sk->sk_peer_cred); 1722 put_pid(sk->sk_peer_pid); 1723 if (likely(sk->sk_net_refcnt)) 1724 put_net(sock_net(sk)); 1725 sk_prot_free(sk->sk_prot_creator, sk); 1726 } 1727 1728 void sk_destruct(struct sock *sk) 1729 { 1730 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 1731 1732 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 1733 reuseport_detach_sock(sk); 1734 use_call_rcu = true; 1735 } 1736 1737 if (use_call_rcu) 1738 call_rcu(&sk->sk_rcu, __sk_destruct); 1739 else 1740 __sk_destruct(&sk->sk_rcu); 1741 } 1742 1743 static void __sk_free(struct sock *sk) 1744 { 1745 if (likely(sk->sk_net_refcnt)) 1746 sock_inuse_add(sock_net(sk), -1); 1747 1748 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1749 sock_diag_broadcast_destroy(sk); 1750 else 1751 sk_destruct(sk); 1752 } 1753 1754 void sk_free(struct sock *sk) 1755 { 1756 /* 1757 * We subtract one from sk_wmem_alloc and can know if 1758 * some packets are still in some tx queue. 1759 * If not null, sock_wfree() will call __sk_free(sk) later 1760 */ 1761 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1762 __sk_free(sk); 1763 } 1764 EXPORT_SYMBOL(sk_free); 1765 1766 static void sk_init_common(struct sock *sk) 1767 { 1768 skb_queue_head_init(&sk->sk_receive_queue); 1769 skb_queue_head_init(&sk->sk_write_queue); 1770 skb_queue_head_init(&sk->sk_error_queue); 1771 1772 rwlock_init(&sk->sk_callback_lock); 1773 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1774 af_rlock_keys + sk->sk_family, 1775 af_family_rlock_key_strings[sk->sk_family]); 1776 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1777 af_wlock_keys + sk->sk_family, 1778 af_family_wlock_key_strings[sk->sk_family]); 1779 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1780 af_elock_keys + sk->sk_family, 1781 af_family_elock_key_strings[sk->sk_family]); 1782 lockdep_set_class_and_name(&sk->sk_callback_lock, 1783 af_callback_keys + sk->sk_family, 1784 af_family_clock_key_strings[sk->sk_family]); 1785 } 1786 1787 /** 1788 * sk_clone_lock - clone a socket, and lock its clone 1789 * @sk: the socket to clone 1790 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1791 * 1792 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1793 */ 1794 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1795 { 1796 struct proto *prot = READ_ONCE(sk->sk_prot); 1797 struct sock *newsk; 1798 bool is_charged = true; 1799 1800 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 1801 if (newsk != NULL) { 1802 struct sk_filter *filter; 1803 1804 sock_copy(newsk, sk); 1805 1806 newsk->sk_prot_creator = prot; 1807 1808 /* SANITY */ 1809 if (likely(newsk->sk_net_refcnt)) 1810 get_net(sock_net(newsk)); 1811 sk_node_init(&newsk->sk_node); 1812 sock_lock_init(newsk); 1813 bh_lock_sock(newsk); 1814 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1815 newsk->sk_backlog.len = 0; 1816 1817 atomic_set(&newsk->sk_rmem_alloc, 0); 1818 /* 1819 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1820 */ 1821 refcount_set(&newsk->sk_wmem_alloc, 1); 1822 atomic_set(&newsk->sk_omem_alloc, 0); 1823 sk_init_common(newsk); 1824 1825 newsk->sk_dst_cache = NULL; 1826 newsk->sk_dst_pending_confirm = 0; 1827 newsk->sk_wmem_queued = 0; 1828 newsk->sk_forward_alloc = 0; 1829 atomic_set(&newsk->sk_drops, 0); 1830 newsk->sk_send_head = NULL; 1831 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1832 atomic_set(&newsk->sk_zckey, 0); 1833 1834 sock_reset_flag(newsk, SOCK_DONE); 1835 mem_cgroup_sk_alloc(newsk); 1836 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1837 1838 rcu_read_lock(); 1839 filter = rcu_dereference(sk->sk_filter); 1840 if (filter != NULL) 1841 /* though it's an empty new sock, the charging may fail 1842 * if sysctl_optmem_max was changed between creation of 1843 * original socket and cloning 1844 */ 1845 is_charged = sk_filter_charge(newsk, filter); 1846 RCU_INIT_POINTER(newsk->sk_filter, filter); 1847 rcu_read_unlock(); 1848 1849 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1850 /* We need to make sure that we don't uncharge the new 1851 * socket if we couldn't charge it in the first place 1852 * as otherwise we uncharge the parent's filter. 1853 */ 1854 if (!is_charged) 1855 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1856 sk_free_unlock_clone(newsk); 1857 newsk = NULL; 1858 goto out; 1859 } 1860 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1861 1862 if (bpf_sk_storage_clone(sk, newsk)) { 1863 sk_free_unlock_clone(newsk); 1864 newsk = NULL; 1865 goto out; 1866 } 1867 1868 /* Clear sk_user_data if parent had the pointer tagged 1869 * as not suitable for copying when cloning. 1870 */ 1871 if (sk_user_data_is_nocopy(newsk)) 1872 RCU_INIT_POINTER(newsk->sk_user_data, NULL); 1873 1874 newsk->sk_err = 0; 1875 newsk->sk_err_soft = 0; 1876 newsk->sk_priority = 0; 1877 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1878 if (likely(newsk->sk_net_refcnt)) 1879 sock_inuse_add(sock_net(newsk), 1); 1880 1881 /* 1882 * Before updating sk_refcnt, we must commit prior changes to memory 1883 * (Documentation/RCU/rculist_nulls.txt for details) 1884 */ 1885 smp_wmb(); 1886 refcount_set(&newsk->sk_refcnt, 2); 1887 1888 /* 1889 * Increment the counter in the same struct proto as the master 1890 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1891 * is the same as sk->sk_prot->socks, as this field was copied 1892 * with memcpy). 1893 * 1894 * This _changes_ the previous behaviour, where 1895 * tcp_create_openreq_child always was incrementing the 1896 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1897 * to be taken into account in all callers. -acme 1898 */ 1899 sk_refcnt_debug_inc(newsk); 1900 sk_set_socket(newsk, NULL); 1901 RCU_INIT_POINTER(newsk->sk_wq, NULL); 1902 1903 if (newsk->sk_prot->sockets_allocated) 1904 sk_sockets_allocated_inc(newsk); 1905 1906 if (sock_needs_netstamp(sk) && 1907 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1908 net_enable_timestamp(); 1909 } 1910 out: 1911 return newsk; 1912 } 1913 EXPORT_SYMBOL_GPL(sk_clone_lock); 1914 1915 void sk_free_unlock_clone(struct sock *sk) 1916 { 1917 /* It is still raw copy of parent, so invalidate 1918 * destructor and make plain sk_free() */ 1919 sk->sk_destruct = NULL; 1920 bh_unlock_sock(sk); 1921 sk_free(sk); 1922 } 1923 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1924 1925 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1926 { 1927 u32 max_segs = 1; 1928 1929 sk_dst_set(sk, dst); 1930 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 1931 if (sk->sk_route_caps & NETIF_F_GSO) 1932 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1933 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1934 if (sk_can_gso(sk)) { 1935 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 1936 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1937 } else { 1938 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1939 sk->sk_gso_max_size = dst->dev->gso_max_size; 1940 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1941 } 1942 } 1943 sk->sk_gso_max_segs = max_segs; 1944 } 1945 EXPORT_SYMBOL_GPL(sk_setup_caps); 1946 1947 /* 1948 * Simple resource managers for sockets. 1949 */ 1950 1951 1952 /* 1953 * Write buffer destructor automatically called from kfree_skb. 1954 */ 1955 void sock_wfree(struct sk_buff *skb) 1956 { 1957 struct sock *sk = skb->sk; 1958 unsigned int len = skb->truesize; 1959 1960 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1961 /* 1962 * Keep a reference on sk_wmem_alloc, this will be released 1963 * after sk_write_space() call 1964 */ 1965 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 1966 sk->sk_write_space(sk); 1967 len = 1; 1968 } 1969 /* 1970 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1971 * could not do because of in-flight packets 1972 */ 1973 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 1974 __sk_free(sk); 1975 } 1976 EXPORT_SYMBOL(sock_wfree); 1977 1978 /* This variant of sock_wfree() is used by TCP, 1979 * since it sets SOCK_USE_WRITE_QUEUE. 1980 */ 1981 void __sock_wfree(struct sk_buff *skb) 1982 { 1983 struct sock *sk = skb->sk; 1984 1985 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1986 __sk_free(sk); 1987 } 1988 1989 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1990 { 1991 skb_orphan(skb); 1992 skb->sk = sk; 1993 #ifdef CONFIG_INET 1994 if (unlikely(!sk_fullsock(sk))) { 1995 skb->destructor = sock_edemux; 1996 sock_hold(sk); 1997 return; 1998 } 1999 #endif 2000 skb->destructor = sock_wfree; 2001 skb_set_hash_from_sk(skb, sk); 2002 /* 2003 * We used to take a refcount on sk, but following operation 2004 * is enough to guarantee sk_free() wont free this sock until 2005 * all in-flight packets are completed 2006 */ 2007 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2008 } 2009 EXPORT_SYMBOL(skb_set_owner_w); 2010 2011 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2012 { 2013 #ifdef CONFIG_TLS_DEVICE 2014 /* Drivers depend on in-order delivery for crypto offload, 2015 * partial orphan breaks out-of-order-OK logic. 2016 */ 2017 if (skb->decrypted) 2018 return false; 2019 #endif 2020 return (skb->destructor == sock_wfree || 2021 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2022 } 2023 2024 /* This helper is used by netem, as it can hold packets in its 2025 * delay queue. We want to allow the owner socket to send more 2026 * packets, as if they were already TX completed by a typical driver. 2027 * But we also want to keep skb->sk set because some packet schedulers 2028 * rely on it (sch_fq for example). 2029 */ 2030 void skb_orphan_partial(struct sk_buff *skb) 2031 { 2032 if (skb_is_tcp_pure_ack(skb)) 2033 return; 2034 2035 if (can_skb_orphan_partial(skb)) { 2036 struct sock *sk = skb->sk; 2037 2038 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 2039 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)); 2040 skb->destructor = sock_efree; 2041 } 2042 } else { 2043 skb_orphan(skb); 2044 } 2045 } 2046 EXPORT_SYMBOL(skb_orphan_partial); 2047 2048 /* 2049 * Read buffer destructor automatically called from kfree_skb. 2050 */ 2051 void sock_rfree(struct sk_buff *skb) 2052 { 2053 struct sock *sk = skb->sk; 2054 unsigned int len = skb->truesize; 2055 2056 atomic_sub(len, &sk->sk_rmem_alloc); 2057 sk_mem_uncharge(sk, len); 2058 } 2059 EXPORT_SYMBOL(sock_rfree); 2060 2061 /* 2062 * Buffer destructor for skbs that are not used directly in read or write 2063 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2064 */ 2065 void sock_efree(struct sk_buff *skb) 2066 { 2067 sock_put(skb->sk); 2068 } 2069 EXPORT_SYMBOL(sock_efree); 2070 2071 kuid_t sock_i_uid(struct sock *sk) 2072 { 2073 kuid_t uid; 2074 2075 read_lock_bh(&sk->sk_callback_lock); 2076 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2077 read_unlock_bh(&sk->sk_callback_lock); 2078 return uid; 2079 } 2080 EXPORT_SYMBOL(sock_i_uid); 2081 2082 unsigned long sock_i_ino(struct sock *sk) 2083 { 2084 unsigned long ino; 2085 2086 read_lock_bh(&sk->sk_callback_lock); 2087 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2088 read_unlock_bh(&sk->sk_callback_lock); 2089 return ino; 2090 } 2091 EXPORT_SYMBOL(sock_i_ino); 2092 2093 /* 2094 * Allocate a skb from the socket's send buffer. 2095 */ 2096 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2097 gfp_t priority) 2098 { 2099 if (force || 2100 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2101 struct sk_buff *skb = alloc_skb(size, priority); 2102 2103 if (skb) { 2104 skb_set_owner_w(skb, sk); 2105 return skb; 2106 } 2107 } 2108 return NULL; 2109 } 2110 EXPORT_SYMBOL(sock_wmalloc); 2111 2112 static void sock_ofree(struct sk_buff *skb) 2113 { 2114 struct sock *sk = skb->sk; 2115 2116 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2117 } 2118 2119 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2120 gfp_t priority) 2121 { 2122 struct sk_buff *skb; 2123 2124 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2125 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2126 sysctl_optmem_max) 2127 return NULL; 2128 2129 skb = alloc_skb(size, priority); 2130 if (!skb) 2131 return NULL; 2132 2133 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2134 skb->sk = sk; 2135 skb->destructor = sock_ofree; 2136 return skb; 2137 } 2138 2139 /* 2140 * Allocate a memory block from the socket's option memory buffer. 2141 */ 2142 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2143 { 2144 if ((unsigned int)size <= sysctl_optmem_max && 2145 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2146 void *mem; 2147 /* First do the add, to avoid the race if kmalloc 2148 * might sleep. 2149 */ 2150 atomic_add(size, &sk->sk_omem_alloc); 2151 mem = kmalloc(size, priority); 2152 if (mem) 2153 return mem; 2154 atomic_sub(size, &sk->sk_omem_alloc); 2155 } 2156 return NULL; 2157 } 2158 EXPORT_SYMBOL(sock_kmalloc); 2159 2160 /* Free an option memory block. Note, we actually want the inline 2161 * here as this allows gcc to detect the nullify and fold away the 2162 * condition entirely. 2163 */ 2164 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2165 const bool nullify) 2166 { 2167 if (WARN_ON_ONCE(!mem)) 2168 return; 2169 if (nullify) 2170 kzfree(mem); 2171 else 2172 kfree(mem); 2173 atomic_sub(size, &sk->sk_omem_alloc); 2174 } 2175 2176 void sock_kfree_s(struct sock *sk, void *mem, int size) 2177 { 2178 __sock_kfree_s(sk, mem, size, false); 2179 } 2180 EXPORT_SYMBOL(sock_kfree_s); 2181 2182 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2183 { 2184 __sock_kfree_s(sk, mem, size, true); 2185 } 2186 EXPORT_SYMBOL(sock_kzfree_s); 2187 2188 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2189 I think, these locks should be removed for datagram sockets. 2190 */ 2191 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2192 { 2193 DEFINE_WAIT(wait); 2194 2195 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2196 for (;;) { 2197 if (!timeo) 2198 break; 2199 if (signal_pending(current)) 2200 break; 2201 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2202 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2203 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2204 break; 2205 if (sk->sk_shutdown & SEND_SHUTDOWN) 2206 break; 2207 if (sk->sk_err) 2208 break; 2209 timeo = schedule_timeout(timeo); 2210 } 2211 finish_wait(sk_sleep(sk), &wait); 2212 return timeo; 2213 } 2214 2215 2216 /* 2217 * Generic send/receive buffer handlers 2218 */ 2219 2220 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2221 unsigned long data_len, int noblock, 2222 int *errcode, int max_page_order) 2223 { 2224 struct sk_buff *skb; 2225 long timeo; 2226 int err; 2227 2228 timeo = sock_sndtimeo(sk, noblock); 2229 for (;;) { 2230 err = sock_error(sk); 2231 if (err != 0) 2232 goto failure; 2233 2234 err = -EPIPE; 2235 if (sk->sk_shutdown & SEND_SHUTDOWN) 2236 goto failure; 2237 2238 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2239 break; 2240 2241 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2242 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2243 err = -EAGAIN; 2244 if (!timeo) 2245 goto failure; 2246 if (signal_pending(current)) 2247 goto interrupted; 2248 timeo = sock_wait_for_wmem(sk, timeo); 2249 } 2250 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2251 errcode, sk->sk_allocation); 2252 if (skb) 2253 skb_set_owner_w(skb, sk); 2254 return skb; 2255 2256 interrupted: 2257 err = sock_intr_errno(timeo); 2258 failure: 2259 *errcode = err; 2260 return NULL; 2261 } 2262 EXPORT_SYMBOL(sock_alloc_send_pskb); 2263 2264 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2265 int noblock, int *errcode) 2266 { 2267 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2268 } 2269 EXPORT_SYMBOL(sock_alloc_send_skb); 2270 2271 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2272 struct sockcm_cookie *sockc) 2273 { 2274 u32 tsflags; 2275 2276 switch (cmsg->cmsg_type) { 2277 case SO_MARK: 2278 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2279 return -EPERM; 2280 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2281 return -EINVAL; 2282 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2283 break; 2284 case SO_TIMESTAMPING_OLD: 2285 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2286 return -EINVAL; 2287 2288 tsflags = *(u32 *)CMSG_DATA(cmsg); 2289 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2290 return -EINVAL; 2291 2292 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2293 sockc->tsflags |= tsflags; 2294 break; 2295 case SCM_TXTIME: 2296 if (!sock_flag(sk, SOCK_TXTIME)) 2297 return -EINVAL; 2298 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2299 return -EINVAL; 2300 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2301 break; 2302 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2303 case SCM_RIGHTS: 2304 case SCM_CREDENTIALS: 2305 break; 2306 default: 2307 return -EINVAL; 2308 } 2309 return 0; 2310 } 2311 EXPORT_SYMBOL(__sock_cmsg_send); 2312 2313 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2314 struct sockcm_cookie *sockc) 2315 { 2316 struct cmsghdr *cmsg; 2317 int ret; 2318 2319 for_each_cmsghdr(cmsg, msg) { 2320 if (!CMSG_OK(msg, cmsg)) 2321 return -EINVAL; 2322 if (cmsg->cmsg_level != SOL_SOCKET) 2323 continue; 2324 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2325 if (ret) 2326 return ret; 2327 } 2328 return 0; 2329 } 2330 EXPORT_SYMBOL(sock_cmsg_send); 2331 2332 static void sk_enter_memory_pressure(struct sock *sk) 2333 { 2334 if (!sk->sk_prot->enter_memory_pressure) 2335 return; 2336 2337 sk->sk_prot->enter_memory_pressure(sk); 2338 } 2339 2340 static void sk_leave_memory_pressure(struct sock *sk) 2341 { 2342 if (sk->sk_prot->leave_memory_pressure) { 2343 sk->sk_prot->leave_memory_pressure(sk); 2344 } else { 2345 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2346 2347 if (memory_pressure && READ_ONCE(*memory_pressure)) 2348 WRITE_ONCE(*memory_pressure, 0); 2349 } 2350 } 2351 2352 /* On 32bit arches, an skb frag is limited to 2^15 */ 2353 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2354 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2355 2356 /** 2357 * skb_page_frag_refill - check that a page_frag contains enough room 2358 * @sz: minimum size of the fragment we want to get 2359 * @pfrag: pointer to page_frag 2360 * @gfp: priority for memory allocation 2361 * 2362 * Note: While this allocator tries to use high order pages, there is 2363 * no guarantee that allocations succeed. Therefore, @sz MUST be 2364 * less or equal than PAGE_SIZE. 2365 */ 2366 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2367 { 2368 if (pfrag->page) { 2369 if (page_ref_count(pfrag->page) == 1) { 2370 pfrag->offset = 0; 2371 return true; 2372 } 2373 if (pfrag->offset + sz <= pfrag->size) 2374 return true; 2375 put_page(pfrag->page); 2376 } 2377 2378 pfrag->offset = 0; 2379 if (SKB_FRAG_PAGE_ORDER && 2380 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2381 /* Avoid direct reclaim but allow kswapd to wake */ 2382 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2383 __GFP_COMP | __GFP_NOWARN | 2384 __GFP_NORETRY, 2385 SKB_FRAG_PAGE_ORDER); 2386 if (likely(pfrag->page)) { 2387 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2388 return true; 2389 } 2390 } 2391 pfrag->page = alloc_page(gfp); 2392 if (likely(pfrag->page)) { 2393 pfrag->size = PAGE_SIZE; 2394 return true; 2395 } 2396 return false; 2397 } 2398 EXPORT_SYMBOL(skb_page_frag_refill); 2399 2400 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2401 { 2402 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2403 return true; 2404 2405 sk_enter_memory_pressure(sk); 2406 sk_stream_moderate_sndbuf(sk); 2407 return false; 2408 } 2409 EXPORT_SYMBOL(sk_page_frag_refill); 2410 2411 static void __lock_sock(struct sock *sk) 2412 __releases(&sk->sk_lock.slock) 2413 __acquires(&sk->sk_lock.slock) 2414 { 2415 DEFINE_WAIT(wait); 2416 2417 for (;;) { 2418 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2419 TASK_UNINTERRUPTIBLE); 2420 spin_unlock_bh(&sk->sk_lock.slock); 2421 schedule(); 2422 spin_lock_bh(&sk->sk_lock.slock); 2423 if (!sock_owned_by_user(sk)) 2424 break; 2425 } 2426 finish_wait(&sk->sk_lock.wq, &wait); 2427 } 2428 2429 void __release_sock(struct sock *sk) 2430 __releases(&sk->sk_lock.slock) 2431 __acquires(&sk->sk_lock.slock) 2432 { 2433 struct sk_buff *skb, *next; 2434 2435 while ((skb = sk->sk_backlog.head) != NULL) { 2436 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2437 2438 spin_unlock_bh(&sk->sk_lock.slock); 2439 2440 do { 2441 next = skb->next; 2442 prefetch(next); 2443 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2444 skb_mark_not_on_list(skb); 2445 sk_backlog_rcv(sk, skb); 2446 2447 cond_resched(); 2448 2449 skb = next; 2450 } while (skb != NULL); 2451 2452 spin_lock_bh(&sk->sk_lock.slock); 2453 } 2454 2455 /* 2456 * Doing the zeroing here guarantee we can not loop forever 2457 * while a wild producer attempts to flood us. 2458 */ 2459 sk->sk_backlog.len = 0; 2460 } 2461 2462 void __sk_flush_backlog(struct sock *sk) 2463 { 2464 spin_lock_bh(&sk->sk_lock.slock); 2465 __release_sock(sk); 2466 spin_unlock_bh(&sk->sk_lock.slock); 2467 } 2468 2469 /** 2470 * sk_wait_data - wait for data to arrive at sk_receive_queue 2471 * @sk: sock to wait on 2472 * @timeo: for how long 2473 * @skb: last skb seen on sk_receive_queue 2474 * 2475 * Now socket state including sk->sk_err is changed only under lock, 2476 * hence we may omit checks after joining wait queue. 2477 * We check receive queue before schedule() only as optimization; 2478 * it is very likely that release_sock() added new data. 2479 */ 2480 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2481 { 2482 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2483 int rc; 2484 2485 add_wait_queue(sk_sleep(sk), &wait); 2486 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2487 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2488 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2489 remove_wait_queue(sk_sleep(sk), &wait); 2490 return rc; 2491 } 2492 EXPORT_SYMBOL(sk_wait_data); 2493 2494 /** 2495 * __sk_mem_raise_allocated - increase memory_allocated 2496 * @sk: socket 2497 * @size: memory size to allocate 2498 * @amt: pages to allocate 2499 * @kind: allocation type 2500 * 2501 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2502 */ 2503 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2504 { 2505 struct proto *prot = sk->sk_prot; 2506 long allocated = sk_memory_allocated_add(sk, amt); 2507 bool charged = true; 2508 2509 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2510 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2511 goto suppress_allocation; 2512 2513 /* Under limit. */ 2514 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2515 sk_leave_memory_pressure(sk); 2516 return 1; 2517 } 2518 2519 /* Under pressure. */ 2520 if (allocated > sk_prot_mem_limits(sk, 1)) 2521 sk_enter_memory_pressure(sk); 2522 2523 /* Over hard limit. */ 2524 if (allocated > sk_prot_mem_limits(sk, 2)) 2525 goto suppress_allocation; 2526 2527 /* guarantee minimum buffer size under pressure */ 2528 if (kind == SK_MEM_RECV) { 2529 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2530 return 1; 2531 2532 } else { /* SK_MEM_SEND */ 2533 int wmem0 = sk_get_wmem0(sk, prot); 2534 2535 if (sk->sk_type == SOCK_STREAM) { 2536 if (sk->sk_wmem_queued < wmem0) 2537 return 1; 2538 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2539 return 1; 2540 } 2541 } 2542 2543 if (sk_has_memory_pressure(sk)) { 2544 u64 alloc; 2545 2546 if (!sk_under_memory_pressure(sk)) 2547 return 1; 2548 alloc = sk_sockets_allocated_read_positive(sk); 2549 if (sk_prot_mem_limits(sk, 2) > alloc * 2550 sk_mem_pages(sk->sk_wmem_queued + 2551 atomic_read(&sk->sk_rmem_alloc) + 2552 sk->sk_forward_alloc)) 2553 return 1; 2554 } 2555 2556 suppress_allocation: 2557 2558 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2559 sk_stream_moderate_sndbuf(sk); 2560 2561 /* Fail only if socket is _under_ its sndbuf. 2562 * In this case we cannot block, so that we have to fail. 2563 */ 2564 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2565 return 1; 2566 } 2567 2568 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2569 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2570 2571 sk_memory_allocated_sub(sk, amt); 2572 2573 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2574 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2575 2576 return 0; 2577 } 2578 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2579 2580 /** 2581 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2582 * @sk: socket 2583 * @size: memory size to allocate 2584 * @kind: allocation type 2585 * 2586 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2587 * rmem allocation. This function assumes that protocols which have 2588 * memory_pressure use sk_wmem_queued as write buffer accounting. 2589 */ 2590 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2591 { 2592 int ret, amt = sk_mem_pages(size); 2593 2594 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2595 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2596 if (!ret) 2597 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2598 return ret; 2599 } 2600 EXPORT_SYMBOL(__sk_mem_schedule); 2601 2602 /** 2603 * __sk_mem_reduce_allocated - reclaim memory_allocated 2604 * @sk: socket 2605 * @amount: number of quanta 2606 * 2607 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2608 */ 2609 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2610 { 2611 sk_memory_allocated_sub(sk, amount); 2612 2613 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2614 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2615 2616 if (sk_under_memory_pressure(sk) && 2617 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2618 sk_leave_memory_pressure(sk); 2619 } 2620 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2621 2622 /** 2623 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2624 * @sk: socket 2625 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2626 */ 2627 void __sk_mem_reclaim(struct sock *sk, int amount) 2628 { 2629 amount >>= SK_MEM_QUANTUM_SHIFT; 2630 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2631 __sk_mem_reduce_allocated(sk, amount); 2632 } 2633 EXPORT_SYMBOL(__sk_mem_reclaim); 2634 2635 int sk_set_peek_off(struct sock *sk, int val) 2636 { 2637 sk->sk_peek_off = val; 2638 return 0; 2639 } 2640 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2641 2642 /* 2643 * Set of default routines for initialising struct proto_ops when 2644 * the protocol does not support a particular function. In certain 2645 * cases where it makes no sense for a protocol to have a "do nothing" 2646 * function, some default processing is provided. 2647 */ 2648 2649 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2650 { 2651 return -EOPNOTSUPP; 2652 } 2653 EXPORT_SYMBOL(sock_no_bind); 2654 2655 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2656 int len, int flags) 2657 { 2658 return -EOPNOTSUPP; 2659 } 2660 EXPORT_SYMBOL(sock_no_connect); 2661 2662 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2663 { 2664 return -EOPNOTSUPP; 2665 } 2666 EXPORT_SYMBOL(sock_no_socketpair); 2667 2668 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2669 bool kern) 2670 { 2671 return -EOPNOTSUPP; 2672 } 2673 EXPORT_SYMBOL(sock_no_accept); 2674 2675 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2676 int peer) 2677 { 2678 return -EOPNOTSUPP; 2679 } 2680 EXPORT_SYMBOL(sock_no_getname); 2681 2682 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2683 { 2684 return -EOPNOTSUPP; 2685 } 2686 EXPORT_SYMBOL(sock_no_ioctl); 2687 2688 int sock_no_listen(struct socket *sock, int backlog) 2689 { 2690 return -EOPNOTSUPP; 2691 } 2692 EXPORT_SYMBOL(sock_no_listen); 2693 2694 int sock_no_shutdown(struct socket *sock, int how) 2695 { 2696 return -EOPNOTSUPP; 2697 } 2698 EXPORT_SYMBOL(sock_no_shutdown); 2699 2700 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2701 char __user *optval, unsigned int optlen) 2702 { 2703 return -EOPNOTSUPP; 2704 } 2705 EXPORT_SYMBOL(sock_no_setsockopt); 2706 2707 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2708 char __user *optval, int __user *optlen) 2709 { 2710 return -EOPNOTSUPP; 2711 } 2712 EXPORT_SYMBOL(sock_no_getsockopt); 2713 2714 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2715 { 2716 return -EOPNOTSUPP; 2717 } 2718 EXPORT_SYMBOL(sock_no_sendmsg); 2719 2720 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2721 { 2722 return -EOPNOTSUPP; 2723 } 2724 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2725 2726 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2727 int flags) 2728 { 2729 return -EOPNOTSUPP; 2730 } 2731 EXPORT_SYMBOL(sock_no_recvmsg); 2732 2733 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2734 { 2735 /* Mirror missing mmap method error code */ 2736 return -ENODEV; 2737 } 2738 EXPORT_SYMBOL(sock_no_mmap); 2739 2740 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2741 { 2742 ssize_t res; 2743 struct msghdr msg = {.msg_flags = flags}; 2744 struct kvec iov; 2745 char *kaddr = kmap(page); 2746 iov.iov_base = kaddr + offset; 2747 iov.iov_len = size; 2748 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2749 kunmap(page); 2750 return res; 2751 } 2752 EXPORT_SYMBOL(sock_no_sendpage); 2753 2754 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2755 int offset, size_t size, int flags) 2756 { 2757 ssize_t res; 2758 struct msghdr msg = {.msg_flags = flags}; 2759 struct kvec iov; 2760 char *kaddr = kmap(page); 2761 2762 iov.iov_base = kaddr + offset; 2763 iov.iov_len = size; 2764 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2765 kunmap(page); 2766 return res; 2767 } 2768 EXPORT_SYMBOL(sock_no_sendpage_locked); 2769 2770 /* 2771 * Default Socket Callbacks 2772 */ 2773 2774 static void sock_def_wakeup(struct sock *sk) 2775 { 2776 struct socket_wq *wq; 2777 2778 rcu_read_lock(); 2779 wq = rcu_dereference(sk->sk_wq); 2780 if (skwq_has_sleeper(wq)) 2781 wake_up_interruptible_all(&wq->wait); 2782 rcu_read_unlock(); 2783 } 2784 2785 static void sock_def_error_report(struct sock *sk) 2786 { 2787 struct socket_wq *wq; 2788 2789 rcu_read_lock(); 2790 wq = rcu_dereference(sk->sk_wq); 2791 if (skwq_has_sleeper(wq)) 2792 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 2793 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2794 rcu_read_unlock(); 2795 } 2796 2797 void sock_def_readable(struct sock *sk) 2798 { 2799 struct socket_wq *wq; 2800 2801 rcu_read_lock(); 2802 wq = rcu_dereference(sk->sk_wq); 2803 if (skwq_has_sleeper(wq)) 2804 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 2805 EPOLLRDNORM | EPOLLRDBAND); 2806 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2807 rcu_read_unlock(); 2808 } 2809 2810 static void sock_def_write_space(struct sock *sk) 2811 { 2812 struct socket_wq *wq; 2813 2814 rcu_read_lock(); 2815 2816 /* Do not wake up a writer until he can make "significant" 2817 * progress. --DaveM 2818 */ 2819 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 2820 wq = rcu_dereference(sk->sk_wq); 2821 if (skwq_has_sleeper(wq)) 2822 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 2823 EPOLLWRNORM | EPOLLWRBAND); 2824 2825 /* Should agree with poll, otherwise some programs break */ 2826 if (sock_writeable(sk)) 2827 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2828 } 2829 2830 rcu_read_unlock(); 2831 } 2832 2833 static void sock_def_destruct(struct sock *sk) 2834 { 2835 } 2836 2837 void sk_send_sigurg(struct sock *sk) 2838 { 2839 if (sk->sk_socket && sk->sk_socket->file) 2840 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2841 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2842 } 2843 EXPORT_SYMBOL(sk_send_sigurg); 2844 2845 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2846 unsigned long expires) 2847 { 2848 if (!mod_timer(timer, expires)) 2849 sock_hold(sk); 2850 } 2851 EXPORT_SYMBOL(sk_reset_timer); 2852 2853 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2854 { 2855 if (del_timer(timer)) 2856 __sock_put(sk); 2857 } 2858 EXPORT_SYMBOL(sk_stop_timer); 2859 2860 void sock_init_data(struct socket *sock, struct sock *sk) 2861 { 2862 sk_init_common(sk); 2863 sk->sk_send_head = NULL; 2864 2865 timer_setup(&sk->sk_timer, NULL, 0); 2866 2867 sk->sk_allocation = GFP_KERNEL; 2868 sk->sk_rcvbuf = sysctl_rmem_default; 2869 sk->sk_sndbuf = sysctl_wmem_default; 2870 sk->sk_state = TCP_CLOSE; 2871 sk_set_socket(sk, sock); 2872 2873 sock_set_flag(sk, SOCK_ZAPPED); 2874 2875 if (sock) { 2876 sk->sk_type = sock->type; 2877 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 2878 sock->sk = sk; 2879 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2880 } else { 2881 RCU_INIT_POINTER(sk->sk_wq, NULL); 2882 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2883 } 2884 2885 rwlock_init(&sk->sk_callback_lock); 2886 if (sk->sk_kern_sock) 2887 lockdep_set_class_and_name( 2888 &sk->sk_callback_lock, 2889 af_kern_callback_keys + sk->sk_family, 2890 af_family_kern_clock_key_strings[sk->sk_family]); 2891 else 2892 lockdep_set_class_and_name( 2893 &sk->sk_callback_lock, 2894 af_callback_keys + sk->sk_family, 2895 af_family_clock_key_strings[sk->sk_family]); 2896 2897 sk->sk_state_change = sock_def_wakeup; 2898 sk->sk_data_ready = sock_def_readable; 2899 sk->sk_write_space = sock_def_write_space; 2900 sk->sk_error_report = sock_def_error_report; 2901 sk->sk_destruct = sock_def_destruct; 2902 2903 sk->sk_frag.page = NULL; 2904 sk->sk_frag.offset = 0; 2905 sk->sk_peek_off = -1; 2906 2907 sk->sk_peer_pid = NULL; 2908 sk->sk_peer_cred = NULL; 2909 sk->sk_write_pending = 0; 2910 sk->sk_rcvlowat = 1; 2911 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2912 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2913 2914 sk->sk_stamp = SK_DEFAULT_STAMP; 2915 #if BITS_PER_LONG==32 2916 seqlock_init(&sk->sk_stamp_seq); 2917 #endif 2918 atomic_set(&sk->sk_zckey, 0); 2919 2920 #ifdef CONFIG_NET_RX_BUSY_POLL 2921 sk->sk_napi_id = 0; 2922 sk->sk_ll_usec = sysctl_net_busy_read; 2923 #endif 2924 2925 sk->sk_max_pacing_rate = ~0UL; 2926 sk->sk_pacing_rate = ~0UL; 2927 WRITE_ONCE(sk->sk_pacing_shift, 10); 2928 sk->sk_incoming_cpu = -1; 2929 2930 sk_rx_queue_clear(sk); 2931 /* 2932 * Before updating sk_refcnt, we must commit prior changes to memory 2933 * (Documentation/RCU/rculist_nulls.txt for details) 2934 */ 2935 smp_wmb(); 2936 refcount_set(&sk->sk_refcnt, 1); 2937 atomic_set(&sk->sk_drops, 0); 2938 } 2939 EXPORT_SYMBOL(sock_init_data); 2940 2941 void lock_sock_nested(struct sock *sk, int subclass) 2942 { 2943 might_sleep(); 2944 spin_lock_bh(&sk->sk_lock.slock); 2945 if (sk->sk_lock.owned) 2946 __lock_sock(sk); 2947 sk->sk_lock.owned = 1; 2948 spin_unlock(&sk->sk_lock.slock); 2949 /* 2950 * The sk_lock has mutex_lock() semantics here: 2951 */ 2952 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2953 local_bh_enable(); 2954 } 2955 EXPORT_SYMBOL(lock_sock_nested); 2956 2957 void release_sock(struct sock *sk) 2958 { 2959 spin_lock_bh(&sk->sk_lock.slock); 2960 if (sk->sk_backlog.tail) 2961 __release_sock(sk); 2962 2963 /* Warning : release_cb() might need to release sk ownership, 2964 * ie call sock_release_ownership(sk) before us. 2965 */ 2966 if (sk->sk_prot->release_cb) 2967 sk->sk_prot->release_cb(sk); 2968 2969 sock_release_ownership(sk); 2970 if (waitqueue_active(&sk->sk_lock.wq)) 2971 wake_up(&sk->sk_lock.wq); 2972 spin_unlock_bh(&sk->sk_lock.slock); 2973 } 2974 EXPORT_SYMBOL(release_sock); 2975 2976 /** 2977 * lock_sock_fast - fast version of lock_sock 2978 * @sk: socket 2979 * 2980 * This version should be used for very small section, where process wont block 2981 * return false if fast path is taken: 2982 * 2983 * sk_lock.slock locked, owned = 0, BH disabled 2984 * 2985 * return true if slow path is taken: 2986 * 2987 * sk_lock.slock unlocked, owned = 1, BH enabled 2988 */ 2989 bool lock_sock_fast(struct sock *sk) 2990 { 2991 might_sleep(); 2992 spin_lock_bh(&sk->sk_lock.slock); 2993 2994 if (!sk->sk_lock.owned) 2995 /* 2996 * Note : We must disable BH 2997 */ 2998 return false; 2999 3000 __lock_sock(sk); 3001 sk->sk_lock.owned = 1; 3002 spin_unlock(&sk->sk_lock.slock); 3003 /* 3004 * The sk_lock has mutex_lock() semantics here: 3005 */ 3006 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 3007 local_bh_enable(); 3008 return true; 3009 } 3010 EXPORT_SYMBOL(lock_sock_fast); 3011 3012 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3013 bool timeval, bool time32) 3014 { 3015 struct sock *sk = sock->sk; 3016 struct timespec64 ts; 3017 3018 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3019 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3020 if (ts.tv_sec == -1) 3021 return -ENOENT; 3022 if (ts.tv_sec == 0) { 3023 ktime_t kt = ktime_get_real(); 3024 sock_write_timestamp(sk, kt); 3025 ts = ktime_to_timespec64(kt); 3026 } 3027 3028 if (timeval) 3029 ts.tv_nsec /= 1000; 3030 3031 #ifdef CONFIG_COMPAT_32BIT_TIME 3032 if (time32) 3033 return put_old_timespec32(&ts, userstamp); 3034 #endif 3035 #ifdef CONFIG_SPARC64 3036 /* beware of padding in sparc64 timeval */ 3037 if (timeval && !in_compat_syscall()) { 3038 struct __kernel_old_timeval __user tv = { 3039 .tv_sec = ts.tv_sec, 3040 .tv_usec = ts.tv_nsec, 3041 }; 3042 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3043 return -EFAULT; 3044 return 0; 3045 } 3046 #endif 3047 return put_timespec64(&ts, userstamp); 3048 } 3049 EXPORT_SYMBOL(sock_gettstamp); 3050 3051 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3052 { 3053 if (!sock_flag(sk, flag)) { 3054 unsigned long previous_flags = sk->sk_flags; 3055 3056 sock_set_flag(sk, flag); 3057 /* 3058 * we just set one of the two flags which require net 3059 * time stamping, but time stamping might have been on 3060 * already because of the other one 3061 */ 3062 if (sock_needs_netstamp(sk) && 3063 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3064 net_enable_timestamp(); 3065 } 3066 } 3067 3068 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3069 int level, int type) 3070 { 3071 struct sock_exterr_skb *serr; 3072 struct sk_buff *skb; 3073 int copied, err; 3074 3075 err = -EAGAIN; 3076 skb = sock_dequeue_err_skb(sk); 3077 if (skb == NULL) 3078 goto out; 3079 3080 copied = skb->len; 3081 if (copied > len) { 3082 msg->msg_flags |= MSG_TRUNC; 3083 copied = len; 3084 } 3085 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3086 if (err) 3087 goto out_free_skb; 3088 3089 sock_recv_timestamp(msg, sk, skb); 3090 3091 serr = SKB_EXT_ERR(skb); 3092 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3093 3094 msg->msg_flags |= MSG_ERRQUEUE; 3095 err = copied; 3096 3097 out_free_skb: 3098 kfree_skb(skb); 3099 out: 3100 return err; 3101 } 3102 EXPORT_SYMBOL(sock_recv_errqueue); 3103 3104 /* 3105 * Get a socket option on an socket. 3106 * 3107 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3108 * asynchronous errors should be reported by getsockopt. We assume 3109 * this means if you specify SO_ERROR (otherwise whats the point of it). 3110 */ 3111 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3112 char __user *optval, int __user *optlen) 3113 { 3114 struct sock *sk = sock->sk; 3115 3116 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3117 } 3118 EXPORT_SYMBOL(sock_common_getsockopt); 3119 3120 #ifdef CONFIG_COMPAT 3121 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 3122 char __user *optval, int __user *optlen) 3123 { 3124 struct sock *sk = sock->sk; 3125 3126 if (sk->sk_prot->compat_getsockopt != NULL) 3127 return sk->sk_prot->compat_getsockopt(sk, level, optname, 3128 optval, optlen); 3129 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3130 } 3131 EXPORT_SYMBOL(compat_sock_common_getsockopt); 3132 #endif 3133 3134 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3135 int flags) 3136 { 3137 struct sock *sk = sock->sk; 3138 int addr_len = 0; 3139 int err; 3140 3141 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3142 flags & ~MSG_DONTWAIT, &addr_len); 3143 if (err >= 0) 3144 msg->msg_namelen = addr_len; 3145 return err; 3146 } 3147 EXPORT_SYMBOL(sock_common_recvmsg); 3148 3149 /* 3150 * Set socket options on an inet socket. 3151 */ 3152 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3153 char __user *optval, unsigned int optlen) 3154 { 3155 struct sock *sk = sock->sk; 3156 3157 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3158 } 3159 EXPORT_SYMBOL(sock_common_setsockopt); 3160 3161 #ifdef CONFIG_COMPAT 3162 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 3163 char __user *optval, unsigned int optlen) 3164 { 3165 struct sock *sk = sock->sk; 3166 3167 if (sk->sk_prot->compat_setsockopt != NULL) 3168 return sk->sk_prot->compat_setsockopt(sk, level, optname, 3169 optval, optlen); 3170 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3171 } 3172 EXPORT_SYMBOL(compat_sock_common_setsockopt); 3173 #endif 3174 3175 void sk_common_release(struct sock *sk) 3176 { 3177 if (sk->sk_prot->destroy) 3178 sk->sk_prot->destroy(sk); 3179 3180 /* 3181 * Observation: when sock_common_release is called, processes have 3182 * no access to socket. But net still has. 3183 * Step one, detach it from networking: 3184 * 3185 * A. Remove from hash tables. 3186 */ 3187 3188 sk->sk_prot->unhash(sk); 3189 3190 /* 3191 * In this point socket cannot receive new packets, but it is possible 3192 * that some packets are in flight because some CPU runs receiver and 3193 * did hash table lookup before we unhashed socket. They will achieve 3194 * receive queue and will be purged by socket destructor. 3195 * 3196 * Also we still have packets pending on receive queue and probably, 3197 * our own packets waiting in device queues. sock_destroy will drain 3198 * receive queue, but transmitted packets will delay socket destruction 3199 * until the last reference will be released. 3200 */ 3201 3202 sock_orphan(sk); 3203 3204 xfrm_sk_free_policy(sk); 3205 3206 sk_refcnt_debug_release(sk); 3207 3208 sock_put(sk); 3209 } 3210 EXPORT_SYMBOL(sk_common_release); 3211 3212 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3213 { 3214 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3215 3216 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3217 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3218 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3219 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3220 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3221 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3222 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3223 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3224 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3225 } 3226 3227 #ifdef CONFIG_PROC_FS 3228 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3229 struct prot_inuse { 3230 int val[PROTO_INUSE_NR]; 3231 }; 3232 3233 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3234 3235 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3236 { 3237 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3238 } 3239 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3240 3241 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3242 { 3243 int cpu, idx = prot->inuse_idx; 3244 int res = 0; 3245 3246 for_each_possible_cpu(cpu) 3247 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3248 3249 return res >= 0 ? res : 0; 3250 } 3251 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3252 3253 static void sock_inuse_add(struct net *net, int val) 3254 { 3255 this_cpu_add(*net->core.sock_inuse, val); 3256 } 3257 3258 int sock_inuse_get(struct net *net) 3259 { 3260 int cpu, res = 0; 3261 3262 for_each_possible_cpu(cpu) 3263 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3264 3265 return res; 3266 } 3267 3268 EXPORT_SYMBOL_GPL(sock_inuse_get); 3269 3270 static int __net_init sock_inuse_init_net(struct net *net) 3271 { 3272 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3273 if (net->core.prot_inuse == NULL) 3274 return -ENOMEM; 3275 3276 net->core.sock_inuse = alloc_percpu(int); 3277 if (net->core.sock_inuse == NULL) 3278 goto out; 3279 3280 return 0; 3281 3282 out: 3283 free_percpu(net->core.prot_inuse); 3284 return -ENOMEM; 3285 } 3286 3287 static void __net_exit sock_inuse_exit_net(struct net *net) 3288 { 3289 free_percpu(net->core.prot_inuse); 3290 free_percpu(net->core.sock_inuse); 3291 } 3292 3293 static struct pernet_operations net_inuse_ops = { 3294 .init = sock_inuse_init_net, 3295 .exit = sock_inuse_exit_net, 3296 }; 3297 3298 static __init int net_inuse_init(void) 3299 { 3300 if (register_pernet_subsys(&net_inuse_ops)) 3301 panic("Cannot initialize net inuse counters"); 3302 3303 return 0; 3304 } 3305 3306 core_initcall(net_inuse_init); 3307 3308 static int assign_proto_idx(struct proto *prot) 3309 { 3310 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3311 3312 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3313 pr_err("PROTO_INUSE_NR exhausted\n"); 3314 return -ENOSPC; 3315 } 3316 3317 set_bit(prot->inuse_idx, proto_inuse_idx); 3318 return 0; 3319 } 3320 3321 static void release_proto_idx(struct proto *prot) 3322 { 3323 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3324 clear_bit(prot->inuse_idx, proto_inuse_idx); 3325 } 3326 #else 3327 static inline int assign_proto_idx(struct proto *prot) 3328 { 3329 return 0; 3330 } 3331 3332 static inline void release_proto_idx(struct proto *prot) 3333 { 3334 } 3335 3336 static void sock_inuse_add(struct net *net, int val) 3337 { 3338 } 3339 #endif 3340 3341 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3342 { 3343 if (!rsk_prot) 3344 return; 3345 kfree(rsk_prot->slab_name); 3346 rsk_prot->slab_name = NULL; 3347 kmem_cache_destroy(rsk_prot->slab); 3348 rsk_prot->slab = NULL; 3349 } 3350 3351 static int req_prot_init(const struct proto *prot) 3352 { 3353 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3354 3355 if (!rsk_prot) 3356 return 0; 3357 3358 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3359 prot->name); 3360 if (!rsk_prot->slab_name) 3361 return -ENOMEM; 3362 3363 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3364 rsk_prot->obj_size, 0, 3365 SLAB_ACCOUNT | prot->slab_flags, 3366 NULL); 3367 3368 if (!rsk_prot->slab) { 3369 pr_crit("%s: Can't create request sock SLAB cache!\n", 3370 prot->name); 3371 return -ENOMEM; 3372 } 3373 return 0; 3374 } 3375 3376 int proto_register(struct proto *prot, int alloc_slab) 3377 { 3378 int ret = -ENOBUFS; 3379 3380 if (alloc_slab) { 3381 prot->slab = kmem_cache_create_usercopy(prot->name, 3382 prot->obj_size, 0, 3383 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3384 prot->slab_flags, 3385 prot->useroffset, prot->usersize, 3386 NULL); 3387 3388 if (prot->slab == NULL) { 3389 pr_crit("%s: Can't create sock SLAB cache!\n", 3390 prot->name); 3391 goto out; 3392 } 3393 3394 if (req_prot_init(prot)) 3395 goto out_free_request_sock_slab; 3396 3397 if (prot->twsk_prot != NULL) { 3398 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3399 3400 if (prot->twsk_prot->twsk_slab_name == NULL) 3401 goto out_free_request_sock_slab; 3402 3403 prot->twsk_prot->twsk_slab = 3404 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3405 prot->twsk_prot->twsk_obj_size, 3406 0, 3407 SLAB_ACCOUNT | 3408 prot->slab_flags, 3409 NULL); 3410 if (prot->twsk_prot->twsk_slab == NULL) 3411 goto out_free_timewait_sock_slab_name; 3412 } 3413 } 3414 3415 mutex_lock(&proto_list_mutex); 3416 ret = assign_proto_idx(prot); 3417 if (ret) { 3418 mutex_unlock(&proto_list_mutex); 3419 goto out_free_timewait_sock_slab_name; 3420 } 3421 list_add(&prot->node, &proto_list); 3422 mutex_unlock(&proto_list_mutex); 3423 return ret; 3424 3425 out_free_timewait_sock_slab_name: 3426 if (alloc_slab && prot->twsk_prot) 3427 kfree(prot->twsk_prot->twsk_slab_name); 3428 out_free_request_sock_slab: 3429 if (alloc_slab) { 3430 req_prot_cleanup(prot->rsk_prot); 3431 3432 kmem_cache_destroy(prot->slab); 3433 prot->slab = NULL; 3434 } 3435 out: 3436 return ret; 3437 } 3438 EXPORT_SYMBOL(proto_register); 3439 3440 void proto_unregister(struct proto *prot) 3441 { 3442 mutex_lock(&proto_list_mutex); 3443 release_proto_idx(prot); 3444 list_del(&prot->node); 3445 mutex_unlock(&proto_list_mutex); 3446 3447 kmem_cache_destroy(prot->slab); 3448 prot->slab = NULL; 3449 3450 req_prot_cleanup(prot->rsk_prot); 3451 3452 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3453 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3454 kfree(prot->twsk_prot->twsk_slab_name); 3455 prot->twsk_prot->twsk_slab = NULL; 3456 } 3457 } 3458 EXPORT_SYMBOL(proto_unregister); 3459 3460 int sock_load_diag_module(int family, int protocol) 3461 { 3462 if (!protocol) { 3463 if (!sock_is_registered(family)) 3464 return -ENOENT; 3465 3466 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3467 NETLINK_SOCK_DIAG, family); 3468 } 3469 3470 #ifdef CONFIG_INET 3471 if (family == AF_INET && 3472 protocol != IPPROTO_RAW && 3473 !rcu_access_pointer(inet_protos[protocol])) 3474 return -ENOENT; 3475 #endif 3476 3477 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3478 NETLINK_SOCK_DIAG, family, protocol); 3479 } 3480 EXPORT_SYMBOL(sock_load_diag_module); 3481 3482 #ifdef CONFIG_PROC_FS 3483 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3484 __acquires(proto_list_mutex) 3485 { 3486 mutex_lock(&proto_list_mutex); 3487 return seq_list_start_head(&proto_list, *pos); 3488 } 3489 3490 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3491 { 3492 return seq_list_next(v, &proto_list, pos); 3493 } 3494 3495 static void proto_seq_stop(struct seq_file *seq, void *v) 3496 __releases(proto_list_mutex) 3497 { 3498 mutex_unlock(&proto_list_mutex); 3499 } 3500 3501 static char proto_method_implemented(const void *method) 3502 { 3503 return method == NULL ? 'n' : 'y'; 3504 } 3505 static long sock_prot_memory_allocated(struct proto *proto) 3506 { 3507 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3508 } 3509 3510 static const char *sock_prot_memory_pressure(struct proto *proto) 3511 { 3512 return proto->memory_pressure != NULL ? 3513 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3514 } 3515 3516 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3517 { 3518 3519 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3520 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3521 proto->name, 3522 proto->obj_size, 3523 sock_prot_inuse_get(seq_file_net(seq), proto), 3524 sock_prot_memory_allocated(proto), 3525 sock_prot_memory_pressure(proto), 3526 proto->max_header, 3527 proto->slab == NULL ? "no" : "yes", 3528 module_name(proto->owner), 3529 proto_method_implemented(proto->close), 3530 proto_method_implemented(proto->connect), 3531 proto_method_implemented(proto->disconnect), 3532 proto_method_implemented(proto->accept), 3533 proto_method_implemented(proto->ioctl), 3534 proto_method_implemented(proto->init), 3535 proto_method_implemented(proto->destroy), 3536 proto_method_implemented(proto->shutdown), 3537 proto_method_implemented(proto->setsockopt), 3538 proto_method_implemented(proto->getsockopt), 3539 proto_method_implemented(proto->sendmsg), 3540 proto_method_implemented(proto->recvmsg), 3541 proto_method_implemented(proto->sendpage), 3542 proto_method_implemented(proto->bind), 3543 proto_method_implemented(proto->backlog_rcv), 3544 proto_method_implemented(proto->hash), 3545 proto_method_implemented(proto->unhash), 3546 proto_method_implemented(proto->get_port), 3547 proto_method_implemented(proto->enter_memory_pressure)); 3548 } 3549 3550 static int proto_seq_show(struct seq_file *seq, void *v) 3551 { 3552 if (v == &proto_list) 3553 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3554 "protocol", 3555 "size", 3556 "sockets", 3557 "memory", 3558 "press", 3559 "maxhdr", 3560 "slab", 3561 "module", 3562 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3563 else 3564 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3565 return 0; 3566 } 3567 3568 static const struct seq_operations proto_seq_ops = { 3569 .start = proto_seq_start, 3570 .next = proto_seq_next, 3571 .stop = proto_seq_stop, 3572 .show = proto_seq_show, 3573 }; 3574 3575 static __net_init int proto_init_net(struct net *net) 3576 { 3577 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3578 sizeof(struct seq_net_private))) 3579 return -ENOMEM; 3580 3581 return 0; 3582 } 3583 3584 static __net_exit void proto_exit_net(struct net *net) 3585 { 3586 remove_proc_entry("protocols", net->proc_net); 3587 } 3588 3589 3590 static __net_initdata struct pernet_operations proto_net_ops = { 3591 .init = proto_init_net, 3592 .exit = proto_exit_net, 3593 }; 3594 3595 static int __init proto_init(void) 3596 { 3597 return register_pernet_subsys(&proto_net_ops); 3598 } 3599 3600 subsys_initcall(proto_init); 3601 3602 #endif /* PROC_FS */ 3603 3604 #ifdef CONFIG_NET_RX_BUSY_POLL 3605 bool sk_busy_loop_end(void *p, unsigned long start_time) 3606 { 3607 struct sock *sk = p; 3608 3609 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3610 sk_busy_loop_timeout(sk, start_time); 3611 } 3612 EXPORT_SYMBOL(sk_busy_loop_end); 3613 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3614