1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Generic socket support routines. Memory allocators, socket lock/release 7 * handler for protocols to use and generic option handler. 8 * 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 * This program is free software; you can redistribute it and/or 87 * modify it under the terms of the GNU General Public License 88 * as published by the Free Software Foundation; either version 89 * 2 of the License, or (at your option) any later version. 90 */ 91 92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 93 94 #include <asm/unaligned.h> 95 #include <linux/capability.h> 96 #include <linux/errno.h> 97 #include <linux/errqueue.h> 98 #include <linux/types.h> 99 #include <linux/socket.h> 100 #include <linux/in.h> 101 #include <linux/kernel.h> 102 #include <linux/module.h> 103 #include <linux/proc_fs.h> 104 #include <linux/seq_file.h> 105 #include <linux/sched.h> 106 #include <linux/sched/mm.h> 107 #include <linux/timer.h> 108 #include <linux/string.h> 109 #include <linux/sockios.h> 110 #include <linux/net.h> 111 #include <linux/mm.h> 112 #include <linux/slab.h> 113 #include <linux/interrupt.h> 114 #include <linux/poll.h> 115 #include <linux/tcp.h> 116 #include <linux/init.h> 117 #include <linux/highmem.h> 118 #include <linux/user_namespace.h> 119 #include <linux/static_key.h> 120 #include <linux/memcontrol.h> 121 #include <linux/prefetch.h> 122 123 #include <linux/uaccess.h> 124 125 #include <linux/netdevice.h> 126 #include <net/protocol.h> 127 #include <linux/skbuff.h> 128 #include <net/net_namespace.h> 129 #include <net/request_sock.h> 130 #include <net/sock.h> 131 #include <linux/net_tstamp.h> 132 #include <net/xfrm.h> 133 #include <linux/ipsec.h> 134 #include <net/cls_cgroup.h> 135 #include <net/netprio_cgroup.h> 136 #include <linux/sock_diag.h> 137 138 #include <linux/filter.h> 139 #include <net/sock_reuseport.h> 140 141 #include <trace/events/sock.h> 142 143 #include <net/tcp.h> 144 #include <net/busy_poll.h> 145 146 static DEFINE_MUTEX(proto_list_mutex); 147 static LIST_HEAD(proto_list); 148 149 static void sock_inuse_add(struct net *net, int val); 150 151 /** 152 * sk_ns_capable - General socket capability test 153 * @sk: Socket to use a capability on or through 154 * @user_ns: The user namespace of the capability to use 155 * @cap: The capability to use 156 * 157 * Test to see if the opener of the socket had when the socket was 158 * created and the current process has the capability @cap in the user 159 * namespace @user_ns. 160 */ 161 bool sk_ns_capable(const struct sock *sk, 162 struct user_namespace *user_ns, int cap) 163 { 164 return file_ns_capable(sk->sk_socket->file, user_ns, cap) && 165 ns_capable(user_ns, cap); 166 } 167 EXPORT_SYMBOL(sk_ns_capable); 168 169 /** 170 * sk_capable - Socket global capability test 171 * @sk: Socket to use a capability on or through 172 * @cap: The global capability to use 173 * 174 * Test to see if the opener of the socket had when the socket was 175 * created and the current process has the capability @cap in all user 176 * namespaces. 177 */ 178 bool sk_capable(const struct sock *sk, int cap) 179 { 180 return sk_ns_capable(sk, &init_user_ns, cap); 181 } 182 EXPORT_SYMBOL(sk_capable); 183 184 /** 185 * sk_net_capable - Network namespace socket capability test 186 * @sk: Socket to use a capability on or through 187 * @cap: The capability to use 188 * 189 * Test to see if the opener of the socket had when the socket was created 190 * and the current process has the capability @cap over the network namespace 191 * the socket is a member of. 192 */ 193 bool sk_net_capable(const struct sock *sk, int cap) 194 { 195 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap); 196 } 197 EXPORT_SYMBOL(sk_net_capable); 198 199 /* 200 * Each address family might have different locking rules, so we have 201 * one slock key per address family and separate keys for internal and 202 * userspace sockets. 203 */ 204 static struct lock_class_key af_family_keys[AF_MAX]; 205 static struct lock_class_key af_family_kern_keys[AF_MAX]; 206 static struct lock_class_key af_family_slock_keys[AF_MAX]; 207 static struct lock_class_key af_family_kern_slock_keys[AF_MAX]; 208 209 /* 210 * Make lock validator output more readable. (we pre-construct these 211 * strings build-time, so that runtime initialization of socket 212 * locks is fast): 213 */ 214 215 #define _sock_locks(x) \ 216 x "AF_UNSPEC", x "AF_UNIX" , x "AF_INET" , \ 217 x "AF_AX25" , x "AF_IPX" , x "AF_APPLETALK", \ 218 x "AF_NETROM", x "AF_BRIDGE" , x "AF_ATMPVC" , \ 219 x "AF_X25" , x "AF_INET6" , x "AF_ROSE" , \ 220 x "AF_DECnet", x "AF_NETBEUI" , x "AF_SECURITY" , \ 221 x "AF_KEY" , x "AF_NETLINK" , x "AF_PACKET" , \ 222 x "AF_ASH" , x "AF_ECONET" , x "AF_ATMSVC" , \ 223 x "AF_RDS" , x "AF_SNA" , x "AF_IRDA" , \ 224 x "AF_PPPOX" , x "AF_WANPIPE" , x "AF_LLC" , \ 225 x "27" , x "28" , x "AF_CAN" , \ 226 x "AF_TIPC" , x "AF_BLUETOOTH", x "IUCV" , \ 227 x "AF_RXRPC" , x "AF_ISDN" , x "AF_PHONET" , \ 228 x "AF_IEEE802154", x "AF_CAIF" , x "AF_ALG" , \ 229 x "AF_NFC" , x "AF_VSOCK" , x "AF_KCM" , \ 230 x "AF_QIPCRTR", x "AF_SMC" , x "AF_XDP" , \ 231 x "AF_MAX" 232 233 static const char *const af_family_key_strings[AF_MAX+1] = { 234 _sock_locks("sk_lock-") 235 }; 236 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 237 _sock_locks("slock-") 238 }; 239 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 240 _sock_locks("clock-") 241 }; 242 243 static const char *const af_family_kern_key_strings[AF_MAX+1] = { 244 _sock_locks("k-sk_lock-") 245 }; 246 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = { 247 _sock_locks("k-slock-") 248 }; 249 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = { 250 _sock_locks("k-clock-") 251 }; 252 static const char *const af_family_rlock_key_strings[AF_MAX+1] = { 253 _sock_locks("rlock-") 254 }; 255 static const char *const af_family_wlock_key_strings[AF_MAX+1] = { 256 _sock_locks("wlock-") 257 }; 258 static const char *const af_family_elock_key_strings[AF_MAX+1] = { 259 _sock_locks("elock-") 260 }; 261 262 /* 263 * sk_callback_lock and sk queues locking rules are per-address-family, 264 * so split the lock classes by using a per-AF key: 265 */ 266 static struct lock_class_key af_callback_keys[AF_MAX]; 267 static struct lock_class_key af_rlock_keys[AF_MAX]; 268 static struct lock_class_key af_wlock_keys[AF_MAX]; 269 static struct lock_class_key af_elock_keys[AF_MAX]; 270 static struct lock_class_key af_kern_callback_keys[AF_MAX]; 271 272 /* Run time adjustable parameters. */ 273 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 274 EXPORT_SYMBOL(sysctl_wmem_max); 275 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 276 EXPORT_SYMBOL(sysctl_rmem_max); 277 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 278 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 279 280 /* Maximal space eaten by iovec or ancillary data plus some space */ 281 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 282 EXPORT_SYMBOL(sysctl_optmem_max); 283 284 int sysctl_tstamp_allow_data __read_mostly = 1; 285 286 DEFINE_STATIC_KEY_FALSE(memalloc_socks_key); 287 EXPORT_SYMBOL_GPL(memalloc_socks_key); 288 289 /** 290 * sk_set_memalloc - sets %SOCK_MEMALLOC 291 * @sk: socket to set it on 292 * 293 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves. 294 * It's the responsibility of the admin to adjust min_free_kbytes 295 * to meet the requirements 296 */ 297 void sk_set_memalloc(struct sock *sk) 298 { 299 sock_set_flag(sk, SOCK_MEMALLOC); 300 sk->sk_allocation |= __GFP_MEMALLOC; 301 static_branch_inc(&memalloc_socks_key); 302 } 303 EXPORT_SYMBOL_GPL(sk_set_memalloc); 304 305 void sk_clear_memalloc(struct sock *sk) 306 { 307 sock_reset_flag(sk, SOCK_MEMALLOC); 308 sk->sk_allocation &= ~__GFP_MEMALLOC; 309 static_branch_dec(&memalloc_socks_key); 310 311 /* 312 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward 313 * progress of swapping. SOCK_MEMALLOC may be cleared while 314 * it has rmem allocations due to the last swapfile being deactivated 315 * but there is a risk that the socket is unusable due to exceeding 316 * the rmem limits. Reclaim the reserves and obey rmem limits again. 317 */ 318 sk_mem_reclaim(sk); 319 } 320 EXPORT_SYMBOL_GPL(sk_clear_memalloc); 321 322 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 323 { 324 int ret; 325 unsigned int noreclaim_flag; 326 327 /* these should have been dropped before queueing */ 328 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC)); 329 330 noreclaim_flag = memalloc_noreclaim_save(); 331 ret = sk->sk_backlog_rcv(sk, skb); 332 memalloc_noreclaim_restore(noreclaim_flag); 333 334 return ret; 335 } 336 EXPORT_SYMBOL(__sk_backlog_rcv); 337 338 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 339 { 340 struct timeval tv; 341 342 if (optlen < sizeof(tv)) 343 return -EINVAL; 344 if (copy_from_user(&tv, optval, sizeof(tv))) 345 return -EFAULT; 346 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 347 return -EDOM; 348 349 if (tv.tv_sec < 0) { 350 static int warned __read_mostly; 351 352 *timeo_p = 0; 353 if (warned < 10 && net_ratelimit()) { 354 warned++; 355 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 356 __func__, current->comm, task_pid_nr(current)); 357 } 358 return 0; 359 } 360 *timeo_p = MAX_SCHEDULE_TIMEOUT; 361 if (tv.tv_sec == 0 && tv.tv_usec == 0) 362 return 0; 363 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 364 *timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ); 365 return 0; 366 } 367 368 static void sock_warn_obsolete_bsdism(const char *name) 369 { 370 static int warned; 371 static char warncomm[TASK_COMM_LEN]; 372 if (strcmp(warncomm, current->comm) && warned < 5) { 373 strcpy(warncomm, current->comm); 374 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 375 warncomm, name); 376 warned++; 377 } 378 } 379 380 static bool sock_needs_netstamp(const struct sock *sk) 381 { 382 switch (sk->sk_family) { 383 case AF_UNSPEC: 384 case AF_UNIX: 385 return false; 386 default: 387 return true; 388 } 389 } 390 391 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 392 { 393 if (sk->sk_flags & flags) { 394 sk->sk_flags &= ~flags; 395 if (sock_needs_netstamp(sk) && 396 !(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 397 net_disable_timestamp(); 398 } 399 } 400 401 402 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 403 { 404 unsigned long flags; 405 struct sk_buff_head *list = &sk->sk_receive_queue; 406 407 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 408 atomic_inc(&sk->sk_drops); 409 trace_sock_rcvqueue_full(sk, skb); 410 return -ENOMEM; 411 } 412 413 if (!sk_rmem_schedule(sk, skb, skb->truesize)) { 414 atomic_inc(&sk->sk_drops); 415 return -ENOBUFS; 416 } 417 418 skb->dev = NULL; 419 skb_set_owner_r(skb, sk); 420 421 /* we escape from rcu protected region, make sure we dont leak 422 * a norefcounted dst 423 */ 424 skb_dst_force(skb); 425 426 spin_lock_irqsave(&list->lock, flags); 427 sock_skb_set_dropcount(sk, skb); 428 __skb_queue_tail(list, skb); 429 spin_unlock_irqrestore(&list->lock, flags); 430 431 if (!sock_flag(sk, SOCK_DEAD)) 432 sk->sk_data_ready(sk); 433 return 0; 434 } 435 EXPORT_SYMBOL(__sock_queue_rcv_skb); 436 437 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 438 { 439 int err; 440 441 err = sk_filter(sk, skb); 442 if (err) 443 return err; 444 445 return __sock_queue_rcv_skb(sk, skb); 446 } 447 EXPORT_SYMBOL(sock_queue_rcv_skb); 448 449 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, 450 const int nested, unsigned int trim_cap, bool refcounted) 451 { 452 int rc = NET_RX_SUCCESS; 453 454 if (sk_filter_trim_cap(sk, skb, trim_cap)) 455 goto discard_and_relse; 456 457 skb->dev = NULL; 458 459 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) { 460 atomic_inc(&sk->sk_drops); 461 goto discard_and_relse; 462 } 463 if (nested) 464 bh_lock_sock_nested(sk); 465 else 466 bh_lock_sock(sk); 467 if (!sock_owned_by_user(sk)) { 468 /* 469 * trylock + unlock semantics: 470 */ 471 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 472 473 rc = sk_backlog_rcv(sk, skb); 474 475 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 476 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 477 bh_unlock_sock(sk); 478 atomic_inc(&sk->sk_drops); 479 goto discard_and_relse; 480 } 481 482 bh_unlock_sock(sk); 483 out: 484 if (refcounted) 485 sock_put(sk); 486 return rc; 487 discard_and_relse: 488 kfree_skb(skb); 489 goto out; 490 } 491 EXPORT_SYMBOL(__sk_receive_skb); 492 493 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 494 { 495 struct dst_entry *dst = __sk_dst_get(sk); 496 497 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 498 sk_tx_queue_clear(sk); 499 sk->sk_dst_pending_confirm = 0; 500 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 501 dst_release(dst); 502 return NULL; 503 } 504 505 return dst; 506 } 507 EXPORT_SYMBOL(__sk_dst_check); 508 509 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 510 { 511 struct dst_entry *dst = sk_dst_get(sk); 512 513 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 514 sk_dst_reset(sk); 515 dst_release(dst); 516 return NULL; 517 } 518 519 return dst; 520 } 521 EXPORT_SYMBOL(sk_dst_check); 522 523 static int sock_setbindtodevice(struct sock *sk, char __user *optval, 524 int optlen) 525 { 526 int ret = -ENOPROTOOPT; 527 #ifdef CONFIG_NETDEVICES 528 struct net *net = sock_net(sk); 529 char devname[IFNAMSIZ]; 530 int index; 531 532 /* Sorry... */ 533 ret = -EPERM; 534 if (!ns_capable(net->user_ns, CAP_NET_RAW)) 535 goto out; 536 537 ret = -EINVAL; 538 if (optlen < 0) 539 goto out; 540 541 /* Bind this socket to a particular device like "eth0", 542 * as specified in the passed interface name. If the 543 * name is "" or the option length is zero the socket 544 * is not bound. 545 */ 546 if (optlen > IFNAMSIZ - 1) 547 optlen = IFNAMSIZ - 1; 548 memset(devname, 0, sizeof(devname)); 549 550 ret = -EFAULT; 551 if (copy_from_user(devname, optval, optlen)) 552 goto out; 553 554 index = 0; 555 if (devname[0] != '\0') { 556 struct net_device *dev; 557 558 rcu_read_lock(); 559 dev = dev_get_by_name_rcu(net, devname); 560 if (dev) 561 index = dev->ifindex; 562 rcu_read_unlock(); 563 ret = -ENODEV; 564 if (!dev) 565 goto out; 566 } 567 568 lock_sock(sk); 569 sk->sk_bound_dev_if = index; 570 if (sk->sk_prot->rehash) 571 sk->sk_prot->rehash(sk); 572 sk_dst_reset(sk); 573 release_sock(sk); 574 575 ret = 0; 576 577 out: 578 #endif 579 580 return ret; 581 } 582 583 static int sock_getbindtodevice(struct sock *sk, char __user *optval, 584 int __user *optlen, int len) 585 { 586 int ret = -ENOPROTOOPT; 587 #ifdef CONFIG_NETDEVICES 588 struct net *net = sock_net(sk); 589 char devname[IFNAMSIZ]; 590 591 if (sk->sk_bound_dev_if == 0) { 592 len = 0; 593 goto zero; 594 } 595 596 ret = -EINVAL; 597 if (len < IFNAMSIZ) 598 goto out; 599 600 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if); 601 if (ret) 602 goto out; 603 604 len = strlen(devname) + 1; 605 606 ret = -EFAULT; 607 if (copy_to_user(optval, devname, len)) 608 goto out; 609 610 zero: 611 ret = -EFAULT; 612 if (put_user(len, optlen)) 613 goto out; 614 615 ret = 0; 616 617 out: 618 #endif 619 620 return ret; 621 } 622 623 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 624 { 625 if (valbool) 626 sock_set_flag(sk, bit); 627 else 628 sock_reset_flag(sk, bit); 629 } 630 631 bool sk_mc_loop(struct sock *sk) 632 { 633 if (dev_recursion_level()) 634 return false; 635 if (!sk) 636 return true; 637 switch (sk->sk_family) { 638 case AF_INET: 639 return inet_sk(sk)->mc_loop; 640 #if IS_ENABLED(CONFIG_IPV6) 641 case AF_INET6: 642 return inet6_sk(sk)->mc_loop; 643 #endif 644 } 645 WARN_ON(1); 646 return true; 647 } 648 EXPORT_SYMBOL(sk_mc_loop); 649 650 /* 651 * This is meant for all protocols to use and covers goings on 652 * at the socket level. Everything here is generic. 653 */ 654 655 int sock_setsockopt(struct socket *sock, int level, int optname, 656 char __user *optval, unsigned int optlen) 657 { 658 struct sock_txtime sk_txtime; 659 struct sock *sk = sock->sk; 660 int val; 661 int valbool; 662 struct linger ling; 663 int ret = 0; 664 665 /* 666 * Options without arguments 667 */ 668 669 if (optname == SO_BINDTODEVICE) 670 return sock_setbindtodevice(sk, optval, optlen); 671 672 if (optlen < sizeof(int)) 673 return -EINVAL; 674 675 if (get_user(val, (int __user *)optval)) 676 return -EFAULT; 677 678 valbool = val ? 1 : 0; 679 680 lock_sock(sk); 681 682 switch (optname) { 683 case SO_DEBUG: 684 if (val && !capable(CAP_NET_ADMIN)) 685 ret = -EACCES; 686 else 687 sock_valbool_flag(sk, SOCK_DBG, valbool); 688 break; 689 case SO_REUSEADDR: 690 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 691 break; 692 case SO_REUSEPORT: 693 sk->sk_reuseport = valbool; 694 break; 695 case SO_TYPE: 696 case SO_PROTOCOL: 697 case SO_DOMAIN: 698 case SO_ERROR: 699 ret = -ENOPROTOOPT; 700 break; 701 case SO_DONTROUTE: 702 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 703 sk_dst_reset(sk); 704 break; 705 case SO_BROADCAST: 706 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 707 break; 708 case SO_SNDBUF: 709 /* Don't error on this BSD doesn't and if you think 710 * about it this is right. Otherwise apps have to 711 * play 'guess the biggest size' games. RCVBUF/SNDBUF 712 * are treated in BSD as hints 713 */ 714 val = min_t(u32, val, sysctl_wmem_max); 715 set_sndbuf: 716 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 717 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 718 /* Wake up sending tasks if we upped the value. */ 719 sk->sk_write_space(sk); 720 break; 721 722 case SO_SNDBUFFORCE: 723 if (!capable(CAP_NET_ADMIN)) { 724 ret = -EPERM; 725 break; 726 } 727 goto set_sndbuf; 728 729 case SO_RCVBUF: 730 /* Don't error on this BSD doesn't and if you think 731 * about it this is right. Otherwise apps have to 732 * play 'guess the biggest size' games. RCVBUF/SNDBUF 733 * are treated in BSD as hints 734 */ 735 val = min_t(u32, val, sysctl_rmem_max); 736 set_rcvbuf: 737 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 738 /* 739 * We double it on the way in to account for 740 * "struct sk_buff" etc. overhead. Applications 741 * assume that the SO_RCVBUF setting they make will 742 * allow that much actual data to be received on that 743 * socket. 744 * 745 * Applications are unaware that "struct sk_buff" and 746 * other overheads allocate from the receive buffer 747 * during socket buffer allocation. 748 * 749 * And after considering the possible alternatives, 750 * returning the value we actually used in getsockopt 751 * is the most desirable behavior. 752 */ 753 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 754 break; 755 756 case SO_RCVBUFFORCE: 757 if (!capable(CAP_NET_ADMIN)) { 758 ret = -EPERM; 759 break; 760 } 761 goto set_rcvbuf; 762 763 case SO_KEEPALIVE: 764 if (sk->sk_prot->keepalive) 765 sk->sk_prot->keepalive(sk, valbool); 766 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 767 break; 768 769 case SO_OOBINLINE: 770 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 771 break; 772 773 case SO_NO_CHECK: 774 sk->sk_no_check_tx = valbool; 775 break; 776 777 case SO_PRIORITY: 778 if ((val >= 0 && val <= 6) || 779 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 780 sk->sk_priority = val; 781 else 782 ret = -EPERM; 783 break; 784 785 case SO_LINGER: 786 if (optlen < sizeof(ling)) { 787 ret = -EINVAL; /* 1003.1g */ 788 break; 789 } 790 if (copy_from_user(&ling, optval, sizeof(ling))) { 791 ret = -EFAULT; 792 break; 793 } 794 if (!ling.l_onoff) 795 sock_reset_flag(sk, SOCK_LINGER); 796 else { 797 #if (BITS_PER_LONG == 32) 798 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 799 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 800 else 801 #endif 802 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 803 sock_set_flag(sk, SOCK_LINGER); 804 } 805 break; 806 807 case SO_BSDCOMPAT: 808 sock_warn_obsolete_bsdism("setsockopt"); 809 break; 810 811 case SO_PASSCRED: 812 if (valbool) 813 set_bit(SOCK_PASSCRED, &sock->flags); 814 else 815 clear_bit(SOCK_PASSCRED, &sock->flags); 816 break; 817 818 case SO_TIMESTAMP: 819 case SO_TIMESTAMPNS: 820 if (valbool) { 821 if (optname == SO_TIMESTAMP) 822 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 823 else 824 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 825 sock_set_flag(sk, SOCK_RCVTSTAMP); 826 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 827 } else { 828 sock_reset_flag(sk, SOCK_RCVTSTAMP); 829 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 830 } 831 break; 832 833 case SO_TIMESTAMPING: 834 if (val & ~SOF_TIMESTAMPING_MASK) { 835 ret = -EINVAL; 836 break; 837 } 838 839 if (val & SOF_TIMESTAMPING_OPT_ID && 840 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) { 841 if (sk->sk_protocol == IPPROTO_TCP && 842 sk->sk_type == SOCK_STREAM) { 843 if ((1 << sk->sk_state) & 844 (TCPF_CLOSE | TCPF_LISTEN)) { 845 ret = -EINVAL; 846 break; 847 } 848 sk->sk_tskey = tcp_sk(sk)->snd_una; 849 } else { 850 sk->sk_tskey = 0; 851 } 852 } 853 854 if (val & SOF_TIMESTAMPING_OPT_STATS && 855 !(val & SOF_TIMESTAMPING_OPT_TSONLY)) { 856 ret = -EINVAL; 857 break; 858 } 859 860 sk->sk_tsflags = val; 861 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 862 sock_enable_timestamp(sk, 863 SOCK_TIMESTAMPING_RX_SOFTWARE); 864 else 865 sock_disable_timestamp(sk, 866 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 867 break; 868 869 case SO_RCVLOWAT: 870 if (val < 0) 871 val = INT_MAX; 872 if (sock->ops->set_rcvlowat) 873 ret = sock->ops->set_rcvlowat(sk, val); 874 else 875 sk->sk_rcvlowat = val ? : 1; 876 break; 877 878 case SO_RCVTIMEO: 879 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 880 break; 881 882 case SO_SNDTIMEO: 883 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 884 break; 885 886 case SO_ATTACH_FILTER: 887 ret = -EINVAL; 888 if (optlen == sizeof(struct sock_fprog)) { 889 struct sock_fprog fprog; 890 891 ret = -EFAULT; 892 if (copy_from_user(&fprog, optval, sizeof(fprog))) 893 break; 894 895 ret = sk_attach_filter(&fprog, sk); 896 } 897 break; 898 899 case SO_ATTACH_BPF: 900 ret = -EINVAL; 901 if (optlen == sizeof(u32)) { 902 u32 ufd; 903 904 ret = -EFAULT; 905 if (copy_from_user(&ufd, optval, sizeof(ufd))) 906 break; 907 908 ret = sk_attach_bpf(ufd, sk); 909 } 910 break; 911 912 case SO_ATTACH_REUSEPORT_CBPF: 913 ret = -EINVAL; 914 if (optlen == sizeof(struct sock_fprog)) { 915 struct sock_fprog fprog; 916 917 ret = -EFAULT; 918 if (copy_from_user(&fprog, optval, sizeof(fprog))) 919 break; 920 921 ret = sk_reuseport_attach_filter(&fprog, sk); 922 } 923 break; 924 925 case SO_ATTACH_REUSEPORT_EBPF: 926 ret = -EINVAL; 927 if (optlen == sizeof(u32)) { 928 u32 ufd; 929 930 ret = -EFAULT; 931 if (copy_from_user(&ufd, optval, sizeof(ufd))) 932 break; 933 934 ret = sk_reuseport_attach_bpf(ufd, sk); 935 } 936 break; 937 938 case SO_DETACH_FILTER: 939 ret = sk_detach_filter(sk); 940 break; 941 942 case SO_LOCK_FILTER: 943 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool) 944 ret = -EPERM; 945 else 946 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool); 947 break; 948 949 case SO_PASSSEC: 950 if (valbool) 951 set_bit(SOCK_PASSSEC, &sock->flags); 952 else 953 clear_bit(SOCK_PASSSEC, &sock->flags); 954 break; 955 case SO_MARK: 956 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 957 ret = -EPERM; 958 } else if (val != sk->sk_mark) { 959 sk->sk_mark = val; 960 sk_dst_reset(sk); 961 } 962 break; 963 964 case SO_RXQ_OVFL: 965 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 966 break; 967 968 case SO_WIFI_STATUS: 969 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 970 break; 971 972 case SO_PEEK_OFF: 973 if (sock->ops->set_peek_off) 974 ret = sock->ops->set_peek_off(sk, val); 975 else 976 ret = -EOPNOTSUPP; 977 break; 978 979 case SO_NOFCS: 980 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 981 break; 982 983 case SO_SELECT_ERR_QUEUE: 984 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool); 985 break; 986 987 #ifdef CONFIG_NET_RX_BUSY_POLL 988 case SO_BUSY_POLL: 989 /* allow unprivileged users to decrease the value */ 990 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN)) 991 ret = -EPERM; 992 else { 993 if (val < 0) 994 ret = -EINVAL; 995 else 996 sk->sk_ll_usec = val; 997 } 998 break; 999 #endif 1000 1001 case SO_MAX_PACING_RATE: 1002 if (val != ~0U) 1003 cmpxchg(&sk->sk_pacing_status, 1004 SK_PACING_NONE, 1005 SK_PACING_NEEDED); 1006 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val; 1007 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 1008 sk->sk_max_pacing_rate); 1009 break; 1010 1011 case SO_INCOMING_CPU: 1012 sk->sk_incoming_cpu = val; 1013 break; 1014 1015 case SO_CNX_ADVICE: 1016 if (val == 1) 1017 dst_negative_advice(sk); 1018 break; 1019 1020 case SO_ZEROCOPY: 1021 if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) { 1022 if (!((sk->sk_type == SOCK_STREAM && 1023 sk->sk_protocol == IPPROTO_TCP) || 1024 (sk->sk_type == SOCK_DGRAM && 1025 sk->sk_protocol == IPPROTO_UDP))) 1026 ret = -ENOTSUPP; 1027 } else if (sk->sk_family != PF_RDS) { 1028 ret = -ENOTSUPP; 1029 } 1030 if (!ret) { 1031 if (val < 0 || val > 1) 1032 ret = -EINVAL; 1033 else 1034 sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool); 1035 } 1036 break; 1037 1038 case SO_TXTIME: 1039 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) { 1040 ret = -EPERM; 1041 } else if (optlen != sizeof(struct sock_txtime)) { 1042 ret = -EINVAL; 1043 } else if (copy_from_user(&sk_txtime, optval, 1044 sizeof(struct sock_txtime))) { 1045 ret = -EFAULT; 1046 } else if (sk_txtime.flags & ~SOF_TXTIME_FLAGS_MASK) { 1047 ret = -EINVAL; 1048 } else { 1049 sock_valbool_flag(sk, SOCK_TXTIME, true); 1050 sk->sk_clockid = sk_txtime.clockid; 1051 sk->sk_txtime_deadline_mode = 1052 !!(sk_txtime.flags & SOF_TXTIME_DEADLINE_MODE); 1053 sk->sk_txtime_report_errors = 1054 !!(sk_txtime.flags & SOF_TXTIME_REPORT_ERRORS); 1055 } 1056 break; 1057 1058 default: 1059 ret = -ENOPROTOOPT; 1060 break; 1061 } 1062 release_sock(sk); 1063 return ret; 1064 } 1065 EXPORT_SYMBOL(sock_setsockopt); 1066 1067 1068 static void cred_to_ucred(struct pid *pid, const struct cred *cred, 1069 struct ucred *ucred) 1070 { 1071 ucred->pid = pid_vnr(pid); 1072 ucred->uid = ucred->gid = -1; 1073 if (cred) { 1074 struct user_namespace *current_ns = current_user_ns(); 1075 1076 ucred->uid = from_kuid_munged(current_ns, cred->euid); 1077 ucred->gid = from_kgid_munged(current_ns, cred->egid); 1078 } 1079 } 1080 1081 static int groups_to_user(gid_t __user *dst, const struct group_info *src) 1082 { 1083 struct user_namespace *user_ns = current_user_ns(); 1084 int i; 1085 1086 for (i = 0; i < src->ngroups; i++) 1087 if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i)) 1088 return -EFAULT; 1089 1090 return 0; 1091 } 1092 1093 int sock_getsockopt(struct socket *sock, int level, int optname, 1094 char __user *optval, int __user *optlen) 1095 { 1096 struct sock *sk = sock->sk; 1097 1098 union { 1099 int val; 1100 u64 val64; 1101 struct linger ling; 1102 struct timeval tm; 1103 struct sock_txtime txtime; 1104 } v; 1105 1106 int lv = sizeof(int); 1107 int len; 1108 1109 if (get_user(len, optlen)) 1110 return -EFAULT; 1111 if (len < 0) 1112 return -EINVAL; 1113 1114 memset(&v, 0, sizeof(v)); 1115 1116 switch (optname) { 1117 case SO_DEBUG: 1118 v.val = sock_flag(sk, SOCK_DBG); 1119 break; 1120 1121 case SO_DONTROUTE: 1122 v.val = sock_flag(sk, SOCK_LOCALROUTE); 1123 break; 1124 1125 case SO_BROADCAST: 1126 v.val = sock_flag(sk, SOCK_BROADCAST); 1127 break; 1128 1129 case SO_SNDBUF: 1130 v.val = sk->sk_sndbuf; 1131 break; 1132 1133 case SO_RCVBUF: 1134 v.val = sk->sk_rcvbuf; 1135 break; 1136 1137 case SO_REUSEADDR: 1138 v.val = sk->sk_reuse; 1139 break; 1140 1141 case SO_REUSEPORT: 1142 v.val = sk->sk_reuseport; 1143 break; 1144 1145 case SO_KEEPALIVE: 1146 v.val = sock_flag(sk, SOCK_KEEPOPEN); 1147 break; 1148 1149 case SO_TYPE: 1150 v.val = sk->sk_type; 1151 break; 1152 1153 case SO_PROTOCOL: 1154 v.val = sk->sk_protocol; 1155 break; 1156 1157 case SO_DOMAIN: 1158 v.val = sk->sk_family; 1159 break; 1160 1161 case SO_ERROR: 1162 v.val = -sock_error(sk); 1163 if (v.val == 0) 1164 v.val = xchg(&sk->sk_err_soft, 0); 1165 break; 1166 1167 case SO_OOBINLINE: 1168 v.val = sock_flag(sk, SOCK_URGINLINE); 1169 break; 1170 1171 case SO_NO_CHECK: 1172 v.val = sk->sk_no_check_tx; 1173 break; 1174 1175 case SO_PRIORITY: 1176 v.val = sk->sk_priority; 1177 break; 1178 1179 case SO_LINGER: 1180 lv = sizeof(v.ling); 1181 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 1182 v.ling.l_linger = sk->sk_lingertime / HZ; 1183 break; 1184 1185 case SO_BSDCOMPAT: 1186 sock_warn_obsolete_bsdism("getsockopt"); 1187 break; 1188 1189 case SO_TIMESTAMP: 1190 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 1191 !sock_flag(sk, SOCK_RCVTSTAMPNS); 1192 break; 1193 1194 case SO_TIMESTAMPNS: 1195 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 1196 break; 1197 1198 case SO_TIMESTAMPING: 1199 v.val = sk->sk_tsflags; 1200 break; 1201 1202 case SO_RCVTIMEO: 1203 lv = sizeof(struct timeval); 1204 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 1205 v.tm.tv_sec = 0; 1206 v.tm.tv_usec = 0; 1207 } else { 1208 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 1209 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ; 1210 } 1211 break; 1212 1213 case SO_SNDTIMEO: 1214 lv = sizeof(struct timeval); 1215 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 1216 v.tm.tv_sec = 0; 1217 v.tm.tv_usec = 0; 1218 } else { 1219 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 1220 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ; 1221 } 1222 break; 1223 1224 case SO_RCVLOWAT: 1225 v.val = sk->sk_rcvlowat; 1226 break; 1227 1228 case SO_SNDLOWAT: 1229 v.val = 1; 1230 break; 1231 1232 case SO_PASSCRED: 1233 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 1234 break; 1235 1236 case SO_PEERCRED: 1237 { 1238 struct ucred peercred; 1239 if (len > sizeof(peercred)) 1240 len = sizeof(peercred); 1241 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 1242 if (copy_to_user(optval, &peercred, len)) 1243 return -EFAULT; 1244 goto lenout; 1245 } 1246 1247 case SO_PEERGROUPS: 1248 { 1249 int ret, n; 1250 1251 if (!sk->sk_peer_cred) 1252 return -ENODATA; 1253 1254 n = sk->sk_peer_cred->group_info->ngroups; 1255 if (len < n * sizeof(gid_t)) { 1256 len = n * sizeof(gid_t); 1257 return put_user(len, optlen) ? -EFAULT : -ERANGE; 1258 } 1259 len = n * sizeof(gid_t); 1260 1261 ret = groups_to_user((gid_t __user *)optval, 1262 sk->sk_peer_cred->group_info); 1263 if (ret) 1264 return ret; 1265 goto lenout; 1266 } 1267 1268 case SO_PEERNAME: 1269 { 1270 char address[128]; 1271 1272 lv = sock->ops->getname(sock, (struct sockaddr *)address, 2); 1273 if (lv < 0) 1274 return -ENOTCONN; 1275 if (lv < len) 1276 return -EINVAL; 1277 if (copy_to_user(optval, address, len)) 1278 return -EFAULT; 1279 goto lenout; 1280 } 1281 1282 /* Dubious BSD thing... Probably nobody even uses it, but 1283 * the UNIX standard wants it for whatever reason... -DaveM 1284 */ 1285 case SO_ACCEPTCONN: 1286 v.val = sk->sk_state == TCP_LISTEN; 1287 break; 1288 1289 case SO_PASSSEC: 1290 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1291 break; 1292 1293 case SO_PEERSEC: 1294 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1295 1296 case SO_MARK: 1297 v.val = sk->sk_mark; 1298 break; 1299 1300 case SO_RXQ_OVFL: 1301 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1302 break; 1303 1304 case SO_WIFI_STATUS: 1305 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1306 break; 1307 1308 case SO_PEEK_OFF: 1309 if (!sock->ops->set_peek_off) 1310 return -EOPNOTSUPP; 1311 1312 v.val = sk->sk_peek_off; 1313 break; 1314 case SO_NOFCS: 1315 v.val = sock_flag(sk, SOCK_NOFCS); 1316 break; 1317 1318 case SO_BINDTODEVICE: 1319 return sock_getbindtodevice(sk, optval, optlen, len); 1320 1321 case SO_GET_FILTER: 1322 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len); 1323 if (len < 0) 1324 return len; 1325 1326 goto lenout; 1327 1328 case SO_LOCK_FILTER: 1329 v.val = sock_flag(sk, SOCK_FILTER_LOCKED); 1330 break; 1331 1332 case SO_BPF_EXTENSIONS: 1333 v.val = bpf_tell_extensions(); 1334 break; 1335 1336 case SO_SELECT_ERR_QUEUE: 1337 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE); 1338 break; 1339 1340 #ifdef CONFIG_NET_RX_BUSY_POLL 1341 case SO_BUSY_POLL: 1342 v.val = sk->sk_ll_usec; 1343 break; 1344 #endif 1345 1346 case SO_MAX_PACING_RATE: 1347 /* 32bit version */ 1348 v.val = min_t(unsigned long, sk->sk_max_pacing_rate, ~0U); 1349 break; 1350 1351 case SO_INCOMING_CPU: 1352 v.val = sk->sk_incoming_cpu; 1353 break; 1354 1355 case SO_MEMINFO: 1356 { 1357 u32 meminfo[SK_MEMINFO_VARS]; 1358 1359 if (get_user(len, optlen)) 1360 return -EFAULT; 1361 1362 sk_get_meminfo(sk, meminfo); 1363 1364 len = min_t(unsigned int, len, sizeof(meminfo)); 1365 if (copy_to_user(optval, &meminfo, len)) 1366 return -EFAULT; 1367 1368 goto lenout; 1369 } 1370 1371 #ifdef CONFIG_NET_RX_BUSY_POLL 1372 case SO_INCOMING_NAPI_ID: 1373 v.val = READ_ONCE(sk->sk_napi_id); 1374 1375 /* aggregate non-NAPI IDs down to 0 */ 1376 if (v.val < MIN_NAPI_ID) 1377 v.val = 0; 1378 1379 break; 1380 #endif 1381 1382 case SO_COOKIE: 1383 lv = sizeof(u64); 1384 if (len < lv) 1385 return -EINVAL; 1386 v.val64 = sock_gen_cookie(sk); 1387 break; 1388 1389 case SO_ZEROCOPY: 1390 v.val = sock_flag(sk, SOCK_ZEROCOPY); 1391 break; 1392 1393 case SO_TXTIME: 1394 lv = sizeof(v.txtime); 1395 v.txtime.clockid = sk->sk_clockid; 1396 v.txtime.flags |= sk->sk_txtime_deadline_mode ? 1397 SOF_TXTIME_DEADLINE_MODE : 0; 1398 v.txtime.flags |= sk->sk_txtime_report_errors ? 1399 SOF_TXTIME_REPORT_ERRORS : 0; 1400 break; 1401 1402 default: 1403 /* We implement the SO_SNDLOWAT etc to not be settable 1404 * (1003.1g 7). 1405 */ 1406 return -ENOPROTOOPT; 1407 } 1408 1409 if (len > lv) 1410 len = lv; 1411 if (copy_to_user(optval, &v, len)) 1412 return -EFAULT; 1413 lenout: 1414 if (put_user(len, optlen)) 1415 return -EFAULT; 1416 return 0; 1417 } 1418 1419 /* 1420 * Initialize an sk_lock. 1421 * 1422 * (We also register the sk_lock with the lock validator.) 1423 */ 1424 static inline void sock_lock_init(struct sock *sk) 1425 { 1426 if (sk->sk_kern_sock) 1427 sock_lock_init_class_and_name( 1428 sk, 1429 af_family_kern_slock_key_strings[sk->sk_family], 1430 af_family_kern_slock_keys + sk->sk_family, 1431 af_family_kern_key_strings[sk->sk_family], 1432 af_family_kern_keys + sk->sk_family); 1433 else 1434 sock_lock_init_class_and_name( 1435 sk, 1436 af_family_slock_key_strings[sk->sk_family], 1437 af_family_slock_keys + sk->sk_family, 1438 af_family_key_strings[sk->sk_family], 1439 af_family_keys + sk->sk_family); 1440 } 1441 1442 /* 1443 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1444 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1445 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1446 */ 1447 static void sock_copy(struct sock *nsk, const struct sock *osk) 1448 { 1449 #ifdef CONFIG_SECURITY_NETWORK 1450 void *sptr = nsk->sk_security; 1451 #endif 1452 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1453 1454 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1455 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1456 1457 #ifdef CONFIG_SECURITY_NETWORK 1458 nsk->sk_security = sptr; 1459 security_sk_clone(osk, nsk); 1460 #endif 1461 } 1462 1463 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1464 int family) 1465 { 1466 struct sock *sk; 1467 struct kmem_cache *slab; 1468 1469 slab = prot->slab; 1470 if (slab != NULL) { 1471 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1472 if (!sk) 1473 return sk; 1474 if (priority & __GFP_ZERO) 1475 sk_prot_clear_nulls(sk, prot->obj_size); 1476 } else 1477 sk = kmalloc(prot->obj_size, priority); 1478 1479 if (sk != NULL) { 1480 if (security_sk_alloc(sk, family, priority)) 1481 goto out_free; 1482 1483 if (!try_module_get(prot->owner)) 1484 goto out_free_sec; 1485 sk_tx_queue_clear(sk); 1486 } 1487 1488 return sk; 1489 1490 out_free_sec: 1491 security_sk_free(sk); 1492 out_free: 1493 if (slab != NULL) 1494 kmem_cache_free(slab, sk); 1495 else 1496 kfree(sk); 1497 return NULL; 1498 } 1499 1500 static void sk_prot_free(struct proto *prot, struct sock *sk) 1501 { 1502 struct kmem_cache *slab; 1503 struct module *owner; 1504 1505 owner = prot->owner; 1506 slab = prot->slab; 1507 1508 cgroup_sk_free(&sk->sk_cgrp_data); 1509 mem_cgroup_sk_free(sk); 1510 security_sk_free(sk); 1511 if (slab != NULL) 1512 kmem_cache_free(slab, sk); 1513 else 1514 kfree(sk); 1515 module_put(owner); 1516 } 1517 1518 /** 1519 * sk_alloc - All socket objects are allocated here 1520 * @net: the applicable net namespace 1521 * @family: protocol family 1522 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1523 * @prot: struct proto associated with this new sock instance 1524 * @kern: is this to be a kernel socket? 1525 */ 1526 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1527 struct proto *prot, int kern) 1528 { 1529 struct sock *sk; 1530 1531 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1532 if (sk) { 1533 sk->sk_family = family; 1534 /* 1535 * See comment in struct sock definition to understand 1536 * why we need sk_prot_creator -acme 1537 */ 1538 sk->sk_prot = sk->sk_prot_creator = prot; 1539 sk->sk_kern_sock = kern; 1540 sock_lock_init(sk); 1541 sk->sk_net_refcnt = kern ? 0 : 1; 1542 if (likely(sk->sk_net_refcnt)) { 1543 get_net(net); 1544 sock_inuse_add(net, 1); 1545 } 1546 1547 sock_net_set(sk, net); 1548 refcount_set(&sk->sk_wmem_alloc, 1); 1549 1550 mem_cgroup_sk_alloc(sk); 1551 cgroup_sk_alloc(&sk->sk_cgrp_data); 1552 sock_update_classid(&sk->sk_cgrp_data); 1553 sock_update_netprioidx(&sk->sk_cgrp_data); 1554 } 1555 1556 return sk; 1557 } 1558 EXPORT_SYMBOL(sk_alloc); 1559 1560 /* Sockets having SOCK_RCU_FREE will call this function after one RCU 1561 * grace period. This is the case for UDP sockets and TCP listeners. 1562 */ 1563 static void __sk_destruct(struct rcu_head *head) 1564 { 1565 struct sock *sk = container_of(head, struct sock, sk_rcu); 1566 struct sk_filter *filter; 1567 1568 if (sk->sk_destruct) 1569 sk->sk_destruct(sk); 1570 1571 filter = rcu_dereference_check(sk->sk_filter, 1572 refcount_read(&sk->sk_wmem_alloc) == 0); 1573 if (filter) { 1574 sk_filter_uncharge(sk, filter); 1575 RCU_INIT_POINTER(sk->sk_filter, NULL); 1576 } 1577 if (rcu_access_pointer(sk->sk_reuseport_cb)) 1578 reuseport_detach_sock(sk); 1579 1580 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1581 1582 if (atomic_read(&sk->sk_omem_alloc)) 1583 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1584 __func__, atomic_read(&sk->sk_omem_alloc)); 1585 1586 if (sk->sk_frag.page) { 1587 put_page(sk->sk_frag.page); 1588 sk->sk_frag.page = NULL; 1589 } 1590 1591 if (sk->sk_peer_cred) 1592 put_cred(sk->sk_peer_cred); 1593 put_pid(sk->sk_peer_pid); 1594 if (likely(sk->sk_net_refcnt)) 1595 put_net(sock_net(sk)); 1596 sk_prot_free(sk->sk_prot_creator, sk); 1597 } 1598 1599 void sk_destruct(struct sock *sk) 1600 { 1601 if (sock_flag(sk, SOCK_RCU_FREE)) 1602 call_rcu(&sk->sk_rcu, __sk_destruct); 1603 else 1604 __sk_destruct(&sk->sk_rcu); 1605 } 1606 1607 static void __sk_free(struct sock *sk) 1608 { 1609 if (likely(sk->sk_net_refcnt)) 1610 sock_inuse_add(sock_net(sk), -1); 1611 1612 if (unlikely(sk->sk_net_refcnt && sock_diag_has_destroy_listeners(sk))) 1613 sock_diag_broadcast_destroy(sk); 1614 else 1615 sk_destruct(sk); 1616 } 1617 1618 void sk_free(struct sock *sk) 1619 { 1620 /* 1621 * We subtract one from sk_wmem_alloc and can know if 1622 * some packets are still in some tx queue. 1623 * If not null, sock_wfree() will call __sk_free(sk) later 1624 */ 1625 if (refcount_dec_and_test(&sk->sk_wmem_alloc)) 1626 __sk_free(sk); 1627 } 1628 EXPORT_SYMBOL(sk_free); 1629 1630 static void sk_init_common(struct sock *sk) 1631 { 1632 skb_queue_head_init(&sk->sk_receive_queue); 1633 skb_queue_head_init(&sk->sk_write_queue); 1634 skb_queue_head_init(&sk->sk_error_queue); 1635 1636 rwlock_init(&sk->sk_callback_lock); 1637 lockdep_set_class_and_name(&sk->sk_receive_queue.lock, 1638 af_rlock_keys + sk->sk_family, 1639 af_family_rlock_key_strings[sk->sk_family]); 1640 lockdep_set_class_and_name(&sk->sk_write_queue.lock, 1641 af_wlock_keys + sk->sk_family, 1642 af_family_wlock_key_strings[sk->sk_family]); 1643 lockdep_set_class_and_name(&sk->sk_error_queue.lock, 1644 af_elock_keys + sk->sk_family, 1645 af_family_elock_key_strings[sk->sk_family]); 1646 lockdep_set_class_and_name(&sk->sk_callback_lock, 1647 af_callback_keys + sk->sk_family, 1648 af_family_clock_key_strings[sk->sk_family]); 1649 } 1650 1651 /** 1652 * sk_clone_lock - clone a socket, and lock its clone 1653 * @sk: the socket to clone 1654 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1655 * 1656 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1657 */ 1658 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1659 { 1660 struct sock *newsk; 1661 bool is_charged = true; 1662 1663 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1664 if (newsk != NULL) { 1665 struct sk_filter *filter; 1666 1667 sock_copy(newsk, sk); 1668 1669 newsk->sk_prot_creator = sk->sk_prot; 1670 1671 /* SANITY */ 1672 if (likely(newsk->sk_net_refcnt)) 1673 get_net(sock_net(newsk)); 1674 sk_node_init(&newsk->sk_node); 1675 sock_lock_init(newsk); 1676 bh_lock_sock(newsk); 1677 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1678 newsk->sk_backlog.len = 0; 1679 1680 atomic_set(&newsk->sk_rmem_alloc, 0); 1681 /* 1682 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1683 */ 1684 refcount_set(&newsk->sk_wmem_alloc, 1); 1685 atomic_set(&newsk->sk_omem_alloc, 0); 1686 sk_init_common(newsk); 1687 1688 newsk->sk_dst_cache = NULL; 1689 newsk->sk_dst_pending_confirm = 0; 1690 newsk->sk_wmem_queued = 0; 1691 newsk->sk_forward_alloc = 0; 1692 atomic_set(&newsk->sk_drops, 0); 1693 newsk->sk_send_head = NULL; 1694 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1695 atomic_set(&newsk->sk_zckey, 0); 1696 1697 sock_reset_flag(newsk, SOCK_DONE); 1698 mem_cgroup_sk_alloc(newsk); 1699 cgroup_sk_alloc(&newsk->sk_cgrp_data); 1700 1701 rcu_read_lock(); 1702 filter = rcu_dereference(sk->sk_filter); 1703 if (filter != NULL) 1704 /* though it's an empty new sock, the charging may fail 1705 * if sysctl_optmem_max was changed between creation of 1706 * original socket and cloning 1707 */ 1708 is_charged = sk_filter_charge(newsk, filter); 1709 RCU_INIT_POINTER(newsk->sk_filter, filter); 1710 rcu_read_unlock(); 1711 1712 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) { 1713 /* We need to make sure that we don't uncharge the new 1714 * socket if we couldn't charge it in the first place 1715 * as otherwise we uncharge the parent's filter. 1716 */ 1717 if (!is_charged) 1718 RCU_INIT_POINTER(newsk->sk_filter, NULL); 1719 sk_free_unlock_clone(newsk); 1720 newsk = NULL; 1721 goto out; 1722 } 1723 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL); 1724 1725 newsk->sk_err = 0; 1726 newsk->sk_err_soft = 0; 1727 newsk->sk_priority = 0; 1728 newsk->sk_incoming_cpu = raw_smp_processor_id(); 1729 atomic64_set(&newsk->sk_cookie, 0); 1730 if (likely(newsk->sk_net_refcnt)) 1731 sock_inuse_add(sock_net(newsk), 1); 1732 1733 /* 1734 * Before updating sk_refcnt, we must commit prior changes to memory 1735 * (Documentation/RCU/rculist_nulls.txt for details) 1736 */ 1737 smp_wmb(); 1738 refcount_set(&newsk->sk_refcnt, 2); 1739 1740 /* 1741 * Increment the counter in the same struct proto as the master 1742 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1743 * is the same as sk->sk_prot->socks, as this field was copied 1744 * with memcpy). 1745 * 1746 * This _changes_ the previous behaviour, where 1747 * tcp_create_openreq_child always was incrementing the 1748 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1749 * to be taken into account in all callers. -acme 1750 */ 1751 sk_refcnt_debug_inc(newsk); 1752 sk_set_socket(newsk, NULL); 1753 newsk->sk_wq = NULL; 1754 1755 if (newsk->sk_prot->sockets_allocated) 1756 sk_sockets_allocated_inc(newsk); 1757 1758 if (sock_needs_netstamp(sk) && 1759 newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1760 net_enable_timestamp(); 1761 } 1762 out: 1763 return newsk; 1764 } 1765 EXPORT_SYMBOL_GPL(sk_clone_lock); 1766 1767 void sk_free_unlock_clone(struct sock *sk) 1768 { 1769 /* It is still raw copy of parent, so invalidate 1770 * destructor and make plain sk_free() */ 1771 sk->sk_destruct = NULL; 1772 bh_unlock_sock(sk); 1773 sk_free(sk); 1774 } 1775 EXPORT_SYMBOL_GPL(sk_free_unlock_clone); 1776 1777 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1778 { 1779 u32 max_segs = 1; 1780 1781 sk_dst_set(sk, dst); 1782 sk->sk_route_caps = dst->dev->features | sk->sk_route_forced_caps; 1783 if (sk->sk_route_caps & NETIF_F_GSO) 1784 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1785 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1786 if (sk_can_gso(sk)) { 1787 if (dst->header_len && !xfrm_dst_offload_ok(dst)) { 1788 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1789 } else { 1790 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1791 sk->sk_gso_max_size = dst->dev->gso_max_size; 1792 max_segs = max_t(u32, dst->dev->gso_max_segs, 1); 1793 } 1794 } 1795 sk->sk_gso_max_segs = max_segs; 1796 } 1797 EXPORT_SYMBOL_GPL(sk_setup_caps); 1798 1799 /* 1800 * Simple resource managers for sockets. 1801 */ 1802 1803 1804 /* 1805 * Write buffer destructor automatically called from kfree_skb. 1806 */ 1807 void sock_wfree(struct sk_buff *skb) 1808 { 1809 struct sock *sk = skb->sk; 1810 unsigned int len = skb->truesize; 1811 1812 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1813 /* 1814 * Keep a reference on sk_wmem_alloc, this will be released 1815 * after sk_write_space() call 1816 */ 1817 WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc)); 1818 sk->sk_write_space(sk); 1819 len = 1; 1820 } 1821 /* 1822 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1823 * could not do because of in-flight packets 1824 */ 1825 if (refcount_sub_and_test(len, &sk->sk_wmem_alloc)) 1826 __sk_free(sk); 1827 } 1828 EXPORT_SYMBOL(sock_wfree); 1829 1830 /* This variant of sock_wfree() is used by TCP, 1831 * since it sets SOCK_USE_WRITE_QUEUE. 1832 */ 1833 void __sock_wfree(struct sk_buff *skb) 1834 { 1835 struct sock *sk = skb->sk; 1836 1837 if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)) 1838 __sk_free(sk); 1839 } 1840 1841 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1842 { 1843 skb_orphan(skb); 1844 skb->sk = sk; 1845 #ifdef CONFIG_INET 1846 if (unlikely(!sk_fullsock(sk))) { 1847 skb->destructor = sock_edemux; 1848 sock_hold(sk); 1849 return; 1850 } 1851 #endif 1852 skb->destructor = sock_wfree; 1853 skb_set_hash_from_sk(skb, sk); 1854 /* 1855 * We used to take a refcount on sk, but following operation 1856 * is enough to guarantee sk_free() wont free this sock until 1857 * all in-flight packets are completed 1858 */ 1859 refcount_add(skb->truesize, &sk->sk_wmem_alloc); 1860 } 1861 EXPORT_SYMBOL(skb_set_owner_w); 1862 1863 /* This helper is used by netem, as it can hold packets in its 1864 * delay queue. We want to allow the owner socket to send more 1865 * packets, as if they were already TX completed by a typical driver. 1866 * But we also want to keep skb->sk set because some packet schedulers 1867 * rely on it (sch_fq for example). 1868 */ 1869 void skb_orphan_partial(struct sk_buff *skb) 1870 { 1871 if (skb_is_tcp_pure_ack(skb)) 1872 return; 1873 1874 if (skb->destructor == sock_wfree 1875 #ifdef CONFIG_INET 1876 || skb->destructor == tcp_wfree 1877 #endif 1878 ) { 1879 struct sock *sk = skb->sk; 1880 1881 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 1882 WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc)); 1883 skb->destructor = sock_efree; 1884 } 1885 } else { 1886 skb_orphan(skb); 1887 } 1888 } 1889 EXPORT_SYMBOL(skb_orphan_partial); 1890 1891 /* 1892 * Read buffer destructor automatically called from kfree_skb. 1893 */ 1894 void sock_rfree(struct sk_buff *skb) 1895 { 1896 struct sock *sk = skb->sk; 1897 unsigned int len = skb->truesize; 1898 1899 atomic_sub(len, &sk->sk_rmem_alloc); 1900 sk_mem_uncharge(sk, len); 1901 } 1902 EXPORT_SYMBOL(sock_rfree); 1903 1904 /* 1905 * Buffer destructor for skbs that are not used directly in read or write 1906 * path, e.g. for error handler skbs. Automatically called from kfree_skb. 1907 */ 1908 void sock_efree(struct sk_buff *skb) 1909 { 1910 sock_put(skb->sk); 1911 } 1912 EXPORT_SYMBOL(sock_efree); 1913 1914 kuid_t sock_i_uid(struct sock *sk) 1915 { 1916 kuid_t uid; 1917 1918 read_lock_bh(&sk->sk_callback_lock); 1919 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID; 1920 read_unlock_bh(&sk->sk_callback_lock); 1921 return uid; 1922 } 1923 EXPORT_SYMBOL(sock_i_uid); 1924 1925 unsigned long sock_i_ino(struct sock *sk) 1926 { 1927 unsigned long ino; 1928 1929 read_lock_bh(&sk->sk_callback_lock); 1930 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1931 read_unlock_bh(&sk->sk_callback_lock); 1932 return ino; 1933 } 1934 EXPORT_SYMBOL(sock_i_ino); 1935 1936 /* 1937 * Allocate a skb from the socket's send buffer. 1938 */ 1939 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1940 gfp_t priority) 1941 { 1942 if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1943 struct sk_buff *skb = alloc_skb(size, priority); 1944 if (skb) { 1945 skb_set_owner_w(skb, sk); 1946 return skb; 1947 } 1948 } 1949 return NULL; 1950 } 1951 EXPORT_SYMBOL(sock_wmalloc); 1952 1953 static void sock_ofree(struct sk_buff *skb) 1954 { 1955 struct sock *sk = skb->sk; 1956 1957 atomic_sub(skb->truesize, &sk->sk_omem_alloc); 1958 } 1959 1960 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1961 gfp_t priority) 1962 { 1963 struct sk_buff *skb; 1964 1965 /* small safe race: SKB_TRUESIZE may differ from final skb->truesize */ 1966 if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) > 1967 sysctl_optmem_max) 1968 return NULL; 1969 1970 skb = alloc_skb(size, priority); 1971 if (!skb) 1972 return NULL; 1973 1974 atomic_add(skb->truesize, &sk->sk_omem_alloc); 1975 skb->sk = sk; 1976 skb->destructor = sock_ofree; 1977 return skb; 1978 } 1979 1980 /* 1981 * Allocate a memory block from the socket's option memory buffer. 1982 */ 1983 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1984 { 1985 if ((unsigned int)size <= sysctl_optmem_max && 1986 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1987 void *mem; 1988 /* First do the add, to avoid the race if kmalloc 1989 * might sleep. 1990 */ 1991 atomic_add(size, &sk->sk_omem_alloc); 1992 mem = kmalloc(size, priority); 1993 if (mem) 1994 return mem; 1995 atomic_sub(size, &sk->sk_omem_alloc); 1996 } 1997 return NULL; 1998 } 1999 EXPORT_SYMBOL(sock_kmalloc); 2000 2001 /* Free an option memory block. Note, we actually want the inline 2002 * here as this allows gcc to detect the nullify and fold away the 2003 * condition entirely. 2004 */ 2005 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size, 2006 const bool nullify) 2007 { 2008 if (WARN_ON_ONCE(!mem)) 2009 return; 2010 if (nullify) 2011 kzfree(mem); 2012 else 2013 kfree(mem); 2014 atomic_sub(size, &sk->sk_omem_alloc); 2015 } 2016 2017 void sock_kfree_s(struct sock *sk, void *mem, int size) 2018 { 2019 __sock_kfree_s(sk, mem, size, false); 2020 } 2021 EXPORT_SYMBOL(sock_kfree_s); 2022 2023 void sock_kzfree_s(struct sock *sk, void *mem, int size) 2024 { 2025 __sock_kfree_s(sk, mem, size, true); 2026 } 2027 EXPORT_SYMBOL(sock_kzfree_s); 2028 2029 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 2030 I think, these locks should be removed for datagram sockets. 2031 */ 2032 static long sock_wait_for_wmem(struct sock *sk, long timeo) 2033 { 2034 DEFINE_WAIT(wait); 2035 2036 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2037 for (;;) { 2038 if (!timeo) 2039 break; 2040 if (signal_pending(current)) 2041 break; 2042 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2043 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 2044 if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 2045 break; 2046 if (sk->sk_shutdown & SEND_SHUTDOWN) 2047 break; 2048 if (sk->sk_err) 2049 break; 2050 timeo = schedule_timeout(timeo); 2051 } 2052 finish_wait(sk_sleep(sk), &wait); 2053 return timeo; 2054 } 2055 2056 2057 /* 2058 * Generic send/receive buffer handlers 2059 */ 2060 2061 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 2062 unsigned long data_len, int noblock, 2063 int *errcode, int max_page_order) 2064 { 2065 struct sk_buff *skb; 2066 long timeo; 2067 int err; 2068 2069 timeo = sock_sndtimeo(sk, noblock); 2070 for (;;) { 2071 err = sock_error(sk); 2072 if (err != 0) 2073 goto failure; 2074 2075 err = -EPIPE; 2076 if (sk->sk_shutdown & SEND_SHUTDOWN) 2077 goto failure; 2078 2079 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf) 2080 break; 2081 2082 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2083 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2084 err = -EAGAIN; 2085 if (!timeo) 2086 goto failure; 2087 if (signal_pending(current)) 2088 goto interrupted; 2089 timeo = sock_wait_for_wmem(sk, timeo); 2090 } 2091 skb = alloc_skb_with_frags(header_len, data_len, max_page_order, 2092 errcode, sk->sk_allocation); 2093 if (skb) 2094 skb_set_owner_w(skb, sk); 2095 return skb; 2096 2097 interrupted: 2098 err = sock_intr_errno(timeo); 2099 failure: 2100 *errcode = err; 2101 return NULL; 2102 } 2103 EXPORT_SYMBOL(sock_alloc_send_pskb); 2104 2105 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 2106 int noblock, int *errcode) 2107 { 2108 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 2109 } 2110 EXPORT_SYMBOL(sock_alloc_send_skb); 2111 2112 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 2113 struct sockcm_cookie *sockc) 2114 { 2115 u32 tsflags; 2116 2117 switch (cmsg->cmsg_type) { 2118 case SO_MARK: 2119 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN)) 2120 return -EPERM; 2121 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2122 return -EINVAL; 2123 sockc->mark = *(u32 *)CMSG_DATA(cmsg); 2124 break; 2125 case SO_TIMESTAMPING: 2126 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32))) 2127 return -EINVAL; 2128 2129 tsflags = *(u32 *)CMSG_DATA(cmsg); 2130 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK) 2131 return -EINVAL; 2132 2133 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK; 2134 sockc->tsflags |= tsflags; 2135 break; 2136 case SCM_TXTIME: 2137 if (!sock_flag(sk, SOCK_TXTIME)) 2138 return -EINVAL; 2139 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u64))) 2140 return -EINVAL; 2141 sockc->transmit_time = get_unaligned((u64 *)CMSG_DATA(cmsg)); 2142 break; 2143 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */ 2144 case SCM_RIGHTS: 2145 case SCM_CREDENTIALS: 2146 break; 2147 default: 2148 return -EINVAL; 2149 } 2150 return 0; 2151 } 2152 EXPORT_SYMBOL(__sock_cmsg_send); 2153 2154 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 2155 struct sockcm_cookie *sockc) 2156 { 2157 struct cmsghdr *cmsg; 2158 int ret; 2159 2160 for_each_cmsghdr(cmsg, msg) { 2161 if (!CMSG_OK(msg, cmsg)) 2162 return -EINVAL; 2163 if (cmsg->cmsg_level != SOL_SOCKET) 2164 continue; 2165 ret = __sock_cmsg_send(sk, msg, cmsg, sockc); 2166 if (ret) 2167 return ret; 2168 } 2169 return 0; 2170 } 2171 EXPORT_SYMBOL(sock_cmsg_send); 2172 2173 static void sk_enter_memory_pressure(struct sock *sk) 2174 { 2175 if (!sk->sk_prot->enter_memory_pressure) 2176 return; 2177 2178 sk->sk_prot->enter_memory_pressure(sk); 2179 } 2180 2181 static void sk_leave_memory_pressure(struct sock *sk) 2182 { 2183 if (sk->sk_prot->leave_memory_pressure) { 2184 sk->sk_prot->leave_memory_pressure(sk); 2185 } else { 2186 unsigned long *memory_pressure = sk->sk_prot->memory_pressure; 2187 2188 if (memory_pressure && *memory_pressure) 2189 *memory_pressure = 0; 2190 } 2191 } 2192 2193 /* On 32bit arches, an skb frag is limited to 2^15 */ 2194 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2195 2196 /** 2197 * skb_page_frag_refill - check that a page_frag contains enough room 2198 * @sz: minimum size of the fragment we want to get 2199 * @pfrag: pointer to page_frag 2200 * @gfp: priority for memory allocation 2201 * 2202 * Note: While this allocator tries to use high order pages, there is 2203 * no guarantee that allocations succeed. Therefore, @sz MUST be 2204 * less or equal than PAGE_SIZE. 2205 */ 2206 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp) 2207 { 2208 if (pfrag->page) { 2209 if (page_ref_count(pfrag->page) == 1) { 2210 pfrag->offset = 0; 2211 return true; 2212 } 2213 if (pfrag->offset + sz <= pfrag->size) 2214 return true; 2215 put_page(pfrag->page); 2216 } 2217 2218 pfrag->offset = 0; 2219 if (SKB_FRAG_PAGE_ORDER) { 2220 /* Avoid direct reclaim but allow kswapd to wake */ 2221 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) | 2222 __GFP_COMP | __GFP_NOWARN | 2223 __GFP_NORETRY, 2224 SKB_FRAG_PAGE_ORDER); 2225 if (likely(pfrag->page)) { 2226 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER; 2227 return true; 2228 } 2229 } 2230 pfrag->page = alloc_page(gfp); 2231 if (likely(pfrag->page)) { 2232 pfrag->size = PAGE_SIZE; 2233 return true; 2234 } 2235 return false; 2236 } 2237 EXPORT_SYMBOL(skb_page_frag_refill); 2238 2239 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag) 2240 { 2241 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation))) 2242 return true; 2243 2244 sk_enter_memory_pressure(sk); 2245 sk_stream_moderate_sndbuf(sk); 2246 return false; 2247 } 2248 EXPORT_SYMBOL(sk_page_frag_refill); 2249 2250 static void __lock_sock(struct sock *sk) 2251 __releases(&sk->sk_lock.slock) 2252 __acquires(&sk->sk_lock.slock) 2253 { 2254 DEFINE_WAIT(wait); 2255 2256 for (;;) { 2257 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 2258 TASK_UNINTERRUPTIBLE); 2259 spin_unlock_bh(&sk->sk_lock.slock); 2260 schedule(); 2261 spin_lock_bh(&sk->sk_lock.slock); 2262 if (!sock_owned_by_user(sk)) 2263 break; 2264 } 2265 finish_wait(&sk->sk_lock.wq, &wait); 2266 } 2267 2268 void __release_sock(struct sock *sk) 2269 __releases(&sk->sk_lock.slock) 2270 __acquires(&sk->sk_lock.slock) 2271 { 2272 struct sk_buff *skb, *next; 2273 2274 while ((skb = sk->sk_backlog.head) != NULL) { 2275 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 2276 2277 spin_unlock_bh(&sk->sk_lock.slock); 2278 2279 do { 2280 next = skb->next; 2281 prefetch(next); 2282 WARN_ON_ONCE(skb_dst_is_noref(skb)); 2283 skb_mark_not_on_list(skb); 2284 sk_backlog_rcv(sk, skb); 2285 2286 cond_resched(); 2287 2288 skb = next; 2289 } while (skb != NULL); 2290 2291 spin_lock_bh(&sk->sk_lock.slock); 2292 } 2293 2294 /* 2295 * Doing the zeroing here guarantee we can not loop forever 2296 * while a wild producer attempts to flood us. 2297 */ 2298 sk->sk_backlog.len = 0; 2299 } 2300 2301 void __sk_flush_backlog(struct sock *sk) 2302 { 2303 spin_lock_bh(&sk->sk_lock.slock); 2304 __release_sock(sk); 2305 spin_unlock_bh(&sk->sk_lock.slock); 2306 } 2307 2308 /** 2309 * sk_wait_data - wait for data to arrive at sk_receive_queue 2310 * @sk: sock to wait on 2311 * @timeo: for how long 2312 * @skb: last skb seen on sk_receive_queue 2313 * 2314 * Now socket state including sk->sk_err is changed only under lock, 2315 * hence we may omit checks after joining wait queue. 2316 * We check receive queue before schedule() only as optimization; 2317 * it is very likely that release_sock() added new data. 2318 */ 2319 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb) 2320 { 2321 DEFINE_WAIT_FUNC(wait, woken_wake_function); 2322 int rc; 2323 2324 add_wait_queue(sk_sleep(sk), &wait); 2325 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2326 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait); 2327 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk); 2328 remove_wait_queue(sk_sleep(sk), &wait); 2329 return rc; 2330 } 2331 EXPORT_SYMBOL(sk_wait_data); 2332 2333 /** 2334 * __sk_mem_raise_allocated - increase memory_allocated 2335 * @sk: socket 2336 * @size: memory size to allocate 2337 * @amt: pages to allocate 2338 * @kind: allocation type 2339 * 2340 * Similar to __sk_mem_schedule(), but does not update sk_forward_alloc 2341 */ 2342 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind) 2343 { 2344 struct proto *prot = sk->sk_prot; 2345 long allocated = sk_memory_allocated_add(sk, amt); 2346 bool charged = true; 2347 2348 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 2349 !(charged = mem_cgroup_charge_skmem(sk->sk_memcg, amt))) 2350 goto suppress_allocation; 2351 2352 /* Under limit. */ 2353 if (allocated <= sk_prot_mem_limits(sk, 0)) { 2354 sk_leave_memory_pressure(sk); 2355 return 1; 2356 } 2357 2358 /* Under pressure. */ 2359 if (allocated > sk_prot_mem_limits(sk, 1)) 2360 sk_enter_memory_pressure(sk); 2361 2362 /* Over hard limit. */ 2363 if (allocated > sk_prot_mem_limits(sk, 2)) 2364 goto suppress_allocation; 2365 2366 /* guarantee minimum buffer size under pressure */ 2367 if (kind == SK_MEM_RECV) { 2368 if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot)) 2369 return 1; 2370 2371 } else { /* SK_MEM_SEND */ 2372 int wmem0 = sk_get_wmem0(sk, prot); 2373 2374 if (sk->sk_type == SOCK_STREAM) { 2375 if (sk->sk_wmem_queued < wmem0) 2376 return 1; 2377 } else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) { 2378 return 1; 2379 } 2380 } 2381 2382 if (sk_has_memory_pressure(sk)) { 2383 u64 alloc; 2384 2385 if (!sk_under_memory_pressure(sk)) 2386 return 1; 2387 alloc = sk_sockets_allocated_read_positive(sk); 2388 if (sk_prot_mem_limits(sk, 2) > alloc * 2389 sk_mem_pages(sk->sk_wmem_queued + 2390 atomic_read(&sk->sk_rmem_alloc) + 2391 sk->sk_forward_alloc)) 2392 return 1; 2393 } 2394 2395 suppress_allocation: 2396 2397 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 2398 sk_stream_moderate_sndbuf(sk); 2399 2400 /* Fail only if socket is _under_ its sndbuf. 2401 * In this case we cannot block, so that we have to fail. 2402 */ 2403 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 2404 return 1; 2405 } 2406 2407 if (kind == SK_MEM_SEND || (kind == SK_MEM_RECV && charged)) 2408 trace_sock_exceed_buf_limit(sk, prot, allocated, kind); 2409 2410 sk_memory_allocated_sub(sk, amt); 2411 2412 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2413 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt); 2414 2415 return 0; 2416 } 2417 EXPORT_SYMBOL(__sk_mem_raise_allocated); 2418 2419 /** 2420 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 2421 * @sk: socket 2422 * @size: memory size to allocate 2423 * @kind: allocation type 2424 * 2425 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 2426 * rmem allocation. This function assumes that protocols which have 2427 * memory_pressure use sk_wmem_queued as write buffer accounting. 2428 */ 2429 int __sk_mem_schedule(struct sock *sk, int size, int kind) 2430 { 2431 int ret, amt = sk_mem_pages(size); 2432 2433 sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT; 2434 ret = __sk_mem_raise_allocated(sk, size, amt, kind); 2435 if (!ret) 2436 sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT; 2437 return ret; 2438 } 2439 EXPORT_SYMBOL(__sk_mem_schedule); 2440 2441 /** 2442 * __sk_mem_reduce_allocated - reclaim memory_allocated 2443 * @sk: socket 2444 * @amount: number of quanta 2445 * 2446 * Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc 2447 */ 2448 void __sk_mem_reduce_allocated(struct sock *sk, int amount) 2449 { 2450 sk_memory_allocated_sub(sk, amount); 2451 2452 if (mem_cgroup_sockets_enabled && sk->sk_memcg) 2453 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount); 2454 2455 if (sk_under_memory_pressure(sk) && 2456 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 2457 sk_leave_memory_pressure(sk); 2458 } 2459 EXPORT_SYMBOL(__sk_mem_reduce_allocated); 2460 2461 /** 2462 * __sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated 2463 * @sk: socket 2464 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple) 2465 */ 2466 void __sk_mem_reclaim(struct sock *sk, int amount) 2467 { 2468 amount >>= SK_MEM_QUANTUM_SHIFT; 2469 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT; 2470 __sk_mem_reduce_allocated(sk, amount); 2471 } 2472 EXPORT_SYMBOL(__sk_mem_reclaim); 2473 2474 int sk_set_peek_off(struct sock *sk, int val) 2475 { 2476 sk->sk_peek_off = val; 2477 return 0; 2478 } 2479 EXPORT_SYMBOL_GPL(sk_set_peek_off); 2480 2481 /* 2482 * Set of default routines for initialising struct proto_ops when 2483 * the protocol does not support a particular function. In certain 2484 * cases where it makes no sense for a protocol to have a "do nothing" 2485 * function, some default processing is provided. 2486 */ 2487 2488 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 2489 { 2490 return -EOPNOTSUPP; 2491 } 2492 EXPORT_SYMBOL(sock_no_bind); 2493 2494 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 2495 int len, int flags) 2496 { 2497 return -EOPNOTSUPP; 2498 } 2499 EXPORT_SYMBOL(sock_no_connect); 2500 2501 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 2502 { 2503 return -EOPNOTSUPP; 2504 } 2505 EXPORT_SYMBOL(sock_no_socketpair); 2506 2507 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags, 2508 bool kern) 2509 { 2510 return -EOPNOTSUPP; 2511 } 2512 EXPORT_SYMBOL(sock_no_accept); 2513 2514 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 2515 int peer) 2516 { 2517 return -EOPNOTSUPP; 2518 } 2519 EXPORT_SYMBOL(sock_no_getname); 2520 2521 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 2522 { 2523 return -EOPNOTSUPP; 2524 } 2525 EXPORT_SYMBOL(sock_no_ioctl); 2526 2527 int sock_no_listen(struct socket *sock, int backlog) 2528 { 2529 return -EOPNOTSUPP; 2530 } 2531 EXPORT_SYMBOL(sock_no_listen); 2532 2533 int sock_no_shutdown(struct socket *sock, int how) 2534 { 2535 return -EOPNOTSUPP; 2536 } 2537 EXPORT_SYMBOL(sock_no_shutdown); 2538 2539 int sock_no_setsockopt(struct socket *sock, int level, int optname, 2540 char __user *optval, unsigned int optlen) 2541 { 2542 return -EOPNOTSUPP; 2543 } 2544 EXPORT_SYMBOL(sock_no_setsockopt); 2545 2546 int sock_no_getsockopt(struct socket *sock, int level, int optname, 2547 char __user *optval, int __user *optlen) 2548 { 2549 return -EOPNOTSUPP; 2550 } 2551 EXPORT_SYMBOL(sock_no_getsockopt); 2552 2553 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len) 2554 { 2555 return -EOPNOTSUPP; 2556 } 2557 EXPORT_SYMBOL(sock_no_sendmsg); 2558 2559 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len) 2560 { 2561 return -EOPNOTSUPP; 2562 } 2563 EXPORT_SYMBOL(sock_no_sendmsg_locked); 2564 2565 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len, 2566 int flags) 2567 { 2568 return -EOPNOTSUPP; 2569 } 2570 EXPORT_SYMBOL(sock_no_recvmsg); 2571 2572 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 2573 { 2574 /* Mirror missing mmap method error code */ 2575 return -ENODEV; 2576 } 2577 EXPORT_SYMBOL(sock_no_mmap); 2578 2579 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 2580 { 2581 ssize_t res; 2582 struct msghdr msg = {.msg_flags = flags}; 2583 struct kvec iov; 2584 char *kaddr = kmap(page); 2585 iov.iov_base = kaddr + offset; 2586 iov.iov_len = size; 2587 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 2588 kunmap(page); 2589 return res; 2590 } 2591 EXPORT_SYMBOL(sock_no_sendpage); 2592 2593 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 2594 int offset, size_t size, int flags) 2595 { 2596 ssize_t res; 2597 struct msghdr msg = {.msg_flags = flags}; 2598 struct kvec iov; 2599 char *kaddr = kmap(page); 2600 2601 iov.iov_base = kaddr + offset; 2602 iov.iov_len = size; 2603 res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size); 2604 kunmap(page); 2605 return res; 2606 } 2607 EXPORT_SYMBOL(sock_no_sendpage_locked); 2608 2609 /* 2610 * Default Socket Callbacks 2611 */ 2612 2613 static void sock_def_wakeup(struct sock *sk) 2614 { 2615 struct socket_wq *wq; 2616 2617 rcu_read_lock(); 2618 wq = rcu_dereference(sk->sk_wq); 2619 if (skwq_has_sleeper(wq)) 2620 wake_up_interruptible_all(&wq->wait); 2621 rcu_read_unlock(); 2622 } 2623 2624 static void sock_def_error_report(struct sock *sk) 2625 { 2626 struct socket_wq *wq; 2627 2628 rcu_read_lock(); 2629 wq = rcu_dereference(sk->sk_wq); 2630 if (skwq_has_sleeper(wq)) 2631 wake_up_interruptible_poll(&wq->wait, EPOLLERR); 2632 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2633 rcu_read_unlock(); 2634 } 2635 2636 static void sock_def_readable(struct sock *sk) 2637 { 2638 struct socket_wq *wq; 2639 2640 rcu_read_lock(); 2641 wq = rcu_dereference(sk->sk_wq); 2642 if (skwq_has_sleeper(wq)) 2643 wake_up_interruptible_sync_poll(&wq->wait, EPOLLIN | EPOLLPRI | 2644 EPOLLRDNORM | EPOLLRDBAND); 2645 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2646 rcu_read_unlock(); 2647 } 2648 2649 static void sock_def_write_space(struct sock *sk) 2650 { 2651 struct socket_wq *wq; 2652 2653 rcu_read_lock(); 2654 2655 /* Do not wake up a writer until he can make "significant" 2656 * progress. --DaveM 2657 */ 2658 if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2659 wq = rcu_dereference(sk->sk_wq); 2660 if (skwq_has_sleeper(wq)) 2661 wake_up_interruptible_sync_poll(&wq->wait, EPOLLOUT | 2662 EPOLLWRNORM | EPOLLWRBAND); 2663 2664 /* Should agree with poll, otherwise some programs break */ 2665 if (sock_writeable(sk)) 2666 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2667 } 2668 2669 rcu_read_unlock(); 2670 } 2671 2672 static void sock_def_destruct(struct sock *sk) 2673 { 2674 } 2675 2676 void sk_send_sigurg(struct sock *sk) 2677 { 2678 if (sk->sk_socket && sk->sk_socket->file) 2679 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2680 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2681 } 2682 EXPORT_SYMBOL(sk_send_sigurg); 2683 2684 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2685 unsigned long expires) 2686 { 2687 if (!mod_timer(timer, expires)) 2688 sock_hold(sk); 2689 } 2690 EXPORT_SYMBOL(sk_reset_timer); 2691 2692 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2693 { 2694 if (del_timer(timer)) 2695 __sock_put(sk); 2696 } 2697 EXPORT_SYMBOL(sk_stop_timer); 2698 2699 void sock_init_data(struct socket *sock, struct sock *sk) 2700 { 2701 sk_init_common(sk); 2702 sk->sk_send_head = NULL; 2703 2704 timer_setup(&sk->sk_timer, NULL, 0); 2705 2706 sk->sk_allocation = GFP_KERNEL; 2707 sk->sk_rcvbuf = sysctl_rmem_default; 2708 sk->sk_sndbuf = sysctl_wmem_default; 2709 sk->sk_state = TCP_CLOSE; 2710 sk_set_socket(sk, sock); 2711 2712 sock_set_flag(sk, SOCK_ZAPPED); 2713 2714 if (sock) { 2715 sk->sk_type = sock->type; 2716 sk->sk_wq = sock->wq; 2717 sock->sk = sk; 2718 sk->sk_uid = SOCK_INODE(sock)->i_uid; 2719 } else { 2720 sk->sk_wq = NULL; 2721 sk->sk_uid = make_kuid(sock_net(sk)->user_ns, 0); 2722 } 2723 2724 rwlock_init(&sk->sk_callback_lock); 2725 if (sk->sk_kern_sock) 2726 lockdep_set_class_and_name( 2727 &sk->sk_callback_lock, 2728 af_kern_callback_keys + sk->sk_family, 2729 af_family_kern_clock_key_strings[sk->sk_family]); 2730 else 2731 lockdep_set_class_and_name( 2732 &sk->sk_callback_lock, 2733 af_callback_keys + sk->sk_family, 2734 af_family_clock_key_strings[sk->sk_family]); 2735 2736 sk->sk_state_change = sock_def_wakeup; 2737 sk->sk_data_ready = sock_def_readable; 2738 sk->sk_write_space = sock_def_write_space; 2739 sk->sk_error_report = sock_def_error_report; 2740 sk->sk_destruct = sock_def_destruct; 2741 2742 sk->sk_frag.page = NULL; 2743 sk->sk_frag.offset = 0; 2744 sk->sk_peek_off = -1; 2745 2746 sk->sk_peer_pid = NULL; 2747 sk->sk_peer_cred = NULL; 2748 sk->sk_write_pending = 0; 2749 sk->sk_rcvlowat = 1; 2750 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2751 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2752 2753 sk->sk_stamp = SK_DEFAULT_STAMP; 2754 #if BITS_PER_LONG==32 2755 seqlock_init(&sk->sk_stamp_seq); 2756 #endif 2757 atomic_set(&sk->sk_zckey, 0); 2758 2759 #ifdef CONFIG_NET_RX_BUSY_POLL 2760 sk->sk_napi_id = 0; 2761 sk->sk_ll_usec = sysctl_net_busy_read; 2762 #endif 2763 2764 sk->sk_max_pacing_rate = ~0UL; 2765 sk->sk_pacing_rate = ~0UL; 2766 sk->sk_pacing_shift = 10; 2767 sk->sk_incoming_cpu = -1; 2768 2769 sk_rx_queue_clear(sk); 2770 /* 2771 * Before updating sk_refcnt, we must commit prior changes to memory 2772 * (Documentation/RCU/rculist_nulls.txt for details) 2773 */ 2774 smp_wmb(); 2775 refcount_set(&sk->sk_refcnt, 1); 2776 atomic_set(&sk->sk_drops, 0); 2777 } 2778 EXPORT_SYMBOL(sock_init_data); 2779 2780 void lock_sock_nested(struct sock *sk, int subclass) 2781 { 2782 might_sleep(); 2783 spin_lock_bh(&sk->sk_lock.slock); 2784 if (sk->sk_lock.owned) 2785 __lock_sock(sk); 2786 sk->sk_lock.owned = 1; 2787 spin_unlock(&sk->sk_lock.slock); 2788 /* 2789 * The sk_lock has mutex_lock() semantics here: 2790 */ 2791 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2792 local_bh_enable(); 2793 } 2794 EXPORT_SYMBOL(lock_sock_nested); 2795 2796 void release_sock(struct sock *sk) 2797 { 2798 spin_lock_bh(&sk->sk_lock.slock); 2799 if (sk->sk_backlog.tail) 2800 __release_sock(sk); 2801 2802 /* Warning : release_cb() might need to release sk ownership, 2803 * ie call sock_release_ownership(sk) before us. 2804 */ 2805 if (sk->sk_prot->release_cb) 2806 sk->sk_prot->release_cb(sk); 2807 2808 sock_release_ownership(sk); 2809 if (waitqueue_active(&sk->sk_lock.wq)) 2810 wake_up(&sk->sk_lock.wq); 2811 spin_unlock_bh(&sk->sk_lock.slock); 2812 } 2813 EXPORT_SYMBOL(release_sock); 2814 2815 /** 2816 * lock_sock_fast - fast version of lock_sock 2817 * @sk: socket 2818 * 2819 * This version should be used for very small section, where process wont block 2820 * return false if fast path is taken: 2821 * 2822 * sk_lock.slock locked, owned = 0, BH disabled 2823 * 2824 * return true if slow path is taken: 2825 * 2826 * sk_lock.slock unlocked, owned = 1, BH enabled 2827 */ 2828 bool lock_sock_fast(struct sock *sk) 2829 { 2830 might_sleep(); 2831 spin_lock_bh(&sk->sk_lock.slock); 2832 2833 if (!sk->sk_lock.owned) 2834 /* 2835 * Note : We must disable BH 2836 */ 2837 return false; 2838 2839 __lock_sock(sk); 2840 sk->sk_lock.owned = 1; 2841 spin_unlock(&sk->sk_lock.slock); 2842 /* 2843 * The sk_lock has mutex_lock() semantics here: 2844 */ 2845 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2846 local_bh_enable(); 2847 return true; 2848 } 2849 EXPORT_SYMBOL(lock_sock_fast); 2850 2851 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2852 { 2853 struct timeval tv; 2854 2855 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2856 tv = ktime_to_timeval(sock_read_timestamp(sk)); 2857 if (tv.tv_sec == -1) 2858 return -ENOENT; 2859 if (tv.tv_sec == 0) { 2860 ktime_t kt = ktime_get_real(); 2861 sock_write_timestamp(sk, kt); 2862 tv = ktime_to_timeval(kt); 2863 } 2864 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2865 } 2866 EXPORT_SYMBOL(sock_get_timestamp); 2867 2868 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2869 { 2870 struct timespec ts; 2871 2872 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2873 ts = ktime_to_timespec(sock_read_timestamp(sk)); 2874 if (ts.tv_sec == -1) 2875 return -ENOENT; 2876 if (ts.tv_sec == 0) { 2877 ktime_t kt = ktime_get_real(); 2878 sock_write_timestamp(sk, kt); 2879 ts = ktime_to_timespec(sk->sk_stamp); 2880 } 2881 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2882 } 2883 EXPORT_SYMBOL(sock_get_timestampns); 2884 2885 void sock_enable_timestamp(struct sock *sk, int flag) 2886 { 2887 if (!sock_flag(sk, flag)) { 2888 unsigned long previous_flags = sk->sk_flags; 2889 2890 sock_set_flag(sk, flag); 2891 /* 2892 * we just set one of the two flags which require net 2893 * time stamping, but time stamping might have been on 2894 * already because of the other one 2895 */ 2896 if (sock_needs_netstamp(sk) && 2897 !(previous_flags & SK_FLAGS_TIMESTAMP)) 2898 net_enable_timestamp(); 2899 } 2900 } 2901 2902 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2903 int level, int type) 2904 { 2905 struct sock_exterr_skb *serr; 2906 struct sk_buff *skb; 2907 int copied, err; 2908 2909 err = -EAGAIN; 2910 skb = sock_dequeue_err_skb(sk); 2911 if (skb == NULL) 2912 goto out; 2913 2914 copied = skb->len; 2915 if (copied > len) { 2916 msg->msg_flags |= MSG_TRUNC; 2917 copied = len; 2918 } 2919 err = skb_copy_datagram_msg(skb, 0, msg, copied); 2920 if (err) 2921 goto out_free_skb; 2922 2923 sock_recv_timestamp(msg, sk, skb); 2924 2925 serr = SKB_EXT_ERR(skb); 2926 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee); 2927 2928 msg->msg_flags |= MSG_ERRQUEUE; 2929 err = copied; 2930 2931 out_free_skb: 2932 kfree_skb(skb); 2933 out: 2934 return err; 2935 } 2936 EXPORT_SYMBOL(sock_recv_errqueue); 2937 2938 /* 2939 * Get a socket option on an socket. 2940 * 2941 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2942 * asynchronous errors should be reported by getsockopt. We assume 2943 * this means if you specify SO_ERROR (otherwise whats the point of it). 2944 */ 2945 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2946 char __user *optval, int __user *optlen) 2947 { 2948 struct sock *sk = sock->sk; 2949 2950 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2951 } 2952 EXPORT_SYMBOL(sock_common_getsockopt); 2953 2954 #ifdef CONFIG_COMPAT 2955 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2956 char __user *optval, int __user *optlen) 2957 { 2958 struct sock *sk = sock->sk; 2959 2960 if (sk->sk_prot->compat_getsockopt != NULL) 2961 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2962 optval, optlen); 2963 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2964 } 2965 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2966 #endif 2967 2968 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 2969 int flags) 2970 { 2971 struct sock *sk = sock->sk; 2972 int addr_len = 0; 2973 int err; 2974 2975 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 2976 flags & ~MSG_DONTWAIT, &addr_len); 2977 if (err >= 0) 2978 msg->msg_namelen = addr_len; 2979 return err; 2980 } 2981 EXPORT_SYMBOL(sock_common_recvmsg); 2982 2983 /* 2984 * Set socket options on an inet socket. 2985 */ 2986 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2987 char __user *optval, unsigned int optlen) 2988 { 2989 struct sock *sk = sock->sk; 2990 2991 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2992 } 2993 EXPORT_SYMBOL(sock_common_setsockopt); 2994 2995 #ifdef CONFIG_COMPAT 2996 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2997 char __user *optval, unsigned int optlen) 2998 { 2999 struct sock *sk = sock->sk; 3000 3001 if (sk->sk_prot->compat_setsockopt != NULL) 3002 return sk->sk_prot->compat_setsockopt(sk, level, optname, 3003 optval, optlen); 3004 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 3005 } 3006 EXPORT_SYMBOL(compat_sock_common_setsockopt); 3007 #endif 3008 3009 void sk_common_release(struct sock *sk) 3010 { 3011 if (sk->sk_prot->destroy) 3012 sk->sk_prot->destroy(sk); 3013 3014 /* 3015 * Observation: when sock_common_release is called, processes have 3016 * no access to socket. But net still has. 3017 * Step one, detach it from networking: 3018 * 3019 * A. Remove from hash tables. 3020 */ 3021 3022 sk->sk_prot->unhash(sk); 3023 3024 /* 3025 * In this point socket cannot receive new packets, but it is possible 3026 * that some packets are in flight because some CPU runs receiver and 3027 * did hash table lookup before we unhashed socket. They will achieve 3028 * receive queue and will be purged by socket destructor. 3029 * 3030 * Also we still have packets pending on receive queue and probably, 3031 * our own packets waiting in device queues. sock_destroy will drain 3032 * receive queue, but transmitted packets will delay socket destruction 3033 * until the last reference will be released. 3034 */ 3035 3036 sock_orphan(sk); 3037 3038 xfrm_sk_free_policy(sk); 3039 3040 sk_refcnt_debug_release(sk); 3041 3042 sock_put(sk); 3043 } 3044 EXPORT_SYMBOL(sk_common_release); 3045 3046 void sk_get_meminfo(const struct sock *sk, u32 *mem) 3047 { 3048 memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS); 3049 3050 mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk); 3051 mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf; 3052 mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk); 3053 mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf; 3054 mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc; 3055 mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued; 3056 mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc); 3057 mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len; 3058 mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops); 3059 } 3060 3061 #ifdef CONFIG_PROC_FS 3062 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 3063 struct prot_inuse { 3064 int val[PROTO_INUSE_NR]; 3065 }; 3066 3067 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 3068 3069 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 3070 { 3071 __this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 3072 } 3073 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 3074 3075 int sock_prot_inuse_get(struct net *net, struct proto *prot) 3076 { 3077 int cpu, idx = prot->inuse_idx; 3078 int res = 0; 3079 3080 for_each_possible_cpu(cpu) 3081 res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx]; 3082 3083 return res >= 0 ? res : 0; 3084 } 3085 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 3086 3087 static void sock_inuse_add(struct net *net, int val) 3088 { 3089 this_cpu_add(*net->core.sock_inuse, val); 3090 } 3091 3092 int sock_inuse_get(struct net *net) 3093 { 3094 int cpu, res = 0; 3095 3096 for_each_possible_cpu(cpu) 3097 res += *per_cpu_ptr(net->core.sock_inuse, cpu); 3098 3099 return res; 3100 } 3101 3102 EXPORT_SYMBOL_GPL(sock_inuse_get); 3103 3104 static int __net_init sock_inuse_init_net(struct net *net) 3105 { 3106 net->core.prot_inuse = alloc_percpu(struct prot_inuse); 3107 if (net->core.prot_inuse == NULL) 3108 return -ENOMEM; 3109 3110 net->core.sock_inuse = alloc_percpu(int); 3111 if (net->core.sock_inuse == NULL) 3112 goto out; 3113 3114 return 0; 3115 3116 out: 3117 free_percpu(net->core.prot_inuse); 3118 return -ENOMEM; 3119 } 3120 3121 static void __net_exit sock_inuse_exit_net(struct net *net) 3122 { 3123 free_percpu(net->core.prot_inuse); 3124 free_percpu(net->core.sock_inuse); 3125 } 3126 3127 static struct pernet_operations net_inuse_ops = { 3128 .init = sock_inuse_init_net, 3129 .exit = sock_inuse_exit_net, 3130 }; 3131 3132 static __init int net_inuse_init(void) 3133 { 3134 if (register_pernet_subsys(&net_inuse_ops)) 3135 panic("Cannot initialize net inuse counters"); 3136 3137 return 0; 3138 } 3139 3140 core_initcall(net_inuse_init); 3141 3142 static void assign_proto_idx(struct proto *prot) 3143 { 3144 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 3145 3146 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 3147 pr_err("PROTO_INUSE_NR exhausted\n"); 3148 return; 3149 } 3150 3151 set_bit(prot->inuse_idx, proto_inuse_idx); 3152 } 3153 3154 static void release_proto_idx(struct proto *prot) 3155 { 3156 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 3157 clear_bit(prot->inuse_idx, proto_inuse_idx); 3158 } 3159 #else 3160 static inline void assign_proto_idx(struct proto *prot) 3161 { 3162 } 3163 3164 static inline void release_proto_idx(struct proto *prot) 3165 { 3166 } 3167 3168 static void sock_inuse_add(struct net *net, int val) 3169 { 3170 } 3171 #endif 3172 3173 static void req_prot_cleanup(struct request_sock_ops *rsk_prot) 3174 { 3175 if (!rsk_prot) 3176 return; 3177 kfree(rsk_prot->slab_name); 3178 rsk_prot->slab_name = NULL; 3179 kmem_cache_destroy(rsk_prot->slab); 3180 rsk_prot->slab = NULL; 3181 } 3182 3183 static int req_prot_init(const struct proto *prot) 3184 { 3185 struct request_sock_ops *rsk_prot = prot->rsk_prot; 3186 3187 if (!rsk_prot) 3188 return 0; 3189 3190 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", 3191 prot->name); 3192 if (!rsk_prot->slab_name) 3193 return -ENOMEM; 3194 3195 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name, 3196 rsk_prot->obj_size, 0, 3197 SLAB_ACCOUNT | prot->slab_flags, 3198 NULL); 3199 3200 if (!rsk_prot->slab) { 3201 pr_crit("%s: Can't create request sock SLAB cache!\n", 3202 prot->name); 3203 return -ENOMEM; 3204 } 3205 return 0; 3206 } 3207 3208 int proto_register(struct proto *prot, int alloc_slab) 3209 { 3210 if (alloc_slab) { 3211 prot->slab = kmem_cache_create_usercopy(prot->name, 3212 prot->obj_size, 0, 3213 SLAB_HWCACHE_ALIGN | SLAB_ACCOUNT | 3214 prot->slab_flags, 3215 prot->useroffset, prot->usersize, 3216 NULL); 3217 3218 if (prot->slab == NULL) { 3219 pr_crit("%s: Can't create sock SLAB cache!\n", 3220 prot->name); 3221 goto out; 3222 } 3223 3224 if (req_prot_init(prot)) 3225 goto out_free_request_sock_slab; 3226 3227 if (prot->twsk_prot != NULL) { 3228 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 3229 3230 if (prot->twsk_prot->twsk_slab_name == NULL) 3231 goto out_free_request_sock_slab; 3232 3233 prot->twsk_prot->twsk_slab = 3234 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 3235 prot->twsk_prot->twsk_obj_size, 3236 0, 3237 SLAB_ACCOUNT | 3238 prot->slab_flags, 3239 NULL); 3240 if (prot->twsk_prot->twsk_slab == NULL) 3241 goto out_free_timewait_sock_slab_name; 3242 } 3243 } 3244 3245 mutex_lock(&proto_list_mutex); 3246 list_add(&prot->node, &proto_list); 3247 assign_proto_idx(prot); 3248 mutex_unlock(&proto_list_mutex); 3249 return 0; 3250 3251 out_free_timewait_sock_slab_name: 3252 kfree(prot->twsk_prot->twsk_slab_name); 3253 out_free_request_sock_slab: 3254 req_prot_cleanup(prot->rsk_prot); 3255 3256 kmem_cache_destroy(prot->slab); 3257 prot->slab = NULL; 3258 out: 3259 return -ENOBUFS; 3260 } 3261 EXPORT_SYMBOL(proto_register); 3262 3263 void proto_unregister(struct proto *prot) 3264 { 3265 mutex_lock(&proto_list_mutex); 3266 release_proto_idx(prot); 3267 list_del(&prot->node); 3268 mutex_unlock(&proto_list_mutex); 3269 3270 kmem_cache_destroy(prot->slab); 3271 prot->slab = NULL; 3272 3273 req_prot_cleanup(prot->rsk_prot); 3274 3275 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 3276 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 3277 kfree(prot->twsk_prot->twsk_slab_name); 3278 prot->twsk_prot->twsk_slab = NULL; 3279 } 3280 } 3281 EXPORT_SYMBOL(proto_unregister); 3282 3283 int sock_load_diag_module(int family, int protocol) 3284 { 3285 if (!protocol) { 3286 if (!sock_is_registered(family)) 3287 return -ENOENT; 3288 3289 return request_module("net-pf-%d-proto-%d-type-%d", PF_NETLINK, 3290 NETLINK_SOCK_DIAG, family); 3291 } 3292 3293 #ifdef CONFIG_INET 3294 if (family == AF_INET && 3295 protocol != IPPROTO_RAW && 3296 !rcu_access_pointer(inet_protos[protocol])) 3297 return -ENOENT; 3298 #endif 3299 3300 return request_module("net-pf-%d-proto-%d-type-%d-%d", PF_NETLINK, 3301 NETLINK_SOCK_DIAG, family, protocol); 3302 } 3303 EXPORT_SYMBOL(sock_load_diag_module); 3304 3305 #ifdef CONFIG_PROC_FS 3306 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 3307 __acquires(proto_list_mutex) 3308 { 3309 mutex_lock(&proto_list_mutex); 3310 return seq_list_start_head(&proto_list, *pos); 3311 } 3312 3313 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 3314 { 3315 return seq_list_next(v, &proto_list, pos); 3316 } 3317 3318 static void proto_seq_stop(struct seq_file *seq, void *v) 3319 __releases(proto_list_mutex) 3320 { 3321 mutex_unlock(&proto_list_mutex); 3322 } 3323 3324 static char proto_method_implemented(const void *method) 3325 { 3326 return method == NULL ? 'n' : 'y'; 3327 } 3328 static long sock_prot_memory_allocated(struct proto *proto) 3329 { 3330 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 3331 } 3332 3333 static char *sock_prot_memory_pressure(struct proto *proto) 3334 { 3335 return proto->memory_pressure != NULL ? 3336 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 3337 } 3338 3339 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 3340 { 3341 3342 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 3343 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 3344 proto->name, 3345 proto->obj_size, 3346 sock_prot_inuse_get(seq_file_net(seq), proto), 3347 sock_prot_memory_allocated(proto), 3348 sock_prot_memory_pressure(proto), 3349 proto->max_header, 3350 proto->slab == NULL ? "no" : "yes", 3351 module_name(proto->owner), 3352 proto_method_implemented(proto->close), 3353 proto_method_implemented(proto->connect), 3354 proto_method_implemented(proto->disconnect), 3355 proto_method_implemented(proto->accept), 3356 proto_method_implemented(proto->ioctl), 3357 proto_method_implemented(proto->init), 3358 proto_method_implemented(proto->destroy), 3359 proto_method_implemented(proto->shutdown), 3360 proto_method_implemented(proto->setsockopt), 3361 proto_method_implemented(proto->getsockopt), 3362 proto_method_implemented(proto->sendmsg), 3363 proto_method_implemented(proto->recvmsg), 3364 proto_method_implemented(proto->sendpage), 3365 proto_method_implemented(proto->bind), 3366 proto_method_implemented(proto->backlog_rcv), 3367 proto_method_implemented(proto->hash), 3368 proto_method_implemented(proto->unhash), 3369 proto_method_implemented(proto->get_port), 3370 proto_method_implemented(proto->enter_memory_pressure)); 3371 } 3372 3373 static int proto_seq_show(struct seq_file *seq, void *v) 3374 { 3375 if (v == &proto_list) 3376 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 3377 "protocol", 3378 "size", 3379 "sockets", 3380 "memory", 3381 "press", 3382 "maxhdr", 3383 "slab", 3384 "module", 3385 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 3386 else 3387 proto_seq_printf(seq, list_entry(v, struct proto, node)); 3388 return 0; 3389 } 3390 3391 static const struct seq_operations proto_seq_ops = { 3392 .start = proto_seq_start, 3393 .next = proto_seq_next, 3394 .stop = proto_seq_stop, 3395 .show = proto_seq_show, 3396 }; 3397 3398 static __net_init int proto_init_net(struct net *net) 3399 { 3400 if (!proc_create_net("protocols", 0444, net->proc_net, &proto_seq_ops, 3401 sizeof(struct seq_net_private))) 3402 return -ENOMEM; 3403 3404 return 0; 3405 } 3406 3407 static __net_exit void proto_exit_net(struct net *net) 3408 { 3409 remove_proc_entry("protocols", net->proc_net); 3410 } 3411 3412 3413 static __net_initdata struct pernet_operations proto_net_ops = { 3414 .init = proto_init_net, 3415 .exit = proto_exit_net, 3416 }; 3417 3418 static int __init proto_init(void) 3419 { 3420 return register_pernet_subsys(&proto_net_ops); 3421 } 3422 3423 subsys_initcall(proto_init); 3424 3425 #endif /* PROC_FS */ 3426 3427 #ifdef CONFIG_NET_RX_BUSY_POLL 3428 bool sk_busy_loop_end(void *p, unsigned long start_time) 3429 { 3430 struct sock *sk = p; 3431 3432 return !skb_queue_empty(&sk->sk_receive_queue) || 3433 sk_busy_loop_timeout(sk, start_time); 3434 } 3435 EXPORT_SYMBOL(sk_busy_loop_end); 3436 #endif /* CONFIG_NET_RX_BUSY_POLL */ 3437