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 <linux/capability.h> 95 #include <linux/errno.h> 96 #include <linux/types.h> 97 #include <linux/socket.h> 98 #include <linux/in.h> 99 #include <linux/kernel.h> 100 #include <linux/module.h> 101 #include <linux/proc_fs.h> 102 #include <linux/seq_file.h> 103 #include <linux/sched.h> 104 #include <linux/timer.h> 105 #include <linux/string.h> 106 #include <linux/sockios.h> 107 #include <linux/net.h> 108 #include <linux/mm.h> 109 #include <linux/slab.h> 110 #include <linux/interrupt.h> 111 #include <linux/poll.h> 112 #include <linux/tcp.h> 113 #include <linux/init.h> 114 #include <linux/highmem.h> 115 #include <linux/user_namespace.h> 116 #include <linux/static_key.h> 117 #include <linux/memcontrol.h> 118 #include <linux/prefetch.h> 119 120 #include <asm/uaccess.h> 121 122 #include <linux/netdevice.h> 123 #include <net/protocol.h> 124 #include <linux/skbuff.h> 125 #include <net/net_namespace.h> 126 #include <net/request_sock.h> 127 #include <net/sock.h> 128 #include <linux/net_tstamp.h> 129 #include <net/xfrm.h> 130 #include <linux/ipsec.h> 131 #include <net/cls_cgroup.h> 132 #include <net/netprio_cgroup.h> 133 134 #include <linux/filter.h> 135 136 #include <trace/events/sock.h> 137 138 #ifdef CONFIG_INET 139 #include <net/tcp.h> 140 #endif 141 142 static DEFINE_MUTEX(proto_list_mutex); 143 static LIST_HEAD(proto_list); 144 145 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM 146 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 147 { 148 struct proto *proto; 149 int ret = 0; 150 151 mutex_lock(&proto_list_mutex); 152 list_for_each_entry(proto, &proto_list, node) { 153 if (proto->init_cgroup) { 154 ret = proto->init_cgroup(memcg, ss); 155 if (ret) 156 goto out; 157 } 158 } 159 160 mutex_unlock(&proto_list_mutex); 161 return ret; 162 out: 163 list_for_each_entry_continue_reverse(proto, &proto_list, node) 164 if (proto->destroy_cgroup) 165 proto->destroy_cgroup(memcg); 166 mutex_unlock(&proto_list_mutex); 167 return ret; 168 } 169 170 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 171 { 172 struct proto *proto; 173 174 mutex_lock(&proto_list_mutex); 175 list_for_each_entry_reverse(proto, &proto_list, node) 176 if (proto->destroy_cgroup) 177 proto->destroy_cgroup(memcg); 178 mutex_unlock(&proto_list_mutex); 179 } 180 #endif 181 182 /* 183 * Each address family might have different locking rules, so we have 184 * one slock key per address family: 185 */ 186 static struct lock_class_key af_family_keys[AF_MAX]; 187 static struct lock_class_key af_family_slock_keys[AF_MAX]; 188 189 struct static_key memcg_socket_limit_enabled; 190 EXPORT_SYMBOL(memcg_socket_limit_enabled); 191 192 /* 193 * Make lock validator output more readable. (we pre-construct these 194 * strings build-time, so that runtime initialization of socket 195 * locks is fast): 196 */ 197 static const char *const af_family_key_strings[AF_MAX+1] = { 198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" , 199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK", 200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" , 201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" , 202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" , 203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" , 204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" , 205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" , 206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" , 207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" , 208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" , 209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" , 210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" , 211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX" 212 }; 213 static const char *const af_family_slock_key_strings[AF_MAX+1] = { 214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" , 215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK", 216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" , 217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" , 218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" , 219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" , 220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" , 221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" , 222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" , 223 "slock-27" , "slock-28" , "slock-AF_CAN" , 224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" , 225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" , 226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" , 227 "slock-AF_NFC" , "slock-AF_MAX" 228 }; 229 static const char *const af_family_clock_key_strings[AF_MAX+1] = { 230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" , 231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK", 232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" , 233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" , 234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" , 235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" , 236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" , 237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" , 238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" , 239 "clock-27" , "clock-28" , "clock-AF_CAN" , 240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" , 241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" , 242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" , 243 "clock-AF_NFC" , "clock-AF_MAX" 244 }; 245 246 /* 247 * sk_callback_lock locking rules are per-address-family, 248 * so split the lock classes by using a per-AF key: 249 */ 250 static struct lock_class_key af_callback_keys[AF_MAX]; 251 252 /* Take into consideration the size of the struct sk_buff overhead in the 253 * determination of these values, since that is non-constant across 254 * platforms. This makes socket queueing behavior and performance 255 * not depend upon such differences. 256 */ 257 #define _SK_MEM_PACKETS 256 258 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 259 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 260 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 261 262 /* Run time adjustable parameters. */ 263 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX; 264 EXPORT_SYMBOL(sysctl_wmem_max); 265 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX; 266 EXPORT_SYMBOL(sysctl_rmem_max); 267 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX; 268 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX; 269 270 /* Maximal space eaten by iovec or ancillary data plus some space */ 271 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512); 272 EXPORT_SYMBOL(sysctl_optmem_max); 273 274 #if defined(CONFIG_CGROUPS) 275 #if !defined(CONFIG_NET_CLS_CGROUP) 276 int net_cls_subsys_id = -1; 277 EXPORT_SYMBOL_GPL(net_cls_subsys_id); 278 #endif 279 #if !defined(CONFIG_NETPRIO_CGROUP) 280 int net_prio_subsys_id = -1; 281 EXPORT_SYMBOL_GPL(net_prio_subsys_id); 282 #endif 283 #endif 284 285 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen) 286 { 287 struct timeval tv; 288 289 if (optlen < sizeof(tv)) 290 return -EINVAL; 291 if (copy_from_user(&tv, optval, sizeof(tv))) 292 return -EFAULT; 293 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC) 294 return -EDOM; 295 296 if (tv.tv_sec < 0) { 297 static int warned __read_mostly; 298 299 *timeo_p = 0; 300 if (warned < 10 && net_ratelimit()) { 301 warned++; 302 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n", 303 __func__, current->comm, task_pid_nr(current)); 304 } 305 return 0; 306 } 307 *timeo_p = MAX_SCHEDULE_TIMEOUT; 308 if (tv.tv_sec == 0 && tv.tv_usec == 0) 309 return 0; 310 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1)) 311 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ); 312 return 0; 313 } 314 315 static void sock_warn_obsolete_bsdism(const char *name) 316 { 317 static int warned; 318 static char warncomm[TASK_COMM_LEN]; 319 if (strcmp(warncomm, current->comm) && warned < 5) { 320 strcpy(warncomm, current->comm); 321 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n", 322 warncomm, name); 323 warned++; 324 } 325 } 326 327 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 328 329 static void sock_disable_timestamp(struct sock *sk, unsigned long flags) 330 { 331 if (sk->sk_flags & flags) { 332 sk->sk_flags &= ~flags; 333 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP)) 334 net_disable_timestamp(); 335 } 336 } 337 338 339 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 340 { 341 int err; 342 int skb_len; 343 unsigned long flags; 344 struct sk_buff_head *list = &sk->sk_receive_queue; 345 346 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) { 347 atomic_inc(&sk->sk_drops); 348 trace_sock_rcvqueue_full(sk, skb); 349 return -ENOMEM; 350 } 351 352 err = sk_filter(sk, skb); 353 if (err) 354 return err; 355 356 if (!sk_rmem_schedule(sk, skb->truesize)) { 357 atomic_inc(&sk->sk_drops); 358 return -ENOBUFS; 359 } 360 361 skb->dev = NULL; 362 skb_set_owner_r(skb, sk); 363 364 /* Cache the SKB length before we tack it onto the receive 365 * queue. Once it is added it no longer belongs to us and 366 * may be freed by other threads of control pulling packets 367 * from the queue. 368 */ 369 skb_len = skb->len; 370 371 /* we escape from rcu protected region, make sure we dont leak 372 * a norefcounted dst 373 */ 374 skb_dst_force(skb); 375 376 spin_lock_irqsave(&list->lock, flags); 377 skb->dropcount = atomic_read(&sk->sk_drops); 378 __skb_queue_tail(list, skb); 379 spin_unlock_irqrestore(&list->lock, flags); 380 381 if (!sock_flag(sk, SOCK_DEAD)) 382 sk->sk_data_ready(sk, skb_len); 383 return 0; 384 } 385 EXPORT_SYMBOL(sock_queue_rcv_skb); 386 387 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested) 388 { 389 int rc = NET_RX_SUCCESS; 390 391 if (sk_filter(sk, skb)) 392 goto discard_and_relse; 393 394 skb->dev = NULL; 395 396 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) { 397 atomic_inc(&sk->sk_drops); 398 goto discard_and_relse; 399 } 400 if (nested) 401 bh_lock_sock_nested(sk); 402 else 403 bh_lock_sock(sk); 404 if (!sock_owned_by_user(sk)) { 405 /* 406 * trylock + unlock semantics: 407 */ 408 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_); 409 410 rc = sk_backlog_rcv(sk, skb); 411 412 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 413 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) { 414 bh_unlock_sock(sk); 415 atomic_inc(&sk->sk_drops); 416 goto discard_and_relse; 417 } 418 419 bh_unlock_sock(sk); 420 out: 421 sock_put(sk); 422 return rc; 423 discard_and_relse: 424 kfree_skb(skb); 425 goto out; 426 } 427 EXPORT_SYMBOL(sk_receive_skb); 428 429 void sk_reset_txq(struct sock *sk) 430 { 431 sk_tx_queue_clear(sk); 432 } 433 EXPORT_SYMBOL(sk_reset_txq); 434 435 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie) 436 { 437 struct dst_entry *dst = __sk_dst_get(sk); 438 439 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 440 sk_tx_queue_clear(sk); 441 RCU_INIT_POINTER(sk->sk_dst_cache, NULL); 442 dst_release(dst); 443 return NULL; 444 } 445 446 return dst; 447 } 448 EXPORT_SYMBOL(__sk_dst_check); 449 450 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie) 451 { 452 struct dst_entry *dst = sk_dst_get(sk); 453 454 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) { 455 sk_dst_reset(sk); 456 dst_release(dst); 457 return NULL; 458 } 459 460 return dst; 461 } 462 EXPORT_SYMBOL(sk_dst_check); 463 464 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen) 465 { 466 int ret = -ENOPROTOOPT; 467 #ifdef CONFIG_NETDEVICES 468 struct net *net = sock_net(sk); 469 char devname[IFNAMSIZ]; 470 int index; 471 472 /* Sorry... */ 473 ret = -EPERM; 474 if (!capable(CAP_NET_RAW)) 475 goto out; 476 477 ret = -EINVAL; 478 if (optlen < 0) 479 goto out; 480 481 /* Bind this socket to a particular device like "eth0", 482 * as specified in the passed interface name. If the 483 * name is "" or the option length is zero the socket 484 * is not bound. 485 */ 486 if (optlen > IFNAMSIZ - 1) 487 optlen = IFNAMSIZ - 1; 488 memset(devname, 0, sizeof(devname)); 489 490 ret = -EFAULT; 491 if (copy_from_user(devname, optval, optlen)) 492 goto out; 493 494 index = 0; 495 if (devname[0] != '\0') { 496 struct net_device *dev; 497 498 rcu_read_lock(); 499 dev = dev_get_by_name_rcu(net, devname); 500 if (dev) 501 index = dev->ifindex; 502 rcu_read_unlock(); 503 ret = -ENODEV; 504 if (!dev) 505 goto out; 506 } 507 508 lock_sock(sk); 509 sk->sk_bound_dev_if = index; 510 sk_dst_reset(sk); 511 release_sock(sk); 512 513 ret = 0; 514 515 out: 516 #endif 517 518 return ret; 519 } 520 521 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool) 522 { 523 if (valbool) 524 sock_set_flag(sk, bit); 525 else 526 sock_reset_flag(sk, bit); 527 } 528 529 /* 530 * This is meant for all protocols to use and covers goings on 531 * at the socket level. Everything here is generic. 532 */ 533 534 int sock_setsockopt(struct socket *sock, int level, int optname, 535 char __user *optval, unsigned int optlen) 536 { 537 struct sock *sk = sock->sk; 538 int val; 539 int valbool; 540 struct linger ling; 541 int ret = 0; 542 543 /* 544 * Options without arguments 545 */ 546 547 if (optname == SO_BINDTODEVICE) 548 return sock_bindtodevice(sk, optval, optlen); 549 550 if (optlen < sizeof(int)) 551 return -EINVAL; 552 553 if (get_user(val, (int __user *)optval)) 554 return -EFAULT; 555 556 valbool = val ? 1 : 0; 557 558 lock_sock(sk); 559 560 switch (optname) { 561 case SO_DEBUG: 562 if (val && !capable(CAP_NET_ADMIN)) 563 ret = -EACCES; 564 else 565 sock_valbool_flag(sk, SOCK_DBG, valbool); 566 break; 567 case SO_REUSEADDR: 568 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE); 569 break; 570 case SO_TYPE: 571 case SO_PROTOCOL: 572 case SO_DOMAIN: 573 case SO_ERROR: 574 ret = -ENOPROTOOPT; 575 break; 576 case SO_DONTROUTE: 577 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool); 578 break; 579 case SO_BROADCAST: 580 sock_valbool_flag(sk, SOCK_BROADCAST, valbool); 581 break; 582 case SO_SNDBUF: 583 /* Don't error on this BSD doesn't and if you think 584 * about it this is right. Otherwise apps have to 585 * play 'guess the biggest size' games. RCVBUF/SNDBUF 586 * are treated in BSD as hints 587 */ 588 val = min_t(u32, val, sysctl_wmem_max); 589 set_sndbuf: 590 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 591 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF); 592 /* Wake up sending tasks if we upped the value. */ 593 sk->sk_write_space(sk); 594 break; 595 596 case SO_SNDBUFFORCE: 597 if (!capable(CAP_NET_ADMIN)) { 598 ret = -EPERM; 599 break; 600 } 601 goto set_sndbuf; 602 603 case SO_RCVBUF: 604 /* Don't error on this BSD doesn't and if you think 605 * about it this is right. Otherwise apps have to 606 * play 'guess the biggest size' games. RCVBUF/SNDBUF 607 * are treated in BSD as hints 608 */ 609 val = min_t(u32, val, sysctl_rmem_max); 610 set_rcvbuf: 611 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 612 /* 613 * We double it on the way in to account for 614 * "struct sk_buff" etc. overhead. Applications 615 * assume that the SO_RCVBUF setting they make will 616 * allow that much actual data to be received on that 617 * socket. 618 * 619 * Applications are unaware that "struct sk_buff" and 620 * other overheads allocate from the receive buffer 621 * during socket buffer allocation. 622 * 623 * And after considering the possible alternatives, 624 * returning the value we actually used in getsockopt 625 * is the most desirable behavior. 626 */ 627 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF); 628 break; 629 630 case SO_RCVBUFFORCE: 631 if (!capable(CAP_NET_ADMIN)) { 632 ret = -EPERM; 633 break; 634 } 635 goto set_rcvbuf; 636 637 case SO_KEEPALIVE: 638 #ifdef CONFIG_INET 639 if (sk->sk_protocol == IPPROTO_TCP) 640 tcp_set_keepalive(sk, valbool); 641 #endif 642 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 643 break; 644 645 case SO_OOBINLINE: 646 sock_valbool_flag(sk, SOCK_URGINLINE, valbool); 647 break; 648 649 case SO_NO_CHECK: 650 sk->sk_no_check = valbool; 651 break; 652 653 case SO_PRIORITY: 654 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN)) 655 sk->sk_priority = val; 656 else 657 ret = -EPERM; 658 break; 659 660 case SO_LINGER: 661 if (optlen < sizeof(ling)) { 662 ret = -EINVAL; /* 1003.1g */ 663 break; 664 } 665 if (copy_from_user(&ling, optval, sizeof(ling))) { 666 ret = -EFAULT; 667 break; 668 } 669 if (!ling.l_onoff) 670 sock_reset_flag(sk, SOCK_LINGER); 671 else { 672 #if (BITS_PER_LONG == 32) 673 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ) 674 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT; 675 else 676 #endif 677 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ; 678 sock_set_flag(sk, SOCK_LINGER); 679 } 680 break; 681 682 case SO_BSDCOMPAT: 683 sock_warn_obsolete_bsdism("setsockopt"); 684 break; 685 686 case SO_PASSCRED: 687 if (valbool) 688 set_bit(SOCK_PASSCRED, &sock->flags); 689 else 690 clear_bit(SOCK_PASSCRED, &sock->flags); 691 break; 692 693 case SO_TIMESTAMP: 694 case SO_TIMESTAMPNS: 695 if (valbool) { 696 if (optname == SO_TIMESTAMP) 697 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 698 else 699 sock_set_flag(sk, SOCK_RCVTSTAMPNS); 700 sock_set_flag(sk, SOCK_RCVTSTAMP); 701 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 702 } else { 703 sock_reset_flag(sk, SOCK_RCVTSTAMP); 704 sock_reset_flag(sk, SOCK_RCVTSTAMPNS); 705 } 706 break; 707 708 case SO_TIMESTAMPING: 709 if (val & ~SOF_TIMESTAMPING_MASK) { 710 ret = -EINVAL; 711 break; 712 } 713 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE, 714 val & SOF_TIMESTAMPING_TX_HARDWARE); 715 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE, 716 val & SOF_TIMESTAMPING_TX_SOFTWARE); 717 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE, 718 val & SOF_TIMESTAMPING_RX_HARDWARE); 719 if (val & SOF_TIMESTAMPING_RX_SOFTWARE) 720 sock_enable_timestamp(sk, 721 SOCK_TIMESTAMPING_RX_SOFTWARE); 722 else 723 sock_disable_timestamp(sk, 724 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)); 725 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE, 726 val & SOF_TIMESTAMPING_SOFTWARE); 727 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE, 728 val & SOF_TIMESTAMPING_SYS_HARDWARE); 729 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE, 730 val & SOF_TIMESTAMPING_RAW_HARDWARE); 731 break; 732 733 case SO_RCVLOWAT: 734 if (val < 0) 735 val = INT_MAX; 736 sk->sk_rcvlowat = val ? : 1; 737 break; 738 739 case SO_RCVTIMEO: 740 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen); 741 break; 742 743 case SO_SNDTIMEO: 744 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen); 745 break; 746 747 case SO_ATTACH_FILTER: 748 ret = -EINVAL; 749 if (optlen == sizeof(struct sock_fprog)) { 750 struct sock_fprog fprog; 751 752 ret = -EFAULT; 753 if (copy_from_user(&fprog, optval, sizeof(fprog))) 754 break; 755 756 ret = sk_attach_filter(&fprog, sk); 757 } 758 break; 759 760 case SO_DETACH_FILTER: 761 ret = sk_detach_filter(sk); 762 break; 763 764 case SO_PASSSEC: 765 if (valbool) 766 set_bit(SOCK_PASSSEC, &sock->flags); 767 else 768 clear_bit(SOCK_PASSSEC, &sock->flags); 769 break; 770 case SO_MARK: 771 if (!capable(CAP_NET_ADMIN)) 772 ret = -EPERM; 773 else 774 sk->sk_mark = val; 775 break; 776 777 /* We implement the SO_SNDLOWAT etc to 778 not be settable (1003.1g 5.3) */ 779 case SO_RXQ_OVFL: 780 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool); 781 break; 782 783 case SO_WIFI_STATUS: 784 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool); 785 break; 786 787 case SO_PEEK_OFF: 788 if (sock->ops->set_peek_off) 789 sock->ops->set_peek_off(sk, val); 790 else 791 ret = -EOPNOTSUPP; 792 break; 793 794 case SO_NOFCS: 795 sock_valbool_flag(sk, SOCK_NOFCS, valbool); 796 break; 797 798 default: 799 ret = -ENOPROTOOPT; 800 break; 801 } 802 release_sock(sk); 803 return ret; 804 } 805 EXPORT_SYMBOL(sock_setsockopt); 806 807 808 void cred_to_ucred(struct pid *pid, const struct cred *cred, 809 struct ucred *ucred) 810 { 811 ucred->pid = pid_vnr(pid); 812 ucred->uid = ucred->gid = -1; 813 if (cred) { 814 struct user_namespace *current_ns = current_user_ns(); 815 816 ucred->uid = from_kuid(current_ns, cred->euid); 817 ucred->gid = from_kgid(current_ns, cred->egid); 818 } 819 } 820 EXPORT_SYMBOL_GPL(cred_to_ucred); 821 822 int sock_getsockopt(struct socket *sock, int level, int optname, 823 char __user *optval, int __user *optlen) 824 { 825 struct sock *sk = sock->sk; 826 827 union { 828 int val; 829 struct linger ling; 830 struct timeval tm; 831 } v; 832 833 int lv = sizeof(int); 834 int len; 835 836 if (get_user(len, optlen)) 837 return -EFAULT; 838 if (len < 0) 839 return -EINVAL; 840 841 memset(&v, 0, sizeof(v)); 842 843 switch (optname) { 844 case SO_DEBUG: 845 v.val = sock_flag(sk, SOCK_DBG); 846 break; 847 848 case SO_DONTROUTE: 849 v.val = sock_flag(sk, SOCK_LOCALROUTE); 850 break; 851 852 case SO_BROADCAST: 853 v.val = sock_flag(sk, SOCK_BROADCAST); 854 break; 855 856 case SO_SNDBUF: 857 v.val = sk->sk_sndbuf; 858 break; 859 860 case SO_RCVBUF: 861 v.val = sk->sk_rcvbuf; 862 break; 863 864 case SO_REUSEADDR: 865 v.val = sk->sk_reuse; 866 break; 867 868 case SO_KEEPALIVE: 869 v.val = sock_flag(sk, SOCK_KEEPOPEN); 870 break; 871 872 case SO_TYPE: 873 v.val = sk->sk_type; 874 break; 875 876 case SO_PROTOCOL: 877 v.val = sk->sk_protocol; 878 break; 879 880 case SO_DOMAIN: 881 v.val = sk->sk_family; 882 break; 883 884 case SO_ERROR: 885 v.val = -sock_error(sk); 886 if (v.val == 0) 887 v.val = xchg(&sk->sk_err_soft, 0); 888 break; 889 890 case SO_OOBINLINE: 891 v.val = sock_flag(sk, SOCK_URGINLINE); 892 break; 893 894 case SO_NO_CHECK: 895 v.val = sk->sk_no_check; 896 break; 897 898 case SO_PRIORITY: 899 v.val = sk->sk_priority; 900 break; 901 902 case SO_LINGER: 903 lv = sizeof(v.ling); 904 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER); 905 v.ling.l_linger = sk->sk_lingertime / HZ; 906 break; 907 908 case SO_BSDCOMPAT: 909 sock_warn_obsolete_bsdism("getsockopt"); 910 break; 911 912 case SO_TIMESTAMP: 913 v.val = sock_flag(sk, SOCK_RCVTSTAMP) && 914 !sock_flag(sk, SOCK_RCVTSTAMPNS); 915 break; 916 917 case SO_TIMESTAMPNS: 918 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS); 919 break; 920 921 case SO_TIMESTAMPING: 922 v.val = 0; 923 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE)) 924 v.val |= SOF_TIMESTAMPING_TX_HARDWARE; 925 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE)) 926 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE; 927 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE)) 928 v.val |= SOF_TIMESTAMPING_RX_HARDWARE; 929 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) 930 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE; 931 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) 932 v.val |= SOF_TIMESTAMPING_SOFTWARE; 933 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)) 934 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE; 935 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) 936 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE; 937 break; 938 939 case SO_RCVTIMEO: 940 lv = sizeof(struct timeval); 941 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) { 942 v.tm.tv_sec = 0; 943 v.tm.tv_usec = 0; 944 } else { 945 v.tm.tv_sec = sk->sk_rcvtimeo / HZ; 946 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ; 947 } 948 break; 949 950 case SO_SNDTIMEO: 951 lv = sizeof(struct timeval); 952 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) { 953 v.tm.tv_sec = 0; 954 v.tm.tv_usec = 0; 955 } else { 956 v.tm.tv_sec = sk->sk_sndtimeo / HZ; 957 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ; 958 } 959 break; 960 961 case SO_RCVLOWAT: 962 v.val = sk->sk_rcvlowat; 963 break; 964 965 case SO_SNDLOWAT: 966 v.val = 1; 967 break; 968 969 case SO_PASSCRED: 970 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags); 971 break; 972 973 case SO_PEERCRED: 974 { 975 struct ucred peercred; 976 if (len > sizeof(peercred)) 977 len = sizeof(peercred); 978 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred); 979 if (copy_to_user(optval, &peercred, len)) 980 return -EFAULT; 981 goto lenout; 982 } 983 984 case SO_PEERNAME: 985 { 986 char address[128]; 987 988 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2)) 989 return -ENOTCONN; 990 if (lv < len) 991 return -EINVAL; 992 if (copy_to_user(optval, address, len)) 993 return -EFAULT; 994 goto lenout; 995 } 996 997 /* Dubious BSD thing... Probably nobody even uses it, but 998 * the UNIX standard wants it for whatever reason... -DaveM 999 */ 1000 case SO_ACCEPTCONN: 1001 v.val = sk->sk_state == TCP_LISTEN; 1002 break; 1003 1004 case SO_PASSSEC: 1005 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags); 1006 break; 1007 1008 case SO_PEERSEC: 1009 return security_socket_getpeersec_stream(sock, optval, optlen, len); 1010 1011 case SO_MARK: 1012 v.val = sk->sk_mark; 1013 break; 1014 1015 case SO_RXQ_OVFL: 1016 v.val = sock_flag(sk, SOCK_RXQ_OVFL); 1017 break; 1018 1019 case SO_WIFI_STATUS: 1020 v.val = sock_flag(sk, SOCK_WIFI_STATUS); 1021 break; 1022 1023 case SO_PEEK_OFF: 1024 if (!sock->ops->set_peek_off) 1025 return -EOPNOTSUPP; 1026 1027 v.val = sk->sk_peek_off; 1028 break; 1029 case SO_NOFCS: 1030 v.val = sock_flag(sk, SOCK_NOFCS); 1031 break; 1032 default: 1033 return -ENOPROTOOPT; 1034 } 1035 1036 if (len > lv) 1037 len = lv; 1038 if (copy_to_user(optval, &v, len)) 1039 return -EFAULT; 1040 lenout: 1041 if (put_user(len, optlen)) 1042 return -EFAULT; 1043 return 0; 1044 } 1045 1046 /* 1047 * Initialize an sk_lock. 1048 * 1049 * (We also register the sk_lock with the lock validator.) 1050 */ 1051 static inline void sock_lock_init(struct sock *sk) 1052 { 1053 sock_lock_init_class_and_name(sk, 1054 af_family_slock_key_strings[sk->sk_family], 1055 af_family_slock_keys + sk->sk_family, 1056 af_family_key_strings[sk->sk_family], 1057 af_family_keys + sk->sk_family); 1058 } 1059 1060 /* 1061 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet, 1062 * even temporarly, because of RCU lookups. sk_node should also be left as is. 1063 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end 1064 */ 1065 static void sock_copy(struct sock *nsk, const struct sock *osk) 1066 { 1067 #ifdef CONFIG_SECURITY_NETWORK 1068 void *sptr = nsk->sk_security; 1069 #endif 1070 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin)); 1071 1072 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end, 1073 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end)); 1074 1075 #ifdef CONFIG_SECURITY_NETWORK 1076 nsk->sk_security = sptr; 1077 security_sk_clone(osk, nsk); 1078 #endif 1079 } 1080 1081 /* 1082 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 1083 * un-modified. Special care is taken when initializing object to zero. 1084 */ 1085 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1086 { 1087 if (offsetof(struct sock, sk_node.next) != 0) 1088 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1089 memset(&sk->sk_node.pprev, 0, 1090 size - offsetof(struct sock, sk_node.pprev)); 1091 } 1092 1093 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size) 1094 { 1095 unsigned long nulls1, nulls2; 1096 1097 nulls1 = offsetof(struct sock, __sk_common.skc_node.next); 1098 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next); 1099 if (nulls1 > nulls2) 1100 swap(nulls1, nulls2); 1101 1102 if (nulls1 != 0) 1103 memset((char *)sk, 0, nulls1); 1104 memset((char *)sk + nulls1 + sizeof(void *), 0, 1105 nulls2 - nulls1 - sizeof(void *)); 1106 memset((char *)sk + nulls2 + sizeof(void *), 0, 1107 size - nulls2 - sizeof(void *)); 1108 } 1109 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls); 1110 1111 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority, 1112 int family) 1113 { 1114 struct sock *sk; 1115 struct kmem_cache *slab; 1116 1117 slab = prot->slab; 1118 if (slab != NULL) { 1119 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO); 1120 if (!sk) 1121 return sk; 1122 if (priority & __GFP_ZERO) { 1123 if (prot->clear_sk) 1124 prot->clear_sk(sk, prot->obj_size); 1125 else 1126 sk_prot_clear_nulls(sk, prot->obj_size); 1127 } 1128 } else 1129 sk = kmalloc(prot->obj_size, priority); 1130 1131 if (sk != NULL) { 1132 kmemcheck_annotate_bitfield(sk, flags); 1133 1134 if (security_sk_alloc(sk, family, priority)) 1135 goto out_free; 1136 1137 if (!try_module_get(prot->owner)) 1138 goto out_free_sec; 1139 sk_tx_queue_clear(sk); 1140 } 1141 1142 return sk; 1143 1144 out_free_sec: 1145 security_sk_free(sk); 1146 out_free: 1147 if (slab != NULL) 1148 kmem_cache_free(slab, sk); 1149 else 1150 kfree(sk); 1151 return NULL; 1152 } 1153 1154 static void sk_prot_free(struct proto *prot, struct sock *sk) 1155 { 1156 struct kmem_cache *slab; 1157 struct module *owner; 1158 1159 owner = prot->owner; 1160 slab = prot->slab; 1161 1162 security_sk_free(sk); 1163 if (slab != NULL) 1164 kmem_cache_free(slab, sk); 1165 else 1166 kfree(sk); 1167 module_put(owner); 1168 } 1169 1170 #ifdef CONFIG_CGROUPS 1171 void sock_update_classid(struct sock *sk) 1172 { 1173 u32 classid; 1174 1175 rcu_read_lock(); /* doing current task, which cannot vanish. */ 1176 classid = task_cls_classid(current); 1177 rcu_read_unlock(); 1178 if (classid && classid != sk->sk_classid) 1179 sk->sk_classid = classid; 1180 } 1181 EXPORT_SYMBOL(sock_update_classid); 1182 1183 void sock_update_netprioidx(struct sock *sk) 1184 { 1185 if (in_interrupt()) 1186 return; 1187 1188 sk->sk_cgrp_prioidx = task_netprioidx(current); 1189 } 1190 EXPORT_SYMBOL_GPL(sock_update_netprioidx); 1191 #endif 1192 1193 /** 1194 * sk_alloc - All socket objects are allocated here 1195 * @net: the applicable net namespace 1196 * @family: protocol family 1197 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1198 * @prot: struct proto associated with this new sock instance 1199 */ 1200 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1201 struct proto *prot) 1202 { 1203 struct sock *sk; 1204 1205 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family); 1206 if (sk) { 1207 sk->sk_family = family; 1208 /* 1209 * See comment in struct sock definition to understand 1210 * why we need sk_prot_creator -acme 1211 */ 1212 sk->sk_prot = sk->sk_prot_creator = prot; 1213 sock_lock_init(sk); 1214 sock_net_set(sk, get_net(net)); 1215 atomic_set(&sk->sk_wmem_alloc, 1); 1216 1217 sock_update_classid(sk); 1218 sock_update_netprioidx(sk); 1219 } 1220 1221 return sk; 1222 } 1223 EXPORT_SYMBOL(sk_alloc); 1224 1225 static void __sk_free(struct sock *sk) 1226 { 1227 struct sk_filter *filter; 1228 1229 if (sk->sk_destruct) 1230 sk->sk_destruct(sk); 1231 1232 filter = rcu_dereference_check(sk->sk_filter, 1233 atomic_read(&sk->sk_wmem_alloc) == 0); 1234 if (filter) { 1235 sk_filter_uncharge(sk, filter); 1236 RCU_INIT_POINTER(sk->sk_filter, NULL); 1237 } 1238 1239 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP); 1240 1241 if (atomic_read(&sk->sk_omem_alloc)) 1242 pr_debug("%s: optmem leakage (%d bytes) detected\n", 1243 __func__, atomic_read(&sk->sk_omem_alloc)); 1244 1245 if (sk->sk_peer_cred) 1246 put_cred(sk->sk_peer_cred); 1247 put_pid(sk->sk_peer_pid); 1248 put_net(sock_net(sk)); 1249 sk_prot_free(sk->sk_prot_creator, sk); 1250 } 1251 1252 void sk_free(struct sock *sk) 1253 { 1254 /* 1255 * We subtract one from sk_wmem_alloc and can know if 1256 * some packets are still in some tx queue. 1257 * If not null, sock_wfree() will call __sk_free(sk) later 1258 */ 1259 if (atomic_dec_and_test(&sk->sk_wmem_alloc)) 1260 __sk_free(sk); 1261 } 1262 EXPORT_SYMBOL(sk_free); 1263 1264 /* 1265 * Last sock_put should drop reference to sk->sk_net. It has already 1266 * been dropped in sk_change_net. Taking reference to stopping namespace 1267 * is not an option. 1268 * Take reference to a socket to remove it from hash _alive_ and after that 1269 * destroy it in the context of init_net. 1270 */ 1271 void sk_release_kernel(struct sock *sk) 1272 { 1273 if (sk == NULL || sk->sk_socket == NULL) 1274 return; 1275 1276 sock_hold(sk); 1277 sock_release(sk->sk_socket); 1278 release_net(sock_net(sk)); 1279 sock_net_set(sk, get_net(&init_net)); 1280 sock_put(sk); 1281 } 1282 EXPORT_SYMBOL(sk_release_kernel); 1283 1284 static void sk_update_clone(const struct sock *sk, struct sock *newsk) 1285 { 1286 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1287 sock_update_memcg(newsk); 1288 } 1289 1290 /** 1291 * sk_clone_lock - clone a socket, and lock its clone 1292 * @sk: the socket to clone 1293 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) 1294 * 1295 * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) 1296 */ 1297 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1298 { 1299 struct sock *newsk; 1300 1301 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family); 1302 if (newsk != NULL) { 1303 struct sk_filter *filter; 1304 1305 sock_copy(newsk, sk); 1306 1307 /* SANITY */ 1308 get_net(sock_net(newsk)); 1309 sk_node_init(&newsk->sk_node); 1310 sock_lock_init(newsk); 1311 bh_lock_sock(newsk); 1312 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL; 1313 newsk->sk_backlog.len = 0; 1314 1315 atomic_set(&newsk->sk_rmem_alloc, 0); 1316 /* 1317 * sk_wmem_alloc set to one (see sk_free() and sock_wfree()) 1318 */ 1319 atomic_set(&newsk->sk_wmem_alloc, 1); 1320 atomic_set(&newsk->sk_omem_alloc, 0); 1321 skb_queue_head_init(&newsk->sk_receive_queue); 1322 skb_queue_head_init(&newsk->sk_write_queue); 1323 #ifdef CONFIG_NET_DMA 1324 skb_queue_head_init(&newsk->sk_async_wait_queue); 1325 #endif 1326 1327 spin_lock_init(&newsk->sk_dst_lock); 1328 rwlock_init(&newsk->sk_callback_lock); 1329 lockdep_set_class_and_name(&newsk->sk_callback_lock, 1330 af_callback_keys + newsk->sk_family, 1331 af_family_clock_key_strings[newsk->sk_family]); 1332 1333 newsk->sk_dst_cache = NULL; 1334 newsk->sk_wmem_queued = 0; 1335 newsk->sk_forward_alloc = 0; 1336 newsk->sk_send_head = NULL; 1337 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK; 1338 1339 sock_reset_flag(newsk, SOCK_DONE); 1340 skb_queue_head_init(&newsk->sk_error_queue); 1341 1342 filter = rcu_dereference_protected(newsk->sk_filter, 1); 1343 if (filter != NULL) 1344 sk_filter_charge(newsk, filter); 1345 1346 if (unlikely(xfrm_sk_clone_policy(newsk))) { 1347 /* It is still raw copy of parent, so invalidate 1348 * destructor and make plain sk_free() */ 1349 newsk->sk_destruct = NULL; 1350 bh_unlock_sock(newsk); 1351 sk_free(newsk); 1352 newsk = NULL; 1353 goto out; 1354 } 1355 1356 newsk->sk_err = 0; 1357 newsk->sk_priority = 0; 1358 /* 1359 * Before updating sk_refcnt, we must commit prior changes to memory 1360 * (Documentation/RCU/rculist_nulls.txt for details) 1361 */ 1362 smp_wmb(); 1363 atomic_set(&newsk->sk_refcnt, 2); 1364 1365 /* 1366 * Increment the counter in the same struct proto as the master 1367 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that 1368 * is the same as sk->sk_prot->socks, as this field was copied 1369 * with memcpy). 1370 * 1371 * This _changes_ the previous behaviour, where 1372 * tcp_create_openreq_child always was incrementing the 1373 * equivalent to tcp_prot->socks (inet_sock_nr), so this have 1374 * to be taken into account in all callers. -acme 1375 */ 1376 sk_refcnt_debug_inc(newsk); 1377 sk_set_socket(newsk, NULL); 1378 newsk->sk_wq = NULL; 1379 1380 sk_update_clone(sk, newsk); 1381 1382 if (newsk->sk_prot->sockets_allocated) 1383 sk_sockets_allocated_inc(newsk); 1384 1385 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP) 1386 net_enable_timestamp(); 1387 } 1388 out: 1389 return newsk; 1390 } 1391 EXPORT_SYMBOL_GPL(sk_clone_lock); 1392 1393 void sk_setup_caps(struct sock *sk, struct dst_entry *dst) 1394 { 1395 __sk_dst_set(sk, dst); 1396 sk->sk_route_caps = dst->dev->features; 1397 if (sk->sk_route_caps & NETIF_F_GSO) 1398 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE; 1399 sk->sk_route_caps &= ~sk->sk_route_nocaps; 1400 if (sk_can_gso(sk)) { 1401 if (dst->header_len) { 1402 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 1403 } else { 1404 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM; 1405 sk->sk_gso_max_size = dst->dev->gso_max_size; 1406 } 1407 } 1408 } 1409 EXPORT_SYMBOL_GPL(sk_setup_caps); 1410 1411 void __init sk_init(void) 1412 { 1413 if (totalram_pages <= 4096) { 1414 sysctl_wmem_max = 32767; 1415 sysctl_rmem_max = 32767; 1416 sysctl_wmem_default = 32767; 1417 sysctl_rmem_default = 32767; 1418 } else if (totalram_pages >= 131072) { 1419 sysctl_wmem_max = 131071; 1420 sysctl_rmem_max = 131071; 1421 } 1422 } 1423 1424 /* 1425 * Simple resource managers for sockets. 1426 */ 1427 1428 1429 /* 1430 * Write buffer destructor automatically called from kfree_skb. 1431 */ 1432 void sock_wfree(struct sk_buff *skb) 1433 { 1434 struct sock *sk = skb->sk; 1435 unsigned int len = skb->truesize; 1436 1437 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) { 1438 /* 1439 * Keep a reference on sk_wmem_alloc, this will be released 1440 * after sk_write_space() call 1441 */ 1442 atomic_sub(len - 1, &sk->sk_wmem_alloc); 1443 sk->sk_write_space(sk); 1444 len = 1; 1445 } 1446 /* 1447 * if sk_wmem_alloc reaches 0, we must finish what sk_free() 1448 * could not do because of in-flight packets 1449 */ 1450 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc)) 1451 __sk_free(sk); 1452 } 1453 EXPORT_SYMBOL(sock_wfree); 1454 1455 /* 1456 * Read buffer destructor automatically called from kfree_skb. 1457 */ 1458 void sock_rfree(struct sk_buff *skb) 1459 { 1460 struct sock *sk = skb->sk; 1461 unsigned int len = skb->truesize; 1462 1463 atomic_sub(len, &sk->sk_rmem_alloc); 1464 sk_mem_uncharge(sk, len); 1465 } 1466 EXPORT_SYMBOL(sock_rfree); 1467 1468 void sock_edemux(struct sk_buff *skb) 1469 { 1470 sock_put(skb->sk); 1471 } 1472 EXPORT_SYMBOL(sock_edemux); 1473 1474 int sock_i_uid(struct sock *sk) 1475 { 1476 int uid; 1477 1478 read_lock_bh(&sk->sk_callback_lock); 1479 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0; 1480 read_unlock_bh(&sk->sk_callback_lock); 1481 return uid; 1482 } 1483 EXPORT_SYMBOL(sock_i_uid); 1484 1485 unsigned long sock_i_ino(struct sock *sk) 1486 { 1487 unsigned long ino; 1488 1489 read_lock_bh(&sk->sk_callback_lock); 1490 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0; 1491 read_unlock_bh(&sk->sk_callback_lock); 1492 return ino; 1493 } 1494 EXPORT_SYMBOL(sock_i_ino); 1495 1496 /* 1497 * Allocate a skb from the socket's send buffer. 1498 */ 1499 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1500 gfp_t priority) 1501 { 1502 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1503 struct sk_buff *skb = alloc_skb(size, priority); 1504 if (skb) { 1505 skb_set_owner_w(skb, sk); 1506 return skb; 1507 } 1508 } 1509 return NULL; 1510 } 1511 EXPORT_SYMBOL(sock_wmalloc); 1512 1513 /* 1514 * Allocate a skb from the socket's receive buffer. 1515 */ 1516 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1517 gfp_t priority) 1518 { 1519 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) { 1520 struct sk_buff *skb = alloc_skb(size, priority); 1521 if (skb) { 1522 skb_set_owner_r(skb, sk); 1523 return skb; 1524 } 1525 } 1526 return NULL; 1527 } 1528 1529 /* 1530 * Allocate a memory block from the socket's option memory buffer. 1531 */ 1532 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority) 1533 { 1534 if ((unsigned int)size <= sysctl_optmem_max && 1535 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) { 1536 void *mem; 1537 /* First do the add, to avoid the race if kmalloc 1538 * might sleep. 1539 */ 1540 atomic_add(size, &sk->sk_omem_alloc); 1541 mem = kmalloc(size, priority); 1542 if (mem) 1543 return mem; 1544 atomic_sub(size, &sk->sk_omem_alloc); 1545 } 1546 return NULL; 1547 } 1548 EXPORT_SYMBOL(sock_kmalloc); 1549 1550 /* 1551 * Free an option memory block. 1552 */ 1553 void sock_kfree_s(struct sock *sk, void *mem, int size) 1554 { 1555 kfree(mem); 1556 atomic_sub(size, &sk->sk_omem_alloc); 1557 } 1558 EXPORT_SYMBOL(sock_kfree_s); 1559 1560 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock. 1561 I think, these locks should be removed for datagram sockets. 1562 */ 1563 static long sock_wait_for_wmem(struct sock *sk, long timeo) 1564 { 1565 DEFINE_WAIT(wait); 1566 1567 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1568 for (;;) { 1569 if (!timeo) 1570 break; 1571 if (signal_pending(current)) 1572 break; 1573 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1574 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1575 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) 1576 break; 1577 if (sk->sk_shutdown & SEND_SHUTDOWN) 1578 break; 1579 if (sk->sk_err) 1580 break; 1581 timeo = schedule_timeout(timeo); 1582 } 1583 finish_wait(sk_sleep(sk), &wait); 1584 return timeo; 1585 } 1586 1587 1588 /* 1589 * Generic send/receive buffer handlers 1590 */ 1591 1592 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1593 unsigned long data_len, int noblock, 1594 int *errcode) 1595 { 1596 struct sk_buff *skb; 1597 gfp_t gfp_mask; 1598 long timeo; 1599 int err; 1600 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; 1601 1602 err = -EMSGSIZE; 1603 if (npages > MAX_SKB_FRAGS) 1604 goto failure; 1605 1606 gfp_mask = sk->sk_allocation; 1607 if (gfp_mask & __GFP_WAIT) 1608 gfp_mask |= __GFP_REPEAT; 1609 1610 timeo = sock_sndtimeo(sk, noblock); 1611 while (1) { 1612 err = sock_error(sk); 1613 if (err != 0) 1614 goto failure; 1615 1616 err = -EPIPE; 1617 if (sk->sk_shutdown & SEND_SHUTDOWN) 1618 goto failure; 1619 1620 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) { 1621 skb = alloc_skb(header_len, gfp_mask); 1622 if (skb) { 1623 int i; 1624 1625 /* No pages, we're done... */ 1626 if (!data_len) 1627 break; 1628 1629 skb->truesize += data_len; 1630 skb_shinfo(skb)->nr_frags = npages; 1631 for (i = 0; i < npages; i++) { 1632 struct page *page; 1633 1634 page = alloc_pages(sk->sk_allocation, 0); 1635 if (!page) { 1636 err = -ENOBUFS; 1637 skb_shinfo(skb)->nr_frags = i; 1638 kfree_skb(skb); 1639 goto failure; 1640 } 1641 1642 __skb_fill_page_desc(skb, i, 1643 page, 0, 1644 (data_len >= PAGE_SIZE ? 1645 PAGE_SIZE : 1646 data_len)); 1647 data_len -= PAGE_SIZE; 1648 } 1649 1650 /* Full success... */ 1651 break; 1652 } 1653 err = -ENOBUFS; 1654 goto failure; 1655 } 1656 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags); 1657 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1658 err = -EAGAIN; 1659 if (!timeo) 1660 goto failure; 1661 if (signal_pending(current)) 1662 goto interrupted; 1663 timeo = sock_wait_for_wmem(sk, timeo); 1664 } 1665 1666 skb_set_owner_w(skb, sk); 1667 return skb; 1668 1669 interrupted: 1670 err = sock_intr_errno(timeo); 1671 failure: 1672 *errcode = err; 1673 return NULL; 1674 } 1675 EXPORT_SYMBOL(sock_alloc_send_pskb); 1676 1677 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1678 int noblock, int *errcode) 1679 { 1680 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode); 1681 } 1682 EXPORT_SYMBOL(sock_alloc_send_skb); 1683 1684 static void __lock_sock(struct sock *sk) 1685 __releases(&sk->sk_lock.slock) 1686 __acquires(&sk->sk_lock.slock) 1687 { 1688 DEFINE_WAIT(wait); 1689 1690 for (;;) { 1691 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait, 1692 TASK_UNINTERRUPTIBLE); 1693 spin_unlock_bh(&sk->sk_lock.slock); 1694 schedule(); 1695 spin_lock_bh(&sk->sk_lock.slock); 1696 if (!sock_owned_by_user(sk)) 1697 break; 1698 } 1699 finish_wait(&sk->sk_lock.wq, &wait); 1700 } 1701 1702 static void __release_sock(struct sock *sk) 1703 __releases(&sk->sk_lock.slock) 1704 __acquires(&sk->sk_lock.slock) 1705 { 1706 struct sk_buff *skb = sk->sk_backlog.head; 1707 1708 do { 1709 sk->sk_backlog.head = sk->sk_backlog.tail = NULL; 1710 bh_unlock_sock(sk); 1711 1712 do { 1713 struct sk_buff *next = skb->next; 1714 1715 prefetch(next); 1716 WARN_ON_ONCE(skb_dst_is_noref(skb)); 1717 skb->next = NULL; 1718 sk_backlog_rcv(sk, skb); 1719 1720 /* 1721 * We are in process context here with softirqs 1722 * disabled, use cond_resched_softirq() to preempt. 1723 * This is safe to do because we've taken the backlog 1724 * queue private: 1725 */ 1726 cond_resched_softirq(); 1727 1728 skb = next; 1729 } while (skb != NULL); 1730 1731 bh_lock_sock(sk); 1732 } while ((skb = sk->sk_backlog.head) != NULL); 1733 1734 /* 1735 * Doing the zeroing here guarantee we can not loop forever 1736 * while a wild producer attempts to flood us. 1737 */ 1738 sk->sk_backlog.len = 0; 1739 } 1740 1741 /** 1742 * sk_wait_data - wait for data to arrive at sk_receive_queue 1743 * @sk: sock to wait on 1744 * @timeo: for how long 1745 * 1746 * Now socket state including sk->sk_err is changed only under lock, 1747 * hence we may omit checks after joining wait queue. 1748 * We check receive queue before schedule() only as optimization; 1749 * it is very likely that release_sock() added new data. 1750 */ 1751 int sk_wait_data(struct sock *sk, long *timeo) 1752 { 1753 int rc; 1754 DEFINE_WAIT(wait); 1755 1756 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 1757 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1758 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue)); 1759 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags); 1760 finish_wait(sk_sleep(sk), &wait); 1761 return rc; 1762 } 1763 EXPORT_SYMBOL(sk_wait_data); 1764 1765 /** 1766 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated 1767 * @sk: socket 1768 * @size: memory size to allocate 1769 * @kind: allocation type 1770 * 1771 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means 1772 * rmem allocation. This function assumes that protocols which have 1773 * memory_pressure use sk_wmem_queued as write buffer accounting. 1774 */ 1775 int __sk_mem_schedule(struct sock *sk, int size, int kind) 1776 { 1777 struct proto *prot = sk->sk_prot; 1778 int amt = sk_mem_pages(size); 1779 long allocated; 1780 int parent_status = UNDER_LIMIT; 1781 1782 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM; 1783 1784 allocated = sk_memory_allocated_add(sk, amt, &parent_status); 1785 1786 /* Under limit. */ 1787 if (parent_status == UNDER_LIMIT && 1788 allocated <= sk_prot_mem_limits(sk, 0)) { 1789 sk_leave_memory_pressure(sk); 1790 return 1; 1791 } 1792 1793 /* Under pressure. (we or our parents) */ 1794 if ((parent_status > SOFT_LIMIT) || 1795 allocated > sk_prot_mem_limits(sk, 1)) 1796 sk_enter_memory_pressure(sk); 1797 1798 /* Over hard limit (we or our parents) */ 1799 if ((parent_status == OVER_LIMIT) || 1800 (allocated > sk_prot_mem_limits(sk, 2))) 1801 goto suppress_allocation; 1802 1803 /* guarantee minimum buffer size under pressure */ 1804 if (kind == SK_MEM_RECV) { 1805 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0]) 1806 return 1; 1807 1808 } else { /* SK_MEM_SEND */ 1809 if (sk->sk_type == SOCK_STREAM) { 1810 if (sk->sk_wmem_queued < prot->sysctl_wmem[0]) 1811 return 1; 1812 } else if (atomic_read(&sk->sk_wmem_alloc) < 1813 prot->sysctl_wmem[0]) 1814 return 1; 1815 } 1816 1817 if (sk_has_memory_pressure(sk)) { 1818 int alloc; 1819 1820 if (!sk_under_memory_pressure(sk)) 1821 return 1; 1822 alloc = sk_sockets_allocated_read_positive(sk); 1823 if (sk_prot_mem_limits(sk, 2) > alloc * 1824 sk_mem_pages(sk->sk_wmem_queued + 1825 atomic_read(&sk->sk_rmem_alloc) + 1826 sk->sk_forward_alloc)) 1827 return 1; 1828 } 1829 1830 suppress_allocation: 1831 1832 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) { 1833 sk_stream_moderate_sndbuf(sk); 1834 1835 /* Fail only if socket is _under_ its sndbuf. 1836 * In this case we cannot block, so that we have to fail. 1837 */ 1838 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf) 1839 return 1; 1840 } 1841 1842 trace_sock_exceed_buf_limit(sk, prot, allocated); 1843 1844 /* Alas. Undo changes. */ 1845 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM; 1846 1847 sk_memory_allocated_sub(sk, amt); 1848 1849 return 0; 1850 } 1851 EXPORT_SYMBOL(__sk_mem_schedule); 1852 1853 /** 1854 * __sk_reclaim - reclaim memory_allocated 1855 * @sk: socket 1856 */ 1857 void __sk_mem_reclaim(struct sock *sk) 1858 { 1859 sk_memory_allocated_sub(sk, 1860 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT); 1861 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1; 1862 1863 if (sk_under_memory_pressure(sk) && 1864 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0))) 1865 sk_leave_memory_pressure(sk); 1866 } 1867 EXPORT_SYMBOL(__sk_mem_reclaim); 1868 1869 1870 /* 1871 * Set of default routines for initialising struct proto_ops when 1872 * the protocol does not support a particular function. In certain 1873 * cases where it makes no sense for a protocol to have a "do nothing" 1874 * function, some default processing is provided. 1875 */ 1876 1877 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len) 1878 { 1879 return -EOPNOTSUPP; 1880 } 1881 EXPORT_SYMBOL(sock_no_bind); 1882 1883 int sock_no_connect(struct socket *sock, struct sockaddr *saddr, 1884 int len, int flags) 1885 { 1886 return -EOPNOTSUPP; 1887 } 1888 EXPORT_SYMBOL(sock_no_connect); 1889 1890 int sock_no_socketpair(struct socket *sock1, struct socket *sock2) 1891 { 1892 return -EOPNOTSUPP; 1893 } 1894 EXPORT_SYMBOL(sock_no_socketpair); 1895 1896 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags) 1897 { 1898 return -EOPNOTSUPP; 1899 } 1900 EXPORT_SYMBOL(sock_no_accept); 1901 1902 int sock_no_getname(struct socket *sock, struct sockaddr *saddr, 1903 int *len, int peer) 1904 { 1905 return -EOPNOTSUPP; 1906 } 1907 EXPORT_SYMBOL(sock_no_getname); 1908 1909 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt) 1910 { 1911 return 0; 1912 } 1913 EXPORT_SYMBOL(sock_no_poll); 1914 1915 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 1916 { 1917 return -EOPNOTSUPP; 1918 } 1919 EXPORT_SYMBOL(sock_no_ioctl); 1920 1921 int sock_no_listen(struct socket *sock, int backlog) 1922 { 1923 return -EOPNOTSUPP; 1924 } 1925 EXPORT_SYMBOL(sock_no_listen); 1926 1927 int sock_no_shutdown(struct socket *sock, int how) 1928 { 1929 return -EOPNOTSUPP; 1930 } 1931 EXPORT_SYMBOL(sock_no_shutdown); 1932 1933 int sock_no_setsockopt(struct socket *sock, int level, int optname, 1934 char __user *optval, unsigned int optlen) 1935 { 1936 return -EOPNOTSUPP; 1937 } 1938 EXPORT_SYMBOL(sock_no_setsockopt); 1939 1940 int sock_no_getsockopt(struct socket *sock, int level, int optname, 1941 char __user *optval, int __user *optlen) 1942 { 1943 return -EOPNOTSUPP; 1944 } 1945 EXPORT_SYMBOL(sock_no_getsockopt); 1946 1947 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1948 size_t len) 1949 { 1950 return -EOPNOTSUPP; 1951 } 1952 EXPORT_SYMBOL(sock_no_sendmsg); 1953 1954 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m, 1955 size_t len, int flags) 1956 { 1957 return -EOPNOTSUPP; 1958 } 1959 EXPORT_SYMBOL(sock_no_recvmsg); 1960 1961 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma) 1962 { 1963 /* Mirror missing mmap method error code */ 1964 return -ENODEV; 1965 } 1966 EXPORT_SYMBOL(sock_no_mmap); 1967 1968 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags) 1969 { 1970 ssize_t res; 1971 struct msghdr msg = {.msg_flags = flags}; 1972 struct kvec iov; 1973 char *kaddr = kmap(page); 1974 iov.iov_base = kaddr + offset; 1975 iov.iov_len = size; 1976 res = kernel_sendmsg(sock, &msg, &iov, 1, size); 1977 kunmap(page); 1978 return res; 1979 } 1980 EXPORT_SYMBOL(sock_no_sendpage); 1981 1982 /* 1983 * Default Socket Callbacks 1984 */ 1985 1986 static void sock_def_wakeup(struct sock *sk) 1987 { 1988 struct socket_wq *wq; 1989 1990 rcu_read_lock(); 1991 wq = rcu_dereference(sk->sk_wq); 1992 if (wq_has_sleeper(wq)) 1993 wake_up_interruptible_all(&wq->wait); 1994 rcu_read_unlock(); 1995 } 1996 1997 static void sock_def_error_report(struct sock *sk) 1998 { 1999 struct socket_wq *wq; 2000 2001 rcu_read_lock(); 2002 wq = rcu_dereference(sk->sk_wq); 2003 if (wq_has_sleeper(wq)) 2004 wake_up_interruptible_poll(&wq->wait, POLLERR); 2005 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR); 2006 rcu_read_unlock(); 2007 } 2008 2009 static void sock_def_readable(struct sock *sk, int len) 2010 { 2011 struct socket_wq *wq; 2012 2013 rcu_read_lock(); 2014 wq = rcu_dereference(sk->sk_wq); 2015 if (wq_has_sleeper(wq)) 2016 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI | 2017 POLLRDNORM | POLLRDBAND); 2018 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN); 2019 rcu_read_unlock(); 2020 } 2021 2022 static void sock_def_write_space(struct sock *sk) 2023 { 2024 struct socket_wq *wq; 2025 2026 rcu_read_lock(); 2027 2028 /* Do not wake up a writer until he can make "significant" 2029 * progress. --DaveM 2030 */ 2031 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) { 2032 wq = rcu_dereference(sk->sk_wq); 2033 if (wq_has_sleeper(wq)) 2034 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT | 2035 POLLWRNORM | POLLWRBAND); 2036 2037 /* Should agree with poll, otherwise some programs break */ 2038 if (sock_writeable(sk)) 2039 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT); 2040 } 2041 2042 rcu_read_unlock(); 2043 } 2044 2045 static void sock_def_destruct(struct sock *sk) 2046 { 2047 kfree(sk->sk_protinfo); 2048 } 2049 2050 void sk_send_sigurg(struct sock *sk) 2051 { 2052 if (sk->sk_socket && sk->sk_socket->file) 2053 if (send_sigurg(&sk->sk_socket->file->f_owner)) 2054 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI); 2055 } 2056 EXPORT_SYMBOL(sk_send_sigurg); 2057 2058 void sk_reset_timer(struct sock *sk, struct timer_list* timer, 2059 unsigned long expires) 2060 { 2061 if (!mod_timer(timer, expires)) 2062 sock_hold(sk); 2063 } 2064 EXPORT_SYMBOL(sk_reset_timer); 2065 2066 void sk_stop_timer(struct sock *sk, struct timer_list* timer) 2067 { 2068 if (timer_pending(timer) && del_timer(timer)) 2069 __sock_put(sk); 2070 } 2071 EXPORT_SYMBOL(sk_stop_timer); 2072 2073 void sock_init_data(struct socket *sock, struct sock *sk) 2074 { 2075 skb_queue_head_init(&sk->sk_receive_queue); 2076 skb_queue_head_init(&sk->sk_write_queue); 2077 skb_queue_head_init(&sk->sk_error_queue); 2078 #ifdef CONFIG_NET_DMA 2079 skb_queue_head_init(&sk->sk_async_wait_queue); 2080 #endif 2081 2082 sk->sk_send_head = NULL; 2083 2084 init_timer(&sk->sk_timer); 2085 2086 sk->sk_allocation = GFP_KERNEL; 2087 sk->sk_rcvbuf = sysctl_rmem_default; 2088 sk->sk_sndbuf = sysctl_wmem_default; 2089 sk->sk_state = TCP_CLOSE; 2090 sk_set_socket(sk, sock); 2091 2092 sock_set_flag(sk, SOCK_ZAPPED); 2093 2094 if (sock) { 2095 sk->sk_type = sock->type; 2096 sk->sk_wq = sock->wq; 2097 sock->sk = sk; 2098 } else 2099 sk->sk_wq = NULL; 2100 2101 spin_lock_init(&sk->sk_dst_lock); 2102 rwlock_init(&sk->sk_callback_lock); 2103 lockdep_set_class_and_name(&sk->sk_callback_lock, 2104 af_callback_keys + sk->sk_family, 2105 af_family_clock_key_strings[sk->sk_family]); 2106 2107 sk->sk_state_change = sock_def_wakeup; 2108 sk->sk_data_ready = sock_def_readable; 2109 sk->sk_write_space = sock_def_write_space; 2110 sk->sk_error_report = sock_def_error_report; 2111 sk->sk_destruct = sock_def_destruct; 2112 2113 sk->sk_sndmsg_page = NULL; 2114 sk->sk_sndmsg_off = 0; 2115 sk->sk_peek_off = -1; 2116 2117 sk->sk_peer_pid = NULL; 2118 sk->sk_peer_cred = NULL; 2119 sk->sk_write_pending = 0; 2120 sk->sk_rcvlowat = 1; 2121 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT; 2122 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT; 2123 2124 sk->sk_stamp = ktime_set(-1L, 0); 2125 2126 /* 2127 * Before updating sk_refcnt, we must commit prior changes to memory 2128 * (Documentation/RCU/rculist_nulls.txt for details) 2129 */ 2130 smp_wmb(); 2131 atomic_set(&sk->sk_refcnt, 1); 2132 atomic_set(&sk->sk_drops, 0); 2133 } 2134 EXPORT_SYMBOL(sock_init_data); 2135 2136 void lock_sock_nested(struct sock *sk, int subclass) 2137 { 2138 might_sleep(); 2139 spin_lock_bh(&sk->sk_lock.slock); 2140 if (sk->sk_lock.owned) 2141 __lock_sock(sk); 2142 sk->sk_lock.owned = 1; 2143 spin_unlock(&sk->sk_lock.slock); 2144 /* 2145 * The sk_lock has mutex_lock() semantics here: 2146 */ 2147 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_); 2148 local_bh_enable(); 2149 } 2150 EXPORT_SYMBOL(lock_sock_nested); 2151 2152 void release_sock(struct sock *sk) 2153 { 2154 /* 2155 * The sk_lock has mutex_unlock() semantics: 2156 */ 2157 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 2158 2159 spin_lock_bh(&sk->sk_lock.slock); 2160 if (sk->sk_backlog.tail) 2161 __release_sock(sk); 2162 2163 if (sk->sk_prot->release_cb) 2164 sk->sk_prot->release_cb(sk); 2165 2166 sk->sk_lock.owned = 0; 2167 if (waitqueue_active(&sk->sk_lock.wq)) 2168 wake_up(&sk->sk_lock.wq); 2169 spin_unlock_bh(&sk->sk_lock.slock); 2170 } 2171 EXPORT_SYMBOL(release_sock); 2172 2173 /** 2174 * lock_sock_fast - fast version of lock_sock 2175 * @sk: socket 2176 * 2177 * This version should be used for very small section, where process wont block 2178 * return false if fast path is taken 2179 * sk_lock.slock locked, owned = 0, BH disabled 2180 * return true if slow path is taken 2181 * sk_lock.slock unlocked, owned = 1, BH enabled 2182 */ 2183 bool lock_sock_fast(struct sock *sk) 2184 { 2185 might_sleep(); 2186 spin_lock_bh(&sk->sk_lock.slock); 2187 2188 if (!sk->sk_lock.owned) 2189 /* 2190 * Note : We must disable BH 2191 */ 2192 return false; 2193 2194 __lock_sock(sk); 2195 sk->sk_lock.owned = 1; 2196 spin_unlock(&sk->sk_lock.slock); 2197 /* 2198 * The sk_lock has mutex_lock() semantics here: 2199 */ 2200 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 2201 local_bh_enable(); 2202 return true; 2203 } 2204 EXPORT_SYMBOL(lock_sock_fast); 2205 2206 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp) 2207 { 2208 struct timeval tv; 2209 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2210 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2211 tv = ktime_to_timeval(sk->sk_stamp); 2212 if (tv.tv_sec == -1) 2213 return -ENOENT; 2214 if (tv.tv_sec == 0) { 2215 sk->sk_stamp = ktime_get_real(); 2216 tv = ktime_to_timeval(sk->sk_stamp); 2217 } 2218 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0; 2219 } 2220 EXPORT_SYMBOL(sock_get_timestamp); 2221 2222 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp) 2223 { 2224 struct timespec ts; 2225 if (!sock_flag(sk, SOCK_TIMESTAMP)) 2226 sock_enable_timestamp(sk, SOCK_TIMESTAMP); 2227 ts = ktime_to_timespec(sk->sk_stamp); 2228 if (ts.tv_sec == -1) 2229 return -ENOENT; 2230 if (ts.tv_sec == 0) { 2231 sk->sk_stamp = ktime_get_real(); 2232 ts = ktime_to_timespec(sk->sk_stamp); 2233 } 2234 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0; 2235 } 2236 EXPORT_SYMBOL(sock_get_timestampns); 2237 2238 void sock_enable_timestamp(struct sock *sk, int flag) 2239 { 2240 if (!sock_flag(sk, flag)) { 2241 unsigned long previous_flags = sk->sk_flags; 2242 2243 sock_set_flag(sk, flag); 2244 /* 2245 * we just set one of the two flags which require net 2246 * time stamping, but time stamping might have been on 2247 * already because of the other one 2248 */ 2249 if (!(previous_flags & SK_FLAGS_TIMESTAMP)) 2250 net_enable_timestamp(); 2251 } 2252 } 2253 2254 /* 2255 * Get a socket option on an socket. 2256 * 2257 * FIX: POSIX 1003.1g is very ambiguous here. It states that 2258 * asynchronous errors should be reported by getsockopt. We assume 2259 * this means if you specify SO_ERROR (otherwise whats the point of it). 2260 */ 2261 int sock_common_getsockopt(struct socket *sock, int level, int optname, 2262 char __user *optval, int __user *optlen) 2263 { 2264 struct sock *sk = sock->sk; 2265 2266 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2267 } 2268 EXPORT_SYMBOL(sock_common_getsockopt); 2269 2270 #ifdef CONFIG_COMPAT 2271 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname, 2272 char __user *optval, int __user *optlen) 2273 { 2274 struct sock *sk = sock->sk; 2275 2276 if (sk->sk_prot->compat_getsockopt != NULL) 2277 return sk->sk_prot->compat_getsockopt(sk, level, optname, 2278 optval, optlen); 2279 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen); 2280 } 2281 EXPORT_SYMBOL(compat_sock_common_getsockopt); 2282 #endif 2283 2284 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 2285 struct msghdr *msg, size_t size, int flags) 2286 { 2287 struct sock *sk = sock->sk; 2288 int addr_len = 0; 2289 int err; 2290 2291 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT, 2292 flags & ~MSG_DONTWAIT, &addr_len); 2293 if (err >= 0) 2294 msg->msg_namelen = addr_len; 2295 return err; 2296 } 2297 EXPORT_SYMBOL(sock_common_recvmsg); 2298 2299 /* 2300 * Set socket options on an inet socket. 2301 */ 2302 int sock_common_setsockopt(struct socket *sock, int level, int optname, 2303 char __user *optval, unsigned int optlen) 2304 { 2305 struct sock *sk = sock->sk; 2306 2307 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2308 } 2309 EXPORT_SYMBOL(sock_common_setsockopt); 2310 2311 #ifdef CONFIG_COMPAT 2312 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname, 2313 char __user *optval, unsigned int optlen) 2314 { 2315 struct sock *sk = sock->sk; 2316 2317 if (sk->sk_prot->compat_setsockopt != NULL) 2318 return sk->sk_prot->compat_setsockopt(sk, level, optname, 2319 optval, optlen); 2320 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen); 2321 } 2322 EXPORT_SYMBOL(compat_sock_common_setsockopt); 2323 #endif 2324 2325 void sk_common_release(struct sock *sk) 2326 { 2327 if (sk->sk_prot->destroy) 2328 sk->sk_prot->destroy(sk); 2329 2330 /* 2331 * Observation: when sock_common_release is called, processes have 2332 * no access to socket. But net still has. 2333 * Step one, detach it from networking: 2334 * 2335 * A. Remove from hash tables. 2336 */ 2337 2338 sk->sk_prot->unhash(sk); 2339 2340 /* 2341 * In this point socket cannot receive new packets, but it is possible 2342 * that some packets are in flight because some CPU runs receiver and 2343 * did hash table lookup before we unhashed socket. They will achieve 2344 * receive queue and will be purged by socket destructor. 2345 * 2346 * Also we still have packets pending on receive queue and probably, 2347 * our own packets waiting in device queues. sock_destroy will drain 2348 * receive queue, but transmitted packets will delay socket destruction 2349 * until the last reference will be released. 2350 */ 2351 2352 sock_orphan(sk); 2353 2354 xfrm_sk_free_policy(sk); 2355 2356 sk_refcnt_debug_release(sk); 2357 sock_put(sk); 2358 } 2359 EXPORT_SYMBOL(sk_common_release); 2360 2361 #ifdef CONFIG_PROC_FS 2362 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 2363 struct prot_inuse { 2364 int val[PROTO_INUSE_NR]; 2365 }; 2366 2367 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR); 2368 2369 #ifdef CONFIG_NET_NS 2370 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2371 { 2372 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val); 2373 } 2374 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2375 2376 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2377 { 2378 int cpu, idx = prot->inuse_idx; 2379 int res = 0; 2380 2381 for_each_possible_cpu(cpu) 2382 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx]; 2383 2384 return res >= 0 ? res : 0; 2385 } 2386 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2387 2388 static int __net_init sock_inuse_init_net(struct net *net) 2389 { 2390 net->core.inuse = alloc_percpu(struct prot_inuse); 2391 return net->core.inuse ? 0 : -ENOMEM; 2392 } 2393 2394 static void __net_exit sock_inuse_exit_net(struct net *net) 2395 { 2396 free_percpu(net->core.inuse); 2397 } 2398 2399 static struct pernet_operations net_inuse_ops = { 2400 .init = sock_inuse_init_net, 2401 .exit = sock_inuse_exit_net, 2402 }; 2403 2404 static __init int net_inuse_init(void) 2405 { 2406 if (register_pernet_subsys(&net_inuse_ops)) 2407 panic("Cannot initialize net inuse counters"); 2408 2409 return 0; 2410 } 2411 2412 core_initcall(net_inuse_init); 2413 #else 2414 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse); 2415 2416 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val) 2417 { 2418 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val); 2419 } 2420 EXPORT_SYMBOL_GPL(sock_prot_inuse_add); 2421 2422 int sock_prot_inuse_get(struct net *net, struct proto *prot) 2423 { 2424 int cpu, idx = prot->inuse_idx; 2425 int res = 0; 2426 2427 for_each_possible_cpu(cpu) 2428 res += per_cpu(prot_inuse, cpu).val[idx]; 2429 2430 return res >= 0 ? res : 0; 2431 } 2432 EXPORT_SYMBOL_GPL(sock_prot_inuse_get); 2433 #endif 2434 2435 static void assign_proto_idx(struct proto *prot) 2436 { 2437 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR); 2438 2439 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) { 2440 pr_err("PROTO_INUSE_NR exhausted\n"); 2441 return; 2442 } 2443 2444 set_bit(prot->inuse_idx, proto_inuse_idx); 2445 } 2446 2447 static void release_proto_idx(struct proto *prot) 2448 { 2449 if (prot->inuse_idx != PROTO_INUSE_NR - 1) 2450 clear_bit(prot->inuse_idx, proto_inuse_idx); 2451 } 2452 #else 2453 static inline void assign_proto_idx(struct proto *prot) 2454 { 2455 } 2456 2457 static inline void release_proto_idx(struct proto *prot) 2458 { 2459 } 2460 #endif 2461 2462 int proto_register(struct proto *prot, int alloc_slab) 2463 { 2464 if (alloc_slab) { 2465 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0, 2466 SLAB_HWCACHE_ALIGN | prot->slab_flags, 2467 NULL); 2468 2469 if (prot->slab == NULL) { 2470 pr_crit("%s: Can't create sock SLAB cache!\n", 2471 prot->name); 2472 goto out; 2473 } 2474 2475 if (prot->rsk_prot != NULL) { 2476 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name); 2477 if (prot->rsk_prot->slab_name == NULL) 2478 goto out_free_sock_slab; 2479 2480 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name, 2481 prot->rsk_prot->obj_size, 0, 2482 SLAB_HWCACHE_ALIGN, NULL); 2483 2484 if (prot->rsk_prot->slab == NULL) { 2485 pr_crit("%s: Can't create request sock SLAB cache!\n", 2486 prot->name); 2487 goto out_free_request_sock_slab_name; 2488 } 2489 } 2490 2491 if (prot->twsk_prot != NULL) { 2492 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name); 2493 2494 if (prot->twsk_prot->twsk_slab_name == NULL) 2495 goto out_free_request_sock_slab; 2496 2497 prot->twsk_prot->twsk_slab = 2498 kmem_cache_create(prot->twsk_prot->twsk_slab_name, 2499 prot->twsk_prot->twsk_obj_size, 2500 0, 2501 SLAB_HWCACHE_ALIGN | 2502 prot->slab_flags, 2503 NULL); 2504 if (prot->twsk_prot->twsk_slab == NULL) 2505 goto out_free_timewait_sock_slab_name; 2506 } 2507 } 2508 2509 mutex_lock(&proto_list_mutex); 2510 list_add(&prot->node, &proto_list); 2511 assign_proto_idx(prot); 2512 mutex_unlock(&proto_list_mutex); 2513 return 0; 2514 2515 out_free_timewait_sock_slab_name: 2516 kfree(prot->twsk_prot->twsk_slab_name); 2517 out_free_request_sock_slab: 2518 if (prot->rsk_prot && prot->rsk_prot->slab) { 2519 kmem_cache_destroy(prot->rsk_prot->slab); 2520 prot->rsk_prot->slab = NULL; 2521 } 2522 out_free_request_sock_slab_name: 2523 if (prot->rsk_prot) 2524 kfree(prot->rsk_prot->slab_name); 2525 out_free_sock_slab: 2526 kmem_cache_destroy(prot->slab); 2527 prot->slab = NULL; 2528 out: 2529 return -ENOBUFS; 2530 } 2531 EXPORT_SYMBOL(proto_register); 2532 2533 void proto_unregister(struct proto *prot) 2534 { 2535 mutex_lock(&proto_list_mutex); 2536 release_proto_idx(prot); 2537 list_del(&prot->node); 2538 mutex_unlock(&proto_list_mutex); 2539 2540 if (prot->slab != NULL) { 2541 kmem_cache_destroy(prot->slab); 2542 prot->slab = NULL; 2543 } 2544 2545 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) { 2546 kmem_cache_destroy(prot->rsk_prot->slab); 2547 kfree(prot->rsk_prot->slab_name); 2548 prot->rsk_prot->slab = NULL; 2549 } 2550 2551 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) { 2552 kmem_cache_destroy(prot->twsk_prot->twsk_slab); 2553 kfree(prot->twsk_prot->twsk_slab_name); 2554 prot->twsk_prot->twsk_slab = NULL; 2555 } 2556 } 2557 EXPORT_SYMBOL(proto_unregister); 2558 2559 #ifdef CONFIG_PROC_FS 2560 static void *proto_seq_start(struct seq_file *seq, loff_t *pos) 2561 __acquires(proto_list_mutex) 2562 { 2563 mutex_lock(&proto_list_mutex); 2564 return seq_list_start_head(&proto_list, *pos); 2565 } 2566 2567 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos) 2568 { 2569 return seq_list_next(v, &proto_list, pos); 2570 } 2571 2572 static void proto_seq_stop(struct seq_file *seq, void *v) 2573 __releases(proto_list_mutex) 2574 { 2575 mutex_unlock(&proto_list_mutex); 2576 } 2577 2578 static char proto_method_implemented(const void *method) 2579 { 2580 return method == NULL ? 'n' : 'y'; 2581 } 2582 static long sock_prot_memory_allocated(struct proto *proto) 2583 { 2584 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L; 2585 } 2586 2587 static char *sock_prot_memory_pressure(struct proto *proto) 2588 { 2589 return proto->memory_pressure != NULL ? 2590 proto_memory_pressure(proto) ? "yes" : "no" : "NI"; 2591 } 2592 2593 static void proto_seq_printf(struct seq_file *seq, struct proto *proto) 2594 { 2595 2596 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s " 2597 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n", 2598 proto->name, 2599 proto->obj_size, 2600 sock_prot_inuse_get(seq_file_net(seq), proto), 2601 sock_prot_memory_allocated(proto), 2602 sock_prot_memory_pressure(proto), 2603 proto->max_header, 2604 proto->slab == NULL ? "no" : "yes", 2605 module_name(proto->owner), 2606 proto_method_implemented(proto->close), 2607 proto_method_implemented(proto->connect), 2608 proto_method_implemented(proto->disconnect), 2609 proto_method_implemented(proto->accept), 2610 proto_method_implemented(proto->ioctl), 2611 proto_method_implemented(proto->init), 2612 proto_method_implemented(proto->destroy), 2613 proto_method_implemented(proto->shutdown), 2614 proto_method_implemented(proto->setsockopt), 2615 proto_method_implemented(proto->getsockopt), 2616 proto_method_implemented(proto->sendmsg), 2617 proto_method_implemented(proto->recvmsg), 2618 proto_method_implemented(proto->sendpage), 2619 proto_method_implemented(proto->bind), 2620 proto_method_implemented(proto->backlog_rcv), 2621 proto_method_implemented(proto->hash), 2622 proto_method_implemented(proto->unhash), 2623 proto_method_implemented(proto->get_port), 2624 proto_method_implemented(proto->enter_memory_pressure)); 2625 } 2626 2627 static int proto_seq_show(struct seq_file *seq, void *v) 2628 { 2629 if (v == &proto_list) 2630 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s", 2631 "protocol", 2632 "size", 2633 "sockets", 2634 "memory", 2635 "press", 2636 "maxhdr", 2637 "slab", 2638 "module", 2639 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n"); 2640 else 2641 proto_seq_printf(seq, list_entry(v, struct proto, node)); 2642 return 0; 2643 } 2644 2645 static const struct seq_operations proto_seq_ops = { 2646 .start = proto_seq_start, 2647 .next = proto_seq_next, 2648 .stop = proto_seq_stop, 2649 .show = proto_seq_show, 2650 }; 2651 2652 static int proto_seq_open(struct inode *inode, struct file *file) 2653 { 2654 return seq_open_net(inode, file, &proto_seq_ops, 2655 sizeof(struct seq_net_private)); 2656 } 2657 2658 static const struct file_operations proto_seq_fops = { 2659 .owner = THIS_MODULE, 2660 .open = proto_seq_open, 2661 .read = seq_read, 2662 .llseek = seq_lseek, 2663 .release = seq_release_net, 2664 }; 2665 2666 static __net_init int proto_init_net(struct net *net) 2667 { 2668 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops)) 2669 return -ENOMEM; 2670 2671 return 0; 2672 } 2673 2674 static __net_exit void proto_exit_net(struct net *net) 2675 { 2676 proc_net_remove(net, "protocols"); 2677 } 2678 2679 2680 static __net_initdata struct pernet_operations proto_net_ops = { 2681 .init = proto_init_net, 2682 .exit = proto_exit_net, 2683 }; 2684 2685 static int __init proto_init(void) 2686 { 2687 return register_pernet_subsys(&proto_net_ops); 2688 } 2689 2690 subsys_initcall(proto_init); 2691 2692 #endif /* PROC_FS */ 2693