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 pid *sk_peer_pid; 299 const struct cred *sk_peer_cred; 300 long sk_rcvtimeo; 301 long sk_sndtimeo; 302 struct sk_filter *sk_filter; 303 void *sk_protinfo; 304 struct timer_list sk_timer; 305 ktime_t sk_stamp; 306 struct socket *sk_socket; 307 void *sk_user_data; 308 struct page *sk_sndmsg_page; 309 struct sk_buff *sk_send_head; 310 __u32 sk_sndmsg_off; 311 int sk_write_pending; 312 #ifdef CONFIG_SECURITY 313 void *sk_security; 314 #endif 315 __u32 sk_mark; 316 u32 sk_classid; 317 void (*sk_state_change)(struct sock *sk); 318 void (*sk_data_ready)(struct sock *sk, int bytes); 319 void (*sk_write_space)(struct sock *sk); 320 void (*sk_error_report)(struct sock *sk); 321 int (*sk_backlog_rcv)(struct sock *sk, 322 struct sk_buff *skb); 323 void (*sk_destruct)(struct sock *sk); 324 }; 325 326 /* 327 * Hashed lists helper routines 328 */ 329 static inline struct sock *sk_entry(const struct hlist_node *node) 330 { 331 return hlist_entry(node, struct sock, sk_node); 332 } 333 334 static inline struct sock *__sk_head(const struct hlist_head *head) 335 { 336 return hlist_entry(head->first, struct sock, sk_node); 337 } 338 339 static inline struct sock *sk_head(const struct hlist_head *head) 340 { 341 return hlist_empty(head) ? NULL : __sk_head(head); 342 } 343 344 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 345 { 346 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 347 } 348 349 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 350 { 351 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 352 } 353 354 static inline struct sock *sk_next(const struct sock *sk) 355 { 356 return sk->sk_node.next ? 357 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 358 } 359 360 static inline struct sock *sk_nulls_next(const struct sock *sk) 361 { 362 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 363 hlist_nulls_entry(sk->sk_nulls_node.next, 364 struct sock, sk_nulls_node) : 365 NULL; 366 } 367 368 static inline int sk_unhashed(const struct sock *sk) 369 { 370 return hlist_unhashed(&sk->sk_node); 371 } 372 373 static inline int sk_hashed(const struct sock *sk) 374 { 375 return !sk_unhashed(sk); 376 } 377 378 static __inline__ void sk_node_init(struct hlist_node *node) 379 { 380 node->pprev = NULL; 381 } 382 383 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node) 384 { 385 node->pprev = NULL; 386 } 387 388 static __inline__ void __sk_del_node(struct sock *sk) 389 { 390 __hlist_del(&sk->sk_node); 391 } 392 393 /* NB: equivalent to hlist_del_init_rcu */ 394 static __inline__ int __sk_del_node_init(struct sock *sk) 395 { 396 if (sk_hashed(sk)) { 397 __sk_del_node(sk); 398 sk_node_init(&sk->sk_node); 399 return 1; 400 } 401 return 0; 402 } 403 404 /* Grab socket reference count. This operation is valid only 405 when sk is ALREADY grabbed f.e. it is found in hash table 406 or a list and the lookup is made under lock preventing hash table 407 modifications. 408 */ 409 410 static inline void sock_hold(struct sock *sk) 411 { 412 atomic_inc(&sk->sk_refcnt); 413 } 414 415 /* Ungrab socket in the context, which assumes that socket refcnt 416 cannot hit zero, f.e. it is true in context of any socketcall. 417 */ 418 static inline void __sock_put(struct sock *sk) 419 { 420 atomic_dec(&sk->sk_refcnt); 421 } 422 423 static __inline__ int sk_del_node_init(struct sock *sk) 424 { 425 int rc = __sk_del_node_init(sk); 426 427 if (rc) { 428 /* paranoid for a while -acme */ 429 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 430 __sock_put(sk); 431 } 432 return rc; 433 } 434 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 435 436 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk) 437 { 438 if (sk_hashed(sk)) { 439 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 440 return 1; 441 } 442 return 0; 443 } 444 445 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk) 446 { 447 int rc = __sk_nulls_del_node_init_rcu(sk); 448 449 if (rc) { 450 /* paranoid for a while -acme */ 451 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 452 __sock_put(sk); 453 } 454 return rc; 455 } 456 457 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list) 458 { 459 hlist_add_head(&sk->sk_node, list); 460 } 461 462 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list) 463 { 464 sock_hold(sk); 465 __sk_add_node(sk, list); 466 } 467 468 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 469 { 470 sock_hold(sk); 471 hlist_add_head_rcu(&sk->sk_node, list); 472 } 473 474 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 475 { 476 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 477 } 478 479 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 480 { 481 sock_hold(sk); 482 __sk_nulls_add_node_rcu(sk, list); 483 } 484 485 static __inline__ void __sk_del_bind_node(struct sock *sk) 486 { 487 __hlist_del(&sk->sk_bind_node); 488 } 489 490 static __inline__ void sk_add_bind_node(struct sock *sk, 491 struct hlist_head *list) 492 { 493 hlist_add_head(&sk->sk_bind_node, list); 494 } 495 496 #define sk_for_each(__sk, node, list) \ 497 hlist_for_each_entry(__sk, node, list, sk_node) 498 #define sk_for_each_rcu(__sk, node, list) \ 499 hlist_for_each_entry_rcu(__sk, node, list, sk_node) 500 #define sk_nulls_for_each(__sk, node, list) \ 501 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 502 #define sk_nulls_for_each_rcu(__sk, node, list) \ 503 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 504 #define sk_for_each_from(__sk, node) \ 505 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 506 hlist_for_each_entry_from(__sk, node, sk_node) 507 #define sk_nulls_for_each_from(__sk, node) \ 508 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 509 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 510 #define sk_for_each_continue(__sk, node) \ 511 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 512 hlist_for_each_entry_continue(__sk, node, sk_node) 513 #define sk_for_each_safe(__sk, node, tmp, list) \ 514 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node) 515 #define sk_for_each_bound(__sk, node, list) \ 516 hlist_for_each_entry(__sk, node, list, sk_bind_node) 517 518 /* Sock flags */ 519 enum sock_flags { 520 SOCK_DEAD, 521 SOCK_DONE, 522 SOCK_URGINLINE, 523 SOCK_KEEPOPEN, 524 SOCK_LINGER, 525 SOCK_DESTROY, 526 SOCK_BROADCAST, 527 SOCK_TIMESTAMP, 528 SOCK_ZAPPED, 529 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 530 SOCK_DBG, /* %SO_DEBUG setting */ 531 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 532 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 533 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 534 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 535 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 536 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 537 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 538 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 539 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 540 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 541 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 542 SOCK_FASYNC, /* fasync() active */ 543 SOCK_RXQ_OVFL, 544 }; 545 546 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 547 { 548 nsk->sk_flags = osk->sk_flags; 549 } 550 551 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 552 { 553 __set_bit(flag, &sk->sk_flags); 554 } 555 556 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 557 { 558 __clear_bit(flag, &sk->sk_flags); 559 } 560 561 static inline int sock_flag(struct sock *sk, enum sock_flags flag) 562 { 563 return test_bit(flag, &sk->sk_flags); 564 } 565 566 static inline void sk_acceptq_removed(struct sock *sk) 567 { 568 sk->sk_ack_backlog--; 569 } 570 571 static inline void sk_acceptq_added(struct sock *sk) 572 { 573 sk->sk_ack_backlog++; 574 } 575 576 static inline int sk_acceptq_is_full(struct sock *sk) 577 { 578 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 579 } 580 581 /* 582 * Compute minimal free write space needed to queue new packets. 583 */ 584 static inline int sk_stream_min_wspace(struct sock *sk) 585 { 586 return sk->sk_wmem_queued >> 1; 587 } 588 589 static inline int sk_stream_wspace(struct sock *sk) 590 { 591 return sk->sk_sndbuf - sk->sk_wmem_queued; 592 } 593 594 extern void sk_stream_write_space(struct sock *sk); 595 596 static inline int sk_stream_memory_free(struct sock *sk) 597 { 598 return sk->sk_wmem_queued < sk->sk_sndbuf; 599 } 600 601 /* OOB backlog add */ 602 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 603 { 604 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 605 skb_dst_force(skb); 606 607 if (!sk->sk_backlog.tail) 608 sk->sk_backlog.head = skb; 609 else 610 sk->sk_backlog.tail->next = skb; 611 612 sk->sk_backlog.tail = skb; 613 skb->next = NULL; 614 } 615 616 /* 617 * Take into account size of receive queue and backlog queue 618 */ 619 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb) 620 { 621 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 622 623 return qsize + skb->truesize > sk->sk_rcvbuf; 624 } 625 626 /* The per-socket spinlock must be held here. */ 627 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb) 628 { 629 if (sk_rcvqueues_full(sk, skb)) 630 return -ENOBUFS; 631 632 __sk_add_backlog(sk, skb); 633 sk->sk_backlog.len += skb->truesize; 634 return 0; 635 } 636 637 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 638 { 639 return sk->sk_backlog_rcv(sk, skb); 640 } 641 642 static inline void sock_rps_record_flow(const struct sock *sk) 643 { 644 #ifdef CONFIG_RPS 645 struct rps_sock_flow_table *sock_flow_table; 646 647 rcu_read_lock(); 648 sock_flow_table = rcu_dereference(rps_sock_flow_table); 649 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 650 rcu_read_unlock(); 651 #endif 652 } 653 654 static inline void sock_rps_reset_flow(const struct sock *sk) 655 { 656 #ifdef CONFIG_RPS 657 struct rps_sock_flow_table *sock_flow_table; 658 659 rcu_read_lock(); 660 sock_flow_table = rcu_dereference(rps_sock_flow_table); 661 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 662 rcu_read_unlock(); 663 #endif 664 } 665 666 static inline void sock_rps_save_rxhash(struct sock *sk, u32 rxhash) 667 { 668 #ifdef CONFIG_RPS 669 if (unlikely(sk->sk_rxhash != rxhash)) { 670 sock_rps_reset_flow(sk); 671 sk->sk_rxhash = rxhash; 672 } 673 #endif 674 } 675 676 #define sk_wait_event(__sk, __timeo, __condition) \ 677 ({ int __rc; \ 678 release_sock(__sk); \ 679 __rc = __condition; \ 680 if (!__rc) { \ 681 *(__timeo) = schedule_timeout(*(__timeo)); \ 682 } \ 683 lock_sock(__sk); \ 684 __rc = __condition; \ 685 __rc; \ 686 }) 687 688 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 689 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 690 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 691 extern int sk_stream_error(struct sock *sk, int flags, int err); 692 extern void sk_stream_kill_queues(struct sock *sk); 693 694 extern int sk_wait_data(struct sock *sk, long *timeo); 695 696 struct request_sock_ops; 697 struct timewait_sock_ops; 698 struct inet_hashinfo; 699 struct raw_hashinfo; 700 701 /* Networking protocol blocks we attach to sockets. 702 * socket layer -> transport layer interface 703 * transport -> network interface is defined by struct inet_proto 704 */ 705 struct proto { 706 void (*close)(struct sock *sk, 707 long timeout); 708 int (*connect)(struct sock *sk, 709 struct sockaddr *uaddr, 710 int addr_len); 711 int (*disconnect)(struct sock *sk, int flags); 712 713 struct sock * (*accept) (struct sock *sk, int flags, int *err); 714 715 int (*ioctl)(struct sock *sk, int cmd, 716 unsigned long arg); 717 int (*init)(struct sock *sk); 718 void (*destroy)(struct sock *sk); 719 void (*shutdown)(struct sock *sk, int how); 720 int (*setsockopt)(struct sock *sk, int level, 721 int optname, char __user *optval, 722 unsigned int optlen); 723 int (*getsockopt)(struct sock *sk, int level, 724 int optname, char __user *optval, 725 int __user *option); 726 #ifdef CONFIG_COMPAT 727 int (*compat_setsockopt)(struct sock *sk, 728 int level, 729 int optname, char __user *optval, 730 unsigned int optlen); 731 int (*compat_getsockopt)(struct sock *sk, 732 int level, 733 int optname, char __user *optval, 734 int __user *option); 735 #endif 736 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 737 struct msghdr *msg, size_t len); 738 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 739 struct msghdr *msg, 740 size_t len, int noblock, int flags, 741 int *addr_len); 742 int (*sendpage)(struct sock *sk, struct page *page, 743 int offset, size_t size, int flags); 744 int (*bind)(struct sock *sk, 745 struct sockaddr *uaddr, int addr_len); 746 747 int (*backlog_rcv) (struct sock *sk, 748 struct sk_buff *skb); 749 750 /* Keeping track of sk's, looking them up, and port selection methods. */ 751 void (*hash)(struct sock *sk); 752 void (*unhash)(struct sock *sk); 753 int (*get_port)(struct sock *sk, unsigned short snum); 754 755 /* Keeping track of sockets in use */ 756 #ifdef CONFIG_PROC_FS 757 unsigned int inuse_idx; 758 #endif 759 760 /* Memory pressure */ 761 void (*enter_memory_pressure)(struct sock *sk); 762 atomic_t *memory_allocated; /* Current allocated memory. */ 763 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 764 /* 765 * Pressure flag: try to collapse. 766 * Technical note: it is used by multiple contexts non atomically. 767 * All the __sk_mem_schedule() is of this nature: accounting 768 * is strict, actions are advisory and have some latency. 769 */ 770 int *memory_pressure; 771 int *sysctl_mem; 772 int *sysctl_wmem; 773 int *sysctl_rmem; 774 int max_header; 775 bool no_autobind; 776 777 struct kmem_cache *slab; 778 unsigned int obj_size; 779 int slab_flags; 780 781 struct percpu_counter *orphan_count; 782 783 struct request_sock_ops *rsk_prot; 784 struct timewait_sock_ops *twsk_prot; 785 786 union { 787 struct inet_hashinfo *hashinfo; 788 struct udp_table *udp_table; 789 struct raw_hashinfo *raw_hash; 790 } h; 791 792 struct module *owner; 793 794 char name[32]; 795 796 struct list_head node; 797 #ifdef SOCK_REFCNT_DEBUG 798 atomic_t socks; 799 #endif 800 }; 801 802 extern int proto_register(struct proto *prot, int alloc_slab); 803 extern void proto_unregister(struct proto *prot); 804 805 #ifdef SOCK_REFCNT_DEBUG 806 static inline void sk_refcnt_debug_inc(struct sock *sk) 807 { 808 atomic_inc(&sk->sk_prot->socks); 809 } 810 811 static inline void sk_refcnt_debug_dec(struct sock *sk) 812 { 813 atomic_dec(&sk->sk_prot->socks); 814 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 815 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 816 } 817 818 static inline void sk_refcnt_debug_release(const struct sock *sk) 819 { 820 if (atomic_read(&sk->sk_refcnt) != 1) 821 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 822 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 823 } 824 #else /* SOCK_REFCNT_DEBUG */ 825 #define sk_refcnt_debug_inc(sk) do { } while (0) 826 #define sk_refcnt_debug_dec(sk) do { } while (0) 827 #define sk_refcnt_debug_release(sk) do { } while (0) 828 #endif /* SOCK_REFCNT_DEBUG */ 829 830 831 #ifdef CONFIG_PROC_FS 832 /* Called with local bh disabled */ 833 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 834 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 835 #else 836 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot, 837 int inc) 838 { 839 } 840 #endif 841 842 843 /* With per-bucket locks this operation is not-atomic, so that 844 * this version is not worse. 845 */ 846 static inline void __sk_prot_rehash(struct sock *sk) 847 { 848 sk->sk_prot->unhash(sk); 849 sk->sk_prot->hash(sk); 850 } 851 852 /* About 10 seconds */ 853 #define SOCK_DESTROY_TIME (10*HZ) 854 855 /* Sockets 0-1023 can't be bound to unless you are superuser */ 856 #define PROT_SOCK 1024 857 858 #define SHUTDOWN_MASK 3 859 #define RCV_SHUTDOWN 1 860 #define SEND_SHUTDOWN 2 861 862 #define SOCK_SNDBUF_LOCK 1 863 #define SOCK_RCVBUF_LOCK 2 864 #define SOCK_BINDADDR_LOCK 4 865 #define SOCK_BINDPORT_LOCK 8 866 867 /* sock_iocb: used to kick off async processing of socket ios */ 868 struct sock_iocb { 869 struct list_head list; 870 871 int flags; 872 int size; 873 struct socket *sock; 874 struct sock *sk; 875 struct scm_cookie *scm; 876 struct msghdr *msg, async_msg; 877 struct kiocb *kiocb; 878 }; 879 880 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 881 { 882 return (struct sock_iocb *)iocb->private; 883 } 884 885 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 886 { 887 return si->kiocb; 888 } 889 890 struct socket_alloc { 891 struct socket socket; 892 struct inode vfs_inode; 893 }; 894 895 static inline struct socket *SOCKET_I(struct inode *inode) 896 { 897 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 898 } 899 900 static inline struct inode *SOCK_INODE(struct socket *socket) 901 { 902 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 903 } 904 905 /* 906 * Functions for memory accounting 907 */ 908 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 909 extern void __sk_mem_reclaim(struct sock *sk); 910 911 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 912 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 913 #define SK_MEM_SEND 0 914 #define SK_MEM_RECV 1 915 916 static inline int sk_mem_pages(int amt) 917 { 918 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 919 } 920 921 static inline int sk_has_account(struct sock *sk) 922 { 923 /* return true if protocol supports memory accounting */ 924 return !!sk->sk_prot->memory_allocated; 925 } 926 927 static inline int sk_wmem_schedule(struct sock *sk, int size) 928 { 929 if (!sk_has_account(sk)) 930 return 1; 931 return size <= sk->sk_forward_alloc || 932 __sk_mem_schedule(sk, size, SK_MEM_SEND); 933 } 934 935 static inline int sk_rmem_schedule(struct sock *sk, int size) 936 { 937 if (!sk_has_account(sk)) 938 return 1; 939 return size <= sk->sk_forward_alloc || 940 __sk_mem_schedule(sk, size, SK_MEM_RECV); 941 } 942 943 static inline void sk_mem_reclaim(struct sock *sk) 944 { 945 if (!sk_has_account(sk)) 946 return; 947 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 948 __sk_mem_reclaim(sk); 949 } 950 951 static inline void sk_mem_reclaim_partial(struct sock *sk) 952 { 953 if (!sk_has_account(sk)) 954 return; 955 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 956 __sk_mem_reclaim(sk); 957 } 958 959 static inline void sk_mem_charge(struct sock *sk, int size) 960 { 961 if (!sk_has_account(sk)) 962 return; 963 sk->sk_forward_alloc -= size; 964 } 965 966 static inline void sk_mem_uncharge(struct sock *sk, int size) 967 { 968 if (!sk_has_account(sk)) 969 return; 970 sk->sk_forward_alloc += size; 971 } 972 973 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 974 { 975 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 976 sk->sk_wmem_queued -= skb->truesize; 977 sk_mem_uncharge(sk, skb->truesize); 978 __kfree_skb(skb); 979 } 980 981 /* Used by processes to "lock" a socket state, so that 982 * interrupts and bottom half handlers won't change it 983 * from under us. It essentially blocks any incoming 984 * packets, so that we won't get any new data or any 985 * packets that change the state of the socket. 986 * 987 * While locked, BH processing will add new packets to 988 * the backlog queue. This queue is processed by the 989 * owner of the socket lock right before it is released. 990 * 991 * Since ~2.3.5 it is also exclusive sleep lock serializing 992 * accesses from user process context. 993 */ 994 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 995 996 /* 997 * Macro so as to not evaluate some arguments when 998 * lockdep is not enabled. 999 * 1000 * Mark both the sk_lock and the sk_lock.slock as a 1001 * per-address-family lock class. 1002 */ 1003 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1004 do { \ 1005 sk->sk_lock.owned = 0; \ 1006 init_waitqueue_head(&sk->sk_lock.wq); \ 1007 spin_lock_init(&(sk)->sk_lock.slock); \ 1008 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1009 sizeof((sk)->sk_lock)); \ 1010 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1011 (skey), (sname)); \ 1012 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1013 } while (0) 1014 1015 extern void lock_sock_nested(struct sock *sk, int subclass); 1016 1017 static inline void lock_sock(struct sock *sk) 1018 { 1019 lock_sock_nested(sk, 0); 1020 } 1021 1022 extern void release_sock(struct sock *sk); 1023 1024 /* BH context may only use the following locking interface. */ 1025 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1026 #define bh_lock_sock_nested(__sk) \ 1027 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1028 SINGLE_DEPTH_NESTING) 1029 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1030 1031 extern bool lock_sock_fast(struct sock *sk); 1032 /** 1033 * unlock_sock_fast - complement of lock_sock_fast 1034 * @sk: socket 1035 * @slow: slow mode 1036 * 1037 * fast unlock socket for user context. 1038 * If slow mode is on, we call regular release_sock() 1039 */ 1040 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1041 { 1042 if (slow) 1043 release_sock(sk); 1044 else 1045 spin_unlock_bh(&sk->sk_lock.slock); 1046 } 1047 1048 1049 extern struct sock *sk_alloc(struct net *net, int family, 1050 gfp_t priority, 1051 struct proto *prot); 1052 extern void sk_free(struct sock *sk); 1053 extern void sk_release_kernel(struct sock *sk); 1054 extern struct sock *sk_clone(const struct sock *sk, 1055 const gfp_t priority); 1056 1057 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1058 unsigned long size, int force, 1059 gfp_t priority); 1060 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1061 unsigned long size, int force, 1062 gfp_t priority); 1063 extern void sock_wfree(struct sk_buff *skb); 1064 extern void sock_rfree(struct sk_buff *skb); 1065 1066 extern int sock_setsockopt(struct socket *sock, int level, 1067 int op, char __user *optval, 1068 unsigned int optlen); 1069 1070 extern int sock_getsockopt(struct socket *sock, int level, 1071 int op, char __user *optval, 1072 int __user *optlen); 1073 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1074 unsigned long size, 1075 int noblock, 1076 int *errcode); 1077 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1078 unsigned long header_len, 1079 unsigned long data_len, 1080 int noblock, 1081 int *errcode); 1082 extern void *sock_kmalloc(struct sock *sk, int size, 1083 gfp_t priority); 1084 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1085 extern void sk_send_sigurg(struct sock *sk); 1086 1087 #ifdef CONFIG_CGROUPS 1088 extern void sock_update_classid(struct sock *sk); 1089 #else 1090 static inline void sock_update_classid(struct sock *sk) 1091 { 1092 } 1093 #endif 1094 1095 /* 1096 * Functions to fill in entries in struct proto_ops when a protocol 1097 * does not implement a particular function. 1098 */ 1099 extern int sock_no_bind(struct socket *, 1100 struct sockaddr *, int); 1101 extern int sock_no_connect(struct socket *, 1102 struct sockaddr *, int, int); 1103 extern int sock_no_socketpair(struct socket *, 1104 struct socket *); 1105 extern int sock_no_accept(struct socket *, 1106 struct socket *, int); 1107 extern int sock_no_getname(struct socket *, 1108 struct sockaddr *, int *, int); 1109 extern unsigned int sock_no_poll(struct file *, struct socket *, 1110 struct poll_table_struct *); 1111 extern int sock_no_ioctl(struct socket *, unsigned int, 1112 unsigned long); 1113 extern int sock_no_listen(struct socket *, int); 1114 extern int sock_no_shutdown(struct socket *, int); 1115 extern int sock_no_getsockopt(struct socket *, int , int, 1116 char __user *, int __user *); 1117 extern int sock_no_setsockopt(struct socket *, int, int, 1118 char __user *, unsigned int); 1119 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1120 struct msghdr *, size_t); 1121 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1122 struct msghdr *, size_t, int); 1123 extern int sock_no_mmap(struct file *file, 1124 struct socket *sock, 1125 struct vm_area_struct *vma); 1126 extern ssize_t sock_no_sendpage(struct socket *sock, 1127 struct page *page, 1128 int offset, size_t size, 1129 int flags); 1130 1131 /* 1132 * Functions to fill in entries in struct proto_ops when a protocol 1133 * uses the inet style. 1134 */ 1135 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1136 char __user *optval, int __user *optlen); 1137 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1138 struct msghdr *msg, size_t size, int flags); 1139 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1140 char __user *optval, unsigned int optlen); 1141 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1142 int optname, char __user *optval, int __user *optlen); 1143 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1144 int optname, char __user *optval, unsigned int optlen); 1145 1146 extern void sk_common_release(struct sock *sk); 1147 1148 /* 1149 * Default socket callbacks and setup code 1150 */ 1151 1152 /* Initialise core socket variables */ 1153 extern void sock_init_data(struct socket *sock, struct sock *sk); 1154 1155 /** 1156 * sk_filter_release - release a socket filter 1157 * @fp: filter to remove 1158 * 1159 * Remove a filter from a socket and release its resources. 1160 */ 1161 1162 static inline void sk_filter_release(struct sk_filter *fp) 1163 { 1164 if (atomic_dec_and_test(&fp->refcnt)) 1165 kfree(fp); 1166 } 1167 1168 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1169 { 1170 unsigned int size = sk_filter_len(fp); 1171 1172 atomic_sub(size, &sk->sk_omem_alloc); 1173 sk_filter_release(fp); 1174 } 1175 1176 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1177 { 1178 atomic_inc(&fp->refcnt); 1179 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1180 } 1181 1182 /* 1183 * Socket reference counting postulates. 1184 * 1185 * * Each user of socket SHOULD hold a reference count. 1186 * * Each access point to socket (an hash table bucket, reference from a list, 1187 * running timer, skb in flight MUST hold a reference count. 1188 * * When reference count hits 0, it means it will never increase back. 1189 * * When reference count hits 0, it means that no references from 1190 * outside exist to this socket and current process on current CPU 1191 * is last user and may/should destroy this socket. 1192 * * sk_free is called from any context: process, BH, IRQ. When 1193 * it is called, socket has no references from outside -> sk_free 1194 * may release descendant resources allocated by the socket, but 1195 * to the time when it is called, socket is NOT referenced by any 1196 * hash tables, lists etc. 1197 * * Packets, delivered from outside (from network or from another process) 1198 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1199 * when they sit in queue. Otherwise, packets will leak to hole, when 1200 * socket is looked up by one cpu and unhasing is made by another CPU. 1201 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1202 * (leak to backlog). Packet socket does all the processing inside 1203 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1204 * use separate SMP lock, so that they are prone too. 1205 */ 1206 1207 /* Ungrab socket and destroy it, if it was the last reference. */ 1208 static inline void sock_put(struct sock *sk) 1209 { 1210 if (atomic_dec_and_test(&sk->sk_refcnt)) 1211 sk_free(sk); 1212 } 1213 1214 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1215 const int nested); 1216 1217 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1218 { 1219 sk->sk_tx_queue_mapping = tx_queue; 1220 } 1221 1222 static inline void sk_tx_queue_clear(struct sock *sk) 1223 { 1224 sk->sk_tx_queue_mapping = -1; 1225 } 1226 1227 static inline int sk_tx_queue_get(const struct sock *sk) 1228 { 1229 return sk ? sk->sk_tx_queue_mapping : -1; 1230 } 1231 1232 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1233 { 1234 sk_tx_queue_clear(sk); 1235 sk->sk_socket = sock; 1236 } 1237 1238 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1239 { 1240 return &sk->sk_wq->wait; 1241 } 1242 /* Detach socket from process context. 1243 * Announce socket dead, detach it from wait queue and inode. 1244 * Note that parent inode held reference count on this struct sock, 1245 * we do not release it in this function, because protocol 1246 * probably wants some additional cleanups or even continuing 1247 * to work with this socket (TCP). 1248 */ 1249 static inline void sock_orphan(struct sock *sk) 1250 { 1251 write_lock_bh(&sk->sk_callback_lock); 1252 sock_set_flag(sk, SOCK_DEAD); 1253 sk_set_socket(sk, NULL); 1254 sk->sk_wq = NULL; 1255 write_unlock_bh(&sk->sk_callback_lock); 1256 } 1257 1258 static inline void sock_graft(struct sock *sk, struct socket *parent) 1259 { 1260 write_lock_bh(&sk->sk_callback_lock); 1261 rcu_assign_pointer(sk->sk_wq, parent->wq); 1262 parent->sk = sk; 1263 sk_set_socket(sk, parent); 1264 security_sock_graft(sk, parent); 1265 write_unlock_bh(&sk->sk_callback_lock); 1266 } 1267 1268 extern int sock_i_uid(struct sock *sk); 1269 extern unsigned long sock_i_ino(struct sock *sk); 1270 1271 static inline struct dst_entry * 1272 __sk_dst_get(struct sock *sk) 1273 { 1274 return rcu_dereference_check(sk->sk_dst_cache, rcu_read_lock_held() || 1275 sock_owned_by_user(sk) || 1276 lockdep_is_held(&sk->sk_lock.slock)); 1277 } 1278 1279 static inline struct dst_entry * 1280 sk_dst_get(struct sock *sk) 1281 { 1282 struct dst_entry *dst; 1283 1284 rcu_read_lock(); 1285 dst = rcu_dereference(sk->sk_dst_cache); 1286 if (dst) 1287 dst_hold(dst); 1288 rcu_read_unlock(); 1289 return dst; 1290 } 1291 1292 extern void sk_reset_txq(struct sock *sk); 1293 1294 static inline void dst_negative_advice(struct sock *sk) 1295 { 1296 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1297 1298 if (dst && dst->ops->negative_advice) { 1299 ndst = dst->ops->negative_advice(dst); 1300 1301 if (ndst != dst) { 1302 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1303 sk_reset_txq(sk); 1304 } 1305 } 1306 } 1307 1308 static inline void 1309 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1310 { 1311 struct dst_entry *old_dst; 1312 1313 sk_tx_queue_clear(sk); 1314 /* 1315 * This can be called while sk is owned by the caller only, 1316 * with no state that can be checked in a rcu_dereference_check() cond 1317 */ 1318 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1319 rcu_assign_pointer(sk->sk_dst_cache, dst); 1320 dst_release(old_dst); 1321 } 1322 1323 static inline void 1324 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1325 { 1326 spin_lock(&sk->sk_dst_lock); 1327 __sk_dst_set(sk, dst); 1328 spin_unlock(&sk->sk_dst_lock); 1329 } 1330 1331 static inline void 1332 __sk_dst_reset(struct sock *sk) 1333 { 1334 __sk_dst_set(sk, NULL); 1335 } 1336 1337 static inline void 1338 sk_dst_reset(struct sock *sk) 1339 { 1340 spin_lock(&sk->sk_dst_lock); 1341 __sk_dst_reset(sk); 1342 spin_unlock(&sk->sk_dst_lock); 1343 } 1344 1345 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1346 1347 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1348 1349 static inline int sk_can_gso(const struct sock *sk) 1350 { 1351 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1352 } 1353 1354 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1355 1356 static inline void sk_nocaps_add(struct sock *sk, int flags) 1357 { 1358 sk->sk_route_nocaps |= flags; 1359 sk->sk_route_caps &= ~flags; 1360 } 1361 1362 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1363 struct sk_buff *skb, struct page *page, 1364 int off, int copy) 1365 { 1366 if (skb->ip_summed == CHECKSUM_NONE) { 1367 int err = 0; 1368 __wsum csum = csum_and_copy_from_user(from, 1369 page_address(page) + off, 1370 copy, 0, &err); 1371 if (err) 1372 return err; 1373 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1374 } else if (copy_from_user(page_address(page) + off, from, copy)) 1375 return -EFAULT; 1376 1377 skb->len += copy; 1378 skb->data_len += copy; 1379 skb->truesize += copy; 1380 sk->sk_wmem_queued += copy; 1381 sk_mem_charge(sk, copy); 1382 return 0; 1383 } 1384 1385 /** 1386 * sk_wmem_alloc_get - returns write allocations 1387 * @sk: socket 1388 * 1389 * Returns sk_wmem_alloc minus initial offset of one 1390 */ 1391 static inline int sk_wmem_alloc_get(const struct sock *sk) 1392 { 1393 return atomic_read(&sk->sk_wmem_alloc) - 1; 1394 } 1395 1396 /** 1397 * sk_rmem_alloc_get - returns read allocations 1398 * @sk: socket 1399 * 1400 * Returns sk_rmem_alloc 1401 */ 1402 static inline int sk_rmem_alloc_get(const struct sock *sk) 1403 { 1404 return atomic_read(&sk->sk_rmem_alloc); 1405 } 1406 1407 /** 1408 * sk_has_allocations - check if allocations are outstanding 1409 * @sk: socket 1410 * 1411 * Returns true if socket has write or read allocations 1412 */ 1413 static inline int sk_has_allocations(const struct sock *sk) 1414 { 1415 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1416 } 1417 1418 /** 1419 * wq_has_sleeper - check if there are any waiting processes 1420 * @wq: struct socket_wq 1421 * 1422 * Returns true if socket_wq has waiting processes 1423 * 1424 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1425 * barrier call. They were added due to the race found within the tcp code. 1426 * 1427 * Consider following tcp code paths: 1428 * 1429 * CPU1 CPU2 1430 * 1431 * sys_select receive packet 1432 * ... ... 1433 * __add_wait_queue update tp->rcv_nxt 1434 * ... ... 1435 * tp->rcv_nxt check sock_def_readable 1436 * ... { 1437 * schedule rcu_read_lock(); 1438 * wq = rcu_dereference(sk->sk_wq); 1439 * if (wq && waitqueue_active(&wq->wait)) 1440 * wake_up_interruptible(&wq->wait) 1441 * ... 1442 * } 1443 * 1444 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1445 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1446 * could then endup calling schedule and sleep forever if there are no more 1447 * data on the socket. 1448 * 1449 */ 1450 static inline bool wq_has_sleeper(struct socket_wq *wq) 1451 { 1452 1453 /* 1454 * We need to be sure we are in sync with the 1455 * add_wait_queue modifications to the wait queue. 1456 * 1457 * This memory barrier is paired in the sock_poll_wait. 1458 */ 1459 smp_mb(); 1460 return wq && waitqueue_active(&wq->wait); 1461 } 1462 1463 /** 1464 * sock_poll_wait - place memory barrier behind the poll_wait call. 1465 * @filp: file 1466 * @wait_address: socket wait queue 1467 * @p: poll_table 1468 * 1469 * See the comments in the wq_has_sleeper function. 1470 */ 1471 static inline void sock_poll_wait(struct file *filp, 1472 wait_queue_head_t *wait_address, poll_table *p) 1473 { 1474 if (p && wait_address) { 1475 poll_wait(filp, wait_address, p); 1476 /* 1477 * We need to be sure we are in sync with the 1478 * socket flags modification. 1479 * 1480 * This memory barrier is paired in the wq_has_sleeper. 1481 */ 1482 smp_mb(); 1483 } 1484 } 1485 1486 /* 1487 * Queue a received datagram if it will fit. Stream and sequenced 1488 * protocols can't normally use this as they need to fit buffers in 1489 * and play with them. 1490 * 1491 * Inlined as it's very short and called for pretty much every 1492 * packet ever received. 1493 */ 1494 1495 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1496 { 1497 skb_orphan(skb); 1498 skb->sk = sk; 1499 skb->destructor = sock_wfree; 1500 /* 1501 * We used to take a refcount on sk, but following operation 1502 * is enough to guarantee sk_free() wont free this sock until 1503 * all in-flight packets are completed 1504 */ 1505 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1506 } 1507 1508 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1509 { 1510 skb_orphan(skb); 1511 skb->sk = sk; 1512 skb->destructor = sock_rfree; 1513 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1514 sk_mem_charge(sk, skb->truesize); 1515 } 1516 1517 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1518 unsigned long expires); 1519 1520 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer); 1521 1522 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1523 1524 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 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 return read_pnet(&sk->sk_net); 1712 } 1713 1714 static inline 1715 void sock_net_set(struct sock *sk, struct net *net) 1716 { 1717 write_pnet(&sk->sk_net, net); 1718 } 1719 1720 /* 1721 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 1722 * They should not hold a referrence to a namespace in order to allow 1723 * to stop it. 1724 * Sockets after sk_change_net should be released using sk_release_kernel 1725 */ 1726 static inline void sk_change_net(struct sock *sk, struct net *net) 1727 { 1728 put_net(sock_net(sk)); 1729 sock_net_set(sk, hold_net(net)); 1730 } 1731 1732 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 1733 { 1734 if (unlikely(skb->sk)) { 1735 struct sock *sk = skb->sk; 1736 1737 skb->destructor = NULL; 1738 skb->sk = NULL; 1739 return sk; 1740 } 1741 return NULL; 1742 } 1743 1744 extern void sock_enable_timestamp(struct sock *sk, int flag); 1745 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 1746 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 1747 1748 /* 1749 * Enable debug/info messages 1750 */ 1751 extern int net_msg_warn; 1752 #define NETDEBUG(fmt, args...) \ 1753 do { if (net_msg_warn) printk(fmt,##args); } while (0) 1754 1755 #define LIMIT_NETDEBUG(fmt, args...) \ 1756 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 1757 1758 extern __u32 sysctl_wmem_max; 1759 extern __u32 sysctl_rmem_max; 1760 1761 extern void sk_init(void); 1762 1763 extern int sysctl_optmem_max; 1764 1765 extern __u32 sysctl_wmem_default; 1766 extern __u32 sysctl_rmem_default; 1767 1768 #endif /* _SOCK_H */ 1769