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 (sk->sk_bound_dev_if && !ns_capable(net->user_ns, CAP_NET_RAW)) 578 goto out; 579 580 ret = -EINVAL; 581 if (ifindex < 0) 582 goto out; 583 584 sk->sk_bound_dev_if = ifindex; 585 if (sk->sk_prot->rehash) 586 sk->sk_prot->rehash(sk); 587 sk_dst_reset(sk); 588 589 ret = 0; 590 591 out: 592 #endif 593 594 return ret; 595 } 596 597 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 (optlen != sizeof(struct sock_txtime)) { 1156 ret = -EINVAL; 1157 break; 1158 } else if (copy_from_user(&sk_txtime, optval, 1159 sizeof(struct sock_txtime))) { 1160 ret = -EFAULT; 1161 break; 1162 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1163 ret = -EINVAL; 1164 break; 1165 } 1166 /* CLOCK_MONOTONIC is only used by sch_fq, and this packet 1167 * scheduler has enough safe guards. 1168 */ 1169 if (sk_txtime.clockid != CLOCK_MONOTONIC && 1170 !ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1171 ret = -EPERM; 1172 break; 1173 } 1174 sock_valbool_flag(sk, SOCK_TXTIME, true); 1175 sk->sk_clockid = sk_txtime.clockid; 1176 sk->sk_txtime_deadline_mode = 1177 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1178 sk->sk_txtime_report_errors = 1179 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1180 break; 1181 1182 case SO_BINDTOIFINDEX: 1183 ret = sock_setbindtodevice_locked(sk, val); 1184 break; 1185 1186 default: 1187 ret = -ENOPROTOOPT; 1188 break; 1189 } 1190 release_sock(sk); 1191 return ret; 1192 } 1193 EXPORT_SYMBOL(sock_setsockopt); 1194 1195 1196 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1197 struct ucred *ucred) 1198 { 1199 ucred->pid = pid_vnr(pid); 1200 ucred->uid = ucred->gid = -1; 1201 if (cred) { 1202 struct user_namespace *current_ns = current_user_ns(); 1203 1204 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1205 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1206 } 1207 } 1208 1209 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1210 { 1211 struct user_namespace *user_ns = current_user_ns(); 1212 int i; 1213 1214 for (i = 0; i < src->ngroups; i++) 1215 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1216 return -EFAULT; 1217 1218 return 0; 1219 } 1220 1221 int sock_getsockopt(struct socket *sock, int level, int optname, 1222 char __user *optval, int __user *optlen) 1223 { 1224 struct sock *sk = sock->sk; 1225 1226 union { 1227 int val; 1228 u64 val64; 1229 unsigned long ulval; 1230 struct linger ling; 1231 struct old_timeval32 tm32; 1232 struct __kernel_old_timeval tm; 1233 struct __kernel_sock_timeval stm; 1234 struct sock_txtime txtime; 1235 } v; 1236 1237 int lv = sizeof(int); 1238 int len; 1239 1240 if (get_user(len, optlen)) 1241 return -EFAULT; 1242 if (len < 0) 1243 return -EINVAL; 1244 1245 memset(&v, 0, sizeof(v)); 1246 1247 switch (optname) { 1248 case SO_DEBUG: 1249 v.val = sock_flag(sk, SOCK_DBG); 1250 break; 1251 1252 case SO_DONTROUTE: 1253 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1254 break; 1255 1256 case SO_BROADCAST: 1257 v.val = sock_flag(sk, SOCK_BROADCAST); 1258 break; 1259 1260 case SO_SNDBUF: 1261 v.val = sk->sk_sndbuf; 1262 break; 1263 1264 case SO_RCVBUF: 1265 v.val = sk->sk_rcvbuf; 1266 break; 1267 1268 case SO_REUSEADDR: 1269 v.val = sk->sk_reuse; 1270 break; 1271 1272 case SO_REUSEPORT: 1273 v.val = sk->sk_reuseport; 1274 break; 1275 1276 case SO_KEEPALIVE: 1277 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1278 break; 1279 1280 case SO_TYPE: 1281 v.val = sk->sk_type; 1282 break; 1283 1284 case SO_PROTOCOL: 1285 v.val = sk->sk_protocol; 1286 break; 1287 1288 case SO_DOMAIN: 1289 v.val = sk->sk_family; 1290 break; 1291 1292 case SO_ERROR: 1293 v.val = -sock_error(sk); 1294 if (v.val == 0) 1295 v.val = xchg(&sk->sk_err_soft, 0); 1296 break; 1297 1298 case SO_OOBINLINE: 1299 v.val = sock_flag(sk, SOCK_URGINLINE); 1300 break; 1301 1302 case SO_NO_CHECK: 1303 v.val = sk->sk_no_check_tx; 1304 break; 1305 1306 case SO_PRIORITY: 1307 v.val = sk->sk_priority; 1308 break; 1309 1310 case SO_LINGER: 1311 lv = sizeof(v.ling); 1312 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1313 v.ling.l_linger = sk->sk_lingertime / HZ; 1314 break; 1315 1316 case SO_BSDCOMPAT: 1317 sock_warn_obsolete_bsdism("getsockopt"); 1318 break; 1319 1320 case SO_TIMESTAMP_OLD: 1321 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1322 !sock_flag(sk, SOCK_TSTAMP_NEW) && 1323 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1324 break; 1325 1326 case SO_TIMESTAMPNS_OLD: 1327 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && !sock_flag(sk, SOCK_TSTAMP_NEW); 1328 break; 1329 1330 case SO_TIMESTAMP_NEW: 1331 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && sock_flag(sk, SOCK_TSTAMP_NEW); 1332 break; 1333 1334 case SO_TIMESTAMPNS_NEW: 1335 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS) && sock_flag(sk, SOCK_TSTAMP_NEW); 1336 break; 1337 1338 case SO_TIMESTAMPING_OLD: 1339 v.val = sk->sk_tsflags; 1340 break; 1341 1342 case SO_RCVTIMEO_OLD: 1343 case SO_RCVTIMEO_NEW: 1344 lv = sock_get_timeout(sk->sk_rcvtimeo, &v, SO_RCVTIMEO_OLD == optname); 1345 break; 1346 1347 case SO_SNDTIMEO_OLD: 1348 case SO_SNDTIMEO_NEW: 1349 lv = sock_get_timeout(sk->sk_sndtimeo, &v, SO_SNDTIMEO_OLD == optname); 1350 break; 1351 1352 case SO_RCVLOWAT: 1353 v.val = sk->sk_rcvlowat; 1354 break; 1355 1356 case SO_SNDLOWAT: 1357 v.val = 1; 1358 break; 1359 1360 case SO_PASSCRED: 1361 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1362 break; 1363 1364 case SO_PEERCRED: 1365 { 1366 struct ucred peercred; 1367 if (len > sizeof(peercred)) 1368 len = sizeof(peercred); 1369 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1370 if (copy_to_user(optval, &peercred, len)) 1371 return -EFAULT; 1372 goto lenout; 1373 } 1374 1375 case SO_PEERGROUPS: 1376 { 1377 int ret, n; 1378 1379 if (!sk->sk_peer_cred) 1380 return -ENODATA; 1381 1382 n = sk->sk_peer_cred->group_info->ngroups; 1383 if (len < n * sizeof(gid_t)) { 1384 len = n * sizeof(gid_t); 1385 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1386 } 1387 len = n * sizeof(gid_t); 1388 1389 ret = groups_to_user((gid_t __user *)optval, 1390 sk->sk_peer_cred->group_info); 1391 if (ret) 1392 return ret; 1393 goto lenout; 1394 } 1395 1396 case SO_PEERNAME: 1397 { 1398 char address[128]; 1399 1400 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1401 if (lv < 0) 1402 return -ENOTCONN; 1403 if (lv < len) 1404 return -EINVAL; 1405 if (copy_to_user(optval, address, len)) 1406 return -EFAULT; 1407 goto lenout; 1408 } 1409 1410 /* Dubious BSD thing... Probably nobody even uses it, but 1411 * the UNIX standard wants it for whatever reason... -DaveM 1412 */ 1413 case SO_ACCEPTCONN: 1414 v.val = sk->sk_state == TCP_LISTEN; 1415 break; 1416 1417 case SO_PASSSEC: 1418 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1419 break; 1420 1421 case SO_PEERSEC: 1422 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1423 1424 case SO_MARK: 1425 v.val = sk->sk_mark; 1426 break; 1427 1428 case SO_RXQ_OVFL: 1429 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1430 break; 1431 1432 case SO_WIFI_STATUS: 1433 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1434 break; 1435 1436 case SO_PEEK_OFF: 1437 if (!sock->ops->set_peek_off) 1438 return -EOPNOTSUPP; 1439 1440 v.val = sk->sk_peek_off; 1441 break; 1442 case SO_NOFCS: 1443 v.val = sock_flag(sk, SOCK_NOFCS); 1444 break; 1445 1446 case SO_BINDTODEVICE: 1447 return sock_getbindtodevice(sk, optval, optlen, len); 1448 1449 case SO_GET_FILTER: 1450 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1451 if (len < 0) 1452 return len; 1453 1454 goto lenout; 1455 1456 case SO_LOCK_FILTER: 1457 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1458 break; 1459 1460 case SO_BPF_EXTENSIONS: 1461 v.val = bpf_tell_extensions(); 1462 break; 1463 1464 case SO_SELECT_ERR_QUEUE: 1465 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1466 break; 1467 1468 #ifdef CONFIG_NET_RX_BUSY_POLL 1469 case SO_BUSY_POLL: 1470 v.val = sk->sk_ll_usec; 1471 break; 1472 #endif 1473 1474 case SO_MAX_PACING_RATE: 1475 if (sizeof(v.ulval) != sizeof(v.val) && len >= sizeof(v.ulval)) { 1476 lv = sizeof(v.ulval); 1477 v.ulval = sk->sk_max_pacing_rate; 1478 } else { 1479 /* 32bit version */ 1480 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1481 } 1482 break; 1483 1484 case SO_INCOMING_CPU: 1485 v.val = READ_ONCE(sk->sk_incoming_cpu); 1486 break; 1487 1488 case SO_MEMINFO: 1489 { 1490 u32 meminfo[SK_MEMINFO_VARS]; 1491 1492 sk_get_meminfo(sk, meminfo); 1493 1494 len = min_t(unsigned int, len, sizeof(meminfo)); 1495 if (copy_to_user(optval, &meminfo, len)) 1496 return -EFAULT; 1497 1498 goto lenout; 1499 } 1500 1501 #ifdef CONFIG_NET_RX_BUSY_POLL 1502 case SO_INCOMING_NAPI_ID: 1503 v.val = READ_ONCE(sk->sk_napi_id); 1504 1505 /* aggregate non-NAPI IDs down to 0 */ 1506 if (v.val < MIN_NAPI_ID) 1507 v.val = 0; 1508 1509 break; 1510 #endif 1511 1512 case SO_COOKIE: 1513 lv = sizeof(u64); 1514 if (len < lv) 1515 return -EINVAL; 1516 v.val64 = sock_gen_cookie(sk); 1517 break; 1518 1519 case SO_ZEROCOPY: 1520 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1521 break; 1522 1523 case SO_TXTIME: 1524 lv = sizeof(v.txtime); 1525 v.txtime.clockid = sk->sk_clockid; 1526 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1527 SOF_TXTIME_DEADLINE_MODE : 0; 1528 v.txtime.flags |= sk->sk_txtime_report_errors ? 1529 SOF_TXTIME_REPORT_ERRORS : 0; 1530 break; 1531 1532 case SO_BINDTOIFINDEX: 1533 v.val = sk->sk_bound_dev_if; 1534 break; 1535 1536 default: 1537 /* We implement the SO_SNDLOWAT etc to not be settable 1538 * (1003.1g 7). 1539 */ 1540 return -ENOPROTOOPT; 1541 } 1542 1543 if (len > lv) 1544 len = lv; 1545 if (copy_to_user(optval, &v, len)) 1546 return -EFAULT; 1547 lenout: 1548 if (put_user(len, optlen)) 1549 return -EFAULT; 1550 return 0; 1551 } 1552 1553 /* 1554 * Initialize an sk_lock. 1555 * 1556 * (We also register the sk_lock with the lock validator.) 1557 */ 1558 static inline void sock_lock_init(struct sock *sk) 1559 { 1560 if (sk->sk_kern_sock) 1561 sock_lock_init_class_and_name( 1562 sk, 1563 af_family_kern_slock_key_strings[sk->sk_family], 1564 af_family_kern_slock_keys + sk->sk_family, 1565 af_family_kern_key_strings[sk->sk_family], 1566 af_family_kern_keys + sk->sk_family); 1567 else 1568 sock_lock_init_class_and_name( 1569 sk, 1570 af_family_slock_key_strings[sk->sk_family], 1571 af_family_slock_keys + sk->sk_family, 1572 af_family_key_strings[sk->sk_family], 1573 af_family_keys + sk->sk_family); 1574 } 1575 1576 /* 1577 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1578 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1579 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1580 */ 1581 static void sock_copy(struct sock *nsk, const struct sock *osk) 1582 { 1583 const struct proto *prot = READ_ONCE(osk->sk_prot); 1584 #ifdef CONFIG_SECURITY_NETWORK 1585 void *sptr = nsk->sk_security; 1586 #endif 1587 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1588 1589 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1590 prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1591 1592 #ifdef CONFIG_SECURITY_NETWORK 1593 nsk->sk_security = sptr; 1594 security_sk_clone(osk, nsk); 1595 #endif 1596 } 1597 1598 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1599 int family) 1600 { 1601 struct sock *sk; 1602 struct kmem_cache *slab; 1603 1604 slab = prot->slab; 1605 if (slab != NULL) { 1606 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1607 if (!sk) 1608 return sk; 1609 if (want_init_on_alloc(priority)) 1610 sk_prot_clear_nulls(sk, prot->obj_size); 1611 } else 1612 sk = kmalloc(prot->obj_size, priority); 1613 1614 if (sk != NULL) { 1615 if (security_sk_alloc(sk, family, priority)) 1616 goto out_free; 1617 1618 if (!try_module_get(prot->owner)) 1619 goto out_free_sec; 1620 sk_tx_queue_clear(sk); 1621 } 1622 1623 return sk; 1624 1625 out_free_sec: 1626 security_sk_free(sk); 1627 out_free: 1628 if (slab != NULL) 1629 kmem_cache_free(slab, sk); 1630 else 1631 kfree(sk); 1632 return NULL; 1633 } 1634 1635 static void sk_prot_free(struct proto *prot, struct sock *sk) 1636 { 1637 struct kmem_cache *slab; 1638 struct module *owner; 1639 1640 owner = prot->owner; 1641 slab = prot->slab; 1642 1643 cgroup_sk_free(&sk->sk_cgrp_data); 1644 mem_cgroup_sk_free(sk); 1645 security_sk_free(sk); 1646 if (slab != NULL) 1647 kmem_cache_free(slab, sk); 1648 else 1649 kfree(sk); 1650 module_put(owner); 1651 } 1652 1653 /** 1654 * sk_alloc - All socket objects are allocated here 1655 * @net: the applicable net namespace 1656 * @family: protocol family 1657 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1658 * @prot: struct proto associated with this new sock instance 1659 * @kern: is this to be a kernel socket? 1660 */ 1661 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1662 struct proto *prot, int kern) 1663 { 1664 struct sock *sk; 1665 1666 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1667 if (sk) { 1668 sk->sk_family = family; 1669 /* 1670 * See comment in struct sock definition to understand 1671 * why we need sk_prot_creator -acme 1672 */ 1673 sk->sk_prot = sk->sk_prot_creator = prot; 1674 sk->sk_kern_sock = kern; 1675 sock_lock_init(sk); 1676 sk->sk_net_refcnt = kern ? 0 : 1; 1677 if (likely(sk->sk_net_refcnt)) { 1678 get_net(net); 1679 sock_inuse_add(net, 1); 1680 } 1681 1682 sock_net_set(sk, net); 1683 refcount_set(&sk->sk_wmem_alloc, 1); 1684 1685 mem_cgroup_sk_alloc(sk); 1686 cgroup_sk_alloc(&sk->sk_cgrp_data); 1687 sock_update_classid(&sk->sk_cgrp_data); 1688 sock_update_netprioidx(&sk->sk_cgrp_data); 1689 } 1690 1691 return sk; 1692 } 1693 EXPORT_SYMBOL(sk_alloc); 1694 1695 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1696 * grace period. This is the case for UDP sockets and TCP listeners. 1697 */ 1698 static void __sk_destruct(struct rcu_head *head) 1699 { 1700 struct sock *sk = container_of(head, struct sock, sk_rcu); 1701 struct sk_filter *filter; 1702 1703 if (sk->sk_destruct) 1704 sk->sk_destruct(sk); 1705 1706 filter = rcu_dereference_check(sk->sk_filter, 1707 refcount_read(&sk->sk_wmem_alloc) == 0); 1708 if (filter) { 1709 sk_filter_uncharge(sk, filter); 1710 RCU_INIT_POINTER(sk->sk_filter, NULL); 1711 } 1712 1713 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1714 1715 #ifdef CONFIG_BPF_SYSCALL 1716 bpf_sk_storage_free(sk); 1717 #endif 1718 1719 if (atomic_read(&sk->sk_omem_alloc)) 1720 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1721 __func__, atomic_read(&sk->sk_omem_alloc)); 1722 1723 if (sk->sk_frag.page) { 1724 put_page(sk->sk_frag.page); 1725 sk->sk_frag.page = NULL; 1726 } 1727 1728 if (sk->sk_peer_cred) 1729 put_cred(sk->sk_peer_cred); 1730 put_pid(sk->sk_peer_pid); 1731 if (likely(sk->sk_net_refcnt)) 1732 put_net(sock_net(sk)); 1733 sk_prot_free(sk->sk_prot_creator, sk); 1734 } 1735 1736 void sk_destruct(struct sock *sk) 1737 { 1738 bool use_call_rcu = sock_flag(sk, SOCK_RCU_FREE); 1739 1740 if (rcu_access_pointer(sk->sk_reuseport_cb)) { 1741 reuseport_detach_sock(sk); 1742 use_call_rcu = true; 1743 } 1744 1745 if (use_call_rcu) 1746 call_rcu(&sk->sk_rcu, __sk_destruct); 1747 else 1748 __sk_destruct(&sk->sk_rcu); 1749 } 1750 1751 static void __sk_free(struct sock *sk) 1752 { 1753 if (likely(sk->sk_net_refcnt)) 1754 sock_inuse_add(sock_net(sk), -1); 1755 1756 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1757 sock_diag_broadcast_destroy(sk); 1758 else 1759 sk_destruct(sk); 1760 } 1761 1762 void sk_free(struct sock *sk) 1763 { 1764 /* 1765 * We subtract one from sk_wmem_alloc and can know if 1766 * some packets are still in some tx queue. 1767 * If not null, sock_wfree() will call __sk_free(sk) later 1768 */ 1769 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1770 __sk_free(sk); 1771 } 1772 EXPORT_SYMBOL(sk_free); 1773 1774 static void sk_init_common(struct sock *sk) 1775 { 1776 skb_queue_head_init(&sk->sk_receive_queue); 1777 skb_queue_head_init(&sk->sk_write_queue); 1778 skb_queue_head_init(&sk->sk_error_queue); 1779 1780 rwlock_init(&sk->sk_callback_lock); 1781 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1782 af_rlock_keys + sk->sk_family, 1783 af_family_rlock_key_strings[sk->sk_family]); 1784 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1785 af_wlock_keys + sk->sk_family, 1786 af_family_wlock_key_strings[sk->sk_family]); 1787 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1788 af_elock_keys + sk->sk_family, 1789 af_family_elock_key_strings[sk->sk_family]); 1790 lockdep_set_class_and_name(&sk->sk_callback_lock, 1791 af_callback_keys + sk->sk_family, 1792 af_family_clock_key_strings[sk->sk_family]); 1793 } 1794 1795 /** 1796 * sk_clone_lock - clone a socket, and lock its clone 1797 * @sk: the socket to clone 1798 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1799 * 1800 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1801 */ 1802 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1803 { 1804 struct proto *prot = READ_ONCE(sk->sk_prot); 1805 struct sock *newsk; 1806 bool is_charged = true; 1807 1808 newsk = sk_prot_alloc(prot, priority, sk->sk_family); 1809 if (newsk != NULL) { 1810 struct sk_filter *filter; 1811 1812 sock_copy(newsk, sk); 1813 1814 newsk->sk_prot_creator = prot; 1815 1816 /* SANITY */ 1817 if (likely(newsk->sk_net_refcnt)) 1818 get_net(sock_net(newsk)); 1819 sk_node_init(&newsk->sk_node); 1820 sock_lock_init(newsk); 1821 bh_lock_sock(newsk); 1822 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1823 newsk->sk_backlog.len = 0; 1824 1825 atomic_set(&newsk->sk_rmem_alloc, 0); 1826 /* 1827 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1828 */ 1829 refcount_set(&newsk->sk_wmem_alloc, 1); 1830 atomic_set(&newsk->sk_omem_alloc, 0); 1831 sk_init_common(newsk); 1832 1833 newsk->sk_dst_cache = NULL; 1834 newsk->sk_dst_pending_confirm = 0; 1835 newsk->sk_wmem_queued = 0; 1836 newsk->sk_forward_alloc = 0; 1837 atomic_set(&newsk->sk_drops, 0); 1838 newsk->sk_send_head = NULL; 1839 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1840 atomic_set(&newsk->sk_zckey, 0); 1841 1842 sock_reset_flag(newsk, SOCK_DONE); 1843 1844 /* sk->sk_memcg will be populated at accept() time */ 1845 newsk->sk_memcg = NULL; 1846 1847 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1848 1849 rcu_read_lock(); 1850 filter = rcu_dereference(sk->sk_filter); 1851 if (filter != NULL) 1852 /* though it's an empty new sock, the charging may fail 1853 * if sysctl_optmem_max was changed between creation of 1854 * original socket and cloning 1855 */ 1856 is_charged = sk_filter_charge(newsk, filter); 1857 RCU_INIT_POINTER(newsk->sk_filter, filter); 1858 rcu_read_unlock(); 1859 1860 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1861 /* We need to make sure that we don't uncharge the new 1862 * socket if we couldn't charge it in the first place 1863 * as otherwise we uncharge the parent's filter. 1864 */ 1865 if (!is_charged) 1866 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1867 sk_free_unlock_clone(newsk); 1868 newsk = NULL; 1869 goto out; 1870 } 1871 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1872 1873 if (bpf_sk_storage_clone(sk, newsk)) { 1874 sk_free_unlock_clone(newsk); 1875 newsk = NULL; 1876 goto out; 1877 } 1878 1879 /* Clear sk_user_data if parent had the pointer tagged 1880 * as not suitable for copying when cloning. 1881 */ 1882 if (sk_user_data_is_nocopy(newsk)) 1883 newsk->sk_user_data = NULL; 1884 1885 newsk->sk_err = 0; 1886 newsk->sk_err_soft = 0; 1887 newsk->sk_priority = 0; 1888 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1889 if (likely(newsk->sk_net_refcnt)) 1890 sock_inuse_add(sock_net(newsk), 1); 1891 1892 /* 1893 * Before updating sk_refcnt, we must commit prior changes to memory 1894 * (Documentation/RCU/rculist_nulls.txt for details) 1895 */ 1896 smp_wmb(); 1897 refcount_set(&newsk->sk_refcnt, 2); 1898 1899 /* 1900 * Increment the counter in the same struct proto as the master 1901 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1902 * is the same as sk->sk_prot->socks, as this field was copied 1903 * with memcpy). 1904 * 1905 * This _changes_ the previous behaviour, where 1906 * tcp_create_openreq_child always was incrementing the 1907 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1908 * to be taken into account in all callers. -acme 1909 */ 1910 sk_refcnt_debug_inc(newsk); 1911 sk_set_socket(newsk, NULL); 1912 RCU_INIT_POINTER(newsk->sk_wq, NULL); 1913 1914 if (newsk->sk_prot->sockets_allocated) 1915 sk_sockets_allocated_inc(newsk); 1916 1917 if (sock_needs_netstamp(sk) && 1918 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1919 net_enable_timestamp(); 1920 } 1921 out: 1922 return newsk; 1923 } 1924 EXPORT_SYMBOL_GPL(sk_clone_lock); 1925 1926 void sk_free_unlock_clone(struct sock *sk) 1927 { 1928 /* It is still raw copy of parent, so invalidate 1929 * destructor and make plain sk_free() */ 1930 sk->sk_destruct = NULL; 1931 bh_unlock_sock(sk); 1932 sk_free(sk); 1933 } 1934 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1935 1936 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1937 { 1938 u32 max_segs = 1; 1939 1940 sk_dst_set(sk, dst); 1941 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 1942 if (sk->sk_route_caps & NETIF_F_GSO) 1943 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1944 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1945 if (sk_can_gso(sk)) { 1946 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 1947 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1948 } else { 1949 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1950 sk->sk_gso_max_size = dst->dev->gso_max_size; 1951 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1952 } 1953 } 1954 sk->sk_gso_max_segs = max_segs; 1955 } 1956 EXPORT_SYMBOL_GPL(sk_setup_caps); 1957 1958 /* 1959 * Simple resource managers for sockets. 1960 */ 1961 1962 1963 /* 1964 * Write buffer destructor automatically called from kfree_skb. 1965 */ 1966 void sock_wfree(struct sk_buff *skb) 1967 { 1968 struct sock *sk = skb->sk; 1969 unsigned int len = skb->truesize; 1970 1971 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1972 /* 1973 * Keep a reference on sk_wmem_alloc, this will be released 1974 * after sk_write_space() call 1975 */ 1976 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 1977 sk->sk_write_space(sk); 1978 len = 1; 1979 } 1980 /* 1981 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1982 * could not do because of in-flight packets 1983 */ 1984 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 1985 __sk_free(sk); 1986 } 1987 EXPORT_SYMBOL(sock_wfree); 1988 1989 /* This variant of sock_wfree() is used by TCP, 1990 * since it sets SOCK_USE_WRITE_QUEUE. 1991 */ 1992 void __sock_wfree(struct sk_buff *skb) 1993 { 1994 struct sock *sk = skb->sk; 1995 1996 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1997 __sk_free(sk); 1998 } 1999 2000 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2001 { 2002 skb_orphan(skb); 2003 skb->sk = sk; 2004 #ifdef CONFIG_INET 2005 if (unlikely(!sk_fullsock(sk))) { 2006 skb->destructor = sock_edemux; 2007 sock_hold(sk); 2008 return; 2009 } 2010 #endif 2011 skb->destructor = sock_wfree; 2012 skb_set_hash_from_sk(skb, sk); 2013 /* 2014 * We used to take a refcount on sk, but following operation 2015 * is enough to guarantee sk_free() wont free this sock until 2016 * all in-flight packets are completed 2017 */ 2018 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 2019 } 2020 EXPORT_SYMBOL(skb_set_owner_w); 2021 2022 static bool can_skb_orphan_partial(const struct sk_buff *skb) 2023 { 2024 #ifdef CONFIG_TLS_DEVICE 2025 /* Drivers depend on in-order delivery for crypto offload, 2026 * partial orphan breaks out-of-order-OK logic. 2027 */ 2028 if (skb->decrypted) 2029 return false; 2030 #endif 2031 return (skb->destructor == sock_wfree || 2032 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree)); 2033 } 2034 2035 /* This helper is used by netem, as it can hold packets in its 2036 * delay queue. We want to allow the owner socket to send more 2037 * packets, as if they were already TX completed by a typical driver. 2038 * But we also want to keep skb->sk set because some packet schedulers 2039 * rely on it (sch_fq for example). 2040 */ 2041 void skb_orphan_partial(struct sk_buff *skb) 2042 { 2043 if (skb_is_tcp_pure_ack(skb)) 2044 return; 2045 2046 if (can_skb_orphan_partial(skb)) { 2047 struct sock *sk = skb->sk; 2048 2049 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 2050 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)); 2051 skb->destructor = sock_efree; 2052 } 2053 } else { 2054 skb_orphan(skb); 2055 } 2056 } 2057 EXPORT_SYMBOL(skb_orphan_partial); 2058 2059 /* 2060 * Read buffer destructor automatically called from kfree_skb. 2061 */ 2062 void sock_rfree(struct sk_buff *skb) 2063 { 2064 struct sock *sk = skb->sk; 2065 unsigned int len = skb->truesize; 2066 2067 atomic_sub(len, &sk->sk_rmem_alloc); 2068 sk_mem_uncharge(sk, len); 2069 } 2070 EXPORT_SYMBOL(sock_rfree); 2071 2072 /* 2073 * Buffer destructor for skbs that are not used directly in read or write 2074 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 2075 */ 2076 void sock_efree(struct sk_buff *skb) 2077 { 2078 sock_put(skb->sk); 2079 } 2080 EXPORT_SYMBOL(sock_efree); 2081 2082 /* Buffer destructor for prefetch/receive path where reference count may 2083 * not be held, e.g. for listen sockets. 2084 */ 2085 #ifdef CONFIG_INET 2086 void sock_pfree(struct sk_buff *skb) 2087 { 2088 if (sk_is_refcounted(skb->sk)) 2089 sock_gen_put(skb->sk); 2090 } 2091 EXPORT_SYMBOL(sock_pfree); 2092 #endif /* CONFIG_INET */ 2093 2094 kuid_t sock_i_uid(struct sock *sk) 2095 { 2096 kuid_t uid; 2097 2098 read_lock_bh(&sk->sk_callback_lock); 2099 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 2100 read_unlock_bh(&sk->sk_callback_lock); 2101 return uid; 2102 } 2103 EXPORT_SYMBOL(sock_i_uid); 2104 2105 unsigned long sock_i_ino(struct sock *sk) 2106 { 2107 unsigned long ino; 2108 2109 read_lock_bh(&sk->sk_callback_lock); 2110 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 2111 read_unlock_bh(&sk->sk_callback_lock); 2112 return ino; 2113 } 2114 EXPORT_SYMBOL(sock_i_ino); 2115 2116 /* 2117 * Allocate a skb from the socket's send buffer. 2118 */ 2119 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 2120 gfp_t priority) 2121 { 2122 if (force || 2123 refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) { 2124 struct sk_buff *skb = alloc_skb(size, priority); 2125 2126 if (skb) { 2127 skb_set_owner_w(skb, sk); 2128 return skb; 2129 } 2130 } 2131 return NULL; 2132 } 2133 EXPORT_SYMBOL(sock_wmalloc); 2134 2135 static void sock_ofree(struct sk_buff *skb) 2136 { 2137 struct sock *sk = skb->sk; 2138 2139 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 2140 } 2141 2142 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 2143 gfp_t priority) 2144 { 2145 struct sk_buff *skb; 2146 2147 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 2148 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 2149 sysctl_optmem_max) 2150 return NULL; 2151 2152 skb = alloc_skb(size, priority); 2153 if (!skb) 2154 return NULL; 2155 2156 atomic_add(skb->truesize, &sk->sk_omem_alloc); 2157 skb->sk = sk; 2158 skb->destructor = sock_ofree; 2159 return skb; 2160 } 2161 2162 /* 2163 * Allocate a memory block from the socket's option memory buffer. 2164 */ 2165 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 2166 { 2167 if ((unsigned int)size <= sysctl_optmem_max && 2168 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 2169 void *mem; 2170 /* First do the add, to avoid the race if kmalloc 2171 * might sleep. 2172 */ 2173 atomic_add(size, &sk->sk_omem_alloc); 2174 mem = kmalloc(size, priority); 2175 if (mem) 2176 return mem; 2177 atomic_sub(size, &sk->sk_omem_alloc); 2178 } 2179 return NULL; 2180 } 2181 EXPORT_SYMBOL(sock_kmalloc); 2182 2183 /* Free an option memory block. Note, we actually want the inline 2184 * here as this allows gcc to detect the nullify and fold away the 2185 * condition entirely. 2186 */ 2187 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2188 const bool nullify) 2189 { 2190 if (WARN_ON_ONCE(!mem)) 2191 return; 2192 if (nullify) 2193 kzfree(mem); 2194 else 2195 kfree(mem); 2196 atomic_sub(size, &sk->sk_omem_alloc); 2197 } 2198 2199 void sock_kfree_s(struct sock *sk, void *mem, int size) 2200 { 2201 __sock_kfree_s(sk, mem, size, false); 2202 } 2203 EXPORT_SYMBOL(sock_kfree_s); 2204 2205 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2206 { 2207 __sock_kfree_s(sk, mem, size, true); 2208 } 2209 EXPORT_SYMBOL(sock_kzfree_s); 2210 2211 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2212 I think, these locks should be removed for datagram sockets. 2213 */ 2214 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2215 { 2216 DEFINE_WAIT(wait); 2217 2218 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2219 for (;;) { 2220 if (!timeo) 2221 break; 2222 if (signal_pending(current)) 2223 break; 2224 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2225 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2226 if (refcount_read(&sk->sk_wmem_alloc) < READ_ONCE(sk->sk_sndbuf)) 2227 break; 2228 if (sk->sk_shutdown & SEND_SHUTDOWN) 2229 break; 2230 if (sk->sk_err) 2231 break; 2232 timeo = schedule_timeout(timeo); 2233 } 2234 finish_wait(sk_sleep(sk), &wait); 2235 return timeo; 2236 } 2237 2238 2239 /* 2240 * Generic send/receive buffer handlers 2241 */ 2242 2243 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2244 unsigned long data_len, int noblock, 2245 int *errcode, int max_page_order) 2246 { 2247 struct sk_buff *skb; 2248 long timeo; 2249 int err; 2250 2251 timeo = sock_sndtimeo(sk, noblock); 2252 for (;;) { 2253 err = sock_error(sk); 2254 if (err != 0) 2255 goto failure; 2256 2257 err = -EPIPE; 2258 if (sk->sk_shutdown & SEND_SHUTDOWN) 2259 goto failure; 2260 2261 if (sk_wmem_alloc_get(sk) < READ_ONCE(sk->sk_sndbuf)) 2262 break; 2263 2264 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2265 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2266 err = -EAGAIN; 2267 if (!timeo) 2268 goto failure; 2269 if (signal_pending(current)) 2270 goto interrupted; 2271 timeo = sock_wait_for_wmem(sk, timeo); 2272 } 2273 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2274 errcode, sk->sk_allocation); 2275 if (skb) 2276 skb_set_owner_w(skb, sk); 2277 return skb; 2278 2279 interrupted: 2280 err = sock_intr_errno(timeo); 2281 failure: 2282 *errcode = err; 2283 return NULL; 2284 } 2285 EXPORT_SYMBOL(sock_alloc_send_pskb); 2286 2287 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2288 int noblock, int *errcode) 2289 { 2290 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2291 } 2292 EXPORT_SYMBOL(sock_alloc_send_skb); 2293 2294 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2295 struct sockcm_cookie *sockc) 2296 { 2297 u32 tsflags; 2298 2299 switch (cmsg->cmsg_type) { 2300 case SO_MARK: 2301 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2302 return -EPERM; 2303 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2304 return -EINVAL; 2305 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2306 break; 2307 case SO_TIMESTAMPING_OLD: 2308 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2309 return -EINVAL; 2310 2311 tsflags = *(u32 *)CMSG_DATA(cmsg); 2312 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2313 return -EINVAL; 2314 2315 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2316 sockc->tsflags |= tsflags; 2317 break; 2318 case SCM_TXTIME: 2319 if (!sock_flag(sk, SOCK_TXTIME)) 2320 return -EINVAL; 2321 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2322 return -EINVAL; 2323 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2324 break; 2325 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2326 case SCM_RIGHTS: 2327 case SCM_CREDENTIALS: 2328 break; 2329 default: 2330 return -EINVAL; 2331 } 2332 return 0; 2333 } 2334 EXPORT_SYMBOL(__sock_cmsg_send); 2335 2336 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2337 struct sockcm_cookie *sockc) 2338 { 2339 struct cmsghdr *cmsg; 2340 int ret; 2341 2342 for_each_cmsghdr(cmsg, msg) { 2343 if (!CMSG_OK(msg, cmsg)) 2344 return -EINVAL; 2345 if (cmsg->cmsg_level != SOL_SOCKET) 2346 continue; 2347 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2348 if (ret) 2349 return ret; 2350 } 2351 return 0; 2352 } 2353 EXPORT_SYMBOL(sock_cmsg_send); 2354 2355 static void sk_enter_memory_pressure(struct sock *sk) 2356 { 2357 if (!sk->sk_prot->enter_memory_pressure) 2358 return; 2359 2360 sk->sk_prot->enter_memory_pressure(sk); 2361 } 2362 2363 static void sk_leave_memory_pressure(struct sock *sk) 2364 { 2365 if (sk->sk_prot->leave_memory_pressure) { 2366 sk->sk_prot->leave_memory_pressure(sk); 2367 } else { 2368 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2369 2370 if (memory_pressure && READ_ONCE(*memory_pressure)) 2371 WRITE_ONCE(*memory_pressure, 0); 2372 } 2373 } 2374 2375 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2376 DEFINE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2377 2378 /** 2379 * skb_page_frag_refill - check that a page_frag contains enough room 2380 * @sz: minimum size of the fragment we want to get 2381 * @pfrag: pointer to page_frag 2382 * @gfp: priority for memory allocation 2383 * 2384 * Note: While this allocator tries to use high order pages, there is 2385 * no guarantee that allocations succeed. Therefore, @sz MUST be 2386 * less or equal than PAGE_SIZE. 2387 */ 2388 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2389 { 2390 if (pfrag->page) { 2391 if (page_ref_count(pfrag->page) == 1) { 2392 pfrag->offset = 0; 2393 return true; 2394 } 2395 if (pfrag->offset + sz <= pfrag->size) 2396 return true; 2397 put_page(pfrag->page); 2398 } 2399 2400 pfrag->offset = 0; 2401 if (SKB_FRAG_PAGE_ORDER && 2402 !static_branch_unlikely(&net_high_order_alloc_disable_key)) { 2403 /* Avoid direct reclaim but allow kswapd to wake */ 2404 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2405 __GFP_COMP | __GFP_NOWARN | 2406 __GFP_NORETRY, 2407 SKB_FRAG_PAGE_ORDER); 2408 if (likely(pfrag->page)) { 2409 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2410 return true; 2411 } 2412 } 2413 pfrag->page = alloc_page(gfp); 2414 if (likely(pfrag->page)) { 2415 pfrag->size = PAGE_SIZE; 2416 return true; 2417 } 2418 return false; 2419 } 2420 EXPORT_SYMBOL(skb_page_frag_refill); 2421 2422 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2423 { 2424 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2425 return true; 2426 2427 sk_enter_memory_pressure(sk); 2428 sk_stream_moderate_sndbuf(sk); 2429 return false; 2430 } 2431 EXPORT_SYMBOL(sk_page_frag_refill); 2432 2433 static void __lock_sock(struct sock *sk) 2434 __releases(&sk->sk_lock.slock) 2435 __acquires(&sk->sk_lock.slock) 2436 { 2437 DEFINE_WAIT(wait); 2438 2439 for (;;) { 2440 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2441 TASK_UNINTERRUPTIBLE); 2442 spin_unlock_bh(&sk->sk_lock.slock); 2443 schedule(); 2444 spin_lock_bh(&sk->sk_lock.slock); 2445 if (!sock_owned_by_user(sk)) 2446 break; 2447 } 2448 finish_wait(&sk->sk_lock.wq, &wait); 2449 } 2450 2451 void __release_sock(struct sock *sk) 2452 __releases(&sk->sk_lock.slock) 2453 __acquires(&sk->sk_lock.slock) 2454 { 2455 struct sk_buff *skb, *next; 2456 2457 while ((skb = sk->sk_backlog.head) != NULL) { 2458 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2459 2460 spin_unlock_bh(&sk->sk_lock.slock); 2461 2462 do { 2463 next = skb->next; 2464 prefetch(next); 2465 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2466 skb_mark_not_on_list(skb); 2467 sk_backlog_rcv(sk, skb); 2468 2469 cond_resched(); 2470 2471 skb = next; 2472 } while (skb != NULL); 2473 2474 spin_lock_bh(&sk->sk_lock.slock); 2475 } 2476 2477 /* 2478 * Doing the zeroing here guarantee we can not loop forever 2479 * while a wild producer attempts to flood us. 2480 */ 2481 sk->sk_backlog.len = 0; 2482 } 2483 2484 void __sk_flush_backlog(struct sock *sk) 2485 { 2486 spin_lock_bh(&sk->sk_lock.slock); 2487 __release_sock(sk); 2488 spin_unlock_bh(&sk->sk_lock.slock); 2489 } 2490 2491 /** 2492 * sk_wait_data - wait for data to arrive at sk_receive_queue 2493 * @sk: sock to wait on 2494 * @timeo: for how long 2495 * @skb: last skb seen on sk_receive_queue 2496 * 2497 * Now socket state including sk->sk_err is changed only under lock, 2498 * hence we may omit checks after joining wait queue. 2499 * We check receive queue before schedule() only as optimization; 2500 * it is very likely that release_sock() added new data. 2501 */ 2502 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2503 { 2504 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2505 int rc; 2506 2507 add_wait_queue(sk_sleep(sk), &wait); 2508 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2509 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2510 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2511 remove_wait_queue(sk_sleep(sk), &wait); 2512 return rc; 2513 } 2514 EXPORT_SYMBOL(sk_wait_data); 2515 2516 /** 2517 * __sk_mem_raise_allocated - increase memory_allocated 2518 * @sk: socket 2519 * @size: memory size to allocate 2520 * @amt: pages to allocate 2521 * @kind: allocation type 2522 * 2523 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2524 */ 2525 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2526 { 2527 struct proto *prot = sk->sk_prot; 2528 long allocated = sk_memory_allocated_add(sk, amt); 2529 bool charged = true; 2530 2531 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2532 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2533 goto suppress_allocation; 2534 2535 /* Under limit. */ 2536 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2537 sk_leave_memory_pressure(sk); 2538 return 1; 2539 } 2540 2541 /* Under pressure. */ 2542 if (allocated > sk_prot_mem_limits(sk, 1)) 2543 sk_enter_memory_pressure(sk); 2544 2545 /* Over hard limit. */ 2546 if (allocated > sk_prot_mem_limits(sk, 2)) 2547 goto suppress_allocation; 2548 2549 /* guarantee minimum buffer size under pressure */ 2550 if (kind == SK_MEM_RECV) { 2551 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2552 return 1; 2553 2554 } else { /* SK_MEM_SEND */ 2555 int wmem0 = sk_get_wmem0(sk, prot); 2556 2557 if (sk->sk_type == SOCK_STREAM) { 2558 if (sk->sk_wmem_queued < wmem0) 2559 return 1; 2560 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2561 return 1; 2562 } 2563 } 2564 2565 if (sk_has_memory_pressure(sk)) { 2566 u64 alloc; 2567 2568 if (!sk_under_memory_pressure(sk)) 2569 return 1; 2570 alloc = sk_sockets_allocated_read_positive(sk); 2571 if (sk_prot_mem_limits(sk, 2) > alloc * 2572 sk_mem_pages(sk->sk_wmem_queued + 2573 atomic_read(&sk->sk_rmem_alloc) + 2574 sk->sk_forward_alloc)) 2575 return 1; 2576 } 2577 2578 suppress_allocation: 2579 2580 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2581 sk_stream_moderate_sndbuf(sk); 2582 2583 /* Fail only if socket is _under_ its sndbuf. 2584 * In this case we cannot block, so that we have to fail. 2585 */ 2586 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2587 return 1; 2588 } 2589 2590 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2591 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2592 2593 sk_memory_allocated_sub(sk, amt); 2594 2595 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2596 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2597 2598 return 0; 2599 } 2600 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2601 2602 /** 2603 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2604 * @sk: socket 2605 * @size: memory size to allocate 2606 * @kind: allocation type 2607 * 2608 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2609 * rmem allocation. This function assumes that protocols which have 2610 * memory_pressure use sk_wmem_queued as write buffer accounting. 2611 */ 2612 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2613 { 2614 int ret, amt = sk_mem_pages(size); 2615 2616 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2617 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2618 if (!ret) 2619 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2620 return ret; 2621 } 2622 EXPORT_SYMBOL(__sk_mem_schedule); 2623 2624 /** 2625 * __sk_mem_reduce_allocated - reclaim memory_allocated 2626 * @sk: socket 2627 * @amount: number of quanta 2628 * 2629 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2630 */ 2631 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2632 { 2633 sk_memory_allocated_sub(sk, amount); 2634 2635 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2636 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2637 2638 if (sk_under_memory_pressure(sk) && 2639 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2640 sk_leave_memory_pressure(sk); 2641 } 2642 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2643 2644 /** 2645 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2646 * @sk: socket 2647 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2648 */ 2649 void __sk_mem_reclaim(struct sock *sk, int amount) 2650 { 2651 amount >>= SK_MEM_QUANTUM_SHIFT; 2652 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2653 __sk_mem_reduce_allocated(sk, amount); 2654 } 2655 EXPORT_SYMBOL(__sk_mem_reclaim); 2656 2657 int sk_set_peek_off(struct sock *sk, int val) 2658 { 2659 sk->sk_peek_off = val; 2660 return 0; 2661 } 2662 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2663 2664 /* 2665 * Set of default routines for initialising struct proto_ops when 2666 * the protocol does not support a particular function. In certain 2667 * cases where it makes no sense for a protocol to have a "do nothing" 2668 * function, some default processing is provided. 2669 */ 2670 2671 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2672 { 2673 return -EOPNOTSUPP; 2674 } 2675 EXPORT_SYMBOL(sock_no_bind); 2676 2677 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2678 int len, int flags) 2679 { 2680 return -EOPNOTSUPP; 2681 } 2682 EXPORT_SYMBOL(sock_no_connect); 2683 2684 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2685 { 2686 return -EOPNOTSUPP; 2687 } 2688 EXPORT_SYMBOL(sock_no_socketpair); 2689 2690 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2691 bool kern) 2692 { 2693 return -EOPNOTSUPP; 2694 } 2695 EXPORT_SYMBOL(sock_no_accept); 2696 2697 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2698 int peer) 2699 { 2700 return -EOPNOTSUPP; 2701 } 2702 EXPORT_SYMBOL(sock_no_getname); 2703 2704 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2705 { 2706 return -EOPNOTSUPP; 2707 } 2708 EXPORT_SYMBOL(sock_no_ioctl); 2709 2710 int sock_no_listen(struct socket *sock, int backlog) 2711 { 2712 return -EOPNOTSUPP; 2713 } 2714 EXPORT_SYMBOL(sock_no_listen); 2715 2716 int sock_no_shutdown(struct socket *sock, int how) 2717 { 2718 return -EOPNOTSUPP; 2719 } 2720 EXPORT_SYMBOL(sock_no_shutdown); 2721 2722 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2723 char __user *optval, unsigned int optlen) 2724 { 2725 return -EOPNOTSUPP; 2726 } 2727 EXPORT_SYMBOL(sock_no_setsockopt); 2728 2729 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2730 char __user *optval, int __user *optlen) 2731 { 2732 return -EOPNOTSUPP; 2733 } 2734 EXPORT_SYMBOL(sock_no_getsockopt); 2735 2736 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2737 { 2738 return -EOPNOTSUPP; 2739 } 2740 EXPORT_SYMBOL(sock_no_sendmsg); 2741 2742 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2743 { 2744 return -EOPNOTSUPP; 2745 } 2746 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2747 2748 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2749 int flags) 2750 { 2751 return -EOPNOTSUPP; 2752 } 2753 EXPORT_SYMBOL(sock_no_recvmsg); 2754 2755 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2756 { 2757 /* Mirror missing mmap method error code */ 2758 return -ENODEV; 2759 } 2760 EXPORT_SYMBOL(sock_no_mmap); 2761 2762 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2763 { 2764 ssize_t res; 2765 struct msghdr msg = {.msg_flags = flags}; 2766 struct kvec iov; 2767 char *kaddr = kmap(page); 2768 iov.iov_base = kaddr + offset; 2769 iov.iov_len = size; 2770 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2771 kunmap(page); 2772 return res; 2773 } 2774 EXPORT_SYMBOL(sock_no_sendpage); 2775 2776 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2777 int offset, size_t size, int flags) 2778 { 2779 ssize_t res; 2780 struct msghdr msg = {.msg_flags = flags}; 2781 struct kvec iov; 2782 char *kaddr = kmap(page); 2783 2784 iov.iov_base = kaddr + offset; 2785 iov.iov_len = size; 2786 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2787 kunmap(page); 2788 return res; 2789 } 2790 EXPORT_SYMBOL(sock_no_sendpage_locked); 2791 2792 /* 2793 * Default Socket Callbacks 2794 */ 2795 2796 static void sock_def_wakeup(struct sock *sk) 2797 { 2798 struct socket_wq *wq; 2799 2800 rcu_read_lock(); 2801 wq = rcu_dereference(sk->sk_wq); 2802 if (skwq_has_sleeper(wq)) 2803 wake_up_interruptible_all(&wq->wait); 2804 rcu_read_unlock(); 2805 } 2806 2807 static void sock_def_error_report(struct sock *sk) 2808 { 2809 struct socket_wq *wq; 2810 2811 rcu_read_lock(); 2812 wq = rcu_dereference(sk->sk_wq); 2813 if (skwq_has_sleeper(wq)) 2814 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 2815 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2816 rcu_read_unlock(); 2817 } 2818 2819 void sock_def_readable(struct sock *sk) 2820 { 2821 struct socket_wq *wq; 2822 2823 rcu_read_lock(); 2824 wq = rcu_dereference(sk->sk_wq); 2825 if (skwq_has_sleeper(wq)) 2826 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 2827 EPOLLRDNORM | EPOLLRDBAND); 2828 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2829 rcu_read_unlock(); 2830 } 2831 2832 static void sock_def_write_space(struct sock *sk) 2833 { 2834 struct socket_wq *wq; 2835 2836 rcu_read_lock(); 2837 2838 /* Do not wake up a writer until he can make "significant" 2839 * progress. --DaveM 2840 */ 2841 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= READ_ONCE(sk->sk_sndbuf)) { 2842 wq = rcu_dereference(sk->sk_wq); 2843 if (skwq_has_sleeper(wq)) 2844 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 2845 EPOLLWRNORM | EPOLLWRBAND); 2846 2847 /* Should agree with poll, otherwise some programs break */ 2848 if (sock_writeable(sk)) 2849 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2850 } 2851 2852 rcu_read_unlock(); 2853 } 2854 2855 static void sock_def_destruct(struct sock *sk) 2856 { 2857 } 2858 2859 void sk_send_sigurg(struct sock *sk) 2860 { 2861 if (sk->sk_socket && sk->sk_socket->file) 2862 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2863 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2864 } 2865 EXPORT_SYMBOL(sk_send_sigurg); 2866 2867 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2868 unsigned long expires) 2869 { 2870 if (!mod_timer(timer, expires)) 2871 sock_hold(sk); 2872 } 2873 EXPORT_SYMBOL(sk_reset_timer); 2874 2875 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2876 { 2877 if (del_timer(timer)) 2878 __sock_put(sk); 2879 } 2880 EXPORT_SYMBOL(sk_stop_timer); 2881 2882 void sock_init_data(struct socket *sock, struct sock *sk) 2883 { 2884 sk_init_common(sk); 2885 sk->sk_send_head = NULL; 2886 2887 timer_setup(&sk->sk_timer, NULL, 0); 2888 2889 sk->sk_allocation = GFP_KERNEL; 2890 sk->sk_rcvbuf = sysctl_rmem_default; 2891 sk->sk_sndbuf = sysctl_wmem_default; 2892 sk->sk_state = TCP_CLOSE; 2893 sk_set_socket(sk, sock); 2894 2895 sock_set_flag(sk, SOCK_ZAPPED); 2896 2897 if (sock) { 2898 sk->sk_type = sock->type; 2899 RCU_INIT_POINTER(sk->sk_wq, &sock->wq); 2900 sock->sk = sk; 2901 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2902 } else { 2903 RCU_INIT_POINTER(sk->sk_wq, NULL); 2904 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2905 } 2906 2907 rwlock_init(&sk->sk_callback_lock); 2908 if (sk->sk_kern_sock) 2909 lockdep_set_class_and_name( 2910 &sk->sk_callback_lock, 2911 af_kern_callback_keys + sk->sk_family, 2912 af_family_kern_clock_key_strings[sk->sk_family]); 2913 else 2914 lockdep_set_class_and_name( 2915 &sk->sk_callback_lock, 2916 af_callback_keys + sk->sk_family, 2917 af_family_clock_key_strings[sk->sk_family]); 2918 2919 sk->sk_state_change = sock_def_wakeup; 2920 sk->sk_data_ready = sock_def_readable; 2921 sk->sk_write_space = sock_def_write_space; 2922 sk->sk_error_report = sock_def_error_report; 2923 sk->sk_destruct = sock_def_destruct; 2924 2925 sk->sk_frag.page = NULL; 2926 sk->sk_frag.offset = 0; 2927 sk->sk_peek_off = -1; 2928 2929 sk->sk_peer_pid = NULL; 2930 sk->sk_peer_cred = NULL; 2931 sk->sk_write_pending = 0; 2932 sk->sk_rcvlowat = 1; 2933 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2934 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2935 2936 sk->sk_stamp = SK_DEFAULT_STAMP; 2937 #if BITS_PER_LONG==32 2938 seqlock_init(&sk->sk_stamp_seq); 2939 #endif 2940 atomic_set(&sk->sk_zckey, 0); 2941 2942 #ifdef CONFIG_NET_RX_BUSY_POLL 2943 sk->sk_napi_id = 0; 2944 sk->sk_ll_usec = sysctl_net_busy_read; 2945 #endif 2946 2947 sk->sk_max_pacing_rate = ~0UL; 2948 sk->sk_pacing_rate = ~0UL; 2949 WRITE_ONCE(sk->sk_pacing_shift, 10); 2950 sk->sk_incoming_cpu = -1; 2951 2952 sk_rx_queue_clear(sk); 2953 /* 2954 * Before updating sk_refcnt, we must commit prior changes to memory 2955 * (Documentation/RCU/rculist_nulls.txt for details) 2956 */ 2957 smp_wmb(); 2958 refcount_set(&sk->sk_refcnt, 1); 2959 atomic_set(&sk->sk_drops, 0); 2960 } 2961 EXPORT_SYMBOL(sock_init_data); 2962 2963 void lock_sock_nested(struct sock *sk, int subclass) 2964 { 2965 might_sleep(); 2966 spin_lock_bh(&sk->sk_lock.slock); 2967 if (sk->sk_lock.owned) 2968 __lock_sock(sk); 2969 sk->sk_lock.owned = 1; 2970 spin_unlock(&sk->sk_lock.slock); 2971 /* 2972 * The sk_lock has mutex_lock() semantics here: 2973 */ 2974 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2975 local_bh_enable(); 2976 } 2977 EXPORT_SYMBOL(lock_sock_nested); 2978 2979 void release_sock(struct sock *sk) 2980 { 2981 spin_lock_bh(&sk->sk_lock.slock); 2982 if (sk->sk_backlog.tail) 2983 __release_sock(sk); 2984 2985 /* Warning : release_cb() might need to release sk ownership, 2986 * ie call sock_release_ownership(sk) before us. 2987 */ 2988 if (sk->sk_prot->release_cb) 2989 sk->sk_prot->release_cb(sk); 2990 2991 sock_release_ownership(sk); 2992 if (waitqueue_active(&sk->sk_lock.wq)) 2993 wake_up(&sk->sk_lock.wq); 2994 spin_unlock_bh(&sk->sk_lock.slock); 2995 } 2996 EXPORT_SYMBOL(release_sock); 2997 2998 /** 2999 * lock_sock_fast - fast version of lock_sock 3000 * @sk: socket 3001 * 3002 * This version should be used for very small section, where process wont block 3003 * return false if fast path is taken: 3004 * 3005 * sk_lock.slock locked, owned = 0, BH disabled 3006 * 3007 * return true if slow path is taken: 3008 * 3009 * sk_lock.slock unlocked, owned = 1, BH enabled 3010 */ 3011 bool lock_sock_fast(struct sock *sk) 3012 { 3013 might_sleep(); 3014 spin_lock_bh(&sk->sk_lock.slock); 3015 3016 if (!sk->sk_lock.owned) 3017 /* 3018 * Note : We must disable BH 3019 */ 3020 return false; 3021 3022 __lock_sock(sk); 3023 sk->sk_lock.owned = 1; 3024 spin_unlock(&sk->sk_lock.slock); 3025 /* 3026 * The sk_lock has mutex_lock() semantics here: 3027 */ 3028 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 3029 local_bh_enable(); 3030 return true; 3031 } 3032 EXPORT_SYMBOL(lock_sock_fast); 3033 3034 int sock_gettstamp(struct socket *sock, void __user *userstamp, 3035 bool timeval, bool time32) 3036 { 3037 struct sock *sk = sock->sk; 3038 struct timespec64 ts; 3039 3040 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 3041 ts = ktime_to_timespec64(sock_read_timestamp(sk)); 3042 if (ts.tv_sec == -1) 3043 return -ENOENT; 3044 if (ts.tv_sec == 0) { 3045 ktime_t kt = ktime_get_real(); 3046 sock_write_timestamp(sk, kt); 3047 ts = ktime_to_timespec64(kt); 3048 } 3049 3050 if (timeval) 3051 ts.tv_nsec /= 1000; 3052 3053 #ifdef CONFIG_COMPAT_32BIT_TIME 3054 if (time32) 3055 return put_old_timespec32(&ts, userstamp); 3056 #endif 3057 #ifdef CONFIG_SPARC64 3058 /* beware of padding in sparc64 timeval */ 3059 if (timeval && !in_compat_syscall()) { 3060 struct __kernel_old_timeval __user tv = { 3061 .tv_sec = ts.tv_sec, 3062 .tv_usec = ts.tv_nsec, 3063 }; 3064 if (copy_to_user(userstamp, &tv, sizeof(tv))) 3065 return -EFAULT; 3066 return 0; 3067 } 3068 #endif 3069 return put_timespec64(&ts, userstamp); 3070 } 3071 EXPORT_SYMBOL(sock_gettstamp); 3072 3073 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag) 3074 { 3075 if (!sock_flag(sk, flag)) { 3076 unsigned long previous_flags = sk->sk_flags; 3077 3078 sock_set_flag(sk, flag); 3079 /* 3080 * we just set one of the two flags which require net 3081 * time stamping, but time stamping might have been on 3082 * already because of the other one 3083 */ 3084 if (sock_needs_netstamp(sk) && 3085 !(previous_flags & SK_FLAGS_TIMESTAMP)) 3086 net_enable_timestamp(); 3087 } 3088 } 3089 3090 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 3091 int level, int type) 3092 { 3093 struct sock_exterr_skb *serr; 3094 struct sk_buff *skb; 3095 int copied, err; 3096 3097 err = -EAGAIN; 3098 skb = sock_dequeue_err_skb(sk); 3099 if (skb == NULL) 3100 goto out; 3101 3102 copied = skb->len; 3103 if (copied > len) { 3104 msg->msg_flags |= MSG_TRUNC; 3105 copied = len; 3106 } 3107 err = skb_copy_datagram_msg(skb, 0, msg, copied); 3108 if (err) 3109 goto out_free_skb; 3110 3111 sock_recv_timestamp(msg, sk, skb); 3112 3113 serr = SKB_EXT_ERR(skb); 3114 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 3115 3116 msg->msg_flags |= MSG_ERRQUEUE; 3117 err = copied; 3118 3119 out_free_skb: 3120 kfree_skb(skb); 3121 out: 3122 return err; 3123 } 3124 EXPORT_SYMBOL(sock_recv_errqueue); 3125 3126 /* 3127 * Get a socket option on an socket. 3128 * 3129 * FIX: POSIX 1003.1g is very ambiguous here. It states that 3130 * asynchronous errors should be reported by getsockopt. We assume 3131 * this means if you specify SO_ERROR (otherwise whats the point of it). 3132 */ 3133 int sock_common_getsockopt(struct socket *sock, int level, int optname, 3134 char __user *optval, int __user *optlen) 3135 { 3136 struct sock *sk = sock->sk; 3137 3138 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3139 } 3140 EXPORT_SYMBOL(sock_common_getsockopt); 3141 3142 #ifdef CONFIG_COMPAT 3143 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 3144 char __user *optval, int __user *optlen) 3145 { 3146 struct sock *sk = sock->sk; 3147 3148 if (sk->sk_prot->compat_getsockopt != NULL) 3149 return sk->sk_prot->compat_getsockopt(sk, level, optname, 3150 optval, optlen); 3151 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 3152 } 3153 EXPORT_SYMBOL(compat_sock_common_getsockopt); 3154 #endif 3155 3156 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 3157 int flags) 3158 { 3159 struct sock *sk = sock->sk; 3160 int addr_len = 0; 3161 int err; 3162 3163 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 3164 flags & ~MSG_DONTWAIT, &addr_len); 3165 if (err >= 0) 3166 msg->msg_namelen = addr_len; 3167 return err; 3168 } 3169 EXPORT_SYMBOL(sock_common_recvmsg); 3170 3171 /* 3172 * Set socket options on an inet socket. 3173 */ 3174 int sock_common_setsockopt(struct socket *sock, int level, int optname, 3175 char __user *optval, unsigned int optlen) 3176 { 3177 struct sock *sk = sock->sk; 3178 3179 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3180 } 3181 EXPORT_SYMBOL(sock_common_setsockopt); 3182 3183 #ifdef CONFIG_COMPAT 3184 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 3185 char __user *optval, unsigned int optlen) 3186 { 3187 struct sock *sk = sock->sk; 3188 3189 if (sk->sk_prot->compat_setsockopt != NULL) 3190 return sk->sk_prot->compat_setsockopt(sk, level, optname, 3191 optval, optlen); 3192 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3193 } 3194 EXPORT_SYMBOL(compat_sock_common_setsockopt); 3195 #endif 3196 3197 void sk_common_release(struct sock *sk) 3198 { 3199 if (sk->sk_prot->destroy) 3200 sk->sk_prot->destroy(sk); 3201 3202 /* 3203 * Observation: when sock_common_release is called, processes have 3204 * no access to socket. But net still has. 3205 * Step one, detach it from networking: 3206 * 3207 * A. Remove from hash tables. 3208 */ 3209 3210 sk->sk_prot->unhash(sk); 3211 3212 /* 3213 * In this point socket cannot receive new packets, but it is possible 3214 * that some packets are in flight because some CPU runs receiver and 3215 * did hash table lookup before we unhashed socket. They will achieve 3216 * receive queue and will be purged by socket destructor. 3217 * 3218 * Also we still have packets pending on receive queue and probably, 3219 * our own packets waiting in device queues. sock_destroy will drain 3220 * receive queue, but transmitted packets will delay socket destruction 3221 * until the last reference will be released. 3222 */ 3223 3224 sock_orphan(sk); 3225 3226 xfrm_sk_free_policy(sk); 3227 3228 sk_refcnt_debug_release(sk); 3229 3230 sock_put(sk); 3231 } 3232 EXPORT_SYMBOL(sk_common_release); 3233 3234 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3235 { 3236 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3237 3238 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3239 mem[SK_MEMINFO_RCVBUF] = READ_ONCE(sk->sk_rcvbuf); 3240 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3241 mem[SK_MEMINFO_SNDBUF] = READ_ONCE(sk->sk_sndbuf); 3242 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3243 mem[SK_MEMINFO_WMEM_QUEUED] = READ_ONCE(sk->sk_wmem_queued); 3244 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3245 mem[SK_MEMINFO_BACKLOG] = READ_ONCE(sk->sk_backlog.len); 3246 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3247 } 3248 3249 #ifdef CONFIG_PROC_FS 3250 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3251 struct prot_inuse { 3252 int val[PROTO_INUSE_NR]; 3253 }; 3254 3255 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3256 3257 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3258 { 3259 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3260 } 3261 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3262 3263 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3264 { 3265 int cpu, idx = prot->inuse_idx; 3266 int res = 0; 3267 3268 for_each_possible_cpu(cpu) 3269 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3270 3271 return res >= 0 ? res : 0; 3272 } 3273 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3274 3275 static void sock_inuse_add(struct net *net, int val) 3276 { 3277 this_cpu_add(*net->core.sock_inuse, val); 3278 } 3279 3280 int sock_inuse_get(struct net *net) 3281 { 3282 int cpu, res = 0; 3283 3284 for_each_possible_cpu(cpu) 3285 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3286 3287 return res; 3288 } 3289 3290 EXPORT_SYMBOL_GPL(sock_inuse_get); 3291 3292 static int __net_init sock_inuse_init_net(struct net *net) 3293 { 3294 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3295 if (net->core.prot_inuse == NULL) 3296 return -ENOMEM; 3297 3298 net->core.sock_inuse = alloc_percpu(int); 3299 if (net->core.sock_inuse == NULL) 3300 goto out; 3301 3302 return 0; 3303 3304 out: 3305 free_percpu(net->core.prot_inuse); 3306 return -ENOMEM; 3307 } 3308 3309 static void __net_exit sock_inuse_exit_net(struct net *net) 3310 { 3311 free_percpu(net->core.prot_inuse); 3312 free_percpu(net->core.sock_inuse); 3313 } 3314 3315 static struct pernet_operations net_inuse_ops = { 3316 .init = sock_inuse_init_net, 3317 .exit = sock_inuse_exit_net, 3318 }; 3319 3320 static __init int net_inuse_init(void) 3321 { 3322 if (register_pernet_subsys(&net_inuse_ops)) 3323 panic("Cannot initialize net inuse counters"); 3324 3325 return 0; 3326 } 3327 3328 core_initcall(net_inuse_init); 3329 3330 static int assign_proto_idx(struct proto *prot) 3331 { 3332 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3333 3334 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3335 pr_err("PROTO_INUSE_NR exhausted\n"); 3336 return -ENOSPC; 3337 } 3338 3339 set_bit(prot->inuse_idx, proto_inuse_idx); 3340 return 0; 3341 } 3342 3343 static void release_proto_idx(struct proto *prot) 3344 { 3345 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3346 clear_bit(prot->inuse_idx, proto_inuse_idx); 3347 } 3348 #else 3349 static inline int assign_proto_idx(struct proto *prot) 3350 { 3351 return 0; 3352 } 3353 3354 static inline void release_proto_idx(struct proto *prot) 3355 { 3356 } 3357 3358 static void sock_inuse_add(struct net *net, int val) 3359 { 3360 } 3361 #endif 3362 3363 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3364 { 3365 if (!rsk_prot) 3366 return; 3367 kfree(rsk_prot->slab_name); 3368 rsk_prot->slab_name = NULL; 3369 kmem_cache_destroy(rsk_prot->slab); 3370 rsk_prot->slab = NULL; 3371 } 3372 3373 static int req_prot_init(const struct proto *prot) 3374 { 3375 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3376 3377 if (!rsk_prot) 3378 return 0; 3379 3380 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3381 prot->name); 3382 if (!rsk_prot->slab_name) 3383 return -ENOMEM; 3384 3385 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3386 rsk_prot->obj_size, 0, 3387 SLAB_ACCOUNT | prot->slab_flags, 3388 NULL); 3389 3390 if (!rsk_prot->slab) { 3391 pr_crit("%s: Can't create request sock SLAB cache!\n", 3392 prot->name); 3393 return -ENOMEM; 3394 } 3395 return 0; 3396 } 3397 3398 int proto_register(struct proto *prot, int alloc_slab) 3399 { 3400 int ret = -ENOBUFS; 3401 3402 if (alloc_slab) { 3403 prot->slab = kmem_cache_create_usercopy(prot->name, 3404 prot->obj_size, 0, 3405 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3406 prot->slab_flags, 3407 prot->useroffset, prot->usersize, 3408 NULL); 3409 3410 if (prot->slab == NULL) { 3411 pr_crit("%s: Can't create sock SLAB cache!\n", 3412 prot->name); 3413 goto out; 3414 } 3415 3416 if (req_prot_init(prot)) 3417 goto out_free_request_sock_slab; 3418 3419 if (prot->twsk_prot != NULL) { 3420 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3421 3422 if (prot->twsk_prot->twsk_slab_name == NULL) 3423 goto out_free_request_sock_slab; 3424 3425 prot->twsk_prot->twsk_slab = 3426 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3427 prot->twsk_prot->twsk_obj_size, 3428 0, 3429 SLAB_ACCOUNT | 3430 prot->slab_flags, 3431 NULL); 3432 if (prot->twsk_prot->twsk_slab == NULL) 3433 goto out_free_timewait_sock_slab_name; 3434 } 3435 } 3436 3437 mutex_lock(&proto_list_mutex); 3438 ret = assign_proto_idx(prot); 3439 if (ret) { 3440 mutex_unlock(&proto_list_mutex); 3441 goto out_free_timewait_sock_slab_name; 3442 } 3443 list_add(&prot->node, &proto_list); 3444 mutex_unlock(&proto_list_mutex); 3445 return ret; 3446 3447 out_free_timewait_sock_slab_name: 3448 if (alloc_slab && prot->twsk_prot) 3449 kfree(prot->twsk_prot->twsk_slab_name); 3450 out_free_request_sock_slab: 3451 if (alloc_slab) { 3452 req_prot_cleanup(prot->rsk_prot); 3453 3454 kmem_cache_destroy(prot->slab); 3455 prot->slab = NULL; 3456 } 3457 out: 3458 return ret; 3459 } 3460 EXPORT_SYMBOL(proto_register); 3461 3462 void proto_unregister(struct proto *prot) 3463 { 3464 mutex_lock(&proto_list_mutex); 3465 release_proto_idx(prot); 3466 list_del(&prot->node); 3467 mutex_unlock(&proto_list_mutex); 3468 3469 kmem_cache_destroy(prot->slab); 3470 prot->slab = NULL; 3471 3472 req_prot_cleanup(prot->rsk_prot); 3473 3474 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3475 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3476 kfree(prot->twsk_prot->twsk_slab_name); 3477 prot->twsk_prot->twsk_slab = NULL; 3478 } 3479 } 3480 EXPORT_SYMBOL(proto_unregister); 3481 3482 int sock_load_diag_module(int family, int protocol) 3483 { 3484 if (!protocol) { 3485 if (!sock_is_registered(family)) 3486 return -ENOENT; 3487 3488 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3489 NETLINK_SOCK_DIAG, family); 3490 } 3491 3492 #ifdef CONFIG_INET 3493 if (family == AF_INET && 3494 protocol != IPPROTO_RAW && 3495 !rcu_access_pointer(inet_protos[protocol])) 3496 return -ENOENT; 3497 #endif 3498 3499 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3500 NETLINK_SOCK_DIAG, family, protocol); 3501 } 3502 EXPORT_SYMBOL(sock_load_diag_module); 3503 3504 #ifdef CONFIG_PROC_FS 3505 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3506 __acquires(proto_list_mutex) 3507 { 3508 mutex_lock(&proto_list_mutex); 3509 return seq_list_start_head(&proto_list, *pos); 3510 } 3511 3512 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3513 { 3514 return seq_list_next(v, &proto_list, pos); 3515 } 3516 3517 static void proto_seq_stop(struct seq_file *seq, void *v) 3518 __releases(proto_list_mutex) 3519 { 3520 mutex_unlock(&proto_list_mutex); 3521 } 3522 3523 static char proto_method_implemented(const void *method) 3524 { 3525 return method == NULL ? 'n' : 'y'; 3526 } 3527 static long sock_prot_memory_allocated(struct proto *proto) 3528 { 3529 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3530 } 3531 3532 static const char *sock_prot_memory_pressure(struct proto *proto) 3533 { 3534 return proto->memory_pressure != NULL ? 3535 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3536 } 3537 3538 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3539 { 3540 3541 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3542 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3543 proto->name, 3544 proto->obj_size, 3545 sock_prot_inuse_get(seq_file_net(seq), proto), 3546 sock_prot_memory_allocated(proto), 3547 sock_prot_memory_pressure(proto), 3548 proto->max_header, 3549 proto->slab == NULL ? "no" : "yes", 3550 module_name(proto->owner), 3551 proto_method_implemented(proto->close), 3552 proto_method_implemented(proto->connect), 3553 proto_method_implemented(proto->disconnect), 3554 proto_method_implemented(proto->accept), 3555 proto_method_implemented(proto->ioctl), 3556 proto_method_implemented(proto->init), 3557 proto_method_implemented(proto->destroy), 3558 proto_method_implemented(proto->shutdown), 3559 proto_method_implemented(proto->setsockopt), 3560 proto_method_implemented(proto->getsockopt), 3561 proto_method_implemented(proto->sendmsg), 3562 proto_method_implemented(proto->recvmsg), 3563 proto_method_implemented(proto->sendpage), 3564 proto_method_implemented(proto->bind), 3565 proto_method_implemented(proto->backlog_rcv), 3566 proto_method_implemented(proto->hash), 3567 proto_method_implemented(proto->unhash), 3568 proto_method_implemented(proto->get_port), 3569 proto_method_implemented(proto->enter_memory_pressure)); 3570 } 3571 3572 static int proto_seq_show(struct seq_file *seq, void *v) 3573 { 3574 if (v == &proto_list) 3575 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3576 "protocol", 3577 "size", 3578 "sockets", 3579 "memory", 3580 "press", 3581 "maxhdr", 3582 "slab", 3583 "module", 3584 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3585 else 3586 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3587 return 0; 3588 } 3589 3590 static const struct seq_operations proto_seq_ops = { 3591 .start = proto_seq_start, 3592 .next = proto_seq_next, 3593 .stop = proto_seq_stop, 3594 .show = proto_seq_show, 3595 }; 3596 3597 static __net_init int proto_init_net(struct net *net) 3598 { 3599 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3600 sizeof(struct seq_net_private))) 3601 return -ENOMEM; 3602 3603 return 0; 3604 } 3605 3606 static __net_exit void proto_exit_net(struct net *net) 3607 { 3608 remove_proc_entry("protocols", net->proc_net); 3609 } 3610 3611 3612 static __net_initdata struct pernet_operations proto_net_ops = { 3613 .init = proto_init_net, 3614 .exit = proto_exit_net, 3615 }; 3616 3617 static int __init proto_init(void) 3618 { 3619 return register_pernet_subsys(&proto_net_ops); 3620 } 3621 3622 subsys_initcall(proto_init); 3623 3624 #endif /* PROC_FS */ 3625 3626 #ifdef CONFIG_NET_RX_BUSY_POLL 3627 bool sk_busy_loop_end(void *p, unsigned long start_time) 3628 { 3629 struct sock *sk = p; 3630 3631 return !skb_queue_empty_lockless(&sk->sk_receive_queue) || 3632 sk_busy_loop_timeout(sk, start_time); 3633 } 3634 EXPORT_SYMBOL(sk_busy_loop_end); 3635 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3636