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 #include <linux/slab.h> 55 56 #include <linux/filter.h> 57 #include <linux/rculist_nulls.h> 58 #include <linux/poll.h> 59 60 #include <linux/atomic.h> 61 #include <net/dst.h> 62 #include <net/checksum.h> 63 64 /* 65 * This structure really needs to be cleaned up. 66 * Most of it is for TCP, and not used by any of 67 * the other protocols. 68 */ 69 70 /* Define this to get the SOCK_DBG debugging facility. */ 71 #define SOCK_DEBUGGING 72 #ifdef SOCK_DEBUGGING 73 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 74 printk(KERN_DEBUG msg); } while (0) 75 #else 76 /* Validate arguments and do nothing */ 77 static inline void __attribute__ ((format (printf, 2, 3))) 78 SOCK_DEBUG(struct sock *sk, const char *msg, ...) 79 { 80 } 81 #endif 82 83 /* This is the per-socket lock. The spinlock provides a synchronization 84 * between user contexts and software interrupt processing, whereas the 85 * mini-semaphore synchronizes multiple users amongst themselves. 86 */ 87 typedef struct { 88 spinlock_t slock; 89 int owned; 90 wait_queue_head_t wq; 91 /* 92 * We express the mutex-alike socket_lock semantics 93 * to the lock validator by explicitly managing 94 * the slock as a lock variant (in addition to 95 * the slock itself): 96 */ 97 #ifdef CONFIG_DEBUG_LOCK_ALLOC 98 struct lockdep_map dep_map; 99 #endif 100 } socket_lock_t; 101 102 struct sock; 103 struct proto; 104 struct net; 105 106 /** 107 * struct sock_common - minimal network layer representation of sockets 108 * @skc_daddr: Foreign IPv4 addr 109 * @skc_rcv_saddr: Bound local IPv4 addr 110 * @skc_hash: hash value used with various protocol lookup tables 111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 112 * @skc_family: network address family 113 * @skc_state: Connection state 114 * @skc_reuse: %SO_REUSEADDR setting 115 * @skc_bound_dev_if: bound device index if != 0 116 * @skc_bind_node: bind hash linkage for various protocol lookup tables 117 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 118 * @skc_prot: protocol handlers inside a network family 119 * @skc_net: reference to the network namespace of this socket 120 * @skc_node: main hash linkage for various protocol lookup tables 121 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 122 * @skc_tx_queue_mapping: tx queue number for this connection 123 * @skc_refcnt: reference count 124 * 125 * This is the minimal network layer representation of sockets, the header 126 * for struct sock and struct inet_timewait_sock. 127 */ 128 struct sock_common { 129 /* skc_daddr and skc_rcv_saddr must be grouped : 130 * cf INET_MATCH() and INET_TW_MATCH() 131 */ 132 __be32 skc_daddr; 133 __be32 skc_rcv_saddr; 134 135 union { 136 unsigned int skc_hash; 137 __u16 skc_u16hashes[2]; 138 }; 139 unsigned short skc_family; 140 volatile unsigned char skc_state; 141 unsigned char skc_reuse; 142 int skc_bound_dev_if; 143 union { 144 struct hlist_node skc_bind_node; 145 struct hlist_nulls_node skc_portaddr_node; 146 }; 147 struct proto *skc_prot; 148 #ifdef CONFIG_NET_NS 149 struct net *skc_net; 150 #endif 151 /* 152 * fields between dontcopy_begin/dontcopy_end 153 * are not copied in sock_copy() 154 */ 155 /* private: */ 156 int skc_dontcopy_begin[0]; 157 /* public: */ 158 union { 159 struct hlist_node skc_node; 160 struct hlist_nulls_node skc_nulls_node; 161 }; 162 int skc_tx_queue_mapping; 163 atomic_t skc_refcnt; 164 /* private: */ 165 int skc_dontcopy_end[0]; 166 /* public: */ 167 }; 168 169 /** 170 * struct sock - network layer representation of sockets 171 * @__sk_common: shared layout with inet_timewait_sock 172 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 173 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 174 * @sk_lock: synchronizer 175 * @sk_rcvbuf: size of receive buffer in bytes 176 * @sk_wq: sock wait queue and async head 177 * @sk_dst_cache: destination cache 178 * @sk_dst_lock: destination cache lock 179 * @sk_policy: flow policy 180 * @sk_rmem_alloc: receive queue bytes committed 181 * @sk_receive_queue: incoming packets 182 * @sk_wmem_alloc: transmit queue bytes committed 183 * @sk_write_queue: Packet sending queue 184 * @sk_async_wait_queue: DMA copied packets 185 * @sk_omem_alloc: "o" is "option" or "other" 186 * @sk_wmem_queued: persistent queue size 187 * @sk_forward_alloc: space allocated forward 188 * @sk_allocation: allocation mode 189 * @sk_sndbuf: size of send buffer in bytes 190 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 191 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 192 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets 193 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 194 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 195 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 196 * @sk_gso_max_size: Maximum GSO segment size to build 197 * @sk_lingertime: %SO_LINGER l_linger setting 198 * @sk_backlog: always used with the per-socket spinlock held 199 * @sk_callback_lock: used with the callbacks in the end of this struct 200 * @sk_error_queue: rarely used 201 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 202 * IPV6_ADDRFORM for instance) 203 * @sk_err: last error 204 * @sk_err_soft: errors that don't cause failure but are the cause of a 205 * persistent failure not just 'timed out' 206 * @sk_drops: raw/udp drops counter 207 * @sk_ack_backlog: current listen backlog 208 * @sk_max_ack_backlog: listen backlog set in listen() 209 * @sk_priority: %SO_PRIORITY setting 210 * @sk_type: socket type (%SOCK_STREAM, etc) 211 * @sk_protocol: which protocol this socket belongs in this network family 212 * @sk_peer_pid: &struct pid for this socket's peer 213 * @sk_peer_cred: %SO_PEERCRED setting 214 * @sk_rcvlowat: %SO_RCVLOWAT setting 215 * @sk_rcvtimeo: %SO_RCVTIMEO setting 216 * @sk_sndtimeo: %SO_SNDTIMEO setting 217 * @sk_rxhash: flow hash received from netif layer 218 * @sk_filter: socket filtering instructions 219 * @sk_protinfo: private area, net family specific, when not using slab 220 * @sk_timer: sock cleanup timer 221 * @sk_stamp: time stamp of last packet received 222 * @sk_socket: Identd and reporting IO signals 223 * @sk_user_data: RPC layer private data 224 * @sk_sndmsg_page: cached page for sendmsg 225 * @sk_sndmsg_off: cached offset for sendmsg 226 * @sk_send_head: front of stuff to transmit 227 * @sk_security: used by security modules 228 * @sk_mark: generic packet mark 229 * @sk_classid: this socket's cgroup classid 230 * @sk_write_pending: a write to stream socket waits to start 231 * @sk_state_change: callback to indicate change in the state of the sock 232 * @sk_data_ready: callback to indicate there is data to be processed 233 * @sk_write_space: callback to indicate there is bf sending space available 234 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 235 * @sk_backlog_rcv: callback to process the backlog 236 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 237 */ 238 struct sock { 239 /* 240 * Now struct inet_timewait_sock also uses sock_common, so please just 241 * don't add nothing before this first member (__sk_common) --acme 242 */ 243 struct sock_common __sk_common; 244 #define sk_node __sk_common.skc_node 245 #define sk_nulls_node __sk_common.skc_nulls_node 246 #define sk_refcnt __sk_common.skc_refcnt 247 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 248 249 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 250 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 251 #define sk_hash __sk_common.skc_hash 252 #define sk_family __sk_common.skc_family 253 #define sk_state __sk_common.skc_state 254 #define sk_reuse __sk_common.skc_reuse 255 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 256 #define sk_bind_node __sk_common.skc_bind_node 257 #define sk_prot __sk_common.skc_prot 258 #define sk_net __sk_common.skc_net 259 socket_lock_t sk_lock; 260 struct sk_buff_head sk_receive_queue; 261 /* 262 * The backlog queue is special, it is always used with 263 * the per-socket spinlock held and requires low latency 264 * access. Therefore we special case it's implementation. 265 * Note : rmem_alloc is in this structure to fill a hole 266 * on 64bit arches, not because its logically part of 267 * backlog. 268 */ 269 struct { 270 atomic_t rmem_alloc; 271 int len; 272 struct sk_buff *head; 273 struct sk_buff *tail; 274 } sk_backlog; 275 #define sk_rmem_alloc sk_backlog.rmem_alloc 276 int sk_forward_alloc; 277 #ifdef CONFIG_RPS 278 __u32 sk_rxhash; 279 #endif 280 atomic_t sk_drops; 281 int sk_rcvbuf; 282 283 struct sk_filter __rcu *sk_filter; 284 struct socket_wq *sk_wq; 285 286 #ifdef CONFIG_NET_DMA 287 struct sk_buff_head sk_async_wait_queue; 288 #endif 289 290 #ifdef CONFIG_XFRM 291 struct xfrm_policy *sk_policy[2]; 292 #endif 293 unsigned long sk_flags; 294 struct dst_entry *sk_dst_cache; 295 spinlock_t sk_dst_lock; 296 atomic_t sk_wmem_alloc; 297 atomic_t sk_omem_alloc; 298 int sk_sndbuf; 299 struct sk_buff_head sk_write_queue; 300 kmemcheck_bitfield_begin(flags); 301 unsigned int sk_shutdown : 2, 302 sk_no_check : 2, 303 sk_userlocks : 4, 304 sk_protocol : 8, 305 sk_type : 16; 306 kmemcheck_bitfield_end(flags); 307 int sk_wmem_queued; 308 gfp_t sk_allocation; 309 int sk_route_caps; 310 int sk_route_nocaps; 311 int sk_gso_type; 312 unsigned int sk_gso_max_size; 313 int sk_rcvlowat; 314 unsigned long sk_lingertime; 315 struct sk_buff_head sk_error_queue; 316 struct proto *sk_prot_creator; 317 rwlock_t sk_callback_lock; 318 int sk_err, 319 sk_err_soft; 320 unsigned short sk_ack_backlog; 321 unsigned short sk_max_ack_backlog; 322 __u32 sk_priority; 323 struct pid *sk_peer_pid; 324 const struct cred *sk_peer_cred; 325 long sk_rcvtimeo; 326 long sk_sndtimeo; 327 void *sk_protinfo; 328 struct timer_list sk_timer; 329 ktime_t sk_stamp; 330 struct socket *sk_socket; 331 void *sk_user_data; 332 struct page *sk_sndmsg_page; 333 struct sk_buff *sk_send_head; 334 __u32 sk_sndmsg_off; 335 int sk_write_pending; 336 #ifdef CONFIG_SECURITY 337 void *sk_security; 338 #endif 339 __u32 sk_mark; 340 u32 sk_classid; 341 void (*sk_state_change)(struct sock *sk); 342 void (*sk_data_ready)(struct sock *sk, int bytes); 343 void (*sk_write_space)(struct sock *sk); 344 void (*sk_error_report)(struct sock *sk); 345 int (*sk_backlog_rcv)(struct sock *sk, 346 struct sk_buff *skb); 347 void (*sk_destruct)(struct sock *sk); 348 }; 349 350 /* 351 * Hashed lists helper routines 352 */ 353 static inline struct sock *sk_entry(const struct hlist_node *node) 354 { 355 return hlist_entry(node, struct sock, sk_node); 356 } 357 358 static inline struct sock *__sk_head(const struct hlist_head *head) 359 { 360 return hlist_entry(head->first, struct sock, sk_node); 361 } 362 363 static inline struct sock *sk_head(const struct hlist_head *head) 364 { 365 return hlist_empty(head) ? NULL : __sk_head(head); 366 } 367 368 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 369 { 370 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 371 } 372 373 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 374 { 375 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 376 } 377 378 static inline struct sock *sk_next(const struct sock *sk) 379 { 380 return sk->sk_node.next ? 381 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 382 } 383 384 static inline struct sock *sk_nulls_next(const struct sock *sk) 385 { 386 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 387 hlist_nulls_entry(sk->sk_nulls_node.next, 388 struct sock, sk_nulls_node) : 389 NULL; 390 } 391 392 static inline int sk_unhashed(const struct sock *sk) 393 { 394 return hlist_unhashed(&sk->sk_node); 395 } 396 397 static inline int sk_hashed(const struct sock *sk) 398 { 399 return !sk_unhashed(sk); 400 } 401 402 static __inline__ void sk_node_init(struct hlist_node *node) 403 { 404 node->pprev = NULL; 405 } 406 407 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node) 408 { 409 node->pprev = NULL; 410 } 411 412 static __inline__ void __sk_del_node(struct sock *sk) 413 { 414 __hlist_del(&sk->sk_node); 415 } 416 417 /* NB: equivalent to hlist_del_init_rcu */ 418 static __inline__ int __sk_del_node_init(struct sock *sk) 419 { 420 if (sk_hashed(sk)) { 421 __sk_del_node(sk); 422 sk_node_init(&sk->sk_node); 423 return 1; 424 } 425 return 0; 426 } 427 428 /* Grab socket reference count. This operation is valid only 429 when sk is ALREADY grabbed f.e. it is found in hash table 430 or a list and the lookup is made under lock preventing hash table 431 modifications. 432 */ 433 434 static inline void sock_hold(struct sock *sk) 435 { 436 atomic_inc(&sk->sk_refcnt); 437 } 438 439 /* Ungrab socket in the context, which assumes that socket refcnt 440 cannot hit zero, f.e. it is true in context of any socketcall. 441 */ 442 static inline void __sock_put(struct sock *sk) 443 { 444 atomic_dec(&sk->sk_refcnt); 445 } 446 447 static __inline__ int sk_del_node_init(struct sock *sk) 448 { 449 int rc = __sk_del_node_init(sk); 450 451 if (rc) { 452 /* paranoid for a while -acme */ 453 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 454 __sock_put(sk); 455 } 456 return rc; 457 } 458 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 459 460 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk) 461 { 462 if (sk_hashed(sk)) { 463 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 464 return 1; 465 } 466 return 0; 467 } 468 469 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk) 470 { 471 int rc = __sk_nulls_del_node_init_rcu(sk); 472 473 if (rc) { 474 /* paranoid for a while -acme */ 475 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 476 __sock_put(sk); 477 } 478 return rc; 479 } 480 481 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list) 482 { 483 hlist_add_head(&sk->sk_node, list); 484 } 485 486 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list) 487 { 488 sock_hold(sk); 489 __sk_add_node(sk, list); 490 } 491 492 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 493 { 494 sock_hold(sk); 495 hlist_add_head_rcu(&sk->sk_node, list); 496 } 497 498 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 499 { 500 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 501 } 502 503 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 504 { 505 sock_hold(sk); 506 __sk_nulls_add_node_rcu(sk, list); 507 } 508 509 static __inline__ void __sk_del_bind_node(struct sock *sk) 510 { 511 __hlist_del(&sk->sk_bind_node); 512 } 513 514 static __inline__ void sk_add_bind_node(struct sock *sk, 515 struct hlist_head *list) 516 { 517 hlist_add_head(&sk->sk_bind_node, list); 518 } 519 520 #define sk_for_each(__sk, node, list) \ 521 hlist_for_each_entry(__sk, node, list, sk_node) 522 #define sk_for_each_rcu(__sk, node, list) \ 523 hlist_for_each_entry_rcu(__sk, node, list, sk_node) 524 #define sk_nulls_for_each(__sk, node, list) \ 525 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 526 #define sk_nulls_for_each_rcu(__sk, node, list) \ 527 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 528 #define sk_for_each_from(__sk, node) \ 529 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 530 hlist_for_each_entry_from(__sk, node, sk_node) 531 #define sk_nulls_for_each_from(__sk, node) \ 532 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 533 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 534 #define sk_for_each_safe(__sk, node, tmp, list) \ 535 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node) 536 #define sk_for_each_bound(__sk, node, list) \ 537 hlist_for_each_entry(__sk, node, list, sk_bind_node) 538 539 /* Sock flags */ 540 enum sock_flags { 541 SOCK_DEAD, 542 SOCK_DONE, 543 SOCK_URGINLINE, 544 SOCK_KEEPOPEN, 545 SOCK_LINGER, 546 SOCK_DESTROY, 547 SOCK_BROADCAST, 548 SOCK_TIMESTAMP, 549 SOCK_ZAPPED, 550 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 551 SOCK_DBG, /* %SO_DEBUG setting */ 552 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 553 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 554 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 555 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 556 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 557 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 558 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 559 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 560 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 561 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 562 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 563 SOCK_FASYNC, /* fasync() active */ 564 SOCK_RXQ_OVFL, 565 }; 566 567 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 568 { 569 nsk->sk_flags = osk->sk_flags; 570 } 571 572 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 573 { 574 __set_bit(flag, &sk->sk_flags); 575 } 576 577 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 578 { 579 __clear_bit(flag, &sk->sk_flags); 580 } 581 582 static inline int sock_flag(struct sock *sk, enum sock_flags flag) 583 { 584 return test_bit(flag, &sk->sk_flags); 585 } 586 587 static inline void sk_acceptq_removed(struct sock *sk) 588 { 589 sk->sk_ack_backlog--; 590 } 591 592 static inline void sk_acceptq_added(struct sock *sk) 593 { 594 sk->sk_ack_backlog++; 595 } 596 597 static inline int sk_acceptq_is_full(struct sock *sk) 598 { 599 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 600 } 601 602 /* 603 * Compute minimal free write space needed to queue new packets. 604 */ 605 static inline int sk_stream_min_wspace(struct sock *sk) 606 { 607 return sk->sk_wmem_queued >> 1; 608 } 609 610 static inline int sk_stream_wspace(struct sock *sk) 611 { 612 return sk->sk_sndbuf - sk->sk_wmem_queued; 613 } 614 615 extern void sk_stream_write_space(struct sock *sk); 616 617 static inline int sk_stream_memory_free(struct sock *sk) 618 { 619 return sk->sk_wmem_queued < sk->sk_sndbuf; 620 } 621 622 /* OOB backlog add */ 623 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 624 { 625 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 626 skb_dst_force(skb); 627 628 if (!sk->sk_backlog.tail) 629 sk->sk_backlog.head = skb; 630 else 631 sk->sk_backlog.tail->next = skb; 632 633 sk->sk_backlog.tail = skb; 634 skb->next = NULL; 635 } 636 637 /* 638 * Take into account size of receive queue and backlog queue 639 */ 640 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb) 641 { 642 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 643 644 return qsize + skb->truesize > sk->sk_rcvbuf; 645 } 646 647 /* The per-socket spinlock must be held here. */ 648 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb) 649 { 650 if (sk_rcvqueues_full(sk, skb)) 651 return -ENOBUFS; 652 653 __sk_add_backlog(sk, skb); 654 sk->sk_backlog.len += skb->truesize; 655 return 0; 656 } 657 658 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 659 { 660 return sk->sk_backlog_rcv(sk, skb); 661 } 662 663 static inline void sock_rps_record_flow(const struct sock *sk) 664 { 665 #ifdef CONFIG_RPS 666 struct rps_sock_flow_table *sock_flow_table; 667 668 rcu_read_lock(); 669 sock_flow_table = rcu_dereference(rps_sock_flow_table); 670 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 671 rcu_read_unlock(); 672 #endif 673 } 674 675 static inline void sock_rps_reset_flow(const struct sock *sk) 676 { 677 #ifdef CONFIG_RPS 678 struct rps_sock_flow_table *sock_flow_table; 679 680 rcu_read_lock(); 681 sock_flow_table = rcu_dereference(rps_sock_flow_table); 682 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 683 rcu_read_unlock(); 684 #endif 685 } 686 687 static inline void sock_rps_save_rxhash(struct sock *sk, u32 rxhash) 688 { 689 #ifdef CONFIG_RPS 690 if (unlikely(sk->sk_rxhash != rxhash)) { 691 sock_rps_reset_flow(sk); 692 sk->sk_rxhash = rxhash; 693 } 694 #endif 695 } 696 697 #define sk_wait_event(__sk, __timeo, __condition) \ 698 ({ int __rc; \ 699 release_sock(__sk); \ 700 __rc = __condition; \ 701 if (!__rc) { \ 702 *(__timeo) = schedule_timeout(*(__timeo)); \ 703 } \ 704 lock_sock(__sk); \ 705 __rc = __condition; \ 706 __rc; \ 707 }) 708 709 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 710 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 711 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 712 extern int sk_stream_error(struct sock *sk, int flags, int err); 713 extern void sk_stream_kill_queues(struct sock *sk); 714 715 extern int sk_wait_data(struct sock *sk, long *timeo); 716 717 struct request_sock_ops; 718 struct timewait_sock_ops; 719 struct inet_hashinfo; 720 struct raw_hashinfo; 721 722 /* Networking protocol blocks we attach to sockets. 723 * socket layer -> transport layer interface 724 * transport -> network interface is defined by struct inet_proto 725 */ 726 struct proto { 727 void (*close)(struct sock *sk, 728 long timeout); 729 int (*connect)(struct sock *sk, 730 struct sockaddr *uaddr, 731 int addr_len); 732 int (*disconnect)(struct sock *sk, int flags); 733 734 struct sock * (*accept) (struct sock *sk, int flags, int *err); 735 736 int (*ioctl)(struct sock *sk, int cmd, 737 unsigned long arg); 738 int (*init)(struct sock *sk); 739 void (*destroy)(struct sock *sk); 740 void (*shutdown)(struct sock *sk, int how); 741 int (*setsockopt)(struct sock *sk, int level, 742 int optname, char __user *optval, 743 unsigned int optlen); 744 int (*getsockopt)(struct sock *sk, int level, 745 int optname, char __user *optval, 746 int __user *option); 747 #ifdef CONFIG_COMPAT 748 int (*compat_setsockopt)(struct sock *sk, 749 int level, 750 int optname, char __user *optval, 751 unsigned int optlen); 752 int (*compat_getsockopt)(struct sock *sk, 753 int level, 754 int optname, char __user *optval, 755 int __user *option); 756 #endif 757 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 758 struct msghdr *msg, size_t len); 759 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 760 struct msghdr *msg, 761 size_t len, int noblock, int flags, 762 int *addr_len); 763 int (*sendpage)(struct sock *sk, struct page *page, 764 int offset, size_t size, int flags); 765 int (*bind)(struct sock *sk, 766 struct sockaddr *uaddr, int addr_len); 767 768 int (*backlog_rcv) (struct sock *sk, 769 struct sk_buff *skb); 770 771 /* Keeping track of sk's, looking them up, and port selection methods. */ 772 void (*hash)(struct sock *sk); 773 void (*unhash)(struct sock *sk); 774 void (*rehash)(struct sock *sk); 775 int (*get_port)(struct sock *sk, unsigned short snum); 776 void (*clear_sk)(struct sock *sk, int size); 777 778 /* Keeping track of sockets in use */ 779 #ifdef CONFIG_PROC_FS 780 unsigned int inuse_idx; 781 #endif 782 783 /* Memory pressure */ 784 void (*enter_memory_pressure)(struct sock *sk); 785 atomic_long_t *memory_allocated; /* Current allocated memory. */ 786 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 787 /* 788 * Pressure flag: try to collapse. 789 * Technical note: it is used by multiple contexts non atomically. 790 * All the __sk_mem_schedule() is of this nature: accounting 791 * is strict, actions are advisory and have some latency. 792 */ 793 int *memory_pressure; 794 long *sysctl_mem; 795 int *sysctl_wmem; 796 int *sysctl_rmem; 797 int max_header; 798 bool no_autobind; 799 800 struct kmem_cache *slab; 801 unsigned int obj_size; 802 int slab_flags; 803 804 struct percpu_counter *orphan_count; 805 806 struct request_sock_ops *rsk_prot; 807 struct timewait_sock_ops *twsk_prot; 808 809 union { 810 struct inet_hashinfo *hashinfo; 811 struct udp_table *udp_table; 812 struct raw_hashinfo *raw_hash; 813 } h; 814 815 struct module *owner; 816 817 char name[32]; 818 819 struct list_head node; 820 #ifdef SOCK_REFCNT_DEBUG 821 atomic_t socks; 822 #endif 823 }; 824 825 extern int proto_register(struct proto *prot, int alloc_slab); 826 extern void proto_unregister(struct proto *prot); 827 828 #ifdef SOCK_REFCNT_DEBUG 829 static inline void sk_refcnt_debug_inc(struct sock *sk) 830 { 831 atomic_inc(&sk->sk_prot->socks); 832 } 833 834 static inline void sk_refcnt_debug_dec(struct sock *sk) 835 { 836 atomic_dec(&sk->sk_prot->socks); 837 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 838 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 839 } 840 841 static inline void sk_refcnt_debug_release(const struct sock *sk) 842 { 843 if (atomic_read(&sk->sk_refcnt) != 1) 844 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 845 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 846 } 847 #else /* SOCK_REFCNT_DEBUG */ 848 #define sk_refcnt_debug_inc(sk) do { } while (0) 849 #define sk_refcnt_debug_dec(sk) do { } while (0) 850 #define sk_refcnt_debug_release(sk) do { } while (0) 851 #endif /* SOCK_REFCNT_DEBUG */ 852 853 854 #ifdef CONFIG_PROC_FS 855 /* Called with local bh disabled */ 856 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 857 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 858 #else 859 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot, 860 int inc) 861 { 862 } 863 #endif 864 865 866 /* With per-bucket locks this operation is not-atomic, so that 867 * this version is not worse. 868 */ 869 static inline void __sk_prot_rehash(struct sock *sk) 870 { 871 sk->sk_prot->unhash(sk); 872 sk->sk_prot->hash(sk); 873 } 874 875 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 876 877 /* About 10 seconds */ 878 #define SOCK_DESTROY_TIME (10*HZ) 879 880 /* Sockets 0-1023 can't be bound to unless you are superuser */ 881 #define PROT_SOCK 1024 882 883 #define SHUTDOWN_MASK 3 884 #define RCV_SHUTDOWN 1 885 #define SEND_SHUTDOWN 2 886 887 #define SOCK_SNDBUF_LOCK 1 888 #define SOCK_RCVBUF_LOCK 2 889 #define SOCK_BINDADDR_LOCK 4 890 #define SOCK_BINDPORT_LOCK 8 891 892 /* sock_iocb: used to kick off async processing of socket ios */ 893 struct sock_iocb { 894 struct list_head list; 895 896 int flags; 897 int size; 898 struct socket *sock; 899 struct sock *sk; 900 struct scm_cookie *scm; 901 struct msghdr *msg, async_msg; 902 struct kiocb *kiocb; 903 }; 904 905 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 906 { 907 return (struct sock_iocb *)iocb->private; 908 } 909 910 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 911 { 912 return si->kiocb; 913 } 914 915 struct socket_alloc { 916 struct socket socket; 917 struct inode vfs_inode; 918 }; 919 920 static inline struct socket *SOCKET_I(struct inode *inode) 921 { 922 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 923 } 924 925 static inline struct inode *SOCK_INODE(struct socket *socket) 926 { 927 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 928 } 929 930 /* 931 * Functions for memory accounting 932 */ 933 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 934 extern void __sk_mem_reclaim(struct sock *sk); 935 936 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 937 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 938 #define SK_MEM_SEND 0 939 #define SK_MEM_RECV 1 940 941 static inline int sk_mem_pages(int amt) 942 { 943 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 944 } 945 946 static inline int sk_has_account(struct sock *sk) 947 { 948 /* return true if protocol supports memory accounting */ 949 return !!sk->sk_prot->memory_allocated; 950 } 951 952 static inline int sk_wmem_schedule(struct sock *sk, int size) 953 { 954 if (!sk_has_account(sk)) 955 return 1; 956 return size <= sk->sk_forward_alloc || 957 __sk_mem_schedule(sk, size, SK_MEM_SEND); 958 } 959 960 static inline int sk_rmem_schedule(struct sock *sk, int size) 961 { 962 if (!sk_has_account(sk)) 963 return 1; 964 return size <= sk->sk_forward_alloc || 965 __sk_mem_schedule(sk, size, SK_MEM_RECV); 966 } 967 968 static inline void sk_mem_reclaim(struct sock *sk) 969 { 970 if (!sk_has_account(sk)) 971 return; 972 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 973 __sk_mem_reclaim(sk); 974 } 975 976 static inline void sk_mem_reclaim_partial(struct sock *sk) 977 { 978 if (!sk_has_account(sk)) 979 return; 980 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 981 __sk_mem_reclaim(sk); 982 } 983 984 static inline void sk_mem_charge(struct sock *sk, int size) 985 { 986 if (!sk_has_account(sk)) 987 return; 988 sk->sk_forward_alloc -= size; 989 } 990 991 static inline void sk_mem_uncharge(struct sock *sk, int size) 992 { 993 if (!sk_has_account(sk)) 994 return; 995 sk->sk_forward_alloc += size; 996 } 997 998 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 999 { 1000 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1001 sk->sk_wmem_queued -= skb->truesize; 1002 sk_mem_uncharge(sk, skb->truesize); 1003 __kfree_skb(skb); 1004 } 1005 1006 /* Used by processes to "lock" a socket state, so that 1007 * interrupts and bottom half handlers won't change it 1008 * from under us. It essentially blocks any incoming 1009 * packets, so that we won't get any new data or any 1010 * packets that change the state of the socket. 1011 * 1012 * While locked, BH processing will add new packets to 1013 * the backlog queue. This queue is processed by the 1014 * owner of the socket lock right before it is released. 1015 * 1016 * Since ~2.3.5 it is also exclusive sleep lock serializing 1017 * accesses from user process context. 1018 */ 1019 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1020 1021 /* 1022 * Macro so as to not evaluate some arguments when 1023 * lockdep is not enabled. 1024 * 1025 * Mark both the sk_lock and the sk_lock.slock as a 1026 * per-address-family lock class. 1027 */ 1028 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1029 do { \ 1030 sk->sk_lock.owned = 0; \ 1031 init_waitqueue_head(&sk->sk_lock.wq); \ 1032 spin_lock_init(&(sk)->sk_lock.slock); \ 1033 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1034 sizeof((sk)->sk_lock)); \ 1035 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1036 (skey), (sname)); \ 1037 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1038 } while (0) 1039 1040 extern void lock_sock_nested(struct sock *sk, int subclass); 1041 1042 static inline void lock_sock(struct sock *sk) 1043 { 1044 lock_sock_nested(sk, 0); 1045 } 1046 1047 extern void release_sock(struct sock *sk); 1048 1049 /* BH context may only use the following locking interface. */ 1050 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1051 #define bh_lock_sock_nested(__sk) \ 1052 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1053 SINGLE_DEPTH_NESTING) 1054 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1055 1056 extern bool lock_sock_fast(struct sock *sk); 1057 /** 1058 * unlock_sock_fast - complement of lock_sock_fast 1059 * @sk: socket 1060 * @slow: slow mode 1061 * 1062 * fast unlock socket for user context. 1063 * If slow mode is on, we call regular release_sock() 1064 */ 1065 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1066 { 1067 if (slow) 1068 release_sock(sk); 1069 else 1070 spin_unlock_bh(&sk->sk_lock.slock); 1071 } 1072 1073 1074 extern struct sock *sk_alloc(struct net *net, int family, 1075 gfp_t priority, 1076 struct proto *prot); 1077 extern void sk_free(struct sock *sk); 1078 extern void sk_release_kernel(struct sock *sk); 1079 extern struct sock *sk_clone(const struct sock *sk, 1080 const gfp_t priority); 1081 1082 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1083 unsigned long size, int force, 1084 gfp_t priority); 1085 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1086 unsigned long size, int force, 1087 gfp_t priority); 1088 extern void sock_wfree(struct sk_buff *skb); 1089 extern void sock_rfree(struct sk_buff *skb); 1090 1091 extern int sock_setsockopt(struct socket *sock, int level, 1092 int op, char __user *optval, 1093 unsigned int optlen); 1094 1095 extern int sock_getsockopt(struct socket *sock, int level, 1096 int op, char __user *optval, 1097 int __user *optlen); 1098 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1099 unsigned long size, 1100 int noblock, 1101 int *errcode); 1102 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1103 unsigned long header_len, 1104 unsigned long data_len, 1105 int noblock, 1106 int *errcode); 1107 extern void *sock_kmalloc(struct sock *sk, int size, 1108 gfp_t priority); 1109 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1110 extern void sk_send_sigurg(struct sock *sk); 1111 1112 #ifdef CONFIG_CGROUPS 1113 extern void sock_update_classid(struct sock *sk); 1114 #else 1115 static inline void sock_update_classid(struct sock *sk) 1116 { 1117 } 1118 #endif 1119 1120 /* 1121 * Functions to fill in entries in struct proto_ops when a protocol 1122 * does not implement a particular function. 1123 */ 1124 extern int sock_no_bind(struct socket *, 1125 struct sockaddr *, int); 1126 extern int sock_no_connect(struct socket *, 1127 struct sockaddr *, int, int); 1128 extern int sock_no_socketpair(struct socket *, 1129 struct socket *); 1130 extern int sock_no_accept(struct socket *, 1131 struct socket *, int); 1132 extern int sock_no_getname(struct socket *, 1133 struct sockaddr *, int *, int); 1134 extern unsigned int sock_no_poll(struct file *, struct socket *, 1135 struct poll_table_struct *); 1136 extern int sock_no_ioctl(struct socket *, unsigned int, 1137 unsigned long); 1138 extern int sock_no_listen(struct socket *, int); 1139 extern int sock_no_shutdown(struct socket *, int); 1140 extern int sock_no_getsockopt(struct socket *, int , int, 1141 char __user *, int __user *); 1142 extern int sock_no_setsockopt(struct socket *, int, int, 1143 char __user *, unsigned int); 1144 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1145 struct msghdr *, size_t); 1146 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1147 struct msghdr *, size_t, int); 1148 extern int sock_no_mmap(struct file *file, 1149 struct socket *sock, 1150 struct vm_area_struct *vma); 1151 extern ssize_t sock_no_sendpage(struct socket *sock, 1152 struct page *page, 1153 int offset, size_t size, 1154 int flags); 1155 1156 /* 1157 * Functions to fill in entries in struct proto_ops when a protocol 1158 * uses the inet style. 1159 */ 1160 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1161 char __user *optval, int __user *optlen); 1162 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1163 struct msghdr *msg, size_t size, int flags); 1164 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1165 char __user *optval, unsigned int optlen); 1166 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1167 int optname, char __user *optval, int __user *optlen); 1168 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1169 int optname, char __user *optval, unsigned int optlen); 1170 1171 extern void sk_common_release(struct sock *sk); 1172 1173 /* 1174 * Default socket callbacks and setup code 1175 */ 1176 1177 /* Initialise core socket variables */ 1178 extern void sock_init_data(struct socket *sock, struct sock *sk); 1179 1180 extern void sk_filter_release_rcu(struct rcu_head *rcu); 1181 1182 /** 1183 * sk_filter_release - release a socket filter 1184 * @fp: filter to remove 1185 * 1186 * Remove a filter from a socket and release its resources. 1187 */ 1188 1189 static inline void sk_filter_release(struct sk_filter *fp) 1190 { 1191 if (atomic_dec_and_test(&fp->refcnt)) 1192 call_rcu_bh(&fp->rcu, sk_filter_release_rcu); 1193 } 1194 1195 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1196 { 1197 unsigned int size = sk_filter_len(fp); 1198 1199 atomic_sub(size, &sk->sk_omem_alloc); 1200 sk_filter_release(fp); 1201 } 1202 1203 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1204 { 1205 atomic_inc(&fp->refcnt); 1206 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1207 } 1208 1209 /* 1210 * Socket reference counting postulates. 1211 * 1212 * * Each user of socket SHOULD hold a reference count. 1213 * * Each access point to socket (an hash table bucket, reference from a list, 1214 * running timer, skb in flight MUST hold a reference count. 1215 * * When reference count hits 0, it means it will never increase back. 1216 * * When reference count hits 0, it means that no references from 1217 * outside exist to this socket and current process on current CPU 1218 * is last user and may/should destroy this socket. 1219 * * sk_free is called from any context: process, BH, IRQ. When 1220 * it is called, socket has no references from outside -> sk_free 1221 * may release descendant resources allocated by the socket, but 1222 * to the time when it is called, socket is NOT referenced by any 1223 * hash tables, lists etc. 1224 * * Packets, delivered from outside (from network or from another process) 1225 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1226 * when they sit in queue. Otherwise, packets will leak to hole, when 1227 * socket is looked up by one cpu and unhasing is made by another CPU. 1228 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1229 * (leak to backlog). Packet socket does all the processing inside 1230 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1231 * use separate SMP lock, so that they are prone too. 1232 */ 1233 1234 /* Ungrab socket and destroy it, if it was the last reference. */ 1235 static inline void sock_put(struct sock *sk) 1236 { 1237 if (atomic_dec_and_test(&sk->sk_refcnt)) 1238 sk_free(sk); 1239 } 1240 1241 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1242 const int nested); 1243 1244 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1245 { 1246 sk->sk_tx_queue_mapping = tx_queue; 1247 } 1248 1249 static inline void sk_tx_queue_clear(struct sock *sk) 1250 { 1251 sk->sk_tx_queue_mapping = -1; 1252 } 1253 1254 static inline int sk_tx_queue_get(const struct sock *sk) 1255 { 1256 return sk ? sk->sk_tx_queue_mapping : -1; 1257 } 1258 1259 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1260 { 1261 sk_tx_queue_clear(sk); 1262 sk->sk_socket = sock; 1263 } 1264 1265 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1266 { 1267 return &sk->sk_wq->wait; 1268 } 1269 /* Detach socket from process context. 1270 * Announce socket dead, detach it from wait queue and inode. 1271 * Note that parent inode held reference count on this struct sock, 1272 * we do not release it in this function, because protocol 1273 * probably wants some additional cleanups or even continuing 1274 * to work with this socket (TCP). 1275 */ 1276 static inline void sock_orphan(struct sock *sk) 1277 { 1278 write_lock_bh(&sk->sk_callback_lock); 1279 sock_set_flag(sk, SOCK_DEAD); 1280 sk_set_socket(sk, NULL); 1281 sk->sk_wq = NULL; 1282 write_unlock_bh(&sk->sk_callback_lock); 1283 } 1284 1285 static inline void sock_graft(struct sock *sk, struct socket *parent) 1286 { 1287 write_lock_bh(&sk->sk_callback_lock); 1288 rcu_assign_pointer(sk->sk_wq, parent->wq); 1289 parent->sk = sk; 1290 sk_set_socket(sk, parent); 1291 security_sock_graft(sk, parent); 1292 write_unlock_bh(&sk->sk_callback_lock); 1293 } 1294 1295 extern int sock_i_uid(struct sock *sk); 1296 extern unsigned long sock_i_ino(struct sock *sk); 1297 1298 static inline struct dst_entry * 1299 __sk_dst_get(struct sock *sk) 1300 { 1301 return rcu_dereference_check(sk->sk_dst_cache, rcu_read_lock_held() || 1302 sock_owned_by_user(sk) || 1303 lockdep_is_held(&sk->sk_lock.slock)); 1304 } 1305 1306 static inline struct dst_entry * 1307 sk_dst_get(struct sock *sk) 1308 { 1309 struct dst_entry *dst; 1310 1311 rcu_read_lock(); 1312 dst = rcu_dereference(sk->sk_dst_cache); 1313 if (dst) 1314 dst_hold(dst); 1315 rcu_read_unlock(); 1316 return dst; 1317 } 1318 1319 extern void sk_reset_txq(struct sock *sk); 1320 1321 static inline void dst_negative_advice(struct sock *sk) 1322 { 1323 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1324 1325 if (dst && dst->ops->negative_advice) { 1326 ndst = dst->ops->negative_advice(dst); 1327 1328 if (ndst != dst) { 1329 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1330 sk_reset_txq(sk); 1331 } 1332 } 1333 } 1334 1335 static inline void 1336 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1337 { 1338 struct dst_entry *old_dst; 1339 1340 sk_tx_queue_clear(sk); 1341 /* 1342 * This can be called while sk is owned by the caller only, 1343 * with no state that can be checked in a rcu_dereference_check() cond 1344 */ 1345 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1346 rcu_assign_pointer(sk->sk_dst_cache, dst); 1347 dst_release(old_dst); 1348 } 1349 1350 static inline void 1351 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1352 { 1353 spin_lock(&sk->sk_dst_lock); 1354 __sk_dst_set(sk, dst); 1355 spin_unlock(&sk->sk_dst_lock); 1356 } 1357 1358 static inline void 1359 __sk_dst_reset(struct sock *sk) 1360 { 1361 __sk_dst_set(sk, NULL); 1362 } 1363 1364 static inline void 1365 sk_dst_reset(struct sock *sk) 1366 { 1367 spin_lock(&sk->sk_dst_lock); 1368 __sk_dst_reset(sk); 1369 spin_unlock(&sk->sk_dst_lock); 1370 } 1371 1372 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1373 1374 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1375 1376 static inline int sk_can_gso(const struct sock *sk) 1377 { 1378 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1379 } 1380 1381 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1382 1383 static inline void sk_nocaps_add(struct sock *sk, int flags) 1384 { 1385 sk->sk_route_nocaps |= flags; 1386 sk->sk_route_caps &= ~flags; 1387 } 1388 1389 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1390 struct sk_buff *skb, struct page *page, 1391 int off, int copy) 1392 { 1393 if (skb->ip_summed == CHECKSUM_NONE) { 1394 int err = 0; 1395 __wsum csum = csum_and_copy_from_user(from, 1396 page_address(page) + off, 1397 copy, 0, &err); 1398 if (err) 1399 return err; 1400 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1401 } else if (copy_from_user(page_address(page) + off, from, copy)) 1402 return -EFAULT; 1403 1404 skb->len += copy; 1405 skb->data_len += copy; 1406 skb->truesize += copy; 1407 sk->sk_wmem_queued += copy; 1408 sk_mem_charge(sk, copy); 1409 return 0; 1410 } 1411 1412 /** 1413 * sk_wmem_alloc_get - returns write allocations 1414 * @sk: socket 1415 * 1416 * Returns sk_wmem_alloc minus initial offset of one 1417 */ 1418 static inline int sk_wmem_alloc_get(const struct sock *sk) 1419 { 1420 return atomic_read(&sk->sk_wmem_alloc) - 1; 1421 } 1422 1423 /** 1424 * sk_rmem_alloc_get - returns read allocations 1425 * @sk: socket 1426 * 1427 * Returns sk_rmem_alloc 1428 */ 1429 static inline int sk_rmem_alloc_get(const struct sock *sk) 1430 { 1431 return atomic_read(&sk->sk_rmem_alloc); 1432 } 1433 1434 /** 1435 * sk_has_allocations - check if allocations are outstanding 1436 * @sk: socket 1437 * 1438 * Returns true if socket has write or read allocations 1439 */ 1440 static inline int sk_has_allocations(const struct sock *sk) 1441 { 1442 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1443 } 1444 1445 /** 1446 * wq_has_sleeper - check if there are any waiting processes 1447 * @wq: struct socket_wq 1448 * 1449 * Returns true if socket_wq has waiting processes 1450 * 1451 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1452 * barrier call. They were added due to the race found within the tcp code. 1453 * 1454 * Consider following tcp code paths: 1455 * 1456 * CPU1 CPU2 1457 * 1458 * sys_select receive packet 1459 * ... ... 1460 * __add_wait_queue update tp->rcv_nxt 1461 * ... ... 1462 * tp->rcv_nxt check sock_def_readable 1463 * ... { 1464 * schedule rcu_read_lock(); 1465 * wq = rcu_dereference(sk->sk_wq); 1466 * if (wq && waitqueue_active(&wq->wait)) 1467 * wake_up_interruptible(&wq->wait) 1468 * ... 1469 * } 1470 * 1471 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1472 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1473 * could then endup calling schedule and sleep forever if there are no more 1474 * data on the socket. 1475 * 1476 */ 1477 static inline bool wq_has_sleeper(struct socket_wq *wq) 1478 { 1479 1480 /* 1481 * We need to be sure we are in sync with the 1482 * add_wait_queue modifications to the wait queue. 1483 * 1484 * This memory barrier is paired in the sock_poll_wait. 1485 */ 1486 smp_mb(); 1487 return wq && waitqueue_active(&wq->wait); 1488 } 1489 1490 /** 1491 * sock_poll_wait - place memory barrier behind the poll_wait call. 1492 * @filp: file 1493 * @wait_address: socket wait queue 1494 * @p: poll_table 1495 * 1496 * See the comments in the wq_has_sleeper function. 1497 */ 1498 static inline void sock_poll_wait(struct file *filp, 1499 wait_queue_head_t *wait_address, poll_table *p) 1500 { 1501 if (p && wait_address) { 1502 poll_wait(filp, wait_address, p); 1503 /* 1504 * We need to be sure we are in sync with the 1505 * socket flags modification. 1506 * 1507 * This memory barrier is paired in the wq_has_sleeper. 1508 */ 1509 smp_mb(); 1510 } 1511 } 1512 1513 /* 1514 * Queue a received datagram if it will fit. Stream and sequenced 1515 * protocols can't normally use this as they need to fit buffers in 1516 * and play with them. 1517 * 1518 * Inlined as it's very short and called for pretty much every 1519 * packet ever received. 1520 */ 1521 1522 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1523 { 1524 skb_orphan(skb); 1525 skb->sk = sk; 1526 skb->destructor = sock_wfree; 1527 /* 1528 * We used to take a refcount on sk, but following operation 1529 * is enough to guarantee sk_free() wont free this sock until 1530 * all in-flight packets are completed 1531 */ 1532 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1533 } 1534 1535 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1536 { 1537 skb_orphan(skb); 1538 skb->sk = sk; 1539 skb->destructor = sock_rfree; 1540 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1541 sk_mem_charge(sk, skb->truesize); 1542 } 1543 1544 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1545 unsigned long expires); 1546 1547 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer); 1548 1549 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1550 1551 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1552 1553 /* 1554 * Recover an error report and clear atomically 1555 */ 1556 1557 static inline int sock_error(struct sock *sk) 1558 { 1559 int err; 1560 if (likely(!sk->sk_err)) 1561 return 0; 1562 err = xchg(&sk->sk_err, 0); 1563 return -err; 1564 } 1565 1566 static inline unsigned long sock_wspace(struct sock *sk) 1567 { 1568 int amt = 0; 1569 1570 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1571 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 1572 if (amt < 0) 1573 amt = 0; 1574 } 1575 return amt; 1576 } 1577 1578 static inline void sk_wake_async(struct sock *sk, int how, int band) 1579 { 1580 if (sock_flag(sk, SOCK_FASYNC)) 1581 sock_wake_async(sk->sk_socket, how, band); 1582 } 1583 1584 #define SOCK_MIN_SNDBUF 2048 1585 /* 1586 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need 1587 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak 1588 */ 1589 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff)) 1590 1591 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 1592 { 1593 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 1594 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 1595 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF); 1596 } 1597 } 1598 1599 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 1600 1601 static inline struct page *sk_stream_alloc_page(struct sock *sk) 1602 { 1603 struct page *page = NULL; 1604 1605 page = alloc_pages(sk->sk_allocation, 0); 1606 if (!page) { 1607 sk->sk_prot->enter_memory_pressure(sk); 1608 sk_stream_moderate_sndbuf(sk); 1609 } 1610 return page; 1611 } 1612 1613 /* 1614 * Default write policy as shown to user space via poll/select/SIGIO 1615 */ 1616 static inline int sock_writeable(const struct sock *sk) 1617 { 1618 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 1619 } 1620 1621 static inline gfp_t gfp_any(void) 1622 { 1623 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 1624 } 1625 1626 static inline long sock_rcvtimeo(const struct sock *sk, int noblock) 1627 { 1628 return noblock ? 0 : sk->sk_rcvtimeo; 1629 } 1630 1631 static inline long sock_sndtimeo(const struct sock *sk, int noblock) 1632 { 1633 return noblock ? 0 : sk->sk_sndtimeo; 1634 } 1635 1636 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 1637 { 1638 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 1639 } 1640 1641 /* Alas, with timeout socket operations are not restartable. 1642 * Compare this to poll(). 1643 */ 1644 static inline int sock_intr_errno(long timeo) 1645 { 1646 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 1647 } 1648 1649 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 1650 struct sk_buff *skb); 1651 1652 static __inline__ void 1653 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 1654 { 1655 ktime_t kt = skb->tstamp; 1656 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 1657 1658 /* 1659 * generate control messages if 1660 * - receive time stamping in software requested (SOCK_RCVTSTAMP 1661 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 1662 * - software time stamp available and wanted 1663 * (SOCK_TIMESTAMPING_SOFTWARE) 1664 * - hardware time stamps available and wanted 1665 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 1666 * SOCK_TIMESTAMPING_RAW_HARDWARE) 1667 */ 1668 if (sock_flag(sk, SOCK_RCVTSTAMP) || 1669 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 1670 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 1671 (hwtstamps->hwtstamp.tv64 && 1672 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 1673 (hwtstamps->syststamp.tv64 && 1674 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 1675 __sock_recv_timestamp(msg, sk, skb); 1676 else 1677 sk->sk_stamp = kt; 1678 } 1679 1680 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 1681 struct sk_buff *skb); 1682 1683 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 1684 struct sk_buff *skb) 1685 { 1686 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 1687 (1UL << SOCK_RCVTSTAMP) | \ 1688 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 1689 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 1690 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 1691 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 1692 1693 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 1694 __sock_recv_ts_and_drops(msg, sk, skb); 1695 else 1696 sk->sk_stamp = skb->tstamp; 1697 } 1698 1699 /** 1700 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 1701 * @sk: socket sending this packet 1702 * @tx_flags: filled with instructions for time stamping 1703 * 1704 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if 1705 * parameters are invalid. 1706 */ 1707 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 1708 1709 /** 1710 * sk_eat_skb - Release a skb if it is no longer needed 1711 * @sk: socket to eat this skb from 1712 * @skb: socket buffer to eat 1713 * @copied_early: flag indicating whether DMA operations copied this data early 1714 * 1715 * This routine must be called with interrupts disabled or with the socket 1716 * locked so that the sk_buff queue operation is ok. 1717 */ 1718 #ifdef CONFIG_NET_DMA 1719 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 1720 { 1721 __skb_unlink(skb, &sk->sk_receive_queue); 1722 if (!copied_early) 1723 __kfree_skb(skb); 1724 else 1725 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 1726 } 1727 #else 1728 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 1729 { 1730 __skb_unlink(skb, &sk->sk_receive_queue); 1731 __kfree_skb(skb); 1732 } 1733 #endif 1734 1735 static inline 1736 struct net *sock_net(const struct sock *sk) 1737 { 1738 return read_pnet(&sk->sk_net); 1739 } 1740 1741 static inline 1742 void sock_net_set(struct sock *sk, struct net *net) 1743 { 1744 write_pnet(&sk->sk_net, net); 1745 } 1746 1747 /* 1748 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 1749 * They should not hold a referrence to a namespace in order to allow 1750 * to stop it. 1751 * Sockets after sk_change_net should be released using sk_release_kernel 1752 */ 1753 static inline void sk_change_net(struct sock *sk, struct net *net) 1754 { 1755 put_net(sock_net(sk)); 1756 sock_net_set(sk, hold_net(net)); 1757 } 1758 1759 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 1760 { 1761 if (unlikely(skb->sk)) { 1762 struct sock *sk = skb->sk; 1763 1764 skb->destructor = NULL; 1765 skb->sk = NULL; 1766 return sk; 1767 } 1768 return NULL; 1769 } 1770 1771 extern void sock_enable_timestamp(struct sock *sk, int flag); 1772 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 1773 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 1774 1775 /* 1776 * Enable debug/info messages 1777 */ 1778 extern int net_msg_warn; 1779 #define NETDEBUG(fmt, args...) \ 1780 do { if (net_msg_warn) printk(fmt,##args); } while (0) 1781 1782 #define LIMIT_NETDEBUG(fmt, args...) \ 1783 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 1784 1785 extern __u32 sysctl_wmem_max; 1786 extern __u32 sysctl_rmem_max; 1787 1788 extern void sk_init(void); 1789 1790 extern int sysctl_optmem_max; 1791 1792 extern __u32 sysctl_wmem_default; 1793 extern __u32 sysctl_rmem_default; 1794 1795 #endif /* _SOCK_H */ 1796