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