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 * Definitions for the AF_INET socket handler. 7 * 8 * Version: @(#)sock.h 1.0.4 05/13/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Corey Minyard <wf-rch!minyard@relay.EU.net> 13 * Florian La Roche <flla@stud.uni-sb.de> 14 * 15 * Fixes: 16 * Alan Cox : Volatiles in skbuff pointers. See 17 * skbuff comments. May be overdone, 18 * better to prove they can be removed 19 * than the reverse. 20 * Alan Cox : Added a zapped field for tcp to note 21 * a socket is reset and must stay shut up 22 * Alan Cox : New fields for options 23 * Pauline Middelink : identd support 24 * Alan Cox : Eliminate low level recv/recvfrom 25 * David S. Miller : New socket lookup architecture. 26 * Steve Whitehouse: Default routines for sock_ops 27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 28 * protinfo be just a void pointer, as the 29 * protocol specific parts were moved to 30 * respective headers and ipv4/v6, etc now 31 * use private slabcaches for its socks 32 * Pedro Hortas : New flags field for socket options 33 * 34 * 35 * This program is free software; you can redistribute it and/or 36 * modify it under the terms of the GNU General Public License 37 * as published by the Free Software Foundation; either version 38 * 2 of the License, or (at your option) any later version. 39 */ 40 #ifndef _SOCK_H 41 #define _SOCK_H 42 43 #include <linux/kernel.h> 44 #include <linux/list.h> 45 #include <linux/list_nulls.h> 46 #include <linux/timer.h> 47 #include <linux/cache.h> 48 #include <linux/module.h> 49 #include <linux/lockdep.h> 50 #include <linux/netdevice.h> 51 #include <linux/skbuff.h> /* struct sk_buff */ 52 #include <linux/mm.h> 53 #include <linux/security.h> 54 55 #include <linux/filter.h> 56 #include <linux/rculist_nulls.h> 57 #include <linux/poll.h> 58 59 #include <asm/atomic.h> 60 #include <net/dst.h> 61 #include <net/checksum.h> 62 63 /* 64 * This structure really needs to be cleaned up. 65 * Most of it is for TCP, and not used by any of 66 * the other protocols. 67 */ 68 69 /* Define this to get the SOCK_DBG debugging facility. */ 70 #define SOCK_DEBUGGING 71 #ifdef SOCK_DEBUGGING 72 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 73 printk(KERN_DEBUG msg); } while (0) 74 #else 75 /* Validate arguments and do nothing */ 76 static void inline int __attribute__ ((format (printf, 2, 3))) 77 SOCK_DEBUG(struct sock *sk, const char *msg, ...) 78 { 79 } 80 #endif 81 82 /* This is the per-socket lock. The spinlock provides a synchronization 83 * between user contexts and software interrupt processing, whereas the 84 * mini-semaphore synchronizes multiple users amongst themselves. 85 */ 86 typedef struct { 87 spinlock_t slock; 88 int owned; 89 wait_queue_head_t wq; 90 /* 91 * We express the mutex-alike socket_lock semantics 92 * to the lock validator by explicitly managing 93 * the slock as a lock variant (in addition to 94 * the slock itself): 95 */ 96 #ifdef CONFIG_DEBUG_LOCK_ALLOC 97 struct lockdep_map dep_map; 98 #endif 99 } socket_lock_t; 100 101 struct sock; 102 struct proto; 103 struct net; 104 105 /** 106 * struct sock_common - minimal network layer representation of sockets 107 * @skc_node: main hash linkage for various protocol lookup tables 108 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 109 * @skc_refcnt: reference count 110 * @skc_tx_queue_mapping: tx queue number for this connection 111 * @skc_hash: hash value used with various protocol lookup tables 112 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 113 * @skc_family: network address family 114 * @skc_state: Connection state 115 * @skc_reuse: %SO_REUSEADDR setting 116 * @skc_bound_dev_if: bound device index if != 0 117 * @skc_bind_node: bind hash linkage for various protocol lookup tables 118 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 119 * @skc_prot: protocol handlers inside a network family 120 * @skc_net: reference to the network namespace of this socket 121 * 122 * This is the minimal network layer representation of sockets, the header 123 * for struct sock and struct inet_timewait_sock. 124 */ 125 struct sock_common { 126 /* 127 * first fields are not copied in sock_copy() 128 */ 129 union { 130 struct hlist_node skc_node; 131 struct hlist_nulls_node skc_nulls_node; 132 }; 133 atomic_t skc_refcnt; 134 int skc_tx_queue_mapping; 135 136 union { 137 unsigned int skc_hash; 138 __u16 skc_u16hashes[2]; 139 }; 140 unsigned short skc_family; 141 volatile unsigned char skc_state; 142 unsigned char skc_reuse; 143 int skc_bound_dev_if; 144 union { 145 struct hlist_node skc_bind_node; 146 struct hlist_nulls_node skc_portaddr_node; 147 }; 148 struct proto *skc_prot; 149 #ifdef CONFIG_NET_NS 150 struct net *skc_net; 151 #endif 152 }; 153 154 /** 155 * struct sock - network layer representation of sockets 156 * @__sk_common: shared layout with inet_timewait_sock 157 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 158 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 159 * @sk_lock: synchronizer 160 * @sk_rcvbuf: size of receive buffer in bytes 161 * @sk_sleep: sock wait queue 162 * @sk_dst_cache: destination cache 163 * @sk_dst_lock: destination cache lock 164 * @sk_policy: flow policy 165 * @sk_rmem_alloc: receive queue bytes committed 166 * @sk_receive_queue: incoming packets 167 * @sk_wmem_alloc: transmit queue bytes committed 168 * @sk_write_queue: Packet sending queue 169 * @sk_async_wait_queue: DMA copied packets 170 * @sk_omem_alloc: "o" is "option" or "other" 171 * @sk_wmem_queued: persistent queue size 172 * @sk_forward_alloc: space allocated forward 173 * @sk_allocation: allocation mode 174 * @sk_sndbuf: size of send buffer in bytes 175 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 176 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 177 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets 178 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 179 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 180 * @sk_gso_max_size: Maximum GSO segment size to build 181 * @sk_lingertime: %SO_LINGER l_linger setting 182 * @sk_backlog: always used with the per-socket spinlock held 183 * @sk_callback_lock: used with the callbacks in the end of this struct 184 * @sk_error_queue: rarely used 185 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 186 * IPV6_ADDRFORM for instance) 187 * @sk_err: last error 188 * @sk_err_soft: errors that don't cause failure but are the cause of a 189 * persistent failure not just 'timed out' 190 * @sk_drops: raw/udp drops counter 191 * @sk_ack_backlog: current listen backlog 192 * @sk_max_ack_backlog: listen backlog set in listen() 193 * @sk_priority: %SO_PRIORITY setting 194 * @sk_type: socket type (%SOCK_STREAM, etc) 195 * @sk_protocol: which protocol this socket belongs in this network family 196 * @sk_peercred: %SO_PEERCRED setting 197 * @sk_rcvlowat: %SO_RCVLOWAT setting 198 * @sk_rcvtimeo: %SO_RCVTIMEO setting 199 * @sk_sndtimeo: %SO_SNDTIMEO setting 200 * @sk_filter: socket filtering instructions 201 * @sk_protinfo: private area, net family specific, when not using slab 202 * @sk_timer: sock cleanup timer 203 * @sk_stamp: time stamp of last packet received 204 * @sk_socket: Identd and reporting IO signals 205 * @sk_user_data: RPC layer private data 206 * @sk_sndmsg_page: cached page for sendmsg 207 * @sk_sndmsg_off: cached offset for sendmsg 208 * @sk_send_head: front of stuff to transmit 209 * @sk_security: used by security modules 210 * @sk_mark: generic packet mark 211 * @sk_write_pending: a write to stream socket waits to start 212 * @sk_state_change: callback to indicate change in the state of the sock 213 * @sk_data_ready: callback to indicate there is data to be processed 214 * @sk_write_space: callback to indicate there is bf sending space available 215 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 216 * @sk_backlog_rcv: callback to process the backlog 217 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 218 */ 219 struct sock { 220 /* 221 * Now struct inet_timewait_sock also uses sock_common, so please just 222 * don't add nothing before this first member (__sk_common) --acme 223 */ 224 struct sock_common __sk_common; 225 #define sk_node __sk_common.skc_node 226 #define sk_nulls_node __sk_common.skc_nulls_node 227 #define sk_refcnt __sk_common.skc_refcnt 228 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 229 230 #define sk_copy_start __sk_common.skc_hash 231 #define sk_hash __sk_common.skc_hash 232 #define sk_family __sk_common.skc_family 233 #define sk_state __sk_common.skc_state 234 #define sk_reuse __sk_common.skc_reuse 235 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 236 #define sk_bind_node __sk_common.skc_bind_node 237 #define sk_prot __sk_common.skc_prot 238 #define sk_net __sk_common.skc_net 239 kmemcheck_bitfield_begin(flags); 240 unsigned int sk_shutdown : 2, 241 sk_no_check : 2, 242 sk_userlocks : 4, 243 sk_protocol : 8, 244 sk_type : 16; 245 kmemcheck_bitfield_end(flags); 246 int sk_rcvbuf; 247 socket_lock_t sk_lock; 248 /* 249 * The backlog queue is special, it is always used with 250 * the per-socket spinlock held and requires low latency 251 * access. Therefore we special case it's implementation. 252 */ 253 struct { 254 struct sk_buff *head; 255 struct sk_buff *tail; 256 } sk_backlog; 257 wait_queue_head_t *sk_sleep; 258 struct dst_entry *sk_dst_cache; 259 #ifdef CONFIG_XFRM 260 struct xfrm_policy *sk_policy[2]; 261 #endif 262 rwlock_t sk_dst_lock; 263 atomic_t sk_rmem_alloc; 264 atomic_t sk_wmem_alloc; 265 atomic_t sk_omem_alloc; 266 int sk_sndbuf; 267 struct sk_buff_head sk_receive_queue; 268 struct sk_buff_head sk_write_queue; 269 #ifdef CONFIG_NET_DMA 270 struct sk_buff_head sk_async_wait_queue; 271 #endif 272 int sk_wmem_queued; 273 int sk_forward_alloc; 274 gfp_t sk_allocation; 275 int sk_route_caps; 276 int sk_gso_type; 277 unsigned int sk_gso_max_size; 278 int sk_rcvlowat; 279 unsigned long sk_flags; 280 unsigned long sk_lingertime; 281 struct sk_buff_head sk_error_queue; 282 struct proto *sk_prot_creator; 283 rwlock_t sk_callback_lock; 284 int sk_err, 285 sk_err_soft; 286 atomic_t sk_drops; 287 unsigned short sk_ack_backlog; 288 unsigned short sk_max_ack_backlog; 289 __u32 sk_priority; 290 struct ucred sk_peercred; 291 long sk_rcvtimeo; 292 long sk_sndtimeo; 293 struct sk_filter *sk_filter; 294 void *sk_protinfo; 295 struct timer_list sk_timer; 296 ktime_t sk_stamp; 297 struct socket *sk_socket; 298 void *sk_user_data; 299 struct page *sk_sndmsg_page; 300 struct sk_buff *sk_send_head; 301 __u32 sk_sndmsg_off; 302 int sk_write_pending; 303 #ifdef CONFIG_SECURITY 304 void *sk_security; 305 #endif 306 __u32 sk_mark; 307 /* XXX 4 bytes hole on 64 bit */ 308 void (*sk_state_change)(struct sock *sk); 309 void (*sk_data_ready)(struct sock *sk, int bytes); 310 void (*sk_write_space)(struct sock *sk); 311 void (*sk_error_report)(struct sock *sk); 312 int (*sk_backlog_rcv)(struct sock *sk, 313 struct sk_buff *skb); 314 void (*sk_destruct)(struct sock *sk); 315 }; 316 317 /* 318 * Hashed lists helper routines 319 */ 320 static inline struct sock *__sk_head(const struct hlist_head *head) 321 { 322 return hlist_entry(head->first, struct sock, sk_node); 323 } 324 325 static inline struct sock *sk_head(const struct hlist_head *head) 326 { 327 return hlist_empty(head) ? NULL : __sk_head(head); 328 } 329 330 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 331 { 332 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 333 } 334 335 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 336 { 337 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 338 } 339 340 static inline struct sock *sk_next(const struct sock *sk) 341 { 342 return sk->sk_node.next ? 343 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 344 } 345 346 static inline struct sock *sk_nulls_next(const struct sock *sk) 347 { 348 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 349 hlist_nulls_entry(sk->sk_nulls_node.next, 350 struct sock, sk_nulls_node) : 351 NULL; 352 } 353 354 static inline int sk_unhashed(const struct sock *sk) 355 { 356 return hlist_unhashed(&sk->sk_node); 357 } 358 359 static inline int sk_hashed(const struct sock *sk) 360 { 361 return !sk_unhashed(sk); 362 } 363 364 static __inline__ void sk_node_init(struct hlist_node *node) 365 { 366 node->pprev = NULL; 367 } 368 369 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node) 370 { 371 node->pprev = NULL; 372 } 373 374 static __inline__ void __sk_del_node(struct sock *sk) 375 { 376 __hlist_del(&sk->sk_node); 377 } 378 379 static __inline__ int __sk_del_node_init(struct sock *sk) 380 { 381 if (sk_hashed(sk)) { 382 __sk_del_node(sk); 383 sk_node_init(&sk->sk_node); 384 return 1; 385 } 386 return 0; 387 } 388 389 /* Grab socket reference count. This operation is valid only 390 when sk is ALREADY grabbed f.e. it is found in hash table 391 or a list and the lookup is made under lock preventing hash table 392 modifications. 393 */ 394 395 static inline void sock_hold(struct sock *sk) 396 { 397 atomic_inc(&sk->sk_refcnt); 398 } 399 400 /* Ungrab socket in the context, which assumes that socket refcnt 401 cannot hit zero, f.e. it is true in context of any socketcall. 402 */ 403 static inline void __sock_put(struct sock *sk) 404 { 405 atomic_dec(&sk->sk_refcnt); 406 } 407 408 static __inline__ int sk_del_node_init(struct sock *sk) 409 { 410 int rc = __sk_del_node_init(sk); 411 412 if (rc) { 413 /* paranoid for a while -acme */ 414 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 415 __sock_put(sk); 416 } 417 return rc; 418 } 419 420 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk) 421 { 422 if (sk_hashed(sk)) { 423 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 424 return 1; 425 } 426 return 0; 427 } 428 429 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk) 430 { 431 int rc = __sk_nulls_del_node_init_rcu(sk); 432 433 if (rc) { 434 /* paranoid for a while -acme */ 435 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 436 __sock_put(sk); 437 } 438 return rc; 439 } 440 441 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list) 442 { 443 hlist_add_head(&sk->sk_node, list); 444 } 445 446 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list) 447 { 448 sock_hold(sk); 449 __sk_add_node(sk, list); 450 } 451 452 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 453 { 454 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 455 } 456 457 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 458 { 459 sock_hold(sk); 460 __sk_nulls_add_node_rcu(sk, list); 461 } 462 463 static __inline__ void __sk_del_bind_node(struct sock *sk) 464 { 465 __hlist_del(&sk->sk_bind_node); 466 } 467 468 static __inline__ void sk_add_bind_node(struct sock *sk, 469 struct hlist_head *list) 470 { 471 hlist_add_head(&sk->sk_bind_node, list); 472 } 473 474 #define sk_for_each(__sk, node, list) \ 475 hlist_for_each_entry(__sk, node, list, sk_node) 476 #define sk_nulls_for_each(__sk, node, list) \ 477 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 478 #define sk_nulls_for_each_rcu(__sk, node, list) \ 479 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 480 #define sk_for_each_from(__sk, node) \ 481 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 482 hlist_for_each_entry_from(__sk, node, sk_node) 483 #define sk_nulls_for_each_from(__sk, node) \ 484 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 485 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 486 #define sk_for_each_continue(__sk, node) \ 487 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 488 hlist_for_each_entry_continue(__sk, node, sk_node) 489 #define sk_for_each_safe(__sk, node, tmp, list) \ 490 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node) 491 #define sk_for_each_bound(__sk, node, list) \ 492 hlist_for_each_entry(__sk, node, list, sk_bind_node) 493 494 /* Sock flags */ 495 enum sock_flags { 496 SOCK_DEAD, 497 SOCK_DONE, 498 SOCK_URGINLINE, 499 SOCK_KEEPOPEN, 500 SOCK_LINGER, 501 SOCK_DESTROY, 502 SOCK_BROADCAST, 503 SOCK_TIMESTAMP, 504 SOCK_ZAPPED, 505 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 506 SOCK_DBG, /* %SO_DEBUG setting */ 507 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 508 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 509 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 510 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 511 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 512 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 513 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 514 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 515 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 516 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 517 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 518 SOCK_FASYNC, /* fasync() active */ 519 SOCK_RXQ_OVFL, 520 }; 521 522 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 523 { 524 nsk->sk_flags = osk->sk_flags; 525 } 526 527 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 528 { 529 __set_bit(flag, &sk->sk_flags); 530 } 531 532 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 533 { 534 __clear_bit(flag, &sk->sk_flags); 535 } 536 537 static inline int sock_flag(struct sock *sk, enum sock_flags flag) 538 { 539 return test_bit(flag, &sk->sk_flags); 540 } 541 542 static inline void sk_acceptq_removed(struct sock *sk) 543 { 544 sk->sk_ack_backlog--; 545 } 546 547 static inline void sk_acceptq_added(struct sock *sk) 548 { 549 sk->sk_ack_backlog++; 550 } 551 552 static inline int sk_acceptq_is_full(struct sock *sk) 553 { 554 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 555 } 556 557 /* 558 * Compute minimal free write space needed to queue new packets. 559 */ 560 static inline int sk_stream_min_wspace(struct sock *sk) 561 { 562 return sk->sk_wmem_queued >> 1; 563 } 564 565 static inline int sk_stream_wspace(struct sock *sk) 566 { 567 return sk->sk_sndbuf - sk->sk_wmem_queued; 568 } 569 570 extern void sk_stream_write_space(struct sock *sk); 571 572 static inline int sk_stream_memory_free(struct sock *sk) 573 { 574 return sk->sk_wmem_queued < sk->sk_sndbuf; 575 } 576 577 /* The per-socket spinlock must be held here. */ 578 static inline void sk_add_backlog(struct sock *sk, struct sk_buff *skb) 579 { 580 if (!sk->sk_backlog.tail) { 581 sk->sk_backlog.head = sk->sk_backlog.tail = skb; 582 } else { 583 sk->sk_backlog.tail->next = skb; 584 sk->sk_backlog.tail = skb; 585 } 586 skb->next = NULL; 587 } 588 589 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 590 { 591 return sk->sk_backlog_rcv(sk, skb); 592 } 593 594 #define sk_wait_event(__sk, __timeo, __condition) \ 595 ({ int __rc; \ 596 release_sock(__sk); \ 597 __rc = __condition; \ 598 if (!__rc) { \ 599 *(__timeo) = schedule_timeout(*(__timeo)); \ 600 } \ 601 lock_sock(__sk); \ 602 __rc = __condition; \ 603 __rc; \ 604 }) 605 606 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 607 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 608 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 609 extern int sk_stream_error(struct sock *sk, int flags, int err); 610 extern void sk_stream_kill_queues(struct sock *sk); 611 612 extern int sk_wait_data(struct sock *sk, long *timeo); 613 614 struct request_sock_ops; 615 struct timewait_sock_ops; 616 struct inet_hashinfo; 617 struct raw_hashinfo; 618 619 /* Networking protocol blocks we attach to sockets. 620 * socket layer -> transport layer interface 621 * transport -> network interface is defined by struct inet_proto 622 */ 623 struct proto { 624 void (*close)(struct sock *sk, 625 long timeout); 626 int (*connect)(struct sock *sk, 627 struct sockaddr *uaddr, 628 int addr_len); 629 int (*disconnect)(struct sock *sk, int flags); 630 631 struct sock * (*accept) (struct sock *sk, int flags, int *err); 632 633 int (*ioctl)(struct sock *sk, int cmd, 634 unsigned long arg); 635 int (*init)(struct sock *sk); 636 void (*destroy)(struct sock *sk); 637 void (*shutdown)(struct sock *sk, int how); 638 int (*setsockopt)(struct sock *sk, int level, 639 int optname, char __user *optval, 640 unsigned int optlen); 641 int (*getsockopt)(struct sock *sk, int level, 642 int optname, char __user *optval, 643 int __user *option); 644 #ifdef CONFIG_COMPAT 645 int (*compat_setsockopt)(struct sock *sk, 646 int level, 647 int optname, char __user *optval, 648 unsigned int optlen); 649 int (*compat_getsockopt)(struct sock *sk, 650 int level, 651 int optname, char __user *optval, 652 int __user *option); 653 #endif 654 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 655 struct msghdr *msg, size_t len); 656 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 657 struct msghdr *msg, 658 size_t len, int noblock, int flags, 659 int *addr_len); 660 int (*sendpage)(struct sock *sk, struct page *page, 661 int offset, size_t size, int flags); 662 int (*bind)(struct sock *sk, 663 struct sockaddr *uaddr, int addr_len); 664 665 int (*backlog_rcv) (struct sock *sk, 666 struct sk_buff *skb); 667 668 /* Keeping track of sk's, looking them up, and port selection methods. */ 669 void (*hash)(struct sock *sk); 670 void (*unhash)(struct sock *sk); 671 int (*get_port)(struct sock *sk, unsigned short snum); 672 673 /* Keeping track of sockets in use */ 674 #ifdef CONFIG_PROC_FS 675 unsigned int inuse_idx; 676 #endif 677 678 /* Memory pressure */ 679 void (*enter_memory_pressure)(struct sock *sk); 680 atomic_t *memory_allocated; /* Current allocated memory. */ 681 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 682 /* 683 * Pressure flag: try to collapse. 684 * Technical note: it is used by multiple contexts non atomically. 685 * All the __sk_mem_schedule() is of this nature: accounting 686 * is strict, actions are advisory and have some latency. 687 */ 688 int *memory_pressure; 689 int *sysctl_mem; 690 int *sysctl_wmem; 691 int *sysctl_rmem; 692 int max_header; 693 694 struct kmem_cache *slab; 695 unsigned int obj_size; 696 int slab_flags; 697 698 struct percpu_counter *orphan_count; 699 700 struct request_sock_ops *rsk_prot; 701 struct timewait_sock_ops *twsk_prot; 702 703 union { 704 struct inet_hashinfo *hashinfo; 705 struct udp_table *udp_table; 706 struct raw_hashinfo *raw_hash; 707 } h; 708 709 struct module *owner; 710 711 char name[32]; 712 713 struct list_head node; 714 #ifdef SOCK_REFCNT_DEBUG 715 atomic_t socks; 716 #endif 717 }; 718 719 extern int proto_register(struct proto *prot, int alloc_slab); 720 extern void proto_unregister(struct proto *prot); 721 722 #ifdef SOCK_REFCNT_DEBUG 723 static inline void sk_refcnt_debug_inc(struct sock *sk) 724 { 725 atomic_inc(&sk->sk_prot->socks); 726 } 727 728 static inline void sk_refcnt_debug_dec(struct sock *sk) 729 { 730 atomic_dec(&sk->sk_prot->socks); 731 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 732 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 733 } 734 735 static inline void sk_refcnt_debug_release(const struct sock *sk) 736 { 737 if (atomic_read(&sk->sk_refcnt) != 1) 738 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 739 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 740 } 741 #else /* SOCK_REFCNT_DEBUG */ 742 #define sk_refcnt_debug_inc(sk) do { } while (0) 743 #define sk_refcnt_debug_dec(sk) do { } while (0) 744 #define sk_refcnt_debug_release(sk) do { } while (0) 745 #endif /* SOCK_REFCNT_DEBUG */ 746 747 748 #ifdef CONFIG_PROC_FS 749 /* Called with local bh disabled */ 750 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 751 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 752 #else 753 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot, 754 int inc) 755 { 756 } 757 #endif 758 759 760 /* With per-bucket locks this operation is not-atomic, so that 761 * this version is not worse. 762 */ 763 static inline void __sk_prot_rehash(struct sock *sk) 764 { 765 sk->sk_prot->unhash(sk); 766 sk->sk_prot->hash(sk); 767 } 768 769 /* About 10 seconds */ 770 #define SOCK_DESTROY_TIME (10*HZ) 771 772 /* Sockets 0-1023 can't be bound to unless you are superuser */ 773 #define PROT_SOCK 1024 774 775 #define SHUTDOWN_MASK 3 776 #define RCV_SHUTDOWN 1 777 #define SEND_SHUTDOWN 2 778 779 #define SOCK_SNDBUF_LOCK 1 780 #define SOCK_RCVBUF_LOCK 2 781 #define SOCK_BINDADDR_LOCK 4 782 #define SOCK_BINDPORT_LOCK 8 783 784 /* sock_iocb: used to kick off async processing of socket ios */ 785 struct sock_iocb { 786 struct list_head list; 787 788 int flags; 789 int size; 790 struct socket *sock; 791 struct sock *sk; 792 struct scm_cookie *scm; 793 struct msghdr *msg, async_msg; 794 struct kiocb *kiocb; 795 }; 796 797 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 798 { 799 return (struct sock_iocb *)iocb->private; 800 } 801 802 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 803 { 804 return si->kiocb; 805 } 806 807 struct socket_alloc { 808 struct socket socket; 809 struct inode vfs_inode; 810 }; 811 812 static inline struct socket *SOCKET_I(struct inode *inode) 813 { 814 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 815 } 816 817 static inline struct inode *SOCK_INODE(struct socket *socket) 818 { 819 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 820 } 821 822 /* 823 * Functions for memory accounting 824 */ 825 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 826 extern void __sk_mem_reclaim(struct sock *sk); 827 828 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 829 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 830 #define SK_MEM_SEND 0 831 #define SK_MEM_RECV 1 832 833 static inline int sk_mem_pages(int amt) 834 { 835 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 836 } 837 838 static inline int sk_has_account(struct sock *sk) 839 { 840 /* return true if protocol supports memory accounting */ 841 return !!sk->sk_prot->memory_allocated; 842 } 843 844 static inline int sk_wmem_schedule(struct sock *sk, int size) 845 { 846 if (!sk_has_account(sk)) 847 return 1; 848 return size <= sk->sk_forward_alloc || 849 __sk_mem_schedule(sk, size, SK_MEM_SEND); 850 } 851 852 static inline int sk_rmem_schedule(struct sock *sk, int size) 853 { 854 if (!sk_has_account(sk)) 855 return 1; 856 return size <= sk->sk_forward_alloc || 857 __sk_mem_schedule(sk, size, SK_MEM_RECV); 858 } 859 860 static inline void sk_mem_reclaim(struct sock *sk) 861 { 862 if (!sk_has_account(sk)) 863 return; 864 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 865 __sk_mem_reclaim(sk); 866 } 867 868 static inline void sk_mem_reclaim_partial(struct sock *sk) 869 { 870 if (!sk_has_account(sk)) 871 return; 872 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 873 __sk_mem_reclaim(sk); 874 } 875 876 static inline void sk_mem_charge(struct sock *sk, int size) 877 { 878 if (!sk_has_account(sk)) 879 return; 880 sk->sk_forward_alloc -= size; 881 } 882 883 static inline void sk_mem_uncharge(struct sock *sk, int size) 884 { 885 if (!sk_has_account(sk)) 886 return; 887 sk->sk_forward_alloc += size; 888 } 889 890 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 891 { 892 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 893 sk->sk_wmem_queued -= skb->truesize; 894 sk_mem_uncharge(sk, skb->truesize); 895 __kfree_skb(skb); 896 } 897 898 /* Used by processes to "lock" a socket state, so that 899 * interrupts and bottom half handlers won't change it 900 * from under us. It essentially blocks any incoming 901 * packets, so that we won't get any new data or any 902 * packets that change the state of the socket. 903 * 904 * While locked, BH processing will add new packets to 905 * the backlog queue. This queue is processed by the 906 * owner of the socket lock right before it is released. 907 * 908 * Since ~2.3.5 it is also exclusive sleep lock serializing 909 * accesses from user process context. 910 */ 911 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 912 913 /* 914 * Macro so as to not evaluate some arguments when 915 * lockdep is not enabled. 916 * 917 * Mark both the sk_lock and the sk_lock.slock as a 918 * per-address-family lock class. 919 */ 920 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 921 do { \ 922 sk->sk_lock.owned = 0; \ 923 init_waitqueue_head(&sk->sk_lock.wq); \ 924 spin_lock_init(&(sk)->sk_lock.slock); \ 925 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 926 sizeof((sk)->sk_lock)); \ 927 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 928 (skey), (sname)); \ 929 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 930 } while (0) 931 932 extern void lock_sock_nested(struct sock *sk, int subclass); 933 934 static inline void lock_sock(struct sock *sk) 935 { 936 lock_sock_nested(sk, 0); 937 } 938 939 extern void release_sock(struct sock *sk); 940 941 /* BH context may only use the following locking interface. */ 942 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 943 #define bh_lock_sock_nested(__sk) \ 944 spin_lock_nested(&((__sk)->sk_lock.slock), \ 945 SINGLE_DEPTH_NESTING) 946 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 947 948 extern struct sock *sk_alloc(struct net *net, int family, 949 gfp_t priority, 950 struct proto *prot); 951 extern void sk_free(struct sock *sk); 952 extern void sk_release_kernel(struct sock *sk); 953 extern struct sock *sk_clone(const struct sock *sk, 954 const gfp_t priority); 955 956 extern struct sk_buff *sock_wmalloc(struct sock *sk, 957 unsigned long size, int force, 958 gfp_t priority); 959 extern struct sk_buff *sock_rmalloc(struct sock *sk, 960 unsigned long size, int force, 961 gfp_t priority); 962 extern void sock_wfree(struct sk_buff *skb); 963 extern void sock_rfree(struct sk_buff *skb); 964 965 extern int sock_setsockopt(struct socket *sock, int level, 966 int op, char __user *optval, 967 unsigned int optlen); 968 969 extern int sock_getsockopt(struct socket *sock, int level, 970 int op, char __user *optval, 971 int __user *optlen); 972 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 973 unsigned long size, 974 int noblock, 975 int *errcode); 976 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 977 unsigned long header_len, 978 unsigned long data_len, 979 int noblock, 980 int *errcode); 981 extern void *sock_kmalloc(struct sock *sk, int size, 982 gfp_t priority); 983 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 984 extern void sk_send_sigurg(struct sock *sk); 985 986 /* 987 * Functions to fill in entries in struct proto_ops when a protocol 988 * does not implement a particular function. 989 */ 990 extern int sock_no_bind(struct socket *, 991 struct sockaddr *, int); 992 extern int sock_no_connect(struct socket *, 993 struct sockaddr *, int, int); 994 extern int sock_no_socketpair(struct socket *, 995 struct socket *); 996 extern int sock_no_accept(struct socket *, 997 struct socket *, int); 998 extern int sock_no_getname(struct socket *, 999 struct sockaddr *, int *, int); 1000 extern unsigned int sock_no_poll(struct file *, struct socket *, 1001 struct poll_table_struct *); 1002 extern int sock_no_ioctl(struct socket *, unsigned int, 1003 unsigned long); 1004 extern int sock_no_listen(struct socket *, int); 1005 extern int sock_no_shutdown(struct socket *, int); 1006 extern int sock_no_getsockopt(struct socket *, int , int, 1007 char __user *, int __user *); 1008 extern int sock_no_setsockopt(struct socket *, int, int, 1009 char __user *, unsigned int); 1010 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1011 struct msghdr *, size_t); 1012 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1013 struct msghdr *, size_t, int); 1014 extern int sock_no_mmap(struct file *file, 1015 struct socket *sock, 1016 struct vm_area_struct *vma); 1017 extern ssize_t sock_no_sendpage(struct socket *sock, 1018 struct page *page, 1019 int offset, size_t size, 1020 int flags); 1021 1022 /* 1023 * Functions to fill in entries in struct proto_ops when a protocol 1024 * uses the inet style. 1025 */ 1026 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1027 char __user *optval, int __user *optlen); 1028 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1029 struct msghdr *msg, size_t size, int flags); 1030 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1031 char __user *optval, unsigned int optlen); 1032 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1033 int optname, char __user *optval, int __user *optlen); 1034 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1035 int optname, char __user *optval, unsigned int optlen); 1036 1037 extern void sk_common_release(struct sock *sk); 1038 1039 /* 1040 * Default socket callbacks and setup code 1041 */ 1042 1043 /* Initialise core socket variables */ 1044 extern void sock_init_data(struct socket *sock, struct sock *sk); 1045 1046 /** 1047 * sk_filter_release: Release a socket filter 1048 * @fp: filter to remove 1049 * 1050 * Remove a filter from a socket and release its resources. 1051 */ 1052 1053 static inline void sk_filter_release(struct sk_filter *fp) 1054 { 1055 if (atomic_dec_and_test(&fp->refcnt)) 1056 kfree(fp); 1057 } 1058 1059 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1060 { 1061 unsigned int size = sk_filter_len(fp); 1062 1063 atomic_sub(size, &sk->sk_omem_alloc); 1064 sk_filter_release(fp); 1065 } 1066 1067 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1068 { 1069 atomic_inc(&fp->refcnt); 1070 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1071 } 1072 1073 /* 1074 * Socket reference counting postulates. 1075 * 1076 * * Each user of socket SHOULD hold a reference count. 1077 * * Each access point to socket (an hash table bucket, reference from a list, 1078 * running timer, skb in flight MUST hold a reference count. 1079 * * When reference count hits 0, it means it will never increase back. 1080 * * When reference count hits 0, it means that no references from 1081 * outside exist to this socket and current process on current CPU 1082 * is last user and may/should destroy this socket. 1083 * * sk_free is called from any context: process, BH, IRQ. When 1084 * it is called, socket has no references from outside -> sk_free 1085 * may release descendant resources allocated by the socket, but 1086 * to the time when it is called, socket is NOT referenced by any 1087 * hash tables, lists etc. 1088 * * Packets, delivered from outside (from network or from another process) 1089 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1090 * when they sit in queue. Otherwise, packets will leak to hole, when 1091 * socket is looked up by one cpu and unhasing is made by another CPU. 1092 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1093 * (leak to backlog). Packet socket does all the processing inside 1094 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1095 * use separate SMP lock, so that they are prone too. 1096 */ 1097 1098 /* Ungrab socket and destroy it, if it was the last reference. */ 1099 static inline void sock_put(struct sock *sk) 1100 { 1101 if (atomic_dec_and_test(&sk->sk_refcnt)) 1102 sk_free(sk); 1103 } 1104 1105 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1106 const int nested); 1107 1108 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1109 { 1110 sk->sk_tx_queue_mapping = tx_queue; 1111 } 1112 1113 static inline void sk_tx_queue_clear(struct sock *sk) 1114 { 1115 sk->sk_tx_queue_mapping = -1; 1116 } 1117 1118 static inline int sk_tx_queue_get(const struct sock *sk) 1119 { 1120 return sk->sk_tx_queue_mapping; 1121 } 1122 1123 static inline bool sk_tx_queue_recorded(const struct sock *sk) 1124 { 1125 return (sk && sk->sk_tx_queue_mapping >= 0); 1126 } 1127 1128 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1129 { 1130 sk_tx_queue_clear(sk); 1131 sk->sk_socket = sock; 1132 } 1133 1134 /* Detach socket from process context. 1135 * Announce socket dead, detach it from wait queue and inode. 1136 * Note that parent inode held reference count on this struct sock, 1137 * we do not release it in this function, because protocol 1138 * probably wants some additional cleanups or even continuing 1139 * to work with this socket (TCP). 1140 */ 1141 static inline void sock_orphan(struct sock *sk) 1142 { 1143 write_lock_bh(&sk->sk_callback_lock); 1144 sock_set_flag(sk, SOCK_DEAD); 1145 sk_set_socket(sk, NULL); 1146 sk->sk_sleep = NULL; 1147 write_unlock_bh(&sk->sk_callback_lock); 1148 } 1149 1150 static inline void sock_graft(struct sock *sk, struct socket *parent) 1151 { 1152 write_lock_bh(&sk->sk_callback_lock); 1153 sk->sk_sleep = &parent->wait; 1154 parent->sk = sk; 1155 sk_set_socket(sk, parent); 1156 security_sock_graft(sk, parent); 1157 write_unlock_bh(&sk->sk_callback_lock); 1158 } 1159 1160 extern int sock_i_uid(struct sock *sk); 1161 extern unsigned long sock_i_ino(struct sock *sk); 1162 1163 static inline struct dst_entry * 1164 __sk_dst_get(struct sock *sk) 1165 { 1166 return sk->sk_dst_cache; 1167 } 1168 1169 static inline struct dst_entry * 1170 sk_dst_get(struct sock *sk) 1171 { 1172 struct dst_entry *dst; 1173 1174 read_lock(&sk->sk_dst_lock); 1175 dst = sk->sk_dst_cache; 1176 if (dst) 1177 dst_hold(dst); 1178 read_unlock(&sk->sk_dst_lock); 1179 return dst; 1180 } 1181 1182 static inline void 1183 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1184 { 1185 struct dst_entry *old_dst; 1186 1187 sk_tx_queue_clear(sk); 1188 old_dst = sk->sk_dst_cache; 1189 sk->sk_dst_cache = dst; 1190 dst_release(old_dst); 1191 } 1192 1193 static inline void 1194 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1195 { 1196 write_lock(&sk->sk_dst_lock); 1197 __sk_dst_set(sk, dst); 1198 write_unlock(&sk->sk_dst_lock); 1199 } 1200 1201 static inline void 1202 __sk_dst_reset(struct sock *sk) 1203 { 1204 struct dst_entry *old_dst; 1205 1206 sk_tx_queue_clear(sk); 1207 old_dst = sk->sk_dst_cache; 1208 sk->sk_dst_cache = NULL; 1209 dst_release(old_dst); 1210 } 1211 1212 static inline void 1213 sk_dst_reset(struct sock *sk) 1214 { 1215 write_lock(&sk->sk_dst_lock); 1216 __sk_dst_reset(sk); 1217 write_unlock(&sk->sk_dst_lock); 1218 } 1219 1220 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1221 1222 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1223 1224 static inline int sk_can_gso(const struct sock *sk) 1225 { 1226 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1227 } 1228 1229 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1230 1231 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1232 struct sk_buff *skb, struct page *page, 1233 int off, int copy) 1234 { 1235 if (skb->ip_summed == CHECKSUM_NONE) { 1236 int err = 0; 1237 __wsum csum = csum_and_copy_from_user(from, 1238 page_address(page) + off, 1239 copy, 0, &err); 1240 if (err) 1241 return err; 1242 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1243 } else if (copy_from_user(page_address(page) + off, from, copy)) 1244 return -EFAULT; 1245 1246 skb->len += copy; 1247 skb->data_len += copy; 1248 skb->truesize += copy; 1249 sk->sk_wmem_queued += copy; 1250 sk_mem_charge(sk, copy); 1251 return 0; 1252 } 1253 1254 /** 1255 * sk_wmem_alloc_get - returns write allocations 1256 * @sk: socket 1257 * 1258 * Returns sk_wmem_alloc minus initial offset of one 1259 */ 1260 static inline int sk_wmem_alloc_get(const struct sock *sk) 1261 { 1262 return atomic_read(&sk->sk_wmem_alloc) - 1; 1263 } 1264 1265 /** 1266 * sk_rmem_alloc_get - returns read allocations 1267 * @sk: socket 1268 * 1269 * Returns sk_rmem_alloc 1270 */ 1271 static inline int sk_rmem_alloc_get(const struct sock *sk) 1272 { 1273 return atomic_read(&sk->sk_rmem_alloc); 1274 } 1275 1276 /** 1277 * sk_has_allocations - check if allocations are outstanding 1278 * @sk: socket 1279 * 1280 * Returns true if socket has write or read allocations 1281 */ 1282 static inline int sk_has_allocations(const struct sock *sk) 1283 { 1284 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1285 } 1286 1287 /** 1288 * sk_has_sleeper - check if there are any waiting processes 1289 * @sk: socket 1290 * 1291 * Returns true if socket has waiting processes 1292 * 1293 * The purpose of the sk_has_sleeper and sock_poll_wait is to wrap the memory 1294 * barrier call. They were added due to the race found within the tcp code. 1295 * 1296 * Consider following tcp code paths: 1297 * 1298 * CPU1 CPU2 1299 * 1300 * sys_select receive packet 1301 * ... ... 1302 * __add_wait_queue update tp->rcv_nxt 1303 * ... ... 1304 * tp->rcv_nxt check sock_def_readable 1305 * ... { 1306 * schedule ... 1307 * if (sk->sk_sleep && waitqueue_active(sk->sk_sleep)) 1308 * wake_up_interruptible(sk->sk_sleep) 1309 * ... 1310 * } 1311 * 1312 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1313 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1314 * could then endup calling schedule and sleep forever if there are no more 1315 * data on the socket. 1316 * 1317 * The sk_has_sleeper is always called right after a call to read_lock, so we 1318 * can use smp_mb__after_lock barrier. 1319 */ 1320 static inline int sk_has_sleeper(struct sock *sk) 1321 { 1322 /* 1323 * We need to be sure we are in sync with the 1324 * add_wait_queue modifications to the wait queue. 1325 * 1326 * This memory barrier is paired in the sock_poll_wait. 1327 */ 1328 smp_mb__after_lock(); 1329 return sk->sk_sleep && waitqueue_active(sk->sk_sleep); 1330 } 1331 1332 /** 1333 * sock_poll_wait - place memory barrier behind the poll_wait call. 1334 * @filp: file 1335 * @wait_address: socket wait queue 1336 * @p: poll_table 1337 * 1338 * See the comments in the sk_has_sleeper function. 1339 */ 1340 static inline void sock_poll_wait(struct file *filp, 1341 wait_queue_head_t *wait_address, poll_table *p) 1342 { 1343 if (p && wait_address) { 1344 poll_wait(filp, wait_address, p); 1345 /* 1346 * We need to be sure we are in sync with the 1347 * socket flags modification. 1348 * 1349 * This memory barrier is paired in the sk_has_sleeper. 1350 */ 1351 smp_mb(); 1352 } 1353 } 1354 1355 /* 1356 * Queue a received datagram if it will fit. Stream and sequenced 1357 * protocols can't normally use this as they need to fit buffers in 1358 * and play with them. 1359 * 1360 * Inlined as it's very short and called for pretty much every 1361 * packet ever received. 1362 */ 1363 1364 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1365 { 1366 skb_orphan(skb); 1367 skb->sk = sk; 1368 skb->destructor = sock_wfree; 1369 /* 1370 * We used to take a refcount on sk, but following operation 1371 * is enough to guarantee sk_free() wont free this sock until 1372 * all in-flight packets are completed 1373 */ 1374 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1375 } 1376 1377 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1378 { 1379 skb_orphan(skb); 1380 skb->sk = sk; 1381 skb->destructor = sock_rfree; 1382 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1383 sk_mem_charge(sk, skb->truesize); 1384 } 1385 1386 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1387 unsigned long expires); 1388 1389 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer); 1390 1391 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1392 1393 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb) 1394 { 1395 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces 1396 number of warnings when compiling with -W --ANK 1397 */ 1398 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >= 1399 (unsigned)sk->sk_rcvbuf) 1400 return -ENOMEM; 1401 skb_set_owner_r(skb, sk); 1402 skb_queue_tail(&sk->sk_error_queue, skb); 1403 if (!sock_flag(sk, SOCK_DEAD)) 1404 sk->sk_data_ready(sk, skb->len); 1405 return 0; 1406 } 1407 1408 /* 1409 * Recover an error report and clear atomically 1410 */ 1411 1412 static inline int sock_error(struct sock *sk) 1413 { 1414 int err; 1415 if (likely(!sk->sk_err)) 1416 return 0; 1417 err = xchg(&sk->sk_err, 0); 1418 return -err; 1419 } 1420 1421 static inline unsigned long sock_wspace(struct sock *sk) 1422 { 1423 int amt = 0; 1424 1425 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1426 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 1427 if (amt < 0) 1428 amt = 0; 1429 } 1430 return amt; 1431 } 1432 1433 static inline void sk_wake_async(struct sock *sk, int how, int band) 1434 { 1435 if (sock_flag(sk, SOCK_FASYNC)) 1436 sock_wake_async(sk->sk_socket, how, band); 1437 } 1438 1439 #define SOCK_MIN_SNDBUF 2048 1440 #define SOCK_MIN_RCVBUF 256 1441 1442 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 1443 { 1444 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 1445 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 1446 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF); 1447 } 1448 } 1449 1450 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 1451 1452 static inline struct page *sk_stream_alloc_page(struct sock *sk) 1453 { 1454 struct page *page = NULL; 1455 1456 page = alloc_pages(sk->sk_allocation, 0); 1457 if (!page) { 1458 sk->sk_prot->enter_memory_pressure(sk); 1459 sk_stream_moderate_sndbuf(sk); 1460 } 1461 return page; 1462 } 1463 1464 /* 1465 * Default write policy as shown to user space via poll/select/SIGIO 1466 */ 1467 static inline int sock_writeable(const struct sock *sk) 1468 { 1469 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 1470 } 1471 1472 static inline gfp_t gfp_any(void) 1473 { 1474 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 1475 } 1476 1477 static inline long sock_rcvtimeo(const struct sock *sk, int noblock) 1478 { 1479 return noblock ? 0 : sk->sk_rcvtimeo; 1480 } 1481 1482 static inline long sock_sndtimeo(const struct sock *sk, int noblock) 1483 { 1484 return noblock ? 0 : sk->sk_sndtimeo; 1485 } 1486 1487 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 1488 { 1489 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 1490 } 1491 1492 /* Alas, with timeout socket operations are not restartable. 1493 * Compare this to poll(). 1494 */ 1495 static inline int sock_intr_errno(long timeo) 1496 { 1497 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 1498 } 1499 1500 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 1501 struct sk_buff *skb); 1502 1503 static __inline__ void 1504 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 1505 { 1506 ktime_t kt = skb->tstamp; 1507 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 1508 1509 /* 1510 * generate control messages if 1511 * - receive time stamping in software requested (SOCK_RCVTSTAMP 1512 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 1513 * - software time stamp available and wanted 1514 * (SOCK_TIMESTAMPING_SOFTWARE) 1515 * - hardware time stamps available and wanted 1516 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 1517 * SOCK_TIMESTAMPING_RAW_HARDWARE) 1518 */ 1519 if (sock_flag(sk, SOCK_RCVTSTAMP) || 1520 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 1521 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 1522 (hwtstamps->hwtstamp.tv64 && 1523 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 1524 (hwtstamps->syststamp.tv64 && 1525 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 1526 __sock_recv_timestamp(msg, sk, skb); 1527 else 1528 sk->sk_stamp = kt; 1529 } 1530 1531 extern void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, struct sk_buff *skb); 1532 1533 /** 1534 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 1535 * @msg: outgoing packet 1536 * @sk: socket sending this packet 1537 * @shtx: filled with instructions for time stamping 1538 * 1539 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if 1540 * parameters are invalid. 1541 */ 1542 extern int sock_tx_timestamp(struct msghdr *msg, 1543 struct sock *sk, 1544 union skb_shared_tx *shtx); 1545 1546 1547 /** 1548 * sk_eat_skb - Release a skb if it is no longer needed 1549 * @sk: socket to eat this skb from 1550 * @skb: socket buffer to eat 1551 * @copied_early: flag indicating whether DMA operations copied this data early 1552 * 1553 * This routine must be called with interrupts disabled or with the socket 1554 * locked so that the sk_buff queue operation is ok. 1555 */ 1556 #ifdef CONFIG_NET_DMA 1557 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 1558 { 1559 __skb_unlink(skb, &sk->sk_receive_queue); 1560 if (!copied_early) 1561 __kfree_skb(skb); 1562 else 1563 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 1564 } 1565 #else 1566 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 1567 { 1568 __skb_unlink(skb, &sk->sk_receive_queue); 1569 __kfree_skb(skb); 1570 } 1571 #endif 1572 1573 static inline 1574 struct net *sock_net(const struct sock *sk) 1575 { 1576 #ifdef CONFIG_NET_NS 1577 return sk->sk_net; 1578 #else 1579 return &init_net; 1580 #endif 1581 } 1582 1583 static inline 1584 void sock_net_set(struct sock *sk, struct net *net) 1585 { 1586 #ifdef CONFIG_NET_NS 1587 sk->sk_net = net; 1588 #endif 1589 } 1590 1591 /* 1592 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 1593 * They should not hold a referrence to a namespace in order to allow 1594 * to stop it. 1595 * Sockets after sk_change_net should be released using sk_release_kernel 1596 */ 1597 static inline void sk_change_net(struct sock *sk, struct net *net) 1598 { 1599 put_net(sock_net(sk)); 1600 sock_net_set(sk, hold_net(net)); 1601 } 1602 1603 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 1604 { 1605 if (unlikely(skb->sk)) { 1606 struct sock *sk = skb->sk; 1607 1608 skb->destructor = NULL; 1609 skb->sk = NULL; 1610 return sk; 1611 } 1612 return NULL; 1613 } 1614 1615 extern void sock_enable_timestamp(struct sock *sk, int flag); 1616 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 1617 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 1618 1619 /* 1620 * Enable debug/info messages 1621 */ 1622 extern int net_msg_warn; 1623 #define NETDEBUG(fmt, args...) \ 1624 do { if (net_msg_warn) printk(fmt,##args); } while (0) 1625 1626 #define LIMIT_NETDEBUG(fmt, args...) \ 1627 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 1628 1629 extern __u32 sysctl_wmem_max; 1630 extern __u32 sysctl_rmem_max; 1631 1632 extern void sk_init(void); 1633 1634 extern int sysctl_optmem_max; 1635 1636 extern __u32 sysctl_wmem_default; 1637 extern __u32 sysctl_rmem_default; 1638 1639 #endif /* _SOCK_H */ 1640