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/bitops.h> 50 #include <linux/lockdep.h> 51 #include <linux/netdevice.h> 52 #include <linux/skbuff.h> /* struct sk_buff */ 53 #include <linux/mm.h> 54 #include <linux/security.h> 55 #include <linux/slab.h> 56 #include <linux/uaccess.h> 57 #include <linux/memcontrol.h> 58 #include <linux/res_counter.h> 59 #include <linux/static_key.h> 60 #include <linux/aio.h> 61 #include <linux/sched.h> 62 63 #include <linux/filter.h> 64 #include <linux/rculist_nulls.h> 65 #include <linux/poll.h> 66 67 #include <linux/atomic.h> 68 #include <net/dst.h> 69 #include <net/checksum.h> 70 71 struct cgroup; 72 struct cgroup_subsys; 73 #ifdef CONFIG_NET 74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss); 75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg); 76 #else 77 static inline 78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 79 { 80 return 0; 81 } 82 static inline 83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 84 { 85 } 86 #endif 87 /* 88 * This structure really needs to be cleaned up. 89 * Most of it is for TCP, and not used by any of 90 * the other protocols. 91 */ 92 93 /* Define this to get the SOCK_DBG debugging facility. */ 94 #define SOCK_DEBUGGING 95 #ifdef SOCK_DEBUGGING 96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 97 printk(KERN_DEBUG msg); } while (0) 98 #else 99 /* Validate arguments and do nothing */ 100 static inline __printf(2, 3) 101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 102 { 103 } 104 #endif 105 106 /* This is the per-socket lock. The spinlock provides a synchronization 107 * between user contexts and software interrupt processing, whereas the 108 * mini-semaphore synchronizes multiple users amongst themselves. 109 */ 110 typedef struct { 111 spinlock_t slock; 112 int owned; 113 wait_queue_head_t wq; 114 /* 115 * We express the mutex-alike socket_lock semantics 116 * to the lock validator by explicitly managing 117 * the slock as a lock variant (in addition to 118 * the slock itself): 119 */ 120 #ifdef CONFIG_DEBUG_LOCK_ALLOC 121 struct lockdep_map dep_map; 122 #endif 123 } socket_lock_t; 124 125 struct sock; 126 struct proto; 127 struct net; 128 129 typedef __u32 __bitwise __portpair; 130 typedef __u64 __bitwise __addrpair; 131 132 /** 133 * struct sock_common - minimal network layer representation of sockets 134 * @skc_daddr: Foreign IPv4 addr 135 * @skc_rcv_saddr: Bound local IPv4 addr 136 * @skc_hash: hash value used with various protocol lookup tables 137 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 138 * @skc_dport: placeholder for inet_dport/tw_dport 139 * @skc_num: placeholder for inet_num/tw_num 140 * @skc_family: network address family 141 * @skc_state: Connection state 142 * @skc_reuse: %SO_REUSEADDR setting 143 * @skc_bound_dev_if: bound device index if != 0 144 * @skc_bind_node: bind hash linkage for various protocol lookup tables 145 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 146 * @skc_prot: protocol handlers inside a network family 147 * @skc_net: reference to the network namespace of this socket 148 * @skc_node: main hash linkage for various protocol lookup tables 149 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 150 * @skc_tx_queue_mapping: tx queue number for this connection 151 * @skc_refcnt: reference count 152 * 153 * This is the minimal network layer representation of sockets, the header 154 * for struct sock and struct inet_timewait_sock. 155 */ 156 struct sock_common { 157 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 158 * address on 64bit arches : cf INET_MATCH() and INET_TW_MATCH() 159 */ 160 union { 161 __addrpair skc_addrpair; 162 struct { 163 __be32 skc_daddr; 164 __be32 skc_rcv_saddr; 165 }; 166 }; 167 union { 168 unsigned int skc_hash; 169 __u16 skc_u16hashes[2]; 170 }; 171 /* skc_dport && skc_num must be grouped as well */ 172 union { 173 __portpair skc_portpair; 174 struct { 175 __be16 skc_dport; 176 __u16 skc_num; 177 }; 178 }; 179 180 unsigned short skc_family; 181 volatile unsigned char skc_state; 182 unsigned char skc_reuse; 183 int skc_bound_dev_if; 184 union { 185 struct hlist_node skc_bind_node; 186 struct hlist_nulls_node skc_portaddr_node; 187 }; 188 struct proto *skc_prot; 189 #ifdef CONFIG_NET_NS 190 struct net *skc_net; 191 #endif 192 /* 193 * fields between dontcopy_begin/dontcopy_end 194 * are not copied in sock_copy() 195 */ 196 /* private: */ 197 int skc_dontcopy_begin[0]; 198 /* public: */ 199 union { 200 struct hlist_node skc_node; 201 struct hlist_nulls_node skc_nulls_node; 202 }; 203 int skc_tx_queue_mapping; 204 atomic_t skc_refcnt; 205 /* private: */ 206 int skc_dontcopy_end[0]; 207 /* public: */ 208 }; 209 210 struct cg_proto; 211 /** 212 * struct sock - network layer representation of sockets 213 * @__sk_common: shared layout with inet_timewait_sock 214 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 215 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 216 * @sk_lock: synchronizer 217 * @sk_rcvbuf: size of receive buffer in bytes 218 * @sk_wq: sock wait queue and async head 219 * @sk_rx_dst: receive input route used by early tcp demux 220 * @sk_dst_cache: destination cache 221 * @sk_dst_lock: destination cache lock 222 * @sk_policy: flow policy 223 * @sk_receive_queue: incoming packets 224 * @sk_wmem_alloc: transmit queue bytes committed 225 * @sk_write_queue: Packet sending queue 226 * @sk_async_wait_queue: DMA copied packets 227 * @sk_omem_alloc: "o" is "option" or "other" 228 * @sk_wmem_queued: persistent queue size 229 * @sk_forward_alloc: space allocated forward 230 * @sk_allocation: allocation mode 231 * @sk_sndbuf: size of send buffer in bytes 232 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 233 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 234 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets 235 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 236 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 237 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 238 * @sk_gso_max_size: Maximum GSO segment size to build 239 * @sk_gso_max_segs: Maximum number of GSO segments 240 * @sk_lingertime: %SO_LINGER l_linger setting 241 * @sk_backlog: always used with the per-socket spinlock held 242 * @sk_callback_lock: used with the callbacks in the end of this struct 243 * @sk_error_queue: rarely used 244 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 245 * IPV6_ADDRFORM for instance) 246 * @sk_err: last error 247 * @sk_err_soft: errors that don't cause failure but are the cause of a 248 * persistent failure not just 'timed out' 249 * @sk_drops: raw/udp drops counter 250 * @sk_ack_backlog: current listen backlog 251 * @sk_max_ack_backlog: listen backlog set in listen() 252 * @sk_priority: %SO_PRIORITY setting 253 * @sk_cgrp_prioidx: socket group's priority map index 254 * @sk_type: socket type (%SOCK_STREAM, etc) 255 * @sk_protocol: which protocol this socket belongs in this network family 256 * @sk_peer_pid: &struct pid for this socket's peer 257 * @sk_peer_cred: %SO_PEERCRED setting 258 * @sk_rcvlowat: %SO_RCVLOWAT setting 259 * @sk_rcvtimeo: %SO_RCVTIMEO setting 260 * @sk_sndtimeo: %SO_SNDTIMEO setting 261 * @sk_rxhash: flow hash received from netif layer 262 * @sk_filter: socket filtering instructions 263 * @sk_protinfo: private area, net family specific, when not using slab 264 * @sk_timer: sock cleanup timer 265 * @sk_stamp: time stamp of last packet received 266 * @sk_socket: Identd and reporting IO signals 267 * @sk_user_data: RPC layer private data 268 * @sk_frag: cached page frag 269 * @sk_peek_off: current peek_offset value 270 * @sk_send_head: front of stuff to transmit 271 * @sk_security: used by security modules 272 * @sk_mark: generic packet mark 273 * @sk_classid: this socket's cgroup classid 274 * @sk_cgrp: this socket's cgroup-specific proto data 275 * @sk_write_pending: a write to stream socket waits to start 276 * @sk_state_change: callback to indicate change in the state of the sock 277 * @sk_data_ready: callback to indicate there is data to be processed 278 * @sk_write_space: callback to indicate there is bf sending space available 279 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 280 * @sk_backlog_rcv: callback to process the backlog 281 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 282 */ 283 struct sock { 284 /* 285 * Now struct inet_timewait_sock also uses sock_common, so please just 286 * don't add nothing before this first member (__sk_common) --acme 287 */ 288 struct sock_common __sk_common; 289 #define sk_node __sk_common.skc_node 290 #define sk_nulls_node __sk_common.skc_nulls_node 291 #define sk_refcnt __sk_common.skc_refcnt 292 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 293 294 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 295 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 296 #define sk_hash __sk_common.skc_hash 297 #define sk_family __sk_common.skc_family 298 #define sk_state __sk_common.skc_state 299 #define sk_reuse __sk_common.skc_reuse 300 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 301 #define sk_bind_node __sk_common.skc_bind_node 302 #define sk_prot __sk_common.skc_prot 303 #define sk_net __sk_common.skc_net 304 socket_lock_t sk_lock; 305 struct sk_buff_head sk_receive_queue; 306 /* 307 * The backlog queue is special, it is always used with 308 * the per-socket spinlock held and requires low latency 309 * access. Therefore we special case it's implementation. 310 * Note : rmem_alloc is in this structure to fill a hole 311 * on 64bit arches, not because its logically part of 312 * backlog. 313 */ 314 struct { 315 atomic_t rmem_alloc; 316 int len; 317 struct sk_buff *head; 318 struct sk_buff *tail; 319 } sk_backlog; 320 #define sk_rmem_alloc sk_backlog.rmem_alloc 321 int sk_forward_alloc; 322 #ifdef CONFIG_RPS 323 __u32 sk_rxhash; 324 #endif 325 atomic_t sk_drops; 326 int sk_rcvbuf; 327 328 struct sk_filter __rcu *sk_filter; 329 struct socket_wq __rcu *sk_wq; 330 331 #ifdef CONFIG_NET_DMA 332 struct sk_buff_head sk_async_wait_queue; 333 #endif 334 335 #ifdef CONFIG_XFRM 336 struct xfrm_policy *sk_policy[2]; 337 #endif 338 unsigned long sk_flags; 339 struct dst_entry *sk_rx_dst; 340 struct dst_entry *sk_dst_cache; 341 spinlock_t sk_dst_lock; 342 atomic_t sk_wmem_alloc; 343 atomic_t sk_omem_alloc; 344 int sk_sndbuf; 345 struct sk_buff_head sk_write_queue; 346 kmemcheck_bitfield_begin(flags); 347 unsigned int sk_shutdown : 2, 348 sk_no_check : 2, 349 sk_userlocks : 4, 350 sk_protocol : 8, 351 sk_type : 16; 352 kmemcheck_bitfield_end(flags); 353 int sk_wmem_queued; 354 gfp_t sk_allocation; 355 netdev_features_t sk_route_caps; 356 netdev_features_t sk_route_nocaps; 357 int sk_gso_type; 358 unsigned int sk_gso_max_size; 359 u16 sk_gso_max_segs; 360 int sk_rcvlowat; 361 unsigned long sk_lingertime; 362 struct sk_buff_head sk_error_queue; 363 struct proto *sk_prot_creator; 364 rwlock_t sk_callback_lock; 365 int sk_err, 366 sk_err_soft; 367 unsigned short sk_ack_backlog; 368 unsigned short sk_max_ack_backlog; 369 __u32 sk_priority; 370 #ifdef CONFIG_CGROUPS 371 __u32 sk_cgrp_prioidx; 372 #endif 373 struct pid *sk_peer_pid; 374 const struct cred *sk_peer_cred; 375 long sk_rcvtimeo; 376 long sk_sndtimeo; 377 void *sk_protinfo; 378 struct timer_list sk_timer; 379 ktime_t sk_stamp; 380 struct socket *sk_socket; 381 void *sk_user_data; 382 struct page_frag sk_frag; 383 struct sk_buff *sk_send_head; 384 __s32 sk_peek_off; 385 int sk_write_pending; 386 #ifdef CONFIG_SECURITY 387 void *sk_security; 388 #endif 389 __u32 sk_mark; 390 u32 sk_classid; 391 struct cg_proto *sk_cgrp; 392 void (*sk_state_change)(struct sock *sk); 393 void (*sk_data_ready)(struct sock *sk, int bytes); 394 void (*sk_write_space)(struct sock *sk); 395 void (*sk_error_report)(struct sock *sk); 396 int (*sk_backlog_rcv)(struct sock *sk, 397 struct sk_buff *skb); 398 void (*sk_destruct)(struct sock *sk); 399 }; 400 401 /* 402 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 403 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 404 * on a socket means that the socket will reuse everybody else's port 405 * without looking at the other's sk_reuse value. 406 */ 407 408 #define SK_NO_REUSE 0 409 #define SK_CAN_REUSE 1 410 #define SK_FORCE_REUSE 2 411 412 static inline int sk_peek_offset(struct sock *sk, int flags) 413 { 414 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 415 return sk->sk_peek_off; 416 else 417 return 0; 418 } 419 420 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 421 { 422 if (sk->sk_peek_off >= 0) { 423 if (sk->sk_peek_off >= val) 424 sk->sk_peek_off -= val; 425 else 426 sk->sk_peek_off = 0; 427 } 428 } 429 430 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 431 { 432 if (sk->sk_peek_off >= 0) 433 sk->sk_peek_off += val; 434 } 435 436 /* 437 * Hashed lists helper routines 438 */ 439 static inline struct sock *sk_entry(const struct hlist_node *node) 440 { 441 return hlist_entry(node, struct sock, sk_node); 442 } 443 444 static inline struct sock *__sk_head(const struct hlist_head *head) 445 { 446 return hlist_entry(head->first, struct sock, sk_node); 447 } 448 449 static inline struct sock *sk_head(const struct hlist_head *head) 450 { 451 return hlist_empty(head) ? NULL : __sk_head(head); 452 } 453 454 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 455 { 456 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 457 } 458 459 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 460 { 461 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 462 } 463 464 static inline struct sock *sk_next(const struct sock *sk) 465 { 466 return sk->sk_node.next ? 467 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 468 } 469 470 static inline struct sock *sk_nulls_next(const struct sock *sk) 471 { 472 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 473 hlist_nulls_entry(sk->sk_nulls_node.next, 474 struct sock, sk_nulls_node) : 475 NULL; 476 } 477 478 static inline bool sk_unhashed(const struct sock *sk) 479 { 480 return hlist_unhashed(&sk->sk_node); 481 } 482 483 static inline bool sk_hashed(const struct sock *sk) 484 { 485 return !sk_unhashed(sk); 486 } 487 488 static inline void sk_node_init(struct hlist_node *node) 489 { 490 node->pprev = NULL; 491 } 492 493 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 494 { 495 node->pprev = NULL; 496 } 497 498 static inline void __sk_del_node(struct sock *sk) 499 { 500 __hlist_del(&sk->sk_node); 501 } 502 503 /* NB: equivalent to hlist_del_init_rcu */ 504 static inline bool __sk_del_node_init(struct sock *sk) 505 { 506 if (sk_hashed(sk)) { 507 __sk_del_node(sk); 508 sk_node_init(&sk->sk_node); 509 return true; 510 } 511 return false; 512 } 513 514 /* Grab socket reference count. This operation is valid only 515 when sk is ALREADY grabbed f.e. it is found in hash table 516 or a list and the lookup is made under lock preventing hash table 517 modifications. 518 */ 519 520 static inline void sock_hold(struct sock *sk) 521 { 522 atomic_inc(&sk->sk_refcnt); 523 } 524 525 /* Ungrab socket in the context, which assumes that socket refcnt 526 cannot hit zero, f.e. it is true in context of any socketcall. 527 */ 528 static inline void __sock_put(struct sock *sk) 529 { 530 atomic_dec(&sk->sk_refcnt); 531 } 532 533 static inline bool sk_del_node_init(struct sock *sk) 534 { 535 bool rc = __sk_del_node_init(sk); 536 537 if (rc) { 538 /* paranoid for a while -acme */ 539 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 540 __sock_put(sk); 541 } 542 return rc; 543 } 544 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 545 546 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 547 { 548 if (sk_hashed(sk)) { 549 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 550 return true; 551 } 552 return false; 553 } 554 555 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 556 { 557 bool rc = __sk_nulls_del_node_init_rcu(sk); 558 559 if (rc) { 560 /* paranoid for a while -acme */ 561 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 562 __sock_put(sk); 563 } 564 return rc; 565 } 566 567 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 568 { 569 hlist_add_head(&sk->sk_node, list); 570 } 571 572 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 573 { 574 sock_hold(sk); 575 __sk_add_node(sk, list); 576 } 577 578 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 579 { 580 sock_hold(sk); 581 hlist_add_head_rcu(&sk->sk_node, list); 582 } 583 584 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 585 { 586 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 587 } 588 589 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 590 { 591 sock_hold(sk); 592 __sk_nulls_add_node_rcu(sk, list); 593 } 594 595 static inline void __sk_del_bind_node(struct sock *sk) 596 { 597 __hlist_del(&sk->sk_bind_node); 598 } 599 600 static inline void sk_add_bind_node(struct sock *sk, 601 struct hlist_head *list) 602 { 603 hlist_add_head(&sk->sk_bind_node, list); 604 } 605 606 #define sk_for_each(__sk, node, list) \ 607 hlist_for_each_entry(__sk, node, list, sk_node) 608 #define sk_for_each_rcu(__sk, node, list) \ 609 hlist_for_each_entry_rcu(__sk, node, list, sk_node) 610 #define sk_nulls_for_each(__sk, node, list) \ 611 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 612 #define sk_nulls_for_each_rcu(__sk, node, list) \ 613 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 614 #define sk_for_each_from(__sk, node) \ 615 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 616 hlist_for_each_entry_from(__sk, node, sk_node) 617 #define sk_nulls_for_each_from(__sk, node) \ 618 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 619 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 620 #define sk_for_each_safe(__sk, node, tmp, list) \ 621 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node) 622 #define sk_for_each_bound(__sk, node, list) \ 623 hlist_for_each_entry(__sk, node, list, sk_bind_node) 624 625 static inline struct user_namespace *sk_user_ns(struct sock *sk) 626 { 627 /* Careful only use this in a context where these parameters 628 * can not change and must all be valid, such as recvmsg from 629 * userspace. 630 */ 631 return sk->sk_socket->file->f_cred->user_ns; 632 } 633 634 /* Sock flags */ 635 enum sock_flags { 636 SOCK_DEAD, 637 SOCK_DONE, 638 SOCK_URGINLINE, 639 SOCK_KEEPOPEN, 640 SOCK_LINGER, 641 SOCK_DESTROY, 642 SOCK_BROADCAST, 643 SOCK_TIMESTAMP, 644 SOCK_ZAPPED, 645 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 646 SOCK_DBG, /* %SO_DEBUG setting */ 647 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 648 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 649 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 650 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 651 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 652 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 653 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 654 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 655 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 656 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 657 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 658 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 659 SOCK_FASYNC, /* fasync() active */ 660 SOCK_RXQ_OVFL, 661 SOCK_ZEROCOPY, /* buffers from userspace */ 662 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 663 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 664 * Will use last 4 bytes of packet sent from 665 * user-space instead. 666 */ 667 }; 668 669 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 670 { 671 nsk->sk_flags = osk->sk_flags; 672 } 673 674 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 675 { 676 __set_bit(flag, &sk->sk_flags); 677 } 678 679 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 680 { 681 __clear_bit(flag, &sk->sk_flags); 682 } 683 684 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 685 { 686 return test_bit(flag, &sk->sk_flags); 687 } 688 689 #ifdef CONFIG_NET 690 extern struct static_key memalloc_socks; 691 static inline int sk_memalloc_socks(void) 692 { 693 return static_key_false(&memalloc_socks); 694 } 695 #else 696 697 static inline int sk_memalloc_socks(void) 698 { 699 return 0; 700 } 701 702 #endif 703 704 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 705 { 706 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 707 } 708 709 static inline void sk_acceptq_removed(struct sock *sk) 710 { 711 sk->sk_ack_backlog--; 712 } 713 714 static inline void sk_acceptq_added(struct sock *sk) 715 { 716 sk->sk_ack_backlog++; 717 } 718 719 static inline bool sk_acceptq_is_full(const struct sock *sk) 720 { 721 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 722 } 723 724 /* 725 * Compute minimal free write space needed to queue new packets. 726 */ 727 static inline int sk_stream_min_wspace(const struct sock *sk) 728 { 729 return sk->sk_wmem_queued >> 1; 730 } 731 732 static inline int sk_stream_wspace(const struct sock *sk) 733 { 734 return sk->sk_sndbuf - sk->sk_wmem_queued; 735 } 736 737 extern void sk_stream_write_space(struct sock *sk); 738 739 static inline bool sk_stream_memory_free(const struct sock *sk) 740 { 741 return sk->sk_wmem_queued < sk->sk_sndbuf; 742 } 743 744 /* OOB backlog add */ 745 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 746 { 747 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 748 skb_dst_force(skb); 749 750 if (!sk->sk_backlog.tail) 751 sk->sk_backlog.head = skb; 752 else 753 sk->sk_backlog.tail->next = skb; 754 755 sk->sk_backlog.tail = skb; 756 skb->next = NULL; 757 } 758 759 /* 760 * Take into account size of receive queue and backlog queue 761 * Do not take into account this skb truesize, 762 * to allow even a single big packet to come. 763 */ 764 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb, 765 unsigned int limit) 766 { 767 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 768 769 return qsize > limit; 770 } 771 772 /* The per-socket spinlock must be held here. */ 773 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 774 unsigned int limit) 775 { 776 if (sk_rcvqueues_full(sk, skb, limit)) 777 return -ENOBUFS; 778 779 __sk_add_backlog(sk, skb); 780 sk->sk_backlog.len += skb->truesize; 781 return 0; 782 } 783 784 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 785 786 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 787 { 788 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 789 return __sk_backlog_rcv(sk, skb); 790 791 return sk->sk_backlog_rcv(sk, skb); 792 } 793 794 static inline void sock_rps_record_flow(const struct sock *sk) 795 { 796 #ifdef CONFIG_RPS 797 struct rps_sock_flow_table *sock_flow_table; 798 799 rcu_read_lock(); 800 sock_flow_table = rcu_dereference(rps_sock_flow_table); 801 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 802 rcu_read_unlock(); 803 #endif 804 } 805 806 static inline void sock_rps_reset_flow(const struct sock *sk) 807 { 808 #ifdef CONFIG_RPS 809 struct rps_sock_flow_table *sock_flow_table; 810 811 rcu_read_lock(); 812 sock_flow_table = rcu_dereference(rps_sock_flow_table); 813 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 814 rcu_read_unlock(); 815 #endif 816 } 817 818 static inline void sock_rps_save_rxhash(struct sock *sk, 819 const struct sk_buff *skb) 820 { 821 #ifdef CONFIG_RPS 822 if (unlikely(sk->sk_rxhash != skb->rxhash)) { 823 sock_rps_reset_flow(sk); 824 sk->sk_rxhash = skb->rxhash; 825 } 826 #endif 827 } 828 829 static inline void sock_rps_reset_rxhash(struct sock *sk) 830 { 831 #ifdef CONFIG_RPS 832 sock_rps_reset_flow(sk); 833 sk->sk_rxhash = 0; 834 #endif 835 } 836 837 #define sk_wait_event(__sk, __timeo, __condition) \ 838 ({ int __rc; \ 839 release_sock(__sk); \ 840 __rc = __condition; \ 841 if (!__rc) { \ 842 *(__timeo) = schedule_timeout(*(__timeo)); \ 843 } \ 844 lock_sock(__sk); \ 845 __rc = __condition; \ 846 __rc; \ 847 }) 848 849 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 850 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 851 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 852 extern int sk_stream_error(struct sock *sk, int flags, int err); 853 extern void sk_stream_kill_queues(struct sock *sk); 854 extern void sk_set_memalloc(struct sock *sk); 855 extern void sk_clear_memalloc(struct sock *sk); 856 857 extern int sk_wait_data(struct sock *sk, long *timeo); 858 859 struct request_sock_ops; 860 struct timewait_sock_ops; 861 struct inet_hashinfo; 862 struct raw_hashinfo; 863 struct module; 864 865 /* Networking protocol blocks we attach to sockets. 866 * socket layer -> transport layer interface 867 * transport -> network interface is defined by struct inet_proto 868 */ 869 struct proto { 870 void (*close)(struct sock *sk, 871 long timeout); 872 int (*connect)(struct sock *sk, 873 struct sockaddr *uaddr, 874 int addr_len); 875 int (*disconnect)(struct sock *sk, int flags); 876 877 struct sock * (*accept)(struct sock *sk, int flags, int *err); 878 879 int (*ioctl)(struct sock *sk, int cmd, 880 unsigned long arg); 881 int (*init)(struct sock *sk); 882 void (*destroy)(struct sock *sk); 883 void (*shutdown)(struct sock *sk, int how); 884 int (*setsockopt)(struct sock *sk, int level, 885 int optname, char __user *optval, 886 unsigned int optlen); 887 int (*getsockopt)(struct sock *sk, int level, 888 int optname, char __user *optval, 889 int __user *option); 890 #ifdef CONFIG_COMPAT 891 int (*compat_setsockopt)(struct sock *sk, 892 int level, 893 int optname, char __user *optval, 894 unsigned int optlen); 895 int (*compat_getsockopt)(struct sock *sk, 896 int level, 897 int optname, char __user *optval, 898 int __user *option); 899 int (*compat_ioctl)(struct sock *sk, 900 unsigned int cmd, unsigned long arg); 901 #endif 902 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 903 struct msghdr *msg, size_t len); 904 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 905 struct msghdr *msg, 906 size_t len, int noblock, int flags, 907 int *addr_len); 908 int (*sendpage)(struct sock *sk, struct page *page, 909 int offset, size_t size, int flags); 910 int (*bind)(struct sock *sk, 911 struct sockaddr *uaddr, int addr_len); 912 913 int (*backlog_rcv) (struct sock *sk, 914 struct sk_buff *skb); 915 916 void (*release_cb)(struct sock *sk); 917 void (*mtu_reduced)(struct sock *sk); 918 919 /* Keeping track of sk's, looking them up, and port selection methods. */ 920 void (*hash)(struct sock *sk); 921 void (*unhash)(struct sock *sk); 922 void (*rehash)(struct sock *sk); 923 int (*get_port)(struct sock *sk, unsigned short snum); 924 void (*clear_sk)(struct sock *sk, int size); 925 926 /* Keeping track of sockets in use */ 927 #ifdef CONFIG_PROC_FS 928 unsigned int inuse_idx; 929 #endif 930 931 /* Memory pressure */ 932 void (*enter_memory_pressure)(struct sock *sk); 933 atomic_long_t *memory_allocated; /* Current allocated memory. */ 934 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 935 /* 936 * Pressure flag: try to collapse. 937 * Technical note: it is used by multiple contexts non atomically. 938 * All the __sk_mem_schedule() is of this nature: accounting 939 * is strict, actions are advisory and have some latency. 940 */ 941 int *memory_pressure; 942 long *sysctl_mem; 943 int *sysctl_wmem; 944 int *sysctl_rmem; 945 int max_header; 946 bool no_autobind; 947 948 struct kmem_cache *slab; 949 unsigned int obj_size; 950 int slab_flags; 951 952 struct percpu_counter *orphan_count; 953 954 struct request_sock_ops *rsk_prot; 955 struct timewait_sock_ops *twsk_prot; 956 957 union { 958 struct inet_hashinfo *hashinfo; 959 struct udp_table *udp_table; 960 struct raw_hashinfo *raw_hash; 961 } h; 962 963 struct module *owner; 964 965 char name[32]; 966 967 struct list_head node; 968 #ifdef SOCK_REFCNT_DEBUG 969 atomic_t socks; 970 #endif 971 #ifdef CONFIG_MEMCG_KMEM 972 /* 973 * cgroup specific init/deinit functions. Called once for all 974 * protocols that implement it, from cgroups populate function. 975 * This function has to setup any files the protocol want to 976 * appear in the kmem cgroup filesystem. 977 */ 978 int (*init_cgroup)(struct mem_cgroup *memcg, 979 struct cgroup_subsys *ss); 980 void (*destroy_cgroup)(struct mem_cgroup *memcg); 981 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 982 #endif 983 }; 984 985 /* 986 * Bits in struct cg_proto.flags 987 */ 988 enum cg_proto_flags { 989 /* Currently active and new sockets should be assigned to cgroups */ 990 MEMCG_SOCK_ACTIVE, 991 /* It was ever activated; we must disarm static keys on destruction */ 992 MEMCG_SOCK_ACTIVATED, 993 }; 994 995 struct cg_proto { 996 void (*enter_memory_pressure)(struct sock *sk); 997 struct res_counter *memory_allocated; /* Current allocated memory. */ 998 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 999 int *memory_pressure; 1000 long *sysctl_mem; 1001 unsigned long flags; 1002 /* 1003 * memcg field is used to find which memcg we belong directly 1004 * Each memcg struct can hold more than one cg_proto, so container_of 1005 * won't really cut. 1006 * 1007 * The elegant solution would be having an inverse function to 1008 * proto_cgroup in struct proto, but that means polluting the structure 1009 * for everybody, instead of just for memcg users. 1010 */ 1011 struct mem_cgroup *memcg; 1012 }; 1013 1014 extern int proto_register(struct proto *prot, int alloc_slab); 1015 extern void proto_unregister(struct proto *prot); 1016 1017 static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1018 { 1019 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1020 } 1021 1022 static inline bool memcg_proto_activated(struct cg_proto *cg_proto) 1023 { 1024 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags); 1025 } 1026 1027 #ifdef SOCK_REFCNT_DEBUG 1028 static inline void sk_refcnt_debug_inc(struct sock *sk) 1029 { 1030 atomic_inc(&sk->sk_prot->socks); 1031 } 1032 1033 static inline void sk_refcnt_debug_dec(struct sock *sk) 1034 { 1035 atomic_dec(&sk->sk_prot->socks); 1036 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1037 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1038 } 1039 1040 inline void sk_refcnt_debug_release(const struct sock *sk) 1041 { 1042 if (atomic_read(&sk->sk_refcnt) != 1) 1043 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1044 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1045 } 1046 #else /* SOCK_REFCNT_DEBUG */ 1047 #define sk_refcnt_debug_inc(sk) do { } while (0) 1048 #define sk_refcnt_debug_dec(sk) do { } while (0) 1049 #define sk_refcnt_debug_release(sk) do { } while (0) 1050 #endif /* SOCK_REFCNT_DEBUG */ 1051 1052 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1053 extern struct static_key memcg_socket_limit_enabled; 1054 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1055 struct cg_proto *cg_proto) 1056 { 1057 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1058 } 1059 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1060 #else 1061 #define mem_cgroup_sockets_enabled 0 1062 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1063 struct cg_proto *cg_proto) 1064 { 1065 return NULL; 1066 } 1067 #endif 1068 1069 1070 static inline bool sk_has_memory_pressure(const struct sock *sk) 1071 { 1072 return sk->sk_prot->memory_pressure != NULL; 1073 } 1074 1075 static inline bool sk_under_memory_pressure(const struct sock *sk) 1076 { 1077 if (!sk->sk_prot->memory_pressure) 1078 return false; 1079 1080 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1081 return !!*sk->sk_cgrp->memory_pressure; 1082 1083 return !!*sk->sk_prot->memory_pressure; 1084 } 1085 1086 static inline void sk_leave_memory_pressure(struct sock *sk) 1087 { 1088 int *memory_pressure = sk->sk_prot->memory_pressure; 1089 1090 if (!memory_pressure) 1091 return; 1092 1093 if (*memory_pressure) 1094 *memory_pressure = 0; 1095 1096 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1097 struct cg_proto *cg_proto = sk->sk_cgrp; 1098 struct proto *prot = sk->sk_prot; 1099 1100 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1101 if (*cg_proto->memory_pressure) 1102 *cg_proto->memory_pressure = 0; 1103 } 1104 1105 } 1106 1107 static inline void sk_enter_memory_pressure(struct sock *sk) 1108 { 1109 if (!sk->sk_prot->enter_memory_pressure) 1110 return; 1111 1112 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1113 struct cg_proto *cg_proto = sk->sk_cgrp; 1114 struct proto *prot = sk->sk_prot; 1115 1116 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1117 cg_proto->enter_memory_pressure(sk); 1118 } 1119 1120 sk->sk_prot->enter_memory_pressure(sk); 1121 } 1122 1123 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1124 { 1125 long *prot = sk->sk_prot->sysctl_mem; 1126 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1127 prot = sk->sk_cgrp->sysctl_mem; 1128 return prot[index]; 1129 } 1130 1131 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1132 unsigned long amt, 1133 int *parent_status) 1134 { 1135 struct res_counter *fail; 1136 int ret; 1137 1138 ret = res_counter_charge_nofail(prot->memory_allocated, 1139 amt << PAGE_SHIFT, &fail); 1140 if (ret < 0) 1141 *parent_status = OVER_LIMIT; 1142 } 1143 1144 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1145 unsigned long amt) 1146 { 1147 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT); 1148 } 1149 1150 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot) 1151 { 1152 u64 ret; 1153 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE); 1154 return ret >> PAGE_SHIFT; 1155 } 1156 1157 static inline long 1158 sk_memory_allocated(const struct sock *sk) 1159 { 1160 struct proto *prot = sk->sk_prot; 1161 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1162 return memcg_memory_allocated_read(sk->sk_cgrp); 1163 1164 return atomic_long_read(prot->memory_allocated); 1165 } 1166 1167 static inline long 1168 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1169 { 1170 struct proto *prot = sk->sk_prot; 1171 1172 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1173 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1174 /* update the root cgroup regardless */ 1175 atomic_long_add_return(amt, prot->memory_allocated); 1176 return memcg_memory_allocated_read(sk->sk_cgrp); 1177 } 1178 1179 return atomic_long_add_return(amt, prot->memory_allocated); 1180 } 1181 1182 static inline void 1183 sk_memory_allocated_sub(struct sock *sk, int amt) 1184 { 1185 struct proto *prot = sk->sk_prot; 1186 1187 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1188 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1189 1190 atomic_long_sub(amt, prot->memory_allocated); 1191 } 1192 1193 static inline void sk_sockets_allocated_dec(struct sock *sk) 1194 { 1195 struct proto *prot = sk->sk_prot; 1196 1197 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1198 struct cg_proto *cg_proto = sk->sk_cgrp; 1199 1200 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1201 percpu_counter_dec(cg_proto->sockets_allocated); 1202 } 1203 1204 percpu_counter_dec(prot->sockets_allocated); 1205 } 1206 1207 static inline void sk_sockets_allocated_inc(struct sock *sk) 1208 { 1209 struct proto *prot = sk->sk_prot; 1210 1211 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1212 struct cg_proto *cg_proto = sk->sk_cgrp; 1213 1214 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1215 percpu_counter_inc(cg_proto->sockets_allocated); 1216 } 1217 1218 percpu_counter_inc(prot->sockets_allocated); 1219 } 1220 1221 static inline int 1222 sk_sockets_allocated_read_positive(struct sock *sk) 1223 { 1224 struct proto *prot = sk->sk_prot; 1225 1226 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1227 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated); 1228 1229 return percpu_counter_read_positive(prot->sockets_allocated); 1230 } 1231 1232 static inline int 1233 proto_sockets_allocated_sum_positive(struct proto *prot) 1234 { 1235 return percpu_counter_sum_positive(prot->sockets_allocated); 1236 } 1237 1238 static inline long 1239 proto_memory_allocated(struct proto *prot) 1240 { 1241 return atomic_long_read(prot->memory_allocated); 1242 } 1243 1244 static inline bool 1245 proto_memory_pressure(struct proto *prot) 1246 { 1247 if (!prot->memory_pressure) 1248 return false; 1249 return !!*prot->memory_pressure; 1250 } 1251 1252 1253 #ifdef CONFIG_PROC_FS 1254 /* Called with local bh disabled */ 1255 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1256 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 1257 #else 1258 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1259 int inc) 1260 { 1261 } 1262 #endif 1263 1264 1265 /* With per-bucket locks this operation is not-atomic, so that 1266 * this version is not worse. 1267 */ 1268 static inline void __sk_prot_rehash(struct sock *sk) 1269 { 1270 sk->sk_prot->unhash(sk); 1271 sk->sk_prot->hash(sk); 1272 } 1273 1274 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1275 1276 /* About 10 seconds */ 1277 #define SOCK_DESTROY_TIME (10*HZ) 1278 1279 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1280 #define PROT_SOCK 1024 1281 1282 #define SHUTDOWN_MASK 3 1283 #define RCV_SHUTDOWN 1 1284 #define SEND_SHUTDOWN 2 1285 1286 #define SOCK_SNDBUF_LOCK 1 1287 #define SOCK_RCVBUF_LOCK 2 1288 #define SOCK_BINDADDR_LOCK 4 1289 #define SOCK_BINDPORT_LOCK 8 1290 1291 /* sock_iocb: used to kick off async processing of socket ios */ 1292 struct sock_iocb { 1293 struct list_head list; 1294 1295 int flags; 1296 int size; 1297 struct socket *sock; 1298 struct sock *sk; 1299 struct scm_cookie *scm; 1300 struct msghdr *msg, async_msg; 1301 struct kiocb *kiocb; 1302 }; 1303 1304 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 1305 { 1306 return (struct sock_iocb *)iocb->private; 1307 } 1308 1309 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 1310 { 1311 return si->kiocb; 1312 } 1313 1314 struct socket_alloc { 1315 struct socket socket; 1316 struct inode vfs_inode; 1317 }; 1318 1319 static inline struct socket *SOCKET_I(struct inode *inode) 1320 { 1321 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1322 } 1323 1324 static inline struct inode *SOCK_INODE(struct socket *socket) 1325 { 1326 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1327 } 1328 1329 /* 1330 * Functions for memory accounting 1331 */ 1332 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 1333 extern void __sk_mem_reclaim(struct sock *sk); 1334 1335 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1336 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1337 #define SK_MEM_SEND 0 1338 #define SK_MEM_RECV 1 1339 1340 static inline int sk_mem_pages(int amt) 1341 { 1342 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1343 } 1344 1345 static inline bool sk_has_account(struct sock *sk) 1346 { 1347 /* return true if protocol supports memory accounting */ 1348 return !!sk->sk_prot->memory_allocated; 1349 } 1350 1351 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1352 { 1353 if (!sk_has_account(sk)) 1354 return true; 1355 return size <= sk->sk_forward_alloc || 1356 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1357 } 1358 1359 static inline bool 1360 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1361 { 1362 if (!sk_has_account(sk)) 1363 return true; 1364 return size<= sk->sk_forward_alloc || 1365 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1366 skb_pfmemalloc(skb); 1367 } 1368 1369 static inline void sk_mem_reclaim(struct sock *sk) 1370 { 1371 if (!sk_has_account(sk)) 1372 return; 1373 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1374 __sk_mem_reclaim(sk); 1375 } 1376 1377 static inline void sk_mem_reclaim_partial(struct sock *sk) 1378 { 1379 if (!sk_has_account(sk)) 1380 return; 1381 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1382 __sk_mem_reclaim(sk); 1383 } 1384 1385 static inline void sk_mem_charge(struct sock *sk, int size) 1386 { 1387 if (!sk_has_account(sk)) 1388 return; 1389 sk->sk_forward_alloc -= size; 1390 } 1391 1392 static inline void sk_mem_uncharge(struct sock *sk, int size) 1393 { 1394 if (!sk_has_account(sk)) 1395 return; 1396 sk->sk_forward_alloc += size; 1397 } 1398 1399 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1400 { 1401 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1402 sk->sk_wmem_queued -= skb->truesize; 1403 sk_mem_uncharge(sk, skb->truesize); 1404 __kfree_skb(skb); 1405 } 1406 1407 /* Used by processes to "lock" a socket state, so that 1408 * interrupts and bottom half handlers won't change it 1409 * from under us. It essentially blocks any incoming 1410 * packets, so that we won't get any new data or any 1411 * packets that change the state of the socket. 1412 * 1413 * While locked, BH processing will add new packets to 1414 * the backlog queue. This queue is processed by the 1415 * owner of the socket lock right before it is released. 1416 * 1417 * Since ~2.3.5 it is also exclusive sleep lock serializing 1418 * accesses from user process context. 1419 */ 1420 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1421 1422 /* 1423 * Macro so as to not evaluate some arguments when 1424 * lockdep is not enabled. 1425 * 1426 * Mark both the sk_lock and the sk_lock.slock as a 1427 * per-address-family lock class. 1428 */ 1429 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1430 do { \ 1431 sk->sk_lock.owned = 0; \ 1432 init_waitqueue_head(&sk->sk_lock.wq); \ 1433 spin_lock_init(&(sk)->sk_lock.slock); \ 1434 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1435 sizeof((sk)->sk_lock)); \ 1436 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1437 (skey), (sname)); \ 1438 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1439 } while (0) 1440 1441 extern void lock_sock_nested(struct sock *sk, int subclass); 1442 1443 static inline void lock_sock(struct sock *sk) 1444 { 1445 lock_sock_nested(sk, 0); 1446 } 1447 1448 extern void release_sock(struct sock *sk); 1449 1450 /* BH context may only use the following locking interface. */ 1451 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1452 #define bh_lock_sock_nested(__sk) \ 1453 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1454 SINGLE_DEPTH_NESTING) 1455 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1456 1457 extern bool lock_sock_fast(struct sock *sk); 1458 /** 1459 * unlock_sock_fast - complement of lock_sock_fast 1460 * @sk: socket 1461 * @slow: slow mode 1462 * 1463 * fast unlock socket for user context. 1464 * If slow mode is on, we call regular release_sock() 1465 */ 1466 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1467 { 1468 if (slow) 1469 release_sock(sk); 1470 else 1471 spin_unlock_bh(&sk->sk_lock.slock); 1472 } 1473 1474 1475 extern struct sock *sk_alloc(struct net *net, int family, 1476 gfp_t priority, 1477 struct proto *prot); 1478 extern void sk_free(struct sock *sk); 1479 extern void sk_release_kernel(struct sock *sk); 1480 extern struct sock *sk_clone_lock(const struct sock *sk, 1481 const gfp_t priority); 1482 1483 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1484 unsigned long size, int force, 1485 gfp_t priority); 1486 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1487 unsigned long size, int force, 1488 gfp_t priority); 1489 extern void sock_wfree(struct sk_buff *skb); 1490 extern void sock_rfree(struct sk_buff *skb); 1491 extern void sock_edemux(struct sk_buff *skb); 1492 1493 extern int sock_setsockopt(struct socket *sock, int level, 1494 int op, char __user *optval, 1495 unsigned int optlen); 1496 1497 extern int sock_getsockopt(struct socket *sock, int level, 1498 int op, char __user *optval, 1499 int __user *optlen); 1500 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1501 unsigned long size, 1502 int noblock, 1503 int *errcode); 1504 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1505 unsigned long header_len, 1506 unsigned long data_len, 1507 int noblock, 1508 int *errcode); 1509 extern void *sock_kmalloc(struct sock *sk, int size, 1510 gfp_t priority); 1511 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1512 extern void sk_send_sigurg(struct sock *sk); 1513 1514 /* 1515 * Functions to fill in entries in struct proto_ops when a protocol 1516 * does not implement a particular function. 1517 */ 1518 extern int sock_no_bind(struct socket *, 1519 struct sockaddr *, int); 1520 extern int sock_no_connect(struct socket *, 1521 struct sockaddr *, int, int); 1522 extern int sock_no_socketpair(struct socket *, 1523 struct socket *); 1524 extern int sock_no_accept(struct socket *, 1525 struct socket *, int); 1526 extern int sock_no_getname(struct socket *, 1527 struct sockaddr *, int *, int); 1528 extern unsigned int sock_no_poll(struct file *, struct socket *, 1529 struct poll_table_struct *); 1530 extern int sock_no_ioctl(struct socket *, unsigned int, 1531 unsigned long); 1532 extern int sock_no_listen(struct socket *, int); 1533 extern int sock_no_shutdown(struct socket *, int); 1534 extern int sock_no_getsockopt(struct socket *, int , int, 1535 char __user *, int __user *); 1536 extern int sock_no_setsockopt(struct socket *, int, int, 1537 char __user *, unsigned int); 1538 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1539 struct msghdr *, size_t); 1540 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1541 struct msghdr *, size_t, int); 1542 extern int sock_no_mmap(struct file *file, 1543 struct socket *sock, 1544 struct vm_area_struct *vma); 1545 extern ssize_t sock_no_sendpage(struct socket *sock, 1546 struct page *page, 1547 int offset, size_t size, 1548 int flags); 1549 1550 /* 1551 * Functions to fill in entries in struct proto_ops when a protocol 1552 * uses the inet style. 1553 */ 1554 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1555 char __user *optval, int __user *optlen); 1556 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1557 struct msghdr *msg, size_t size, int flags); 1558 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1559 char __user *optval, unsigned int optlen); 1560 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1561 int optname, char __user *optval, int __user *optlen); 1562 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1563 int optname, char __user *optval, unsigned int optlen); 1564 1565 extern void sk_common_release(struct sock *sk); 1566 1567 /* 1568 * Default socket callbacks and setup code 1569 */ 1570 1571 /* Initialise core socket variables */ 1572 extern void sock_init_data(struct socket *sock, struct sock *sk); 1573 1574 extern void sk_filter_release_rcu(struct rcu_head *rcu); 1575 1576 /** 1577 * sk_filter_release - release a socket filter 1578 * @fp: filter to remove 1579 * 1580 * Remove a filter from a socket and release its resources. 1581 */ 1582 1583 static inline void sk_filter_release(struct sk_filter *fp) 1584 { 1585 if (atomic_dec_and_test(&fp->refcnt)) 1586 call_rcu(&fp->rcu, sk_filter_release_rcu); 1587 } 1588 1589 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1590 { 1591 unsigned int size = sk_filter_len(fp); 1592 1593 atomic_sub(size, &sk->sk_omem_alloc); 1594 sk_filter_release(fp); 1595 } 1596 1597 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1598 { 1599 atomic_inc(&fp->refcnt); 1600 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1601 } 1602 1603 /* 1604 * Socket reference counting postulates. 1605 * 1606 * * Each user of socket SHOULD hold a reference count. 1607 * * Each access point to socket (an hash table bucket, reference from a list, 1608 * running timer, skb in flight MUST hold a reference count. 1609 * * When reference count hits 0, it means it will never increase back. 1610 * * When reference count hits 0, it means that no references from 1611 * outside exist to this socket and current process on current CPU 1612 * is last user and may/should destroy this socket. 1613 * * sk_free is called from any context: process, BH, IRQ. When 1614 * it is called, socket has no references from outside -> sk_free 1615 * may release descendant resources allocated by the socket, but 1616 * to the time when it is called, socket is NOT referenced by any 1617 * hash tables, lists etc. 1618 * * Packets, delivered from outside (from network or from another process) 1619 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1620 * when they sit in queue. Otherwise, packets will leak to hole, when 1621 * socket is looked up by one cpu and unhasing is made by another CPU. 1622 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1623 * (leak to backlog). Packet socket does all the processing inside 1624 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1625 * use separate SMP lock, so that they are prone too. 1626 */ 1627 1628 /* Ungrab socket and destroy it, if it was the last reference. */ 1629 static inline void sock_put(struct sock *sk) 1630 { 1631 if (atomic_dec_and_test(&sk->sk_refcnt)) 1632 sk_free(sk); 1633 } 1634 1635 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1636 const int nested); 1637 1638 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1639 { 1640 sk->sk_tx_queue_mapping = tx_queue; 1641 } 1642 1643 static inline void sk_tx_queue_clear(struct sock *sk) 1644 { 1645 sk->sk_tx_queue_mapping = -1; 1646 } 1647 1648 static inline int sk_tx_queue_get(const struct sock *sk) 1649 { 1650 return sk ? sk->sk_tx_queue_mapping : -1; 1651 } 1652 1653 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1654 { 1655 sk_tx_queue_clear(sk); 1656 sk->sk_socket = sock; 1657 } 1658 1659 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1660 { 1661 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1662 return &rcu_dereference_raw(sk->sk_wq)->wait; 1663 } 1664 /* Detach socket from process context. 1665 * Announce socket dead, detach it from wait queue and inode. 1666 * Note that parent inode held reference count on this struct sock, 1667 * we do not release it in this function, because protocol 1668 * probably wants some additional cleanups or even continuing 1669 * to work with this socket (TCP). 1670 */ 1671 static inline void sock_orphan(struct sock *sk) 1672 { 1673 write_lock_bh(&sk->sk_callback_lock); 1674 sock_set_flag(sk, SOCK_DEAD); 1675 sk_set_socket(sk, NULL); 1676 sk->sk_wq = NULL; 1677 write_unlock_bh(&sk->sk_callback_lock); 1678 } 1679 1680 static inline void sock_graft(struct sock *sk, struct socket *parent) 1681 { 1682 write_lock_bh(&sk->sk_callback_lock); 1683 sk->sk_wq = parent->wq; 1684 parent->sk = sk; 1685 sk_set_socket(sk, parent); 1686 security_sock_graft(sk, parent); 1687 write_unlock_bh(&sk->sk_callback_lock); 1688 } 1689 1690 extern kuid_t sock_i_uid(struct sock *sk); 1691 extern unsigned long sock_i_ino(struct sock *sk); 1692 1693 static inline struct dst_entry * 1694 __sk_dst_get(struct sock *sk) 1695 { 1696 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1697 lockdep_is_held(&sk->sk_lock.slock)); 1698 } 1699 1700 static inline struct dst_entry * 1701 sk_dst_get(struct sock *sk) 1702 { 1703 struct dst_entry *dst; 1704 1705 rcu_read_lock(); 1706 dst = rcu_dereference(sk->sk_dst_cache); 1707 if (dst) 1708 dst_hold(dst); 1709 rcu_read_unlock(); 1710 return dst; 1711 } 1712 1713 extern void sk_reset_txq(struct sock *sk); 1714 1715 static inline void dst_negative_advice(struct sock *sk) 1716 { 1717 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1718 1719 if (dst && dst->ops->negative_advice) { 1720 ndst = dst->ops->negative_advice(dst); 1721 1722 if (ndst != dst) { 1723 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1724 sk_reset_txq(sk); 1725 } 1726 } 1727 } 1728 1729 static inline void 1730 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1731 { 1732 struct dst_entry *old_dst; 1733 1734 sk_tx_queue_clear(sk); 1735 /* 1736 * This can be called while sk is owned by the caller only, 1737 * with no state that can be checked in a rcu_dereference_check() cond 1738 */ 1739 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1740 rcu_assign_pointer(sk->sk_dst_cache, dst); 1741 dst_release(old_dst); 1742 } 1743 1744 static inline void 1745 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1746 { 1747 spin_lock(&sk->sk_dst_lock); 1748 __sk_dst_set(sk, dst); 1749 spin_unlock(&sk->sk_dst_lock); 1750 } 1751 1752 static inline void 1753 __sk_dst_reset(struct sock *sk) 1754 { 1755 __sk_dst_set(sk, NULL); 1756 } 1757 1758 static inline void 1759 sk_dst_reset(struct sock *sk) 1760 { 1761 spin_lock(&sk->sk_dst_lock); 1762 __sk_dst_reset(sk); 1763 spin_unlock(&sk->sk_dst_lock); 1764 } 1765 1766 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1767 1768 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1769 1770 static inline bool sk_can_gso(const struct sock *sk) 1771 { 1772 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1773 } 1774 1775 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1776 1777 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1778 { 1779 sk->sk_route_nocaps |= flags; 1780 sk->sk_route_caps &= ~flags; 1781 } 1782 1783 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1784 char __user *from, char *to, 1785 int copy, int offset) 1786 { 1787 if (skb->ip_summed == CHECKSUM_NONE) { 1788 int err = 0; 1789 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1790 if (err) 1791 return err; 1792 skb->csum = csum_block_add(skb->csum, csum, offset); 1793 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1794 if (!access_ok(VERIFY_READ, from, copy) || 1795 __copy_from_user_nocache(to, from, copy)) 1796 return -EFAULT; 1797 } else if (copy_from_user(to, from, copy)) 1798 return -EFAULT; 1799 1800 return 0; 1801 } 1802 1803 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1804 char __user *from, int copy) 1805 { 1806 int err, offset = skb->len; 1807 1808 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1809 copy, offset); 1810 if (err) 1811 __skb_trim(skb, offset); 1812 1813 return err; 1814 } 1815 1816 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from, 1817 struct sk_buff *skb, 1818 struct page *page, 1819 int off, int copy) 1820 { 1821 int err; 1822 1823 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1824 copy, skb->len); 1825 if (err) 1826 return err; 1827 1828 skb->len += copy; 1829 skb->data_len += copy; 1830 skb->truesize += copy; 1831 sk->sk_wmem_queued += copy; 1832 sk_mem_charge(sk, copy); 1833 return 0; 1834 } 1835 1836 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1837 struct sk_buff *skb, struct page *page, 1838 int off, int copy) 1839 { 1840 if (skb->ip_summed == CHECKSUM_NONE) { 1841 int err = 0; 1842 __wsum csum = csum_and_copy_from_user(from, 1843 page_address(page) + off, 1844 copy, 0, &err); 1845 if (err) 1846 return err; 1847 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1848 } else if (copy_from_user(page_address(page) + off, from, copy)) 1849 return -EFAULT; 1850 1851 skb->len += copy; 1852 skb->data_len += copy; 1853 skb->truesize += copy; 1854 sk->sk_wmem_queued += copy; 1855 sk_mem_charge(sk, copy); 1856 return 0; 1857 } 1858 1859 /** 1860 * sk_wmem_alloc_get - returns write allocations 1861 * @sk: socket 1862 * 1863 * Returns sk_wmem_alloc minus initial offset of one 1864 */ 1865 static inline int sk_wmem_alloc_get(const struct sock *sk) 1866 { 1867 return atomic_read(&sk->sk_wmem_alloc) - 1; 1868 } 1869 1870 /** 1871 * sk_rmem_alloc_get - returns read allocations 1872 * @sk: socket 1873 * 1874 * Returns sk_rmem_alloc 1875 */ 1876 static inline int sk_rmem_alloc_get(const struct sock *sk) 1877 { 1878 return atomic_read(&sk->sk_rmem_alloc); 1879 } 1880 1881 /** 1882 * sk_has_allocations - check if allocations are outstanding 1883 * @sk: socket 1884 * 1885 * Returns true if socket has write or read allocations 1886 */ 1887 static inline bool sk_has_allocations(const struct sock *sk) 1888 { 1889 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1890 } 1891 1892 /** 1893 * wq_has_sleeper - check if there are any waiting processes 1894 * @wq: struct socket_wq 1895 * 1896 * Returns true if socket_wq has waiting processes 1897 * 1898 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1899 * barrier call. They were added due to the race found within the tcp code. 1900 * 1901 * Consider following tcp code paths: 1902 * 1903 * CPU1 CPU2 1904 * 1905 * sys_select receive packet 1906 * ... ... 1907 * __add_wait_queue update tp->rcv_nxt 1908 * ... ... 1909 * tp->rcv_nxt check sock_def_readable 1910 * ... { 1911 * schedule rcu_read_lock(); 1912 * wq = rcu_dereference(sk->sk_wq); 1913 * if (wq && waitqueue_active(&wq->wait)) 1914 * wake_up_interruptible(&wq->wait) 1915 * ... 1916 * } 1917 * 1918 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1919 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1920 * could then endup calling schedule and sleep forever if there are no more 1921 * data on the socket. 1922 * 1923 */ 1924 static inline bool wq_has_sleeper(struct socket_wq *wq) 1925 { 1926 /* We need to be sure we are in sync with the 1927 * add_wait_queue modifications to the wait queue. 1928 * 1929 * This memory barrier is paired in the sock_poll_wait. 1930 */ 1931 smp_mb(); 1932 return wq && waitqueue_active(&wq->wait); 1933 } 1934 1935 /** 1936 * sock_poll_wait - place memory barrier behind the poll_wait call. 1937 * @filp: file 1938 * @wait_address: socket wait queue 1939 * @p: poll_table 1940 * 1941 * See the comments in the wq_has_sleeper function. 1942 */ 1943 static inline void sock_poll_wait(struct file *filp, 1944 wait_queue_head_t *wait_address, poll_table *p) 1945 { 1946 if (!poll_does_not_wait(p) && wait_address) { 1947 poll_wait(filp, wait_address, p); 1948 /* We need to be sure we are in sync with the 1949 * socket flags modification. 1950 * 1951 * This memory barrier is paired in the wq_has_sleeper. 1952 */ 1953 smp_mb(); 1954 } 1955 } 1956 1957 /* 1958 * Queue a received datagram if it will fit. Stream and sequenced 1959 * protocols can't normally use this as they need to fit buffers in 1960 * and play with them. 1961 * 1962 * Inlined as it's very short and called for pretty much every 1963 * packet ever received. 1964 */ 1965 1966 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1967 { 1968 skb_orphan(skb); 1969 skb->sk = sk; 1970 skb->destructor = sock_wfree; 1971 /* 1972 * We used to take a refcount on sk, but following operation 1973 * is enough to guarantee sk_free() wont free this sock until 1974 * all in-flight packets are completed 1975 */ 1976 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1977 } 1978 1979 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1980 { 1981 skb_orphan(skb); 1982 skb->sk = sk; 1983 skb->destructor = sock_rfree; 1984 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1985 sk_mem_charge(sk, skb->truesize); 1986 } 1987 1988 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1989 unsigned long expires); 1990 1991 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1992 1993 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1994 1995 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1996 1997 /* 1998 * Recover an error report and clear atomically 1999 */ 2000 2001 static inline int sock_error(struct sock *sk) 2002 { 2003 int err; 2004 if (likely(!sk->sk_err)) 2005 return 0; 2006 err = xchg(&sk->sk_err, 0); 2007 return -err; 2008 } 2009 2010 static inline unsigned long sock_wspace(struct sock *sk) 2011 { 2012 int amt = 0; 2013 2014 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2015 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2016 if (amt < 0) 2017 amt = 0; 2018 } 2019 return amt; 2020 } 2021 2022 static inline void sk_wake_async(struct sock *sk, int how, int band) 2023 { 2024 if (sock_flag(sk, SOCK_FASYNC)) 2025 sock_wake_async(sk->sk_socket, how, band); 2026 } 2027 2028 #define SOCK_MIN_SNDBUF 2048 2029 /* 2030 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need 2031 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak 2032 */ 2033 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff)) 2034 2035 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2036 { 2037 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2038 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2039 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2040 } 2041 } 2042 2043 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 2044 2045 /** 2046 * sk_page_frag - return an appropriate page_frag 2047 * @sk: socket 2048 * 2049 * If socket allocation mode allows current thread to sleep, it means its 2050 * safe to use the per task page_frag instead of the per socket one. 2051 */ 2052 static inline struct page_frag *sk_page_frag(struct sock *sk) 2053 { 2054 if (sk->sk_allocation & __GFP_WAIT) 2055 return ¤t->task_frag; 2056 2057 return &sk->sk_frag; 2058 } 2059 2060 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2061 2062 /* 2063 * Default write policy as shown to user space via poll/select/SIGIO 2064 */ 2065 static inline bool sock_writeable(const struct sock *sk) 2066 { 2067 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2068 } 2069 2070 static inline gfp_t gfp_any(void) 2071 { 2072 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2073 } 2074 2075 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2076 { 2077 return noblock ? 0 : sk->sk_rcvtimeo; 2078 } 2079 2080 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2081 { 2082 return noblock ? 0 : sk->sk_sndtimeo; 2083 } 2084 2085 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2086 { 2087 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2088 } 2089 2090 /* Alas, with timeout socket operations are not restartable. 2091 * Compare this to poll(). 2092 */ 2093 static inline int sock_intr_errno(long timeo) 2094 { 2095 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2096 } 2097 2098 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2099 struct sk_buff *skb); 2100 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2101 struct sk_buff *skb); 2102 2103 static inline void 2104 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2105 { 2106 ktime_t kt = skb->tstamp; 2107 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2108 2109 /* 2110 * generate control messages if 2111 * - receive time stamping in software requested (SOCK_RCVTSTAMP 2112 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 2113 * - software time stamp available and wanted 2114 * (SOCK_TIMESTAMPING_SOFTWARE) 2115 * - hardware time stamps available and wanted 2116 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 2117 * SOCK_TIMESTAMPING_RAW_HARDWARE) 2118 */ 2119 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2120 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 2121 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 2122 (hwtstamps->hwtstamp.tv64 && 2123 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 2124 (hwtstamps->syststamp.tv64 && 2125 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 2126 __sock_recv_timestamp(msg, sk, skb); 2127 else 2128 sk->sk_stamp = kt; 2129 2130 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2131 __sock_recv_wifi_status(msg, sk, skb); 2132 } 2133 2134 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2135 struct sk_buff *skb); 2136 2137 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2138 struct sk_buff *skb) 2139 { 2140 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2141 (1UL << SOCK_RCVTSTAMP) | \ 2142 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 2143 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 2144 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 2145 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 2146 2147 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 2148 __sock_recv_ts_and_drops(msg, sk, skb); 2149 else 2150 sk->sk_stamp = skb->tstamp; 2151 } 2152 2153 /** 2154 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2155 * @sk: socket sending this packet 2156 * @tx_flags: filled with instructions for time stamping 2157 * 2158 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if 2159 * parameters are invalid. 2160 */ 2161 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 2162 2163 /** 2164 * sk_eat_skb - Release a skb if it is no longer needed 2165 * @sk: socket to eat this skb from 2166 * @skb: socket buffer to eat 2167 * @copied_early: flag indicating whether DMA operations copied this data early 2168 * 2169 * This routine must be called with interrupts disabled or with the socket 2170 * locked so that the sk_buff queue operation is ok. 2171 */ 2172 #ifdef CONFIG_NET_DMA 2173 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2174 { 2175 __skb_unlink(skb, &sk->sk_receive_queue); 2176 if (!copied_early) 2177 __kfree_skb(skb); 2178 else 2179 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 2180 } 2181 #else 2182 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2183 { 2184 __skb_unlink(skb, &sk->sk_receive_queue); 2185 __kfree_skb(skb); 2186 } 2187 #endif 2188 2189 static inline 2190 struct net *sock_net(const struct sock *sk) 2191 { 2192 return read_pnet(&sk->sk_net); 2193 } 2194 2195 static inline 2196 void sock_net_set(struct sock *sk, struct net *net) 2197 { 2198 write_pnet(&sk->sk_net, net); 2199 } 2200 2201 /* 2202 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2203 * They should not hold a reference to a namespace in order to allow 2204 * to stop it. 2205 * Sockets after sk_change_net should be released using sk_release_kernel 2206 */ 2207 static inline void sk_change_net(struct sock *sk, struct net *net) 2208 { 2209 put_net(sock_net(sk)); 2210 sock_net_set(sk, hold_net(net)); 2211 } 2212 2213 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2214 { 2215 if (skb->sk) { 2216 struct sock *sk = skb->sk; 2217 2218 skb->destructor = NULL; 2219 skb->sk = NULL; 2220 return sk; 2221 } 2222 return NULL; 2223 } 2224 2225 extern void sock_enable_timestamp(struct sock *sk, int flag); 2226 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 2227 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 2228 2229 /* 2230 * Enable debug/info messages 2231 */ 2232 extern int net_msg_warn; 2233 #define NETDEBUG(fmt, args...) \ 2234 do { if (net_msg_warn) printk(fmt,##args); } while (0) 2235 2236 #define LIMIT_NETDEBUG(fmt, args...) \ 2237 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 2238 2239 extern __u32 sysctl_wmem_max; 2240 extern __u32 sysctl_rmem_max; 2241 2242 extern int sysctl_optmem_max; 2243 2244 extern __u32 sysctl_wmem_default; 2245 extern __u32 sysctl_rmem_default; 2246 2247 #endif /* _SOCK_H */ 2248