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/hardirq.h> 44 #include <linux/kernel.h> 45 #include <linux/list.h> 46 #include <linux/list_nulls.h> 47 #include <linux/timer.h> 48 #include <linux/cache.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 #include <linux/uaccess.h> 56 57 #include <linux/filter.h> 58 #include <linux/rculist_nulls.h> 59 #include <linux/poll.h> 60 61 #include <linux/atomic.h> 62 #include <net/dst.h> 63 #include <net/checksum.h> 64 65 /* 66 * This structure really needs to be cleaned up. 67 * Most of it is for TCP, and not used by any of 68 * the other protocols. 69 */ 70 71 /* Define this to get the SOCK_DBG debugging facility. */ 72 #define SOCK_DEBUGGING 73 #ifdef SOCK_DEBUGGING 74 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 75 printk(KERN_DEBUG msg); } while (0) 76 #else 77 /* Validate arguments and do nothing */ 78 static inline __printf(2, 3) 79 void SOCK_DEBUG(struct sock *sk, const char *msg, ...) 80 { 81 } 82 #endif 83 84 /* This is the per-socket lock. The spinlock provides a synchronization 85 * between user contexts and software interrupt processing, whereas the 86 * mini-semaphore synchronizes multiple users amongst themselves. 87 */ 88 typedef struct { 89 spinlock_t slock; 90 int owned; 91 wait_queue_head_t wq; 92 /* 93 * We express the mutex-alike socket_lock semantics 94 * to the lock validator by explicitly managing 95 * the slock as a lock variant (in addition to 96 * the slock itself): 97 */ 98 #ifdef CONFIG_DEBUG_LOCK_ALLOC 99 struct lockdep_map dep_map; 100 #endif 101 } socket_lock_t; 102 103 struct sock; 104 struct proto; 105 struct net; 106 107 /** 108 * struct sock_common - minimal network layer representation of sockets 109 * @skc_daddr: Foreign IPv4 addr 110 * @skc_rcv_saddr: Bound local IPv4 addr 111 * @skc_hash: hash value used with various protocol lookup tables 112 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 113 * @skc_family: network address family 114 * @skc_state: Connection state 115 * @skc_reuse: %SO_REUSEADDR setting 116 * @skc_bound_dev_if: bound device index if != 0 117 * @skc_bind_node: bind hash linkage for various protocol lookup tables 118 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 119 * @skc_prot: protocol handlers inside a network family 120 * @skc_net: reference to the network namespace of this socket 121 * @skc_node: main hash linkage for various protocol lookup tables 122 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 123 * @skc_tx_queue_mapping: tx queue number for this connection 124 * @skc_refcnt: reference count 125 * 126 * This is the minimal network layer representation of sockets, the header 127 * for struct sock and struct inet_timewait_sock. 128 */ 129 struct sock_common { 130 /* skc_daddr and skc_rcv_saddr must be grouped : 131 * cf INET_MATCH() and INET_TW_MATCH() 132 */ 133 __be32 skc_daddr; 134 __be32 skc_rcv_saddr; 135 136 union { 137 unsigned int skc_hash; 138 __u16 skc_u16hashes[2]; 139 }; 140 unsigned short skc_family; 141 volatile unsigned char skc_state; 142 unsigned char skc_reuse; 143 int skc_bound_dev_if; 144 union { 145 struct hlist_node skc_bind_node; 146 struct hlist_nulls_node skc_portaddr_node; 147 }; 148 struct proto *skc_prot; 149 #ifdef CONFIG_NET_NS 150 struct net *skc_net; 151 #endif 152 /* 153 * fields between dontcopy_begin/dontcopy_end 154 * are not copied in sock_copy() 155 */ 156 /* private: */ 157 int skc_dontcopy_begin[0]; 158 /* public: */ 159 union { 160 struct hlist_node skc_node; 161 struct hlist_nulls_node skc_nulls_node; 162 }; 163 int skc_tx_queue_mapping; 164 atomic_t skc_refcnt; 165 /* private: */ 166 int skc_dontcopy_end[0]; 167 /* public: */ 168 }; 169 170 /** 171 * struct sock - network layer representation of sockets 172 * @__sk_common: shared layout with inet_timewait_sock 173 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 174 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 175 * @sk_lock: synchronizer 176 * @sk_rcvbuf: size of receive buffer in bytes 177 * @sk_wq: sock wait queue and async head 178 * @sk_dst_cache: destination cache 179 * @sk_dst_lock: destination cache lock 180 * @sk_policy: flow policy 181 * @sk_receive_queue: incoming packets 182 * @sk_wmem_alloc: transmit queue bytes committed 183 * @sk_write_queue: Packet sending queue 184 * @sk_async_wait_queue: DMA copied packets 185 * @sk_omem_alloc: "o" is "option" or "other" 186 * @sk_wmem_queued: persistent queue size 187 * @sk_forward_alloc: space allocated forward 188 * @sk_allocation: allocation mode 189 * @sk_sndbuf: size of send buffer in bytes 190 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 191 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 192 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets 193 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 194 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 195 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 196 * @sk_gso_max_size: Maximum GSO segment size to build 197 * @sk_lingertime: %SO_LINGER l_linger setting 198 * @sk_backlog: always used with the per-socket spinlock held 199 * @sk_callback_lock: used with the callbacks in the end of this struct 200 * @sk_error_queue: rarely used 201 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 202 * IPV6_ADDRFORM for instance) 203 * @sk_err: last error 204 * @sk_err_soft: errors that don't cause failure but are the cause of a 205 * persistent failure not just 'timed out' 206 * @sk_drops: raw/udp drops counter 207 * @sk_ack_backlog: current listen backlog 208 * @sk_max_ack_backlog: listen backlog set in listen() 209 * @sk_priority: %SO_PRIORITY setting 210 * @sk_type: socket type (%SOCK_STREAM, etc) 211 * @sk_protocol: which protocol this socket belongs in this network family 212 * @sk_peer_pid: &struct pid for this socket's peer 213 * @sk_peer_cred: %SO_PEERCRED setting 214 * @sk_rcvlowat: %SO_RCVLOWAT setting 215 * @sk_rcvtimeo: %SO_RCVTIMEO setting 216 * @sk_sndtimeo: %SO_SNDTIMEO setting 217 * @sk_rxhash: flow hash received from netif layer 218 * @sk_filter: socket filtering instructions 219 * @sk_protinfo: private area, net family specific, when not using slab 220 * @sk_timer: sock cleanup timer 221 * @sk_stamp: time stamp of last packet received 222 * @sk_socket: Identd and reporting IO signals 223 * @sk_user_data: RPC layer private data 224 * @sk_sndmsg_page: cached page for sendmsg 225 * @sk_sndmsg_off: cached offset for sendmsg 226 * @sk_send_head: front of stuff to transmit 227 * @sk_security: used by security modules 228 * @sk_mark: generic packet mark 229 * @sk_classid: this socket's cgroup classid 230 * @sk_write_pending: a write to stream socket waits to start 231 * @sk_state_change: callback to indicate change in the state of the sock 232 * @sk_data_ready: callback to indicate there is data to be processed 233 * @sk_write_space: callback to indicate there is bf sending space available 234 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 235 * @sk_backlog_rcv: callback to process the backlog 236 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 237 */ 238 struct sock { 239 /* 240 * Now struct inet_timewait_sock also uses sock_common, so please just 241 * don't add nothing before this first member (__sk_common) --acme 242 */ 243 struct sock_common __sk_common; 244 #define sk_node __sk_common.skc_node 245 #define sk_nulls_node __sk_common.skc_nulls_node 246 #define sk_refcnt __sk_common.skc_refcnt 247 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 248 249 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 250 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 251 #define sk_hash __sk_common.skc_hash 252 #define sk_family __sk_common.skc_family 253 #define sk_state __sk_common.skc_state 254 #define sk_reuse __sk_common.skc_reuse 255 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 256 #define sk_bind_node __sk_common.skc_bind_node 257 #define sk_prot __sk_common.skc_prot 258 #define sk_net __sk_common.skc_net 259 socket_lock_t sk_lock; 260 struct sk_buff_head sk_receive_queue; 261 /* 262 * The backlog queue is special, it is always used with 263 * the per-socket spinlock held and requires low latency 264 * access. Therefore we special case it's implementation. 265 * Note : rmem_alloc is in this structure to fill a hole 266 * on 64bit arches, not because its logically part of 267 * backlog. 268 */ 269 struct { 270 atomic_t rmem_alloc; 271 int len; 272 struct sk_buff *head; 273 struct sk_buff *tail; 274 } sk_backlog; 275 #define sk_rmem_alloc sk_backlog.rmem_alloc 276 int sk_forward_alloc; 277 #ifdef CONFIG_RPS 278 __u32 sk_rxhash; 279 #endif 280 atomic_t sk_drops; 281 int sk_rcvbuf; 282 283 struct sk_filter __rcu *sk_filter; 284 struct socket_wq __rcu *sk_wq; 285 286 #ifdef CONFIG_NET_DMA 287 struct sk_buff_head sk_async_wait_queue; 288 #endif 289 290 #ifdef CONFIG_XFRM 291 struct xfrm_policy *sk_policy[2]; 292 #endif 293 unsigned long sk_flags; 294 struct dst_entry *sk_dst_cache; 295 spinlock_t sk_dst_lock; 296 atomic_t sk_wmem_alloc; 297 atomic_t sk_omem_alloc; 298 int sk_sndbuf; 299 struct sk_buff_head sk_write_queue; 300 kmemcheck_bitfield_begin(flags); 301 unsigned int sk_shutdown : 2, 302 sk_no_check : 2, 303 sk_userlocks : 4, 304 sk_protocol : 8, 305 sk_type : 16; 306 kmemcheck_bitfield_end(flags); 307 int sk_wmem_queued; 308 gfp_t sk_allocation; 309 int sk_route_caps; 310 int sk_route_nocaps; 311 int sk_gso_type; 312 unsigned int sk_gso_max_size; 313 int sk_rcvlowat; 314 unsigned long sk_lingertime; 315 struct sk_buff_head sk_error_queue; 316 struct proto *sk_prot_creator; 317 rwlock_t sk_callback_lock; 318 int sk_err, 319 sk_err_soft; 320 unsigned short sk_ack_backlog; 321 unsigned short sk_max_ack_backlog; 322 __u32 sk_priority; 323 struct pid *sk_peer_pid; 324 const struct cred *sk_peer_cred; 325 long sk_rcvtimeo; 326 long sk_sndtimeo; 327 void *sk_protinfo; 328 struct timer_list sk_timer; 329 ktime_t sk_stamp; 330 struct socket *sk_socket; 331 void *sk_user_data; 332 struct page *sk_sndmsg_page; 333 struct sk_buff *sk_send_head; 334 __u32 sk_sndmsg_off; 335 int sk_write_pending; 336 #ifdef CONFIG_SECURITY 337 void *sk_security; 338 #endif 339 __u32 sk_mark; 340 u32 sk_classid; 341 void (*sk_state_change)(struct sock *sk); 342 void (*sk_data_ready)(struct sock *sk, int bytes); 343 void (*sk_write_space)(struct sock *sk); 344 void (*sk_error_report)(struct sock *sk); 345 int (*sk_backlog_rcv)(struct sock *sk, 346 struct sk_buff *skb); 347 void (*sk_destruct)(struct sock *sk); 348 }; 349 350 /* 351 * Hashed lists helper routines 352 */ 353 static inline struct sock *sk_entry(const struct hlist_node *node) 354 { 355 return hlist_entry(node, struct sock, sk_node); 356 } 357 358 static inline struct sock *__sk_head(const struct hlist_head *head) 359 { 360 return hlist_entry(head->first, struct sock, sk_node); 361 } 362 363 static inline struct sock *sk_head(const struct hlist_head *head) 364 { 365 return hlist_empty(head) ? NULL : __sk_head(head); 366 } 367 368 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 369 { 370 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 371 } 372 373 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 374 { 375 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 376 } 377 378 static inline struct sock *sk_next(const struct sock *sk) 379 { 380 return sk->sk_node.next ? 381 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 382 } 383 384 static inline struct sock *sk_nulls_next(const struct sock *sk) 385 { 386 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 387 hlist_nulls_entry(sk->sk_nulls_node.next, 388 struct sock, sk_nulls_node) : 389 NULL; 390 } 391 392 static inline int sk_unhashed(const struct sock *sk) 393 { 394 return hlist_unhashed(&sk->sk_node); 395 } 396 397 static inline int sk_hashed(const struct sock *sk) 398 { 399 return !sk_unhashed(sk); 400 } 401 402 static __inline__ void sk_node_init(struct hlist_node *node) 403 { 404 node->pprev = NULL; 405 } 406 407 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node) 408 { 409 node->pprev = NULL; 410 } 411 412 static __inline__ void __sk_del_node(struct sock *sk) 413 { 414 __hlist_del(&sk->sk_node); 415 } 416 417 /* NB: equivalent to hlist_del_init_rcu */ 418 static __inline__ int __sk_del_node_init(struct sock *sk) 419 { 420 if (sk_hashed(sk)) { 421 __sk_del_node(sk); 422 sk_node_init(&sk->sk_node); 423 return 1; 424 } 425 return 0; 426 } 427 428 /* Grab socket reference count. This operation is valid only 429 when sk is ALREADY grabbed f.e. it is found in hash table 430 or a list and the lookup is made under lock preventing hash table 431 modifications. 432 */ 433 434 static inline void sock_hold(struct sock *sk) 435 { 436 atomic_inc(&sk->sk_refcnt); 437 } 438 439 /* Ungrab socket in the context, which assumes that socket refcnt 440 cannot hit zero, f.e. it is true in context of any socketcall. 441 */ 442 static inline void __sock_put(struct sock *sk) 443 { 444 atomic_dec(&sk->sk_refcnt); 445 } 446 447 static __inline__ int sk_del_node_init(struct sock *sk) 448 { 449 int rc = __sk_del_node_init(sk); 450 451 if (rc) { 452 /* paranoid for a while -acme */ 453 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 454 __sock_put(sk); 455 } 456 return rc; 457 } 458 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 459 460 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk) 461 { 462 if (sk_hashed(sk)) { 463 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 464 return 1; 465 } 466 return 0; 467 } 468 469 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk) 470 { 471 int rc = __sk_nulls_del_node_init_rcu(sk); 472 473 if (rc) { 474 /* paranoid for a while -acme */ 475 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 476 __sock_put(sk); 477 } 478 return rc; 479 } 480 481 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list) 482 { 483 hlist_add_head(&sk->sk_node, list); 484 } 485 486 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list) 487 { 488 sock_hold(sk); 489 __sk_add_node(sk, list); 490 } 491 492 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 493 { 494 sock_hold(sk); 495 hlist_add_head_rcu(&sk->sk_node, list); 496 } 497 498 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 499 { 500 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 501 } 502 503 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 504 { 505 sock_hold(sk); 506 __sk_nulls_add_node_rcu(sk, list); 507 } 508 509 static __inline__ void __sk_del_bind_node(struct sock *sk) 510 { 511 __hlist_del(&sk->sk_bind_node); 512 } 513 514 static __inline__ void sk_add_bind_node(struct sock *sk, 515 struct hlist_head *list) 516 { 517 hlist_add_head(&sk->sk_bind_node, list); 518 } 519 520 #define sk_for_each(__sk, node, list) \ 521 hlist_for_each_entry(__sk, node, list, sk_node) 522 #define sk_for_each_rcu(__sk, node, list) \ 523 hlist_for_each_entry_rcu(__sk, node, list, sk_node) 524 #define sk_nulls_for_each(__sk, node, list) \ 525 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 526 #define sk_nulls_for_each_rcu(__sk, node, list) \ 527 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 528 #define sk_for_each_from(__sk, node) \ 529 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 530 hlist_for_each_entry_from(__sk, node, sk_node) 531 #define sk_nulls_for_each_from(__sk, node) \ 532 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 533 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 534 #define sk_for_each_safe(__sk, node, tmp, list) \ 535 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node) 536 #define sk_for_each_bound(__sk, node, list) \ 537 hlist_for_each_entry(__sk, node, list, sk_bind_node) 538 539 /* Sock flags */ 540 enum sock_flags { 541 SOCK_DEAD, 542 SOCK_DONE, 543 SOCK_URGINLINE, 544 SOCK_KEEPOPEN, 545 SOCK_LINGER, 546 SOCK_DESTROY, 547 SOCK_BROADCAST, 548 SOCK_TIMESTAMP, 549 SOCK_ZAPPED, 550 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 551 SOCK_DBG, /* %SO_DEBUG setting */ 552 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 553 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 554 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 555 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 556 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 557 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 558 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 559 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 560 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 561 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 562 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 563 SOCK_FASYNC, /* fasync() active */ 564 SOCK_RXQ_OVFL, 565 SOCK_ZEROCOPY, /* buffers from userspace */ 566 }; 567 568 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 569 { 570 nsk->sk_flags = osk->sk_flags; 571 } 572 573 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 574 { 575 __set_bit(flag, &sk->sk_flags); 576 } 577 578 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 579 { 580 __clear_bit(flag, &sk->sk_flags); 581 } 582 583 static inline int sock_flag(struct sock *sk, enum sock_flags flag) 584 { 585 return test_bit(flag, &sk->sk_flags); 586 } 587 588 static inline void sk_acceptq_removed(struct sock *sk) 589 { 590 sk->sk_ack_backlog--; 591 } 592 593 static inline void sk_acceptq_added(struct sock *sk) 594 { 595 sk->sk_ack_backlog++; 596 } 597 598 static inline int sk_acceptq_is_full(struct sock *sk) 599 { 600 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 601 } 602 603 /* 604 * Compute minimal free write space needed to queue new packets. 605 */ 606 static inline int sk_stream_min_wspace(struct sock *sk) 607 { 608 return sk->sk_wmem_queued >> 1; 609 } 610 611 static inline int sk_stream_wspace(struct sock *sk) 612 { 613 return sk->sk_sndbuf - sk->sk_wmem_queued; 614 } 615 616 extern void sk_stream_write_space(struct sock *sk); 617 618 static inline int sk_stream_memory_free(struct sock *sk) 619 { 620 return sk->sk_wmem_queued < sk->sk_sndbuf; 621 } 622 623 /* OOB backlog add */ 624 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 625 { 626 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 627 skb_dst_force(skb); 628 629 if (!sk->sk_backlog.tail) 630 sk->sk_backlog.head = skb; 631 else 632 sk->sk_backlog.tail->next = skb; 633 634 sk->sk_backlog.tail = skb; 635 skb->next = NULL; 636 } 637 638 /* 639 * Take into account size of receive queue and backlog queue 640 * Do not take into account this skb truesize, 641 * to allow even a single big packet to come. 642 */ 643 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb) 644 { 645 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 646 647 return qsize > sk->sk_rcvbuf; 648 } 649 650 /* The per-socket spinlock must be held here. */ 651 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb) 652 { 653 if (sk_rcvqueues_full(sk, skb)) 654 return -ENOBUFS; 655 656 __sk_add_backlog(sk, skb); 657 sk->sk_backlog.len += skb->truesize; 658 return 0; 659 } 660 661 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 662 { 663 return sk->sk_backlog_rcv(sk, skb); 664 } 665 666 static inline void sock_rps_record_flow(const struct sock *sk) 667 { 668 #ifdef CONFIG_RPS 669 struct rps_sock_flow_table *sock_flow_table; 670 671 rcu_read_lock(); 672 sock_flow_table = rcu_dereference(rps_sock_flow_table); 673 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 674 rcu_read_unlock(); 675 #endif 676 } 677 678 static inline void sock_rps_reset_flow(const struct sock *sk) 679 { 680 #ifdef CONFIG_RPS 681 struct rps_sock_flow_table *sock_flow_table; 682 683 rcu_read_lock(); 684 sock_flow_table = rcu_dereference(rps_sock_flow_table); 685 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 686 rcu_read_unlock(); 687 #endif 688 } 689 690 static inline void sock_rps_save_rxhash(struct sock *sk, 691 const struct sk_buff *skb) 692 { 693 #ifdef CONFIG_RPS 694 if (unlikely(sk->sk_rxhash != skb->rxhash)) { 695 sock_rps_reset_flow(sk); 696 sk->sk_rxhash = skb->rxhash; 697 } 698 #endif 699 } 700 701 static inline void sock_rps_reset_rxhash(struct sock *sk) 702 { 703 #ifdef CONFIG_RPS 704 sock_rps_reset_flow(sk); 705 sk->sk_rxhash = 0; 706 #endif 707 } 708 709 #define sk_wait_event(__sk, __timeo, __condition) \ 710 ({ int __rc; \ 711 release_sock(__sk); \ 712 __rc = __condition; \ 713 if (!__rc) { \ 714 *(__timeo) = schedule_timeout(*(__timeo)); \ 715 } \ 716 lock_sock(__sk); \ 717 __rc = __condition; \ 718 __rc; \ 719 }) 720 721 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 722 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 723 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 724 extern int sk_stream_error(struct sock *sk, int flags, int err); 725 extern void sk_stream_kill_queues(struct sock *sk); 726 727 extern int sk_wait_data(struct sock *sk, long *timeo); 728 729 struct request_sock_ops; 730 struct timewait_sock_ops; 731 struct inet_hashinfo; 732 struct raw_hashinfo; 733 struct module; 734 735 /* Networking protocol blocks we attach to sockets. 736 * socket layer -> transport layer interface 737 * transport -> network interface is defined by struct inet_proto 738 */ 739 struct proto { 740 void (*close)(struct sock *sk, 741 long timeout); 742 int (*connect)(struct sock *sk, 743 struct sockaddr *uaddr, 744 int addr_len); 745 int (*disconnect)(struct sock *sk, int flags); 746 747 struct sock * (*accept) (struct sock *sk, int flags, int *err); 748 749 int (*ioctl)(struct sock *sk, int cmd, 750 unsigned long arg); 751 int (*init)(struct sock *sk); 752 void (*destroy)(struct sock *sk); 753 void (*shutdown)(struct sock *sk, int how); 754 int (*setsockopt)(struct sock *sk, int level, 755 int optname, char __user *optval, 756 unsigned int optlen); 757 int (*getsockopt)(struct sock *sk, int level, 758 int optname, char __user *optval, 759 int __user *option); 760 #ifdef CONFIG_COMPAT 761 int (*compat_setsockopt)(struct sock *sk, 762 int level, 763 int optname, char __user *optval, 764 unsigned int optlen); 765 int (*compat_getsockopt)(struct sock *sk, 766 int level, 767 int optname, char __user *optval, 768 int __user *option); 769 int (*compat_ioctl)(struct sock *sk, 770 unsigned int cmd, unsigned long arg); 771 #endif 772 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 773 struct msghdr *msg, size_t len); 774 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 775 struct msghdr *msg, 776 size_t len, int noblock, int flags, 777 int *addr_len); 778 int (*sendpage)(struct sock *sk, struct page *page, 779 int offset, size_t size, int flags); 780 int (*bind)(struct sock *sk, 781 struct sockaddr *uaddr, int addr_len); 782 783 int (*backlog_rcv) (struct sock *sk, 784 struct sk_buff *skb); 785 786 /* Keeping track of sk's, looking them up, and port selection methods. */ 787 void (*hash)(struct sock *sk); 788 void (*unhash)(struct sock *sk); 789 void (*rehash)(struct sock *sk); 790 int (*get_port)(struct sock *sk, unsigned short snum); 791 void (*clear_sk)(struct sock *sk, int size); 792 793 /* Keeping track of sockets in use */ 794 #ifdef CONFIG_PROC_FS 795 unsigned int inuse_idx; 796 #endif 797 798 /* Memory pressure */ 799 void (*enter_memory_pressure)(struct sock *sk); 800 atomic_long_t *memory_allocated; /* Current allocated memory. */ 801 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 802 /* 803 * Pressure flag: try to collapse. 804 * Technical note: it is used by multiple contexts non atomically. 805 * All the __sk_mem_schedule() is of this nature: accounting 806 * is strict, actions are advisory and have some latency. 807 */ 808 int *memory_pressure; 809 long *sysctl_mem; 810 int *sysctl_wmem; 811 int *sysctl_rmem; 812 int max_header; 813 bool no_autobind; 814 815 struct kmem_cache *slab; 816 unsigned int obj_size; 817 int slab_flags; 818 819 struct percpu_counter *orphan_count; 820 821 struct request_sock_ops *rsk_prot; 822 struct timewait_sock_ops *twsk_prot; 823 824 union { 825 struct inet_hashinfo *hashinfo; 826 struct udp_table *udp_table; 827 struct raw_hashinfo *raw_hash; 828 } h; 829 830 struct module *owner; 831 832 char name[32]; 833 834 struct list_head node; 835 #ifdef SOCK_REFCNT_DEBUG 836 atomic_t socks; 837 #endif 838 }; 839 840 extern int proto_register(struct proto *prot, int alloc_slab); 841 extern void proto_unregister(struct proto *prot); 842 843 #ifdef SOCK_REFCNT_DEBUG 844 static inline void sk_refcnt_debug_inc(struct sock *sk) 845 { 846 atomic_inc(&sk->sk_prot->socks); 847 } 848 849 static inline void sk_refcnt_debug_dec(struct sock *sk) 850 { 851 atomic_dec(&sk->sk_prot->socks); 852 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 853 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 854 } 855 856 static inline void sk_refcnt_debug_release(const struct sock *sk) 857 { 858 if (atomic_read(&sk->sk_refcnt) != 1) 859 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 860 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 861 } 862 #else /* SOCK_REFCNT_DEBUG */ 863 #define sk_refcnt_debug_inc(sk) do { } while (0) 864 #define sk_refcnt_debug_dec(sk) do { } while (0) 865 #define sk_refcnt_debug_release(sk) do { } while (0) 866 #endif /* SOCK_REFCNT_DEBUG */ 867 868 869 #ifdef CONFIG_PROC_FS 870 /* Called with local bh disabled */ 871 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 872 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 873 #else 874 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot, 875 int inc) 876 { 877 } 878 #endif 879 880 881 /* With per-bucket locks this operation is not-atomic, so that 882 * this version is not worse. 883 */ 884 static inline void __sk_prot_rehash(struct sock *sk) 885 { 886 sk->sk_prot->unhash(sk); 887 sk->sk_prot->hash(sk); 888 } 889 890 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 891 892 /* About 10 seconds */ 893 #define SOCK_DESTROY_TIME (10*HZ) 894 895 /* Sockets 0-1023 can't be bound to unless you are superuser */ 896 #define PROT_SOCK 1024 897 898 #define SHUTDOWN_MASK 3 899 #define RCV_SHUTDOWN 1 900 #define SEND_SHUTDOWN 2 901 902 #define SOCK_SNDBUF_LOCK 1 903 #define SOCK_RCVBUF_LOCK 2 904 #define SOCK_BINDADDR_LOCK 4 905 #define SOCK_BINDPORT_LOCK 8 906 907 /* sock_iocb: used to kick off async processing of socket ios */ 908 struct sock_iocb { 909 struct list_head list; 910 911 int flags; 912 int size; 913 struct socket *sock; 914 struct sock *sk; 915 struct scm_cookie *scm; 916 struct msghdr *msg, async_msg; 917 struct kiocb *kiocb; 918 }; 919 920 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 921 { 922 return (struct sock_iocb *)iocb->private; 923 } 924 925 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 926 { 927 return si->kiocb; 928 } 929 930 struct socket_alloc { 931 struct socket socket; 932 struct inode vfs_inode; 933 }; 934 935 static inline struct socket *SOCKET_I(struct inode *inode) 936 { 937 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 938 } 939 940 static inline struct inode *SOCK_INODE(struct socket *socket) 941 { 942 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 943 } 944 945 /* 946 * Functions for memory accounting 947 */ 948 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 949 extern void __sk_mem_reclaim(struct sock *sk); 950 951 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 952 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 953 #define SK_MEM_SEND 0 954 #define SK_MEM_RECV 1 955 956 static inline int sk_mem_pages(int amt) 957 { 958 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 959 } 960 961 static inline int sk_has_account(struct sock *sk) 962 { 963 /* return true if protocol supports memory accounting */ 964 return !!sk->sk_prot->memory_allocated; 965 } 966 967 static inline int sk_wmem_schedule(struct sock *sk, int size) 968 { 969 if (!sk_has_account(sk)) 970 return 1; 971 return size <= sk->sk_forward_alloc || 972 __sk_mem_schedule(sk, size, SK_MEM_SEND); 973 } 974 975 static inline int sk_rmem_schedule(struct sock *sk, int size) 976 { 977 if (!sk_has_account(sk)) 978 return 1; 979 return size <= sk->sk_forward_alloc || 980 __sk_mem_schedule(sk, size, SK_MEM_RECV); 981 } 982 983 static inline void sk_mem_reclaim(struct sock *sk) 984 { 985 if (!sk_has_account(sk)) 986 return; 987 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 988 __sk_mem_reclaim(sk); 989 } 990 991 static inline void sk_mem_reclaim_partial(struct sock *sk) 992 { 993 if (!sk_has_account(sk)) 994 return; 995 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 996 __sk_mem_reclaim(sk); 997 } 998 999 static inline void sk_mem_charge(struct sock *sk, int size) 1000 { 1001 if (!sk_has_account(sk)) 1002 return; 1003 sk->sk_forward_alloc -= size; 1004 } 1005 1006 static inline void sk_mem_uncharge(struct sock *sk, int size) 1007 { 1008 if (!sk_has_account(sk)) 1009 return; 1010 sk->sk_forward_alloc += size; 1011 } 1012 1013 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1014 { 1015 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1016 sk->sk_wmem_queued -= skb->truesize; 1017 sk_mem_uncharge(sk, skb->truesize); 1018 __kfree_skb(skb); 1019 } 1020 1021 /* Used by processes to "lock" a socket state, so that 1022 * interrupts and bottom half handlers won't change it 1023 * from under us. It essentially blocks any incoming 1024 * packets, so that we won't get any new data or any 1025 * packets that change the state of the socket. 1026 * 1027 * While locked, BH processing will add new packets to 1028 * the backlog queue. This queue is processed by the 1029 * owner of the socket lock right before it is released. 1030 * 1031 * Since ~2.3.5 it is also exclusive sleep lock serializing 1032 * accesses from user process context. 1033 */ 1034 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1035 1036 /* 1037 * Macro so as to not evaluate some arguments when 1038 * lockdep is not enabled. 1039 * 1040 * Mark both the sk_lock and the sk_lock.slock as a 1041 * per-address-family lock class. 1042 */ 1043 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1044 do { \ 1045 sk->sk_lock.owned = 0; \ 1046 init_waitqueue_head(&sk->sk_lock.wq); \ 1047 spin_lock_init(&(sk)->sk_lock.slock); \ 1048 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1049 sizeof((sk)->sk_lock)); \ 1050 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1051 (skey), (sname)); \ 1052 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1053 } while (0) 1054 1055 extern void lock_sock_nested(struct sock *sk, int subclass); 1056 1057 static inline void lock_sock(struct sock *sk) 1058 { 1059 lock_sock_nested(sk, 0); 1060 } 1061 1062 extern void release_sock(struct sock *sk); 1063 1064 /* BH context may only use the following locking interface. */ 1065 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1066 #define bh_lock_sock_nested(__sk) \ 1067 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1068 SINGLE_DEPTH_NESTING) 1069 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1070 1071 extern bool lock_sock_fast(struct sock *sk); 1072 /** 1073 * unlock_sock_fast - complement of lock_sock_fast 1074 * @sk: socket 1075 * @slow: slow mode 1076 * 1077 * fast unlock socket for user context. 1078 * If slow mode is on, we call regular release_sock() 1079 */ 1080 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1081 { 1082 if (slow) 1083 release_sock(sk); 1084 else 1085 spin_unlock_bh(&sk->sk_lock.slock); 1086 } 1087 1088 1089 extern struct sock *sk_alloc(struct net *net, int family, 1090 gfp_t priority, 1091 struct proto *prot); 1092 extern void sk_free(struct sock *sk); 1093 extern void sk_release_kernel(struct sock *sk); 1094 extern struct sock *sk_clone(const struct sock *sk, 1095 const gfp_t priority); 1096 1097 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1098 unsigned long size, int force, 1099 gfp_t priority); 1100 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1101 unsigned long size, int force, 1102 gfp_t priority); 1103 extern void sock_wfree(struct sk_buff *skb); 1104 extern void sock_rfree(struct sk_buff *skb); 1105 1106 extern int sock_setsockopt(struct socket *sock, int level, 1107 int op, char __user *optval, 1108 unsigned int optlen); 1109 1110 extern int sock_getsockopt(struct socket *sock, int level, 1111 int op, char __user *optval, 1112 int __user *optlen); 1113 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1114 unsigned long size, 1115 int noblock, 1116 int *errcode); 1117 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1118 unsigned long header_len, 1119 unsigned long data_len, 1120 int noblock, 1121 int *errcode); 1122 extern void *sock_kmalloc(struct sock *sk, int size, 1123 gfp_t priority); 1124 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1125 extern void sk_send_sigurg(struct sock *sk); 1126 1127 #ifdef CONFIG_CGROUPS 1128 extern void sock_update_classid(struct sock *sk); 1129 #else 1130 static inline void sock_update_classid(struct sock *sk) 1131 { 1132 } 1133 #endif 1134 1135 /* 1136 * Functions to fill in entries in struct proto_ops when a protocol 1137 * does not implement a particular function. 1138 */ 1139 extern int sock_no_bind(struct socket *, 1140 struct sockaddr *, int); 1141 extern int sock_no_connect(struct socket *, 1142 struct sockaddr *, int, int); 1143 extern int sock_no_socketpair(struct socket *, 1144 struct socket *); 1145 extern int sock_no_accept(struct socket *, 1146 struct socket *, int); 1147 extern int sock_no_getname(struct socket *, 1148 struct sockaddr *, int *, int); 1149 extern unsigned int sock_no_poll(struct file *, struct socket *, 1150 struct poll_table_struct *); 1151 extern int sock_no_ioctl(struct socket *, unsigned int, 1152 unsigned long); 1153 extern int sock_no_listen(struct socket *, int); 1154 extern int sock_no_shutdown(struct socket *, int); 1155 extern int sock_no_getsockopt(struct socket *, int , int, 1156 char __user *, int __user *); 1157 extern int sock_no_setsockopt(struct socket *, int, int, 1158 char __user *, unsigned int); 1159 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1160 struct msghdr *, size_t); 1161 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1162 struct msghdr *, size_t, int); 1163 extern int sock_no_mmap(struct file *file, 1164 struct socket *sock, 1165 struct vm_area_struct *vma); 1166 extern ssize_t sock_no_sendpage(struct socket *sock, 1167 struct page *page, 1168 int offset, size_t size, 1169 int flags); 1170 1171 /* 1172 * Functions to fill in entries in struct proto_ops when a protocol 1173 * uses the inet style. 1174 */ 1175 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1176 char __user *optval, int __user *optlen); 1177 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1178 struct msghdr *msg, size_t size, int flags); 1179 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1180 char __user *optval, unsigned int optlen); 1181 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1182 int optname, char __user *optval, int __user *optlen); 1183 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1184 int optname, char __user *optval, unsigned int optlen); 1185 1186 extern void sk_common_release(struct sock *sk); 1187 1188 /* 1189 * Default socket callbacks and setup code 1190 */ 1191 1192 /* Initialise core socket variables */ 1193 extern void sock_init_data(struct socket *sock, struct sock *sk); 1194 1195 extern void sk_filter_release_rcu(struct rcu_head *rcu); 1196 1197 /** 1198 * sk_filter_release - release a socket filter 1199 * @fp: filter to remove 1200 * 1201 * Remove a filter from a socket and release its resources. 1202 */ 1203 1204 static inline void sk_filter_release(struct sk_filter *fp) 1205 { 1206 if (atomic_dec_and_test(&fp->refcnt)) 1207 call_rcu(&fp->rcu, sk_filter_release_rcu); 1208 } 1209 1210 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1211 { 1212 unsigned int size = sk_filter_len(fp); 1213 1214 atomic_sub(size, &sk->sk_omem_alloc); 1215 sk_filter_release(fp); 1216 } 1217 1218 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1219 { 1220 atomic_inc(&fp->refcnt); 1221 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1222 } 1223 1224 /* 1225 * Socket reference counting postulates. 1226 * 1227 * * Each user of socket SHOULD hold a reference count. 1228 * * Each access point to socket (an hash table bucket, reference from a list, 1229 * running timer, skb in flight MUST hold a reference count. 1230 * * When reference count hits 0, it means it will never increase back. 1231 * * When reference count hits 0, it means that no references from 1232 * outside exist to this socket and current process on current CPU 1233 * is last user and may/should destroy this socket. 1234 * * sk_free is called from any context: process, BH, IRQ. When 1235 * it is called, socket has no references from outside -> sk_free 1236 * may release descendant resources allocated by the socket, but 1237 * to the time when it is called, socket is NOT referenced by any 1238 * hash tables, lists etc. 1239 * * Packets, delivered from outside (from network or from another process) 1240 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1241 * when they sit in queue. Otherwise, packets will leak to hole, when 1242 * socket is looked up by one cpu and unhasing is made by another CPU. 1243 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1244 * (leak to backlog). Packet socket does all the processing inside 1245 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1246 * use separate SMP lock, so that they are prone too. 1247 */ 1248 1249 /* Ungrab socket and destroy it, if it was the last reference. */ 1250 static inline void sock_put(struct sock *sk) 1251 { 1252 if (atomic_dec_and_test(&sk->sk_refcnt)) 1253 sk_free(sk); 1254 } 1255 1256 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1257 const int nested); 1258 1259 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1260 { 1261 sk->sk_tx_queue_mapping = tx_queue; 1262 } 1263 1264 static inline void sk_tx_queue_clear(struct sock *sk) 1265 { 1266 sk->sk_tx_queue_mapping = -1; 1267 } 1268 1269 static inline int sk_tx_queue_get(const struct sock *sk) 1270 { 1271 return sk ? sk->sk_tx_queue_mapping : -1; 1272 } 1273 1274 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1275 { 1276 sk_tx_queue_clear(sk); 1277 sk->sk_socket = sock; 1278 } 1279 1280 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1281 { 1282 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1283 return &rcu_dereference_raw(sk->sk_wq)->wait; 1284 } 1285 /* Detach socket from process context. 1286 * Announce socket dead, detach it from wait queue and inode. 1287 * Note that parent inode held reference count on this struct sock, 1288 * we do not release it in this function, because protocol 1289 * probably wants some additional cleanups or even continuing 1290 * to work with this socket (TCP). 1291 */ 1292 static inline void sock_orphan(struct sock *sk) 1293 { 1294 write_lock_bh(&sk->sk_callback_lock); 1295 sock_set_flag(sk, SOCK_DEAD); 1296 sk_set_socket(sk, NULL); 1297 sk->sk_wq = NULL; 1298 write_unlock_bh(&sk->sk_callback_lock); 1299 } 1300 1301 static inline void sock_graft(struct sock *sk, struct socket *parent) 1302 { 1303 write_lock_bh(&sk->sk_callback_lock); 1304 sk->sk_wq = parent->wq; 1305 parent->sk = sk; 1306 sk_set_socket(sk, parent); 1307 security_sock_graft(sk, parent); 1308 write_unlock_bh(&sk->sk_callback_lock); 1309 } 1310 1311 extern int sock_i_uid(struct sock *sk); 1312 extern unsigned long sock_i_ino(struct sock *sk); 1313 1314 static inline struct dst_entry * 1315 __sk_dst_get(struct sock *sk) 1316 { 1317 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1318 lockdep_is_held(&sk->sk_lock.slock)); 1319 } 1320 1321 static inline struct dst_entry * 1322 sk_dst_get(struct sock *sk) 1323 { 1324 struct dst_entry *dst; 1325 1326 rcu_read_lock(); 1327 dst = rcu_dereference(sk->sk_dst_cache); 1328 if (dst) 1329 dst_hold(dst); 1330 rcu_read_unlock(); 1331 return dst; 1332 } 1333 1334 extern void sk_reset_txq(struct sock *sk); 1335 1336 static inline void dst_negative_advice(struct sock *sk) 1337 { 1338 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1339 1340 if (dst && dst->ops->negative_advice) { 1341 ndst = dst->ops->negative_advice(dst); 1342 1343 if (ndst != dst) { 1344 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1345 sk_reset_txq(sk); 1346 } 1347 } 1348 } 1349 1350 static inline void 1351 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1352 { 1353 struct dst_entry *old_dst; 1354 1355 sk_tx_queue_clear(sk); 1356 /* 1357 * This can be called while sk is owned by the caller only, 1358 * with no state that can be checked in a rcu_dereference_check() cond 1359 */ 1360 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1361 rcu_assign_pointer(sk->sk_dst_cache, dst); 1362 dst_release(old_dst); 1363 } 1364 1365 static inline void 1366 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1367 { 1368 spin_lock(&sk->sk_dst_lock); 1369 __sk_dst_set(sk, dst); 1370 spin_unlock(&sk->sk_dst_lock); 1371 } 1372 1373 static inline void 1374 __sk_dst_reset(struct sock *sk) 1375 { 1376 __sk_dst_set(sk, NULL); 1377 } 1378 1379 static inline void 1380 sk_dst_reset(struct sock *sk) 1381 { 1382 spin_lock(&sk->sk_dst_lock); 1383 __sk_dst_reset(sk); 1384 spin_unlock(&sk->sk_dst_lock); 1385 } 1386 1387 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1388 1389 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1390 1391 static inline int sk_can_gso(const struct sock *sk) 1392 { 1393 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1394 } 1395 1396 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1397 1398 static inline void sk_nocaps_add(struct sock *sk, int flags) 1399 { 1400 sk->sk_route_nocaps |= flags; 1401 sk->sk_route_caps &= ~flags; 1402 } 1403 1404 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1405 char __user *from, char *to, 1406 int copy, int offset) 1407 { 1408 if (skb->ip_summed == CHECKSUM_NONE) { 1409 int err = 0; 1410 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1411 if (err) 1412 return err; 1413 skb->csum = csum_block_add(skb->csum, csum, offset); 1414 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1415 if (!access_ok(VERIFY_READ, from, copy) || 1416 __copy_from_user_nocache(to, from, copy)) 1417 return -EFAULT; 1418 } else if (copy_from_user(to, from, copy)) 1419 return -EFAULT; 1420 1421 return 0; 1422 } 1423 1424 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1425 char __user *from, int copy) 1426 { 1427 int err, offset = skb->len; 1428 1429 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1430 copy, offset); 1431 if (err) 1432 __skb_trim(skb, offset); 1433 1434 return err; 1435 } 1436 1437 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from, 1438 struct sk_buff *skb, 1439 struct page *page, 1440 int off, int copy) 1441 { 1442 int err; 1443 1444 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1445 copy, skb->len); 1446 if (err) 1447 return err; 1448 1449 skb->len += copy; 1450 skb->data_len += copy; 1451 skb->truesize += copy; 1452 sk->sk_wmem_queued += copy; 1453 sk_mem_charge(sk, copy); 1454 return 0; 1455 } 1456 1457 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1458 struct sk_buff *skb, struct page *page, 1459 int off, int copy) 1460 { 1461 if (skb->ip_summed == CHECKSUM_NONE) { 1462 int err = 0; 1463 __wsum csum = csum_and_copy_from_user(from, 1464 page_address(page) + off, 1465 copy, 0, &err); 1466 if (err) 1467 return err; 1468 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1469 } else if (copy_from_user(page_address(page) + off, from, copy)) 1470 return -EFAULT; 1471 1472 skb->len += copy; 1473 skb->data_len += copy; 1474 skb->truesize += copy; 1475 sk->sk_wmem_queued += copy; 1476 sk_mem_charge(sk, copy); 1477 return 0; 1478 } 1479 1480 /** 1481 * sk_wmem_alloc_get - returns write allocations 1482 * @sk: socket 1483 * 1484 * Returns sk_wmem_alloc minus initial offset of one 1485 */ 1486 static inline int sk_wmem_alloc_get(const struct sock *sk) 1487 { 1488 return atomic_read(&sk->sk_wmem_alloc) - 1; 1489 } 1490 1491 /** 1492 * sk_rmem_alloc_get - returns read allocations 1493 * @sk: socket 1494 * 1495 * Returns sk_rmem_alloc 1496 */ 1497 static inline int sk_rmem_alloc_get(const struct sock *sk) 1498 { 1499 return atomic_read(&sk->sk_rmem_alloc); 1500 } 1501 1502 /** 1503 * sk_has_allocations - check if allocations are outstanding 1504 * @sk: socket 1505 * 1506 * Returns true if socket has write or read allocations 1507 */ 1508 static inline int sk_has_allocations(const struct sock *sk) 1509 { 1510 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1511 } 1512 1513 /** 1514 * wq_has_sleeper - check if there are any waiting processes 1515 * @wq: struct socket_wq 1516 * 1517 * Returns true if socket_wq has waiting processes 1518 * 1519 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1520 * barrier call. They were added due to the race found within the tcp code. 1521 * 1522 * Consider following tcp code paths: 1523 * 1524 * CPU1 CPU2 1525 * 1526 * sys_select receive packet 1527 * ... ... 1528 * __add_wait_queue update tp->rcv_nxt 1529 * ... ... 1530 * tp->rcv_nxt check sock_def_readable 1531 * ... { 1532 * schedule rcu_read_lock(); 1533 * wq = rcu_dereference(sk->sk_wq); 1534 * if (wq && waitqueue_active(&wq->wait)) 1535 * wake_up_interruptible(&wq->wait) 1536 * ... 1537 * } 1538 * 1539 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1540 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1541 * could then endup calling schedule and sleep forever if there are no more 1542 * data on the socket. 1543 * 1544 */ 1545 static inline bool wq_has_sleeper(struct socket_wq *wq) 1546 { 1547 1548 /* 1549 * We need to be sure we are in sync with the 1550 * add_wait_queue modifications to the wait queue. 1551 * 1552 * This memory barrier is paired in the sock_poll_wait. 1553 */ 1554 smp_mb(); 1555 return wq && waitqueue_active(&wq->wait); 1556 } 1557 1558 /** 1559 * sock_poll_wait - place memory barrier behind the poll_wait call. 1560 * @filp: file 1561 * @wait_address: socket wait queue 1562 * @p: poll_table 1563 * 1564 * See the comments in the wq_has_sleeper function. 1565 */ 1566 static inline void sock_poll_wait(struct file *filp, 1567 wait_queue_head_t *wait_address, poll_table *p) 1568 { 1569 if (p && wait_address) { 1570 poll_wait(filp, wait_address, p); 1571 /* 1572 * We need to be sure we are in sync with the 1573 * socket flags modification. 1574 * 1575 * This memory barrier is paired in the wq_has_sleeper. 1576 */ 1577 smp_mb(); 1578 } 1579 } 1580 1581 /* 1582 * Queue a received datagram if it will fit. Stream and sequenced 1583 * protocols can't normally use this as they need to fit buffers in 1584 * and play with them. 1585 * 1586 * Inlined as it's very short and called for pretty much every 1587 * packet ever received. 1588 */ 1589 1590 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1591 { 1592 skb_orphan(skb); 1593 skb->sk = sk; 1594 skb->destructor = sock_wfree; 1595 /* 1596 * We used to take a refcount on sk, but following operation 1597 * is enough to guarantee sk_free() wont free this sock until 1598 * all in-flight packets are completed 1599 */ 1600 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1601 } 1602 1603 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1604 { 1605 skb_orphan(skb); 1606 skb->sk = sk; 1607 skb->destructor = sock_rfree; 1608 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1609 sk_mem_charge(sk, skb->truesize); 1610 } 1611 1612 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1613 unsigned long expires); 1614 1615 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer); 1616 1617 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1618 1619 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1620 1621 /* 1622 * Recover an error report and clear atomically 1623 */ 1624 1625 static inline int sock_error(struct sock *sk) 1626 { 1627 int err; 1628 if (likely(!sk->sk_err)) 1629 return 0; 1630 err = xchg(&sk->sk_err, 0); 1631 return -err; 1632 } 1633 1634 static inline unsigned long sock_wspace(struct sock *sk) 1635 { 1636 int amt = 0; 1637 1638 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1639 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 1640 if (amt < 0) 1641 amt = 0; 1642 } 1643 return amt; 1644 } 1645 1646 static inline void sk_wake_async(struct sock *sk, int how, int band) 1647 { 1648 if (sock_flag(sk, SOCK_FASYNC)) 1649 sock_wake_async(sk->sk_socket, how, band); 1650 } 1651 1652 #define SOCK_MIN_SNDBUF 2048 1653 /* 1654 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need 1655 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak 1656 */ 1657 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff)) 1658 1659 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 1660 { 1661 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 1662 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 1663 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF); 1664 } 1665 } 1666 1667 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 1668 1669 static inline struct page *sk_stream_alloc_page(struct sock *sk) 1670 { 1671 struct page *page = NULL; 1672 1673 page = alloc_pages(sk->sk_allocation, 0); 1674 if (!page) { 1675 sk->sk_prot->enter_memory_pressure(sk); 1676 sk_stream_moderate_sndbuf(sk); 1677 } 1678 return page; 1679 } 1680 1681 /* 1682 * Default write policy as shown to user space via poll/select/SIGIO 1683 */ 1684 static inline int sock_writeable(const struct sock *sk) 1685 { 1686 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 1687 } 1688 1689 static inline gfp_t gfp_any(void) 1690 { 1691 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 1692 } 1693 1694 static inline long sock_rcvtimeo(const struct sock *sk, int noblock) 1695 { 1696 return noblock ? 0 : sk->sk_rcvtimeo; 1697 } 1698 1699 static inline long sock_sndtimeo(const struct sock *sk, int noblock) 1700 { 1701 return noblock ? 0 : sk->sk_sndtimeo; 1702 } 1703 1704 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 1705 { 1706 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 1707 } 1708 1709 /* Alas, with timeout socket operations are not restartable. 1710 * Compare this to poll(). 1711 */ 1712 static inline int sock_intr_errno(long timeo) 1713 { 1714 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 1715 } 1716 1717 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 1718 struct sk_buff *skb); 1719 1720 static __inline__ void 1721 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 1722 { 1723 ktime_t kt = skb->tstamp; 1724 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 1725 1726 /* 1727 * generate control messages if 1728 * - receive time stamping in software requested (SOCK_RCVTSTAMP 1729 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 1730 * - software time stamp available and wanted 1731 * (SOCK_TIMESTAMPING_SOFTWARE) 1732 * - hardware time stamps available and wanted 1733 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 1734 * SOCK_TIMESTAMPING_RAW_HARDWARE) 1735 */ 1736 if (sock_flag(sk, SOCK_RCVTSTAMP) || 1737 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 1738 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 1739 (hwtstamps->hwtstamp.tv64 && 1740 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 1741 (hwtstamps->syststamp.tv64 && 1742 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 1743 __sock_recv_timestamp(msg, sk, skb); 1744 else 1745 sk->sk_stamp = kt; 1746 } 1747 1748 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 1749 struct sk_buff *skb); 1750 1751 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 1752 struct sk_buff *skb) 1753 { 1754 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 1755 (1UL << SOCK_RCVTSTAMP) | \ 1756 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 1757 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 1758 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 1759 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 1760 1761 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 1762 __sock_recv_ts_and_drops(msg, sk, skb); 1763 else 1764 sk->sk_stamp = skb->tstamp; 1765 } 1766 1767 /** 1768 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 1769 * @sk: socket sending this packet 1770 * @tx_flags: filled with instructions for time stamping 1771 * 1772 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if 1773 * parameters are invalid. 1774 */ 1775 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 1776 1777 /** 1778 * sk_eat_skb - Release a skb if it is no longer needed 1779 * @sk: socket to eat this skb from 1780 * @skb: socket buffer to eat 1781 * @copied_early: flag indicating whether DMA operations copied this data early 1782 * 1783 * This routine must be called with interrupts disabled or with the socket 1784 * locked so that the sk_buff queue operation is ok. 1785 */ 1786 #ifdef CONFIG_NET_DMA 1787 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 1788 { 1789 __skb_unlink(skb, &sk->sk_receive_queue); 1790 if (!copied_early) 1791 __kfree_skb(skb); 1792 else 1793 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 1794 } 1795 #else 1796 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 1797 { 1798 __skb_unlink(skb, &sk->sk_receive_queue); 1799 __kfree_skb(skb); 1800 } 1801 #endif 1802 1803 static inline 1804 struct net *sock_net(const struct sock *sk) 1805 { 1806 return read_pnet(&sk->sk_net); 1807 } 1808 1809 static inline 1810 void sock_net_set(struct sock *sk, struct net *net) 1811 { 1812 write_pnet(&sk->sk_net, net); 1813 } 1814 1815 /* 1816 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 1817 * They should not hold a reference to a namespace in order to allow 1818 * to stop it. 1819 * Sockets after sk_change_net should be released using sk_release_kernel 1820 */ 1821 static inline void sk_change_net(struct sock *sk, struct net *net) 1822 { 1823 put_net(sock_net(sk)); 1824 sock_net_set(sk, hold_net(net)); 1825 } 1826 1827 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 1828 { 1829 if (unlikely(skb->sk)) { 1830 struct sock *sk = skb->sk; 1831 1832 skb->destructor = NULL; 1833 skb->sk = NULL; 1834 return sk; 1835 } 1836 return NULL; 1837 } 1838 1839 extern void sock_enable_timestamp(struct sock *sk, int flag); 1840 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 1841 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 1842 1843 /* 1844 * Enable debug/info messages 1845 */ 1846 extern int net_msg_warn; 1847 #define NETDEBUG(fmt, args...) \ 1848 do { if (net_msg_warn) printk(fmt,##args); } while (0) 1849 1850 #define LIMIT_NETDEBUG(fmt, args...) \ 1851 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 1852 1853 extern __u32 sysctl_wmem_max; 1854 extern __u32 sysctl_rmem_max; 1855 1856 extern void sk_init(void); 1857 1858 extern int sysctl_optmem_max; 1859 1860 extern __u32 sysctl_wmem_default; 1861 extern __u32 sysctl_rmem_default; 1862 1863 #endif /* _SOCK_H */ 1864