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/page_counter.h> 58#include <linux/memcontrol.h> 59#include <linux/static_key.h> 60#include <linux/sched.h> 61 62#include <linux/filter.h> 63#include <linux/rculist_nulls.h> 64#include <linux/poll.h> 65 66#include <linux/atomic.h> 67#include <net/dst.h> 68#include <net/checksum.h> 69#include <net/tcp_states.h> 70#include <linux/net_tstamp.h> 71 72struct cgroup; 73struct cgroup_subsys; 74#ifdef CONFIG_NET 75int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss); 76void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg); 77#else 78static inline 79int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 80{ 81 return 0; 82} 83static inline 84void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 85{ 86} 87#endif 88/* 89 * This structure really needs to be cleaned up. 90 * Most of it is for TCP, and not used by any of 91 * the other protocols. 92 */ 93 94/* Define this to get the SOCK_DBG debugging facility. */ 95#define SOCK_DEBUGGING 96#ifdef SOCK_DEBUGGING 97#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 98 printk(KERN_DEBUG msg); } while (0) 99#else 100/* Validate arguments and do nothing */ 101static inline __printf(2, 3) 102void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 103{ 104} 105#endif 106 107/* This is the per-socket lock. The spinlock provides a synchronization 108 * between user contexts and software interrupt processing, whereas the 109 * mini-semaphore synchronizes multiple users amongst themselves. 110 */ 111typedef struct { 112 spinlock_t slock; 113 int owned; 114 wait_queue_head_t wq; 115 /* 116 * We express the mutex-alike socket_lock semantics 117 * to the lock validator by explicitly managing 118 * the slock as a lock variant (in addition to 119 * the slock itself): 120 */ 121#ifdef CONFIG_DEBUG_LOCK_ALLOC 122 struct lockdep_map dep_map; 123#endif 124} socket_lock_t; 125 126struct sock; 127struct proto; 128struct net; 129 130typedef __u32 __bitwise __portpair; 131typedef __u64 __bitwise __addrpair; 132 133/** 134 * struct sock_common - minimal network layer representation of sockets 135 * @skc_daddr: Foreign IPv4 addr 136 * @skc_rcv_saddr: Bound local IPv4 addr 137 * @skc_hash: hash value used with various protocol lookup tables 138 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 139 * @skc_dport: placeholder for inet_dport/tw_dport 140 * @skc_num: placeholder for inet_num/tw_num 141 * @skc_family: network address family 142 * @skc_state: Connection state 143 * @skc_reuse: %SO_REUSEADDR setting 144 * @skc_reuseport: %SO_REUSEPORT setting 145 * @skc_bound_dev_if: bound device index if != 0 146 * @skc_bind_node: bind hash linkage for various protocol lookup tables 147 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 148 * @skc_prot: protocol handlers inside a network family 149 * @skc_net: reference to the network namespace of this socket 150 * @skc_node: main hash linkage for various protocol lookup tables 151 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 152 * @skc_tx_queue_mapping: tx queue number for this connection 153 * @skc_refcnt: reference count 154 * 155 * This is the minimal network layer representation of sockets, the header 156 * for struct sock and struct inet_timewait_sock. 157 */ 158struct sock_common { 159 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 160 * address on 64bit arches : cf INET_MATCH() 161 */ 162 union { 163 __addrpair skc_addrpair; 164 struct { 165 __be32 skc_daddr; 166 __be32 skc_rcv_saddr; 167 }; 168 }; 169 union { 170 unsigned int skc_hash; 171 __u16 skc_u16hashes[2]; 172 }; 173 /* skc_dport && skc_num must be grouped as well */ 174 union { 175 __portpair skc_portpair; 176 struct { 177 __be16 skc_dport; 178 __u16 skc_num; 179 }; 180 }; 181 182 unsigned short skc_family; 183 volatile unsigned char skc_state; 184 unsigned char skc_reuse:4; 185 unsigned char skc_reuseport:1; 186 unsigned char skc_ipv6only:1; 187 int skc_bound_dev_if; 188 union { 189 struct hlist_node skc_bind_node; 190 struct hlist_nulls_node skc_portaddr_node; 191 }; 192 struct proto *skc_prot; 193 possible_net_t skc_net; 194 195#if IS_ENABLED(CONFIG_IPV6) 196 struct in6_addr skc_v6_daddr; 197 struct in6_addr skc_v6_rcv_saddr; 198#endif 199 200 atomic64_t skc_cookie; 201 202 /* 203 * fields between dontcopy_begin/dontcopy_end 204 * are not copied in sock_copy() 205 */ 206 /* private: */ 207 int skc_dontcopy_begin[0]; 208 /* public: */ 209 union { 210 struct hlist_node skc_node; 211 struct hlist_nulls_node skc_nulls_node; 212 }; 213 int skc_tx_queue_mapping; 214 atomic_t skc_refcnt; 215 /* private: */ 216 int skc_dontcopy_end[0]; 217 /* public: */ 218}; 219 220struct cg_proto; 221/** 222 * struct sock - network layer representation of sockets 223 * @__sk_common: shared layout with inet_timewait_sock 224 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 225 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 226 * @sk_lock: synchronizer 227 * @sk_rcvbuf: size of receive buffer in bytes 228 * @sk_wq: sock wait queue and async head 229 * @sk_rx_dst: receive input route used by early demux 230 * @sk_dst_cache: destination cache 231 * @sk_dst_lock: destination cache lock 232 * @sk_policy: flow policy 233 * @sk_receive_queue: incoming packets 234 * @sk_wmem_alloc: transmit queue bytes committed 235 * @sk_write_queue: Packet sending queue 236 * @sk_omem_alloc: "o" is "option" or "other" 237 * @sk_wmem_queued: persistent queue size 238 * @sk_forward_alloc: space allocated forward 239 * @sk_napi_id: id of the last napi context to receive data for sk 240 * @sk_ll_usec: usecs to busypoll when there is no data 241 * @sk_allocation: allocation mode 242 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 243 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 244 * @sk_sndbuf: size of send buffer in bytes 245 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 246 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 247 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 248 * @sk_no_check_rx: allow zero checksum in RX packets 249 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 250 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 251 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 252 * @sk_gso_max_size: Maximum GSO segment size to build 253 * @sk_gso_max_segs: Maximum number of GSO segments 254 * @sk_lingertime: %SO_LINGER l_linger setting 255 * @sk_backlog: always used with the per-socket spinlock held 256 * @sk_callback_lock: used with the callbacks in the end of this struct 257 * @sk_error_queue: rarely used 258 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 259 * IPV6_ADDRFORM for instance) 260 * @sk_err: last error 261 * @sk_err_soft: errors that don't cause failure but are the cause of a 262 * persistent failure not just 'timed out' 263 * @sk_drops: raw/udp drops counter 264 * @sk_ack_backlog: current listen backlog 265 * @sk_max_ack_backlog: listen backlog set in listen() 266 * @sk_priority: %SO_PRIORITY setting 267 * @sk_cgrp_prioidx: socket group's priority map index 268 * @sk_type: socket type (%SOCK_STREAM, etc) 269 * @sk_protocol: which protocol this socket belongs in this network family 270 * @sk_peer_pid: &struct pid for this socket's peer 271 * @sk_peer_cred: %SO_PEERCRED setting 272 * @sk_rcvlowat: %SO_RCVLOWAT setting 273 * @sk_rcvtimeo: %SO_RCVTIMEO setting 274 * @sk_sndtimeo: %SO_SNDTIMEO setting 275 * @sk_rxhash: flow hash received from netif layer 276 * @sk_incoming_cpu: record cpu processing incoming packets 277 * @sk_txhash: computed flow hash for use on transmit 278 * @sk_filter: socket filtering instructions 279 * @sk_protinfo: private area, net family specific, when not using slab 280 * @sk_timer: sock cleanup timer 281 * @sk_stamp: time stamp of last packet received 282 * @sk_tsflags: SO_TIMESTAMPING socket options 283 * @sk_tskey: counter to disambiguate concurrent tstamp requests 284 * @sk_socket: Identd and reporting IO signals 285 * @sk_user_data: RPC layer private data 286 * @sk_frag: cached page frag 287 * @sk_peek_off: current peek_offset value 288 * @sk_send_head: front of stuff to transmit 289 * @sk_security: used by security modules 290 * @sk_mark: generic packet mark 291 * @sk_classid: this socket's cgroup classid 292 * @sk_cgrp: this socket's cgroup-specific proto data 293 * @sk_write_pending: a write to stream socket waits to start 294 * @sk_state_change: callback to indicate change in the state of the sock 295 * @sk_data_ready: callback to indicate there is data to be processed 296 * @sk_write_space: callback to indicate there is bf sending space available 297 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 298 * @sk_backlog_rcv: callback to process the backlog 299 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 300 */ 301struct sock { 302 /* 303 * Now struct inet_timewait_sock also uses sock_common, so please just 304 * don't add nothing before this first member (__sk_common) --acme 305 */ 306 struct sock_common __sk_common; 307#define sk_node __sk_common.skc_node 308#define sk_nulls_node __sk_common.skc_nulls_node 309#define sk_refcnt __sk_common.skc_refcnt 310#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 311 312#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 313#define sk_dontcopy_end __sk_common.skc_dontcopy_end 314#define sk_hash __sk_common.skc_hash 315#define sk_portpair __sk_common.skc_portpair 316#define sk_num __sk_common.skc_num 317#define sk_dport __sk_common.skc_dport 318#define sk_addrpair __sk_common.skc_addrpair 319#define sk_daddr __sk_common.skc_daddr 320#define sk_rcv_saddr __sk_common.skc_rcv_saddr 321#define sk_family __sk_common.skc_family 322#define sk_state __sk_common.skc_state 323#define sk_reuse __sk_common.skc_reuse 324#define sk_reuseport __sk_common.skc_reuseport 325#define sk_ipv6only __sk_common.skc_ipv6only 326#define sk_bound_dev_if __sk_common.skc_bound_dev_if 327#define sk_bind_node __sk_common.skc_bind_node 328#define sk_prot __sk_common.skc_prot 329#define sk_net __sk_common.skc_net 330#define sk_v6_daddr __sk_common.skc_v6_daddr 331#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 332#define sk_cookie __sk_common.skc_cookie 333 334 socket_lock_t sk_lock; 335 struct sk_buff_head sk_receive_queue; 336 /* 337 * The backlog queue is special, it is always used with 338 * the per-socket spinlock held and requires low latency 339 * access. Therefore we special case it's implementation. 340 * Note : rmem_alloc is in this structure to fill a hole 341 * on 64bit arches, not because its logically part of 342 * backlog. 343 */ 344 struct { 345 atomic_t rmem_alloc; 346 int len; 347 struct sk_buff *head; 348 struct sk_buff *tail; 349 } sk_backlog; 350#define sk_rmem_alloc sk_backlog.rmem_alloc 351 int sk_forward_alloc; 352#ifdef CONFIG_RPS 353 __u32 sk_rxhash; 354#endif 355 u16 sk_incoming_cpu; 356 /* 16bit hole 357 * Warned : sk_incoming_cpu can be set from softirq, 358 * Do not use this hole without fully understanding possible issues. 359 */ 360 361 __u32 sk_txhash; 362#ifdef CONFIG_NET_RX_BUSY_POLL 363 unsigned int sk_napi_id; 364 unsigned int sk_ll_usec; 365#endif 366 atomic_t sk_drops; 367 int sk_rcvbuf; 368 369 struct sk_filter __rcu *sk_filter; 370 struct socket_wq __rcu *sk_wq; 371 372#ifdef CONFIG_XFRM 373 struct xfrm_policy *sk_policy[2]; 374#endif 375 unsigned long sk_flags; 376 struct dst_entry *sk_rx_dst; 377 struct dst_entry __rcu *sk_dst_cache; 378 spinlock_t sk_dst_lock; 379 atomic_t sk_wmem_alloc; 380 atomic_t sk_omem_alloc; 381 int sk_sndbuf; 382 struct sk_buff_head sk_write_queue; 383 kmemcheck_bitfield_begin(flags); 384 unsigned int sk_shutdown : 2, 385 sk_no_check_tx : 1, 386 sk_no_check_rx : 1, 387 sk_userlocks : 4, 388 sk_protocol : 8, 389#define SK_PROTOCOL_MAX U8_MAX 390 sk_type : 16; 391 kmemcheck_bitfield_end(flags); 392 int sk_wmem_queued; 393 gfp_t sk_allocation; 394 u32 sk_pacing_rate; /* bytes per second */ 395 u32 sk_max_pacing_rate; 396 netdev_features_t sk_route_caps; 397 netdev_features_t sk_route_nocaps; 398 int sk_gso_type; 399 unsigned int sk_gso_max_size; 400 u16 sk_gso_max_segs; 401 int sk_rcvlowat; 402 unsigned long sk_lingertime; 403 struct sk_buff_head sk_error_queue; 404 struct proto *sk_prot_creator; 405 rwlock_t sk_callback_lock; 406 int sk_err, 407 sk_err_soft; 408 u32 sk_ack_backlog; 409 u32 sk_max_ack_backlog; 410 __u32 sk_priority; 411#if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 412 __u32 sk_cgrp_prioidx; 413#endif 414 struct pid *sk_peer_pid; 415 const struct cred *sk_peer_cred; 416 long sk_rcvtimeo; 417 long sk_sndtimeo; 418 void *sk_protinfo; 419 struct timer_list sk_timer; 420 ktime_t sk_stamp; 421 u16 sk_tsflags; 422 u32 sk_tskey; 423 struct socket *sk_socket; 424 void *sk_user_data; 425 struct page_frag sk_frag; 426 struct sk_buff *sk_send_head; 427 __s32 sk_peek_off; 428 int sk_write_pending; 429#ifdef CONFIG_SECURITY 430 void *sk_security; 431#endif 432 __u32 sk_mark; 433 u32 sk_classid; 434 struct cg_proto *sk_cgrp; 435 void (*sk_state_change)(struct sock *sk); 436 void (*sk_data_ready)(struct sock *sk); 437 void (*sk_write_space)(struct sock *sk); 438 void (*sk_error_report)(struct sock *sk); 439 int (*sk_backlog_rcv)(struct sock *sk, 440 struct sk_buff *skb); 441 void (*sk_destruct)(struct sock *sk); 442}; 443 444#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 445 446#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 447#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 448 449/* 450 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 451 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 452 * on a socket means that the socket will reuse everybody else's port 453 * without looking at the other's sk_reuse value. 454 */ 455 456#define SK_NO_REUSE 0 457#define SK_CAN_REUSE 1 458#define SK_FORCE_REUSE 2 459 460static inline int sk_peek_offset(struct sock *sk, int flags) 461{ 462 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 463 return sk->sk_peek_off; 464 else 465 return 0; 466} 467 468static inline void sk_peek_offset_bwd(struct sock *sk, int val) 469{ 470 if (sk->sk_peek_off >= 0) { 471 if (sk->sk_peek_off >= val) 472 sk->sk_peek_off -= val; 473 else 474 sk->sk_peek_off = 0; 475 } 476} 477 478static inline void sk_peek_offset_fwd(struct sock *sk, int val) 479{ 480 if (sk->sk_peek_off >= 0) 481 sk->sk_peek_off += val; 482} 483 484/* 485 * Hashed lists helper routines 486 */ 487static inline struct sock *sk_entry(const struct hlist_node *node) 488{ 489 return hlist_entry(node, struct sock, sk_node); 490} 491 492static inline struct sock *__sk_head(const struct hlist_head *head) 493{ 494 return hlist_entry(head->first, struct sock, sk_node); 495} 496 497static inline struct sock *sk_head(const struct hlist_head *head) 498{ 499 return hlist_empty(head) ? NULL : __sk_head(head); 500} 501 502static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 503{ 504 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 505} 506 507static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 508{ 509 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 510} 511 512static inline struct sock *sk_next(const struct sock *sk) 513{ 514 return sk->sk_node.next ? 515 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 516} 517 518static inline struct sock *sk_nulls_next(const struct sock *sk) 519{ 520 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 521 hlist_nulls_entry(sk->sk_nulls_node.next, 522 struct sock, sk_nulls_node) : 523 NULL; 524} 525 526static inline bool sk_unhashed(const struct sock *sk) 527{ 528 return hlist_unhashed(&sk->sk_node); 529} 530 531static inline bool sk_hashed(const struct sock *sk) 532{ 533 return !sk_unhashed(sk); 534} 535 536static inline void sk_node_init(struct hlist_node *node) 537{ 538 node->pprev = NULL; 539} 540 541static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 542{ 543 node->pprev = NULL; 544} 545 546static inline void __sk_del_node(struct sock *sk) 547{ 548 __hlist_del(&sk->sk_node); 549} 550 551/* NB: equivalent to hlist_del_init_rcu */ 552static inline bool __sk_del_node_init(struct sock *sk) 553{ 554 if (sk_hashed(sk)) { 555 __sk_del_node(sk); 556 sk_node_init(&sk->sk_node); 557 return true; 558 } 559 return false; 560} 561 562/* Grab socket reference count. This operation is valid only 563 when sk is ALREADY grabbed f.e. it is found in hash table 564 or a list and the lookup is made under lock preventing hash table 565 modifications. 566 */ 567 568static inline void sock_hold(struct sock *sk) 569{ 570 atomic_inc(&sk->sk_refcnt); 571} 572 573/* Ungrab socket in the context, which assumes that socket refcnt 574 cannot hit zero, f.e. it is true in context of any socketcall. 575 */ 576static inline void __sock_put(struct sock *sk) 577{ 578 atomic_dec(&sk->sk_refcnt); 579} 580 581static inline bool sk_del_node_init(struct sock *sk) 582{ 583 bool rc = __sk_del_node_init(sk); 584 585 if (rc) { 586 /* paranoid for a while -acme */ 587 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 588 __sock_put(sk); 589 } 590 return rc; 591} 592#define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 593 594static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 595{ 596 if (sk_hashed(sk)) { 597 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 598 return true; 599 } 600 return false; 601} 602 603static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 604{ 605 bool rc = __sk_nulls_del_node_init_rcu(sk); 606 607 if (rc) { 608 /* paranoid for a while -acme */ 609 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 610 __sock_put(sk); 611 } 612 return rc; 613} 614 615static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 616{ 617 hlist_add_head(&sk->sk_node, list); 618} 619 620static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 621{ 622 sock_hold(sk); 623 __sk_add_node(sk, list); 624} 625 626static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 627{ 628 sock_hold(sk); 629 hlist_add_head_rcu(&sk->sk_node, list); 630} 631 632static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 633{ 634 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 635} 636 637static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 638{ 639 sock_hold(sk); 640 __sk_nulls_add_node_rcu(sk, list); 641} 642 643static inline void __sk_del_bind_node(struct sock *sk) 644{ 645 __hlist_del(&sk->sk_bind_node); 646} 647 648static inline void sk_add_bind_node(struct sock *sk, 649 struct hlist_head *list) 650{ 651 hlist_add_head(&sk->sk_bind_node, list); 652} 653 654#define sk_for_each(__sk, list) \ 655 hlist_for_each_entry(__sk, list, sk_node) 656#define sk_for_each_rcu(__sk, list) \ 657 hlist_for_each_entry_rcu(__sk, list, sk_node) 658#define sk_nulls_for_each(__sk, node, list) \ 659 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 660#define sk_nulls_for_each_rcu(__sk, node, list) \ 661 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 662#define sk_for_each_from(__sk) \ 663 hlist_for_each_entry_from(__sk, sk_node) 664#define sk_nulls_for_each_from(__sk, node) \ 665 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 666 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 667#define sk_for_each_safe(__sk, tmp, list) \ 668 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 669#define sk_for_each_bound(__sk, list) \ 670 hlist_for_each_entry(__sk, list, sk_bind_node) 671 672/** 673 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset 674 * @tpos: the type * to use as a loop cursor. 675 * @pos: the &struct hlist_node to use as a loop cursor. 676 * @head: the head for your list. 677 * @offset: offset of hlist_node within the struct. 678 * 679 */ 680#define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \ 681 for (pos = (head)->first; \ 682 (!is_a_nulls(pos)) && \ 683 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 684 pos = pos->next) 685 686static inline struct user_namespace *sk_user_ns(struct sock *sk) 687{ 688 /* Careful only use this in a context where these parameters 689 * can not change and must all be valid, such as recvmsg from 690 * userspace. 691 */ 692 return sk->sk_socket->file->f_cred->user_ns; 693} 694 695/* Sock flags */ 696enum sock_flags { 697 SOCK_DEAD, 698 SOCK_DONE, 699 SOCK_URGINLINE, 700 SOCK_KEEPOPEN, 701 SOCK_LINGER, 702 SOCK_DESTROY, 703 SOCK_BROADCAST, 704 SOCK_TIMESTAMP, 705 SOCK_ZAPPED, 706 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 707 SOCK_DBG, /* %SO_DEBUG setting */ 708 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 709 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 710 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 711 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 712 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 713 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 714 SOCK_FASYNC, /* fasync() active */ 715 SOCK_RXQ_OVFL, 716 SOCK_ZEROCOPY, /* buffers from userspace */ 717 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 718 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 719 * Will use last 4 bytes of packet sent from 720 * user-space instead. 721 */ 722 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 723 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 724}; 725 726#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 727 728static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 729{ 730 nsk->sk_flags = osk->sk_flags; 731} 732 733static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 734{ 735 __set_bit(flag, &sk->sk_flags); 736} 737 738static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 739{ 740 __clear_bit(flag, &sk->sk_flags); 741} 742 743static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 744{ 745 return test_bit(flag, &sk->sk_flags); 746} 747 748#ifdef CONFIG_NET 749extern struct static_key memalloc_socks; 750static inline int sk_memalloc_socks(void) 751{ 752 return static_key_false(&memalloc_socks); 753} 754#else 755 756static inline int sk_memalloc_socks(void) 757{ 758 return 0; 759} 760 761#endif 762 763static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 764{ 765 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 766} 767 768static inline void sk_acceptq_removed(struct sock *sk) 769{ 770 sk->sk_ack_backlog--; 771} 772 773static inline void sk_acceptq_added(struct sock *sk) 774{ 775 sk->sk_ack_backlog++; 776} 777 778static inline bool sk_acceptq_is_full(const struct sock *sk) 779{ 780 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 781} 782 783/* 784 * Compute minimal free write space needed to queue new packets. 785 */ 786static inline int sk_stream_min_wspace(const struct sock *sk) 787{ 788 return sk->sk_wmem_queued >> 1; 789} 790 791static inline int sk_stream_wspace(const struct sock *sk) 792{ 793 return sk->sk_sndbuf - sk->sk_wmem_queued; 794} 795 796void sk_stream_write_space(struct sock *sk); 797 798/* OOB backlog add */ 799static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 800{ 801 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 802 skb_dst_force_safe(skb); 803 804 if (!sk->sk_backlog.tail) 805 sk->sk_backlog.head = skb; 806 else 807 sk->sk_backlog.tail->next = skb; 808 809 sk->sk_backlog.tail = skb; 810 skb->next = NULL; 811} 812 813/* 814 * Take into account size of receive queue and backlog queue 815 * Do not take into account this skb truesize, 816 * to allow even a single big packet to come. 817 */ 818static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 819{ 820 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 821 822 return qsize > limit; 823} 824 825/* The per-socket spinlock must be held here. */ 826static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 827 unsigned int limit) 828{ 829 if (sk_rcvqueues_full(sk, limit)) 830 return -ENOBUFS; 831 832 /* 833 * If the skb was allocated from pfmemalloc reserves, only 834 * allow SOCK_MEMALLOC sockets to use it as this socket is 835 * helping free memory 836 */ 837 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 838 return -ENOMEM; 839 840 __sk_add_backlog(sk, skb); 841 sk->sk_backlog.len += skb->truesize; 842 return 0; 843} 844 845int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 846 847static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 848{ 849 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 850 return __sk_backlog_rcv(sk, skb); 851 852 return sk->sk_backlog_rcv(sk, skb); 853} 854 855static inline void sk_incoming_cpu_update(struct sock *sk) 856{ 857 sk->sk_incoming_cpu = raw_smp_processor_id(); 858} 859 860static inline void sock_rps_record_flow_hash(__u32 hash) 861{ 862#ifdef CONFIG_RPS 863 struct rps_sock_flow_table *sock_flow_table; 864 865 rcu_read_lock(); 866 sock_flow_table = rcu_dereference(rps_sock_flow_table); 867 rps_record_sock_flow(sock_flow_table, hash); 868 rcu_read_unlock(); 869#endif 870} 871 872static inline void sock_rps_record_flow(const struct sock *sk) 873{ 874#ifdef CONFIG_RPS 875 sock_rps_record_flow_hash(sk->sk_rxhash); 876#endif 877} 878 879static inline void sock_rps_save_rxhash(struct sock *sk, 880 const struct sk_buff *skb) 881{ 882#ifdef CONFIG_RPS 883 if (unlikely(sk->sk_rxhash != skb->hash)) 884 sk->sk_rxhash = skb->hash; 885#endif 886} 887 888static inline void sock_rps_reset_rxhash(struct sock *sk) 889{ 890#ifdef CONFIG_RPS 891 sk->sk_rxhash = 0; 892#endif 893} 894 895#define sk_wait_event(__sk, __timeo, __condition) \ 896 ({ int __rc; \ 897 release_sock(__sk); \ 898 __rc = __condition; \ 899 if (!__rc) { \ 900 *(__timeo) = schedule_timeout(*(__timeo)); \ 901 } \ 902 sched_annotate_sleep(); \ 903 lock_sock(__sk); \ 904 __rc = __condition; \ 905 __rc; \ 906 }) 907 908int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 909int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 910void sk_stream_wait_close(struct sock *sk, long timeo_p); 911int sk_stream_error(struct sock *sk, int flags, int err); 912void sk_stream_kill_queues(struct sock *sk); 913void sk_set_memalloc(struct sock *sk); 914void sk_clear_memalloc(struct sock *sk); 915 916int sk_wait_data(struct sock *sk, long *timeo); 917 918struct request_sock_ops; 919struct timewait_sock_ops; 920struct inet_hashinfo; 921struct raw_hashinfo; 922struct module; 923 924/* 925 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 926 * un-modified. Special care is taken when initializing object to zero. 927 */ 928static inline void sk_prot_clear_nulls(struct sock *sk, int size) 929{ 930 if (offsetof(struct sock, sk_node.next) != 0) 931 memset(sk, 0, offsetof(struct sock, sk_node.next)); 932 memset(&sk->sk_node.pprev, 0, 933 size - offsetof(struct sock, sk_node.pprev)); 934} 935 936/* Networking protocol blocks we attach to sockets. 937 * socket layer -> transport layer interface 938 * transport -> network interface is defined by struct inet_proto 939 */ 940struct proto { 941 void (*close)(struct sock *sk, 942 long timeout); 943 int (*connect)(struct sock *sk, 944 struct sockaddr *uaddr, 945 int addr_len); 946 int (*disconnect)(struct sock *sk, int flags); 947 948 struct sock * (*accept)(struct sock *sk, int flags, int *err); 949 950 int (*ioctl)(struct sock *sk, int cmd, 951 unsigned long arg); 952 int (*init)(struct sock *sk); 953 void (*destroy)(struct sock *sk); 954 void (*shutdown)(struct sock *sk, int how); 955 int (*setsockopt)(struct sock *sk, int level, 956 int optname, char __user *optval, 957 unsigned int optlen); 958 int (*getsockopt)(struct sock *sk, int level, 959 int optname, char __user *optval, 960 int __user *option); 961#ifdef CONFIG_COMPAT 962 int (*compat_setsockopt)(struct sock *sk, 963 int level, 964 int optname, char __user *optval, 965 unsigned int optlen); 966 int (*compat_getsockopt)(struct sock *sk, 967 int level, 968 int optname, char __user *optval, 969 int __user *option); 970 int (*compat_ioctl)(struct sock *sk, 971 unsigned int cmd, unsigned long arg); 972#endif 973 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 974 size_t len); 975 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 976 size_t len, int noblock, int flags, 977 int *addr_len); 978 int (*sendpage)(struct sock *sk, struct page *page, 979 int offset, size_t size, int flags); 980 int (*bind)(struct sock *sk, 981 struct sockaddr *uaddr, int addr_len); 982 983 int (*backlog_rcv) (struct sock *sk, 984 struct sk_buff *skb); 985 986 void (*release_cb)(struct sock *sk); 987 988 /* Keeping track of sk's, looking them up, and port selection methods. */ 989 void (*hash)(struct sock *sk); 990 void (*unhash)(struct sock *sk); 991 void (*rehash)(struct sock *sk); 992 int (*get_port)(struct sock *sk, unsigned short snum); 993 void (*clear_sk)(struct sock *sk, int size); 994 995 /* Keeping track of sockets in use */ 996#ifdef CONFIG_PROC_FS 997 unsigned int inuse_idx; 998#endif 999 1000 bool (*stream_memory_free)(const struct sock *sk); 1001 /* Memory pressure */ 1002 void (*enter_memory_pressure)(struct sock *sk); 1003 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1004 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1005 /* 1006 * Pressure flag: try to collapse. 1007 * Technical note: it is used by multiple contexts non atomically. 1008 * All the __sk_mem_schedule() is of this nature: accounting 1009 * is strict, actions are advisory and have some latency. 1010 */ 1011 int *memory_pressure; 1012 long *sysctl_mem; 1013 int *sysctl_wmem; 1014 int *sysctl_rmem; 1015 int max_header; 1016 bool no_autobind; 1017 1018 struct kmem_cache *slab; 1019 unsigned int obj_size; 1020 int slab_flags; 1021 1022 struct percpu_counter *orphan_count; 1023 1024 struct request_sock_ops *rsk_prot; 1025 struct timewait_sock_ops *twsk_prot; 1026 1027 union { 1028 struct inet_hashinfo *hashinfo; 1029 struct udp_table *udp_table; 1030 struct raw_hashinfo *raw_hash; 1031 } h; 1032 1033 struct module *owner; 1034 1035 char name[32]; 1036 1037 struct list_head node; 1038#ifdef SOCK_REFCNT_DEBUG 1039 atomic_t socks; 1040#endif 1041#ifdef CONFIG_MEMCG_KMEM 1042 /* 1043 * cgroup specific init/deinit functions. Called once for all 1044 * protocols that implement it, from cgroups populate function. 1045 * This function has to setup any files the protocol want to 1046 * appear in the kmem cgroup filesystem. 1047 */ 1048 int (*init_cgroup)(struct mem_cgroup *memcg, 1049 struct cgroup_subsys *ss); 1050 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1051 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1052#endif 1053}; 1054 1055/* 1056 * Bits in struct cg_proto.flags 1057 */ 1058enum cg_proto_flags { 1059 /* Currently active and new sockets should be assigned to cgroups */ 1060 MEMCG_SOCK_ACTIVE, 1061 /* It was ever activated; we must disarm static keys on destruction */ 1062 MEMCG_SOCK_ACTIVATED, 1063}; 1064 1065struct cg_proto { 1066 struct page_counter memory_allocated; /* Current allocated memory. */ 1067 struct percpu_counter sockets_allocated; /* Current number of sockets. */ 1068 int memory_pressure; 1069 long sysctl_mem[3]; 1070 unsigned long flags; 1071 /* 1072 * memcg field is used to find which memcg we belong directly 1073 * Each memcg struct can hold more than one cg_proto, so container_of 1074 * won't really cut. 1075 * 1076 * The elegant solution would be having an inverse function to 1077 * proto_cgroup in struct proto, but that means polluting the structure 1078 * for everybody, instead of just for memcg users. 1079 */ 1080 struct mem_cgroup *memcg; 1081}; 1082 1083int proto_register(struct proto *prot, int alloc_slab); 1084void proto_unregister(struct proto *prot); 1085 1086static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1087{ 1088 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1089} 1090 1091#ifdef SOCK_REFCNT_DEBUG 1092static inline void sk_refcnt_debug_inc(struct sock *sk) 1093{ 1094 atomic_inc(&sk->sk_prot->socks); 1095} 1096 1097static inline void sk_refcnt_debug_dec(struct sock *sk) 1098{ 1099 atomic_dec(&sk->sk_prot->socks); 1100 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1101 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1102} 1103 1104static inline void sk_refcnt_debug_release(const struct sock *sk) 1105{ 1106 if (atomic_read(&sk->sk_refcnt) != 1) 1107 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1108 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1109} 1110#else /* SOCK_REFCNT_DEBUG */ 1111#define sk_refcnt_debug_inc(sk) do { } while (0) 1112#define sk_refcnt_debug_dec(sk) do { } while (0) 1113#define sk_refcnt_debug_release(sk) do { } while (0) 1114#endif /* SOCK_REFCNT_DEBUG */ 1115 1116#if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1117extern struct static_key memcg_socket_limit_enabled; 1118static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1119 struct cg_proto *cg_proto) 1120{ 1121 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1122} 1123#define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1124#else 1125#define mem_cgroup_sockets_enabled 0 1126static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1127 struct cg_proto *cg_proto) 1128{ 1129 return NULL; 1130} 1131#endif 1132 1133static inline bool sk_stream_memory_free(const struct sock *sk) 1134{ 1135 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1136 return false; 1137 1138 return sk->sk_prot->stream_memory_free ? 1139 sk->sk_prot->stream_memory_free(sk) : true; 1140} 1141 1142static inline bool sk_stream_is_writeable(const struct sock *sk) 1143{ 1144 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1145 sk_stream_memory_free(sk); 1146} 1147 1148 1149static inline bool sk_has_memory_pressure(const struct sock *sk) 1150{ 1151 return sk->sk_prot->memory_pressure != NULL; 1152} 1153 1154static inline bool sk_under_memory_pressure(const struct sock *sk) 1155{ 1156 if (!sk->sk_prot->memory_pressure) 1157 return false; 1158 1159 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1160 return !!sk->sk_cgrp->memory_pressure; 1161 1162 return !!*sk->sk_prot->memory_pressure; 1163} 1164 1165static inline void sk_leave_memory_pressure(struct sock *sk) 1166{ 1167 int *memory_pressure = sk->sk_prot->memory_pressure; 1168 1169 if (!memory_pressure) 1170 return; 1171 1172 if (*memory_pressure) 1173 *memory_pressure = 0; 1174 1175 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1176 struct cg_proto *cg_proto = sk->sk_cgrp; 1177 struct proto *prot = sk->sk_prot; 1178 1179 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1180 cg_proto->memory_pressure = 0; 1181 } 1182 1183} 1184 1185static inline void sk_enter_memory_pressure(struct sock *sk) 1186{ 1187 if (!sk->sk_prot->enter_memory_pressure) 1188 return; 1189 1190 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1191 struct cg_proto *cg_proto = sk->sk_cgrp; 1192 struct proto *prot = sk->sk_prot; 1193 1194 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1195 cg_proto->memory_pressure = 1; 1196 } 1197 1198 sk->sk_prot->enter_memory_pressure(sk); 1199} 1200 1201static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1202{ 1203 long *prot = sk->sk_prot->sysctl_mem; 1204 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1205 prot = sk->sk_cgrp->sysctl_mem; 1206 return prot[index]; 1207} 1208 1209static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1210 unsigned long amt, 1211 int *parent_status) 1212{ 1213 page_counter_charge(&prot->memory_allocated, amt); 1214 1215 if (page_counter_read(&prot->memory_allocated) > 1216 prot->memory_allocated.limit) 1217 *parent_status = OVER_LIMIT; 1218} 1219 1220static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1221 unsigned long amt) 1222{ 1223 page_counter_uncharge(&prot->memory_allocated, amt); 1224} 1225 1226static inline long 1227sk_memory_allocated(const struct sock *sk) 1228{ 1229 struct proto *prot = sk->sk_prot; 1230 1231 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1232 return page_counter_read(&sk->sk_cgrp->memory_allocated); 1233 1234 return atomic_long_read(prot->memory_allocated); 1235} 1236 1237static inline long 1238sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1239{ 1240 struct proto *prot = sk->sk_prot; 1241 1242 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1243 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1244 /* update the root cgroup regardless */ 1245 atomic_long_add_return(amt, prot->memory_allocated); 1246 return page_counter_read(&sk->sk_cgrp->memory_allocated); 1247 } 1248 1249 return atomic_long_add_return(amt, prot->memory_allocated); 1250} 1251 1252static inline void 1253sk_memory_allocated_sub(struct sock *sk, int amt) 1254{ 1255 struct proto *prot = sk->sk_prot; 1256 1257 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1258 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1259 1260 atomic_long_sub(amt, prot->memory_allocated); 1261} 1262 1263static inline void sk_sockets_allocated_dec(struct sock *sk) 1264{ 1265 struct proto *prot = sk->sk_prot; 1266 1267 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1268 struct cg_proto *cg_proto = sk->sk_cgrp; 1269 1270 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1271 percpu_counter_dec(&cg_proto->sockets_allocated); 1272 } 1273 1274 percpu_counter_dec(prot->sockets_allocated); 1275} 1276 1277static inline void sk_sockets_allocated_inc(struct sock *sk) 1278{ 1279 struct proto *prot = sk->sk_prot; 1280 1281 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1282 struct cg_proto *cg_proto = sk->sk_cgrp; 1283 1284 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1285 percpu_counter_inc(&cg_proto->sockets_allocated); 1286 } 1287 1288 percpu_counter_inc(prot->sockets_allocated); 1289} 1290 1291static inline int 1292sk_sockets_allocated_read_positive(struct sock *sk) 1293{ 1294 struct proto *prot = sk->sk_prot; 1295 1296 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1297 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated); 1298 1299 return percpu_counter_read_positive(prot->sockets_allocated); 1300} 1301 1302static inline int 1303proto_sockets_allocated_sum_positive(struct proto *prot) 1304{ 1305 return percpu_counter_sum_positive(prot->sockets_allocated); 1306} 1307 1308static inline long 1309proto_memory_allocated(struct proto *prot) 1310{ 1311 return atomic_long_read(prot->memory_allocated); 1312} 1313 1314static inline bool 1315proto_memory_pressure(struct proto *prot) 1316{ 1317 if (!prot->memory_pressure) 1318 return false; 1319 return !!*prot->memory_pressure; 1320} 1321 1322 1323#ifdef CONFIG_PROC_FS 1324/* Called with local bh disabled */ 1325void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1326int sock_prot_inuse_get(struct net *net, struct proto *proto); 1327#else 1328static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1329 int inc) 1330{ 1331} 1332#endif 1333 1334 1335/* With per-bucket locks this operation is not-atomic, so that 1336 * this version is not worse. 1337 */ 1338static inline void __sk_prot_rehash(struct sock *sk) 1339{ 1340 sk->sk_prot->unhash(sk); 1341 sk->sk_prot->hash(sk); 1342} 1343 1344void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1345 1346/* About 10 seconds */ 1347#define SOCK_DESTROY_TIME (10*HZ) 1348 1349/* Sockets 0-1023 can't be bound to unless you are superuser */ 1350#define PROT_SOCK 1024 1351 1352#define SHUTDOWN_MASK 3 1353#define RCV_SHUTDOWN 1 1354#define SEND_SHUTDOWN 2 1355 1356#define SOCK_SNDBUF_LOCK 1 1357#define SOCK_RCVBUF_LOCK 2 1358#define SOCK_BINDADDR_LOCK 4 1359#define SOCK_BINDPORT_LOCK 8 1360 1361struct socket_alloc { 1362 struct socket socket; 1363 struct inode vfs_inode; 1364}; 1365 1366static inline struct socket *SOCKET_I(struct inode *inode) 1367{ 1368 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1369} 1370 1371static inline struct inode *SOCK_INODE(struct socket *socket) 1372{ 1373 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1374} 1375 1376/* 1377 * Functions for memory accounting 1378 */ 1379int __sk_mem_schedule(struct sock *sk, int size, int kind); 1380void __sk_mem_reclaim(struct sock *sk); 1381 1382#define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1383#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1384#define SK_MEM_SEND 0 1385#define SK_MEM_RECV 1 1386 1387static inline int sk_mem_pages(int amt) 1388{ 1389 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1390} 1391 1392static inline bool sk_has_account(struct sock *sk) 1393{ 1394 /* return true if protocol supports memory accounting */ 1395 return !!sk->sk_prot->memory_allocated; 1396} 1397 1398static inline bool sk_wmem_schedule(struct sock *sk, int size) 1399{ 1400 if (!sk_has_account(sk)) 1401 return true; 1402 return size <= sk->sk_forward_alloc || 1403 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1404} 1405 1406static inline bool 1407sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1408{ 1409 if (!sk_has_account(sk)) 1410 return true; 1411 return size<= sk->sk_forward_alloc || 1412 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1413 skb_pfmemalloc(skb); 1414} 1415 1416static inline void sk_mem_reclaim(struct sock *sk) 1417{ 1418 if (!sk_has_account(sk)) 1419 return; 1420 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1421 __sk_mem_reclaim(sk); 1422} 1423 1424static inline void sk_mem_reclaim_partial(struct sock *sk) 1425{ 1426 if (!sk_has_account(sk)) 1427 return; 1428 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1429 __sk_mem_reclaim(sk); 1430} 1431 1432static inline void sk_mem_charge(struct sock *sk, int size) 1433{ 1434 if (!sk_has_account(sk)) 1435 return; 1436 sk->sk_forward_alloc -= size; 1437} 1438 1439static inline void sk_mem_uncharge(struct sock *sk, int size) 1440{ 1441 if (!sk_has_account(sk)) 1442 return; 1443 sk->sk_forward_alloc += size; 1444} 1445 1446static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1447{ 1448 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1449 sk->sk_wmem_queued -= skb->truesize; 1450 sk_mem_uncharge(sk, skb->truesize); 1451 __kfree_skb(skb); 1452} 1453 1454/* Used by processes to "lock" a socket state, so that 1455 * interrupts and bottom half handlers won't change it 1456 * from under us. It essentially blocks any incoming 1457 * packets, so that we won't get any new data or any 1458 * packets that change the state of the socket. 1459 * 1460 * While locked, BH processing will add new packets to 1461 * the backlog queue. This queue is processed by the 1462 * owner of the socket lock right before it is released. 1463 * 1464 * Since ~2.3.5 it is also exclusive sleep lock serializing 1465 * accesses from user process context. 1466 */ 1467#define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1468 1469static inline void sock_release_ownership(struct sock *sk) 1470{ 1471 sk->sk_lock.owned = 0; 1472} 1473 1474/* 1475 * Macro so as to not evaluate some arguments when 1476 * lockdep is not enabled. 1477 * 1478 * Mark both the sk_lock and the sk_lock.slock as a 1479 * per-address-family lock class. 1480 */ 1481#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1482do { \ 1483 sk->sk_lock.owned = 0; \ 1484 init_waitqueue_head(&sk->sk_lock.wq); \ 1485 spin_lock_init(&(sk)->sk_lock.slock); \ 1486 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1487 sizeof((sk)->sk_lock)); \ 1488 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1489 (skey), (sname)); \ 1490 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1491} while (0) 1492 1493void lock_sock_nested(struct sock *sk, int subclass); 1494 1495static inline void lock_sock(struct sock *sk) 1496{ 1497 lock_sock_nested(sk, 0); 1498} 1499 1500void release_sock(struct sock *sk); 1501 1502/* BH context may only use the following locking interface. */ 1503#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1504#define bh_lock_sock_nested(__sk) \ 1505 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1506 SINGLE_DEPTH_NESTING) 1507#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1508 1509bool lock_sock_fast(struct sock *sk); 1510/** 1511 * unlock_sock_fast - complement of lock_sock_fast 1512 * @sk: socket 1513 * @slow: slow mode 1514 * 1515 * fast unlock socket for user context. 1516 * If slow mode is on, we call regular release_sock() 1517 */ 1518static inline void unlock_sock_fast(struct sock *sk, bool slow) 1519{ 1520 if (slow) 1521 release_sock(sk); 1522 else 1523 spin_unlock_bh(&sk->sk_lock.slock); 1524} 1525 1526 1527struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1528 struct proto *prot); 1529void sk_free(struct sock *sk); 1530void sk_release_kernel(struct sock *sk); 1531struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1532 1533struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1534 gfp_t priority); 1535void sock_wfree(struct sk_buff *skb); 1536void skb_orphan_partial(struct sk_buff *skb); 1537void sock_rfree(struct sk_buff *skb); 1538void sock_efree(struct sk_buff *skb); 1539#ifdef CONFIG_INET 1540void sock_edemux(struct sk_buff *skb); 1541#else 1542#define sock_edemux(skb) sock_efree(skb) 1543#endif 1544 1545int sock_setsockopt(struct socket *sock, int level, int op, 1546 char __user *optval, unsigned int optlen); 1547 1548int sock_getsockopt(struct socket *sock, int level, int op, 1549 char __user *optval, int __user *optlen); 1550struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1551 int noblock, int *errcode); 1552struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1553 unsigned long data_len, int noblock, 1554 int *errcode, int max_page_order); 1555void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1556void sock_kfree_s(struct sock *sk, void *mem, int size); 1557void sock_kzfree_s(struct sock *sk, void *mem, int size); 1558void sk_send_sigurg(struct sock *sk); 1559 1560/* 1561 * Functions to fill in entries in struct proto_ops when a protocol 1562 * does not implement a particular function. 1563 */ 1564int sock_no_bind(struct socket *, struct sockaddr *, int); 1565int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1566int sock_no_socketpair(struct socket *, struct socket *); 1567int sock_no_accept(struct socket *, struct socket *, int); 1568int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1569unsigned int sock_no_poll(struct file *, struct socket *, 1570 struct poll_table_struct *); 1571int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1572int sock_no_listen(struct socket *, int); 1573int sock_no_shutdown(struct socket *, int); 1574int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1575int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1576int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1577int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1578int sock_no_mmap(struct file *file, struct socket *sock, 1579 struct vm_area_struct *vma); 1580ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1581 size_t size, int flags); 1582 1583/* 1584 * Functions to fill in entries in struct proto_ops when a protocol 1585 * uses the inet style. 1586 */ 1587int sock_common_getsockopt(struct socket *sock, int level, int optname, 1588 char __user *optval, int __user *optlen); 1589int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1590 int flags); 1591int sock_common_setsockopt(struct socket *sock, int level, int optname, 1592 char __user *optval, unsigned int optlen); 1593int compat_sock_common_getsockopt(struct socket *sock, int level, 1594 int optname, char __user *optval, int __user *optlen); 1595int compat_sock_common_setsockopt(struct socket *sock, int level, 1596 int optname, char __user *optval, unsigned int optlen); 1597 1598void sk_common_release(struct sock *sk); 1599 1600/* 1601 * Default socket callbacks and setup code 1602 */ 1603 1604/* Initialise core socket variables */ 1605void sock_init_data(struct socket *sock, struct sock *sk); 1606 1607/* 1608 * Socket reference counting postulates. 1609 * 1610 * * Each user of socket SHOULD hold a reference count. 1611 * * Each access point to socket (an hash table bucket, reference from a list, 1612 * running timer, skb in flight MUST hold a reference count. 1613 * * When reference count hits 0, it means it will never increase back. 1614 * * When reference count hits 0, it means that no references from 1615 * outside exist to this socket and current process on current CPU 1616 * is last user and may/should destroy this socket. 1617 * * sk_free is called from any context: process, BH, IRQ. When 1618 * it is called, socket has no references from outside -> sk_free 1619 * may release descendant resources allocated by the socket, but 1620 * to the time when it is called, socket is NOT referenced by any 1621 * hash tables, lists etc. 1622 * * Packets, delivered from outside (from network or from another process) 1623 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1624 * when they sit in queue. Otherwise, packets will leak to hole, when 1625 * socket is looked up by one cpu and unhasing is made by another CPU. 1626 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1627 * (leak to backlog). Packet socket does all the processing inside 1628 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1629 * use separate SMP lock, so that they are prone too. 1630 */ 1631 1632/* Ungrab socket and destroy it, if it was the last reference. */ 1633static inline void sock_put(struct sock *sk) 1634{ 1635 if (atomic_dec_and_test(&sk->sk_refcnt)) 1636 sk_free(sk); 1637} 1638/* Generic version of sock_put(), dealing with all sockets 1639 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1640 */ 1641void sock_gen_put(struct sock *sk); 1642 1643int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested); 1644 1645static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1646{ 1647 sk->sk_tx_queue_mapping = tx_queue; 1648} 1649 1650static inline void sk_tx_queue_clear(struct sock *sk) 1651{ 1652 sk->sk_tx_queue_mapping = -1; 1653} 1654 1655static inline int sk_tx_queue_get(const struct sock *sk) 1656{ 1657 return sk ? sk->sk_tx_queue_mapping : -1; 1658} 1659 1660static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1661{ 1662 sk_tx_queue_clear(sk); 1663 sk->sk_socket = sock; 1664} 1665 1666static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1667{ 1668 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1669 return &rcu_dereference_raw(sk->sk_wq)->wait; 1670} 1671/* Detach socket from process context. 1672 * Announce socket dead, detach it from wait queue and inode. 1673 * Note that parent inode held reference count on this struct sock, 1674 * we do not release it in this function, because protocol 1675 * probably wants some additional cleanups or even continuing 1676 * to work with this socket (TCP). 1677 */ 1678static inline void sock_orphan(struct sock *sk) 1679{ 1680 write_lock_bh(&sk->sk_callback_lock); 1681 sock_set_flag(sk, SOCK_DEAD); 1682 sk_set_socket(sk, NULL); 1683 sk->sk_wq = NULL; 1684 write_unlock_bh(&sk->sk_callback_lock); 1685} 1686 1687static inline void sock_graft(struct sock *sk, struct socket *parent) 1688{ 1689 write_lock_bh(&sk->sk_callback_lock); 1690 sk->sk_wq = parent->wq; 1691 parent->sk = sk; 1692 sk_set_socket(sk, parent); 1693 security_sock_graft(sk, parent); 1694 write_unlock_bh(&sk->sk_callback_lock); 1695} 1696 1697kuid_t sock_i_uid(struct sock *sk); 1698unsigned long sock_i_ino(struct sock *sk); 1699 1700static inline struct dst_entry * 1701__sk_dst_get(struct sock *sk) 1702{ 1703 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1704 lockdep_is_held(&sk->sk_lock.slock)); 1705} 1706 1707static inline struct dst_entry * 1708sk_dst_get(struct sock *sk) 1709{ 1710 struct dst_entry *dst; 1711 1712 rcu_read_lock(); 1713 dst = rcu_dereference(sk->sk_dst_cache); 1714 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1715 dst = NULL; 1716 rcu_read_unlock(); 1717 return dst; 1718} 1719 1720static inline void dst_negative_advice(struct sock *sk) 1721{ 1722 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1723 1724 if (dst && dst->ops->negative_advice) { 1725 ndst = dst->ops->negative_advice(dst); 1726 1727 if (ndst != dst) { 1728 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1729 sk_tx_queue_clear(sk); 1730 } 1731 } 1732} 1733 1734static inline void 1735__sk_dst_set(struct sock *sk, struct dst_entry *dst) 1736{ 1737 struct dst_entry *old_dst; 1738 1739 sk_tx_queue_clear(sk); 1740 /* 1741 * This can be called while sk is owned by the caller only, 1742 * with no state that can be checked in a rcu_dereference_check() cond 1743 */ 1744 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1745 rcu_assign_pointer(sk->sk_dst_cache, dst); 1746 dst_release(old_dst); 1747} 1748 1749static inline void 1750sk_dst_set(struct sock *sk, struct dst_entry *dst) 1751{ 1752 struct dst_entry *old_dst; 1753 1754 sk_tx_queue_clear(sk); 1755 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1756 dst_release(old_dst); 1757} 1758 1759static inline void 1760__sk_dst_reset(struct sock *sk) 1761{ 1762 __sk_dst_set(sk, NULL); 1763} 1764 1765static inline void 1766sk_dst_reset(struct sock *sk) 1767{ 1768 sk_dst_set(sk, NULL); 1769} 1770 1771struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1772 1773struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1774 1775bool sk_mc_loop(struct sock *sk); 1776 1777static inline bool sk_can_gso(const struct sock *sk) 1778{ 1779 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1780} 1781 1782void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1783 1784static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1785{ 1786 sk->sk_route_nocaps |= flags; 1787 sk->sk_route_caps &= ~flags; 1788} 1789 1790static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1791 struct iov_iter *from, char *to, 1792 int copy, int offset) 1793{ 1794 if (skb->ip_summed == CHECKSUM_NONE) { 1795 __wsum csum = 0; 1796 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy) 1797 return -EFAULT; 1798 skb->csum = csum_block_add(skb->csum, csum, offset); 1799 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1800 if (copy_from_iter_nocache(to, copy, from) != copy) 1801 return -EFAULT; 1802 } else if (copy_from_iter(to, copy, from) != copy) 1803 return -EFAULT; 1804 1805 return 0; 1806} 1807 1808static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1809 struct iov_iter *from, int copy) 1810{ 1811 int err, offset = skb->len; 1812 1813 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1814 copy, offset); 1815 if (err) 1816 __skb_trim(skb, offset); 1817 1818 return err; 1819} 1820 1821static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1822 struct sk_buff *skb, 1823 struct page *page, 1824 int off, int copy) 1825{ 1826 int err; 1827 1828 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1829 copy, skb->len); 1830 if (err) 1831 return err; 1832 1833 skb->len += copy; 1834 skb->data_len += copy; 1835 skb->truesize += copy; 1836 sk->sk_wmem_queued += copy; 1837 sk_mem_charge(sk, copy); 1838 return 0; 1839} 1840 1841/** 1842 * sk_wmem_alloc_get - returns write allocations 1843 * @sk: socket 1844 * 1845 * Returns sk_wmem_alloc minus initial offset of one 1846 */ 1847static inline int sk_wmem_alloc_get(const struct sock *sk) 1848{ 1849 return atomic_read(&sk->sk_wmem_alloc) - 1; 1850} 1851 1852/** 1853 * sk_rmem_alloc_get - returns read allocations 1854 * @sk: socket 1855 * 1856 * Returns sk_rmem_alloc 1857 */ 1858static inline int sk_rmem_alloc_get(const struct sock *sk) 1859{ 1860 return atomic_read(&sk->sk_rmem_alloc); 1861} 1862 1863/** 1864 * sk_has_allocations - check if allocations are outstanding 1865 * @sk: socket 1866 * 1867 * Returns true if socket has write or read allocations 1868 */ 1869static inline bool sk_has_allocations(const struct sock *sk) 1870{ 1871 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1872} 1873 1874/** 1875 * wq_has_sleeper - check if there are any waiting processes 1876 * @wq: struct socket_wq 1877 * 1878 * Returns true if socket_wq has waiting processes 1879 * 1880 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1881 * barrier call. They were added due to the race found within the tcp code. 1882 * 1883 * Consider following tcp code paths: 1884 * 1885 * CPU1 CPU2 1886 * 1887 * sys_select receive packet 1888 * ... ... 1889 * __add_wait_queue update tp->rcv_nxt 1890 * ... ... 1891 * tp->rcv_nxt check sock_def_readable 1892 * ... { 1893 * schedule rcu_read_lock(); 1894 * wq = rcu_dereference(sk->sk_wq); 1895 * if (wq && waitqueue_active(&wq->wait)) 1896 * wake_up_interruptible(&wq->wait) 1897 * ... 1898 * } 1899 * 1900 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1901 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1902 * could then endup calling schedule and sleep forever if there are no more 1903 * data on the socket. 1904 * 1905 */ 1906static inline bool wq_has_sleeper(struct socket_wq *wq) 1907{ 1908 /* We need to be sure we are in sync with the 1909 * add_wait_queue modifications to the wait queue. 1910 * 1911 * This memory barrier is paired in the sock_poll_wait. 1912 */ 1913 smp_mb(); 1914 return wq && waitqueue_active(&wq->wait); 1915} 1916 1917/** 1918 * sock_poll_wait - place memory barrier behind the poll_wait call. 1919 * @filp: file 1920 * @wait_address: socket wait queue 1921 * @p: poll_table 1922 * 1923 * See the comments in the wq_has_sleeper function. 1924 */ 1925static inline void sock_poll_wait(struct file *filp, 1926 wait_queue_head_t *wait_address, poll_table *p) 1927{ 1928 if (!poll_does_not_wait(p) && wait_address) { 1929 poll_wait(filp, wait_address, p); 1930 /* We need to be sure we are in sync with the 1931 * socket flags modification. 1932 * 1933 * This memory barrier is paired in the wq_has_sleeper. 1934 */ 1935 smp_mb(); 1936 } 1937} 1938 1939static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1940{ 1941 if (sk->sk_txhash) { 1942 skb->l4_hash = 1; 1943 skb->hash = sk->sk_txhash; 1944 } 1945} 1946 1947/* 1948 * Queue a received datagram if it will fit. Stream and sequenced 1949 * protocols can't normally use this as they need to fit buffers in 1950 * and play with them. 1951 * 1952 * Inlined as it's very short and called for pretty much every 1953 * packet ever received. 1954 */ 1955 1956static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1957{ 1958 skb_orphan(skb); 1959 skb->sk = sk; 1960 skb->destructor = sock_wfree; 1961 skb_set_hash_from_sk(skb, sk); 1962 /* 1963 * We used to take a refcount on sk, but following operation 1964 * is enough to guarantee sk_free() wont free this sock until 1965 * all in-flight packets are completed 1966 */ 1967 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1968} 1969 1970static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1971{ 1972 skb_orphan(skb); 1973 skb->sk = sk; 1974 skb->destructor = sock_rfree; 1975 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1976 sk_mem_charge(sk, skb->truesize); 1977} 1978 1979void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1980 unsigned long expires); 1981 1982void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1983 1984int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1985 1986int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1987struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 1988 1989/* 1990 * Recover an error report and clear atomically 1991 */ 1992 1993static inline int sock_error(struct sock *sk) 1994{ 1995 int err; 1996 if (likely(!sk->sk_err)) 1997 return 0; 1998 err = xchg(&sk->sk_err, 0); 1999 return -err; 2000} 2001 2002static inline unsigned long sock_wspace(struct sock *sk) 2003{ 2004 int amt = 0; 2005 2006 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2007 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2008 if (amt < 0) 2009 amt = 0; 2010 } 2011 return amt; 2012} 2013 2014static inline void sk_wake_async(struct sock *sk, int how, int band) 2015{ 2016 if (sock_flag(sk, SOCK_FASYNC)) 2017 sock_wake_async(sk->sk_socket, how, band); 2018} 2019 2020/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2021 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2022 * Note: for send buffers, TCP works better if we can build two skbs at 2023 * minimum. 2024 */ 2025#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2026 2027#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2028#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2029 2030static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2031{ 2032 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2033 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2034 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2035 } 2036} 2037 2038struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 2039 2040/** 2041 * sk_page_frag - return an appropriate page_frag 2042 * @sk: socket 2043 * 2044 * If socket allocation mode allows current thread to sleep, it means its 2045 * safe to use the per task page_frag instead of the per socket one. 2046 */ 2047static inline struct page_frag *sk_page_frag(struct sock *sk) 2048{ 2049 if (sk->sk_allocation & __GFP_WAIT) 2050 return ¤t->task_frag; 2051 2052 return &sk->sk_frag; 2053} 2054 2055bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2056 2057/* 2058 * Default write policy as shown to user space via poll/select/SIGIO 2059 */ 2060static inline bool sock_writeable(const struct sock *sk) 2061{ 2062 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2063} 2064 2065static inline gfp_t gfp_any(void) 2066{ 2067 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2068} 2069 2070static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2071{ 2072 return noblock ? 0 : sk->sk_rcvtimeo; 2073} 2074 2075static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2076{ 2077 return noblock ? 0 : sk->sk_sndtimeo; 2078} 2079 2080static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2081{ 2082 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2083} 2084 2085/* Alas, with timeout socket operations are not restartable. 2086 * Compare this to poll(). 2087 */ 2088static inline int sock_intr_errno(long timeo) 2089{ 2090 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2091} 2092 2093struct sock_skb_cb { 2094 u32 dropcount; 2095}; 2096 2097/* Store sock_skb_cb at the end of skb->cb[] so protocol families 2098 * using skb->cb[] would keep using it directly and utilize its 2099 * alignement guarantee. 2100 */ 2101#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2102 sizeof(struct sock_skb_cb))) 2103 2104#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2105 SOCK_SKB_CB_OFFSET)) 2106 2107#define sock_skb_cb_check_size(size) \ 2108 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2109 2110static inline void 2111sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2112{ 2113 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops); 2114} 2115 2116void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2117 struct sk_buff *skb); 2118void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2119 struct sk_buff *skb); 2120 2121static inline void 2122sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2123{ 2124 ktime_t kt = skb->tstamp; 2125 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2126 2127 /* 2128 * generate control messages if 2129 * - receive time stamping in software requested 2130 * - software time stamp available and wanted 2131 * - hardware time stamps available and wanted 2132 */ 2133 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2134 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2135 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2136 (hwtstamps->hwtstamp.tv64 && 2137 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2138 __sock_recv_timestamp(msg, sk, skb); 2139 else 2140 sk->sk_stamp = kt; 2141 2142 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2143 __sock_recv_wifi_status(msg, sk, skb); 2144} 2145 2146void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2147 struct sk_buff *skb); 2148 2149static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2150 struct sk_buff *skb) 2151{ 2152#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2153 (1UL << SOCK_RCVTSTAMP)) 2154#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2155 SOF_TIMESTAMPING_RAW_HARDWARE) 2156 2157 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2158 __sock_recv_ts_and_drops(msg, sk, skb); 2159 else 2160 sk->sk_stamp = skb->tstamp; 2161} 2162 2163void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags); 2164 2165/** 2166 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2167 * @sk: socket sending this packet 2168 * @tx_flags: completed with instructions for time stamping 2169 * 2170 * Note : callers should take care of initial *tx_flags value (usually 0) 2171 */ 2172static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags) 2173{ 2174 if (unlikely(sk->sk_tsflags)) 2175 __sock_tx_timestamp(sk, tx_flags); 2176 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2177 *tx_flags |= SKBTX_WIFI_STATUS; 2178} 2179 2180/** 2181 * sk_eat_skb - Release a skb if it is no longer needed 2182 * @sk: socket to eat this skb from 2183 * @skb: socket buffer to eat 2184 * 2185 * This routine must be called with interrupts disabled or with the socket 2186 * locked so that the sk_buff queue operation is ok. 2187*/ 2188static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2189{ 2190 __skb_unlink(skb, &sk->sk_receive_queue); 2191 __kfree_skb(skb); 2192} 2193 2194static inline 2195struct net *sock_net(const struct sock *sk) 2196{ 2197 return read_pnet(&sk->sk_net); 2198} 2199 2200static inline 2201void sock_net_set(struct sock *sk, struct net *net) 2202{ 2203 write_pnet(&sk->sk_net, net); 2204} 2205 2206/* 2207 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2208 * They should not hold a reference to a namespace in order to allow 2209 * to stop it. 2210 * Sockets after sk_change_net should be released using sk_release_kernel 2211 */ 2212static inline void sk_change_net(struct sock *sk, struct net *net) 2213{ 2214 struct net *current_net = sock_net(sk); 2215 2216 if (!net_eq(current_net, net)) { 2217 put_net(current_net); 2218 sock_net_set(sk, net); 2219 } 2220} 2221 2222static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2223{ 2224 if (skb->sk) { 2225 struct sock *sk = skb->sk; 2226 2227 skb->destructor = NULL; 2228 skb->sk = NULL; 2229 return sk; 2230 } 2231 return NULL; 2232} 2233 2234/* This helper checks if a socket is a full socket, 2235 * ie _not_ a timewait or request socket. 2236 */ 2237static inline bool sk_fullsock(const struct sock *sk) 2238{ 2239 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2240} 2241 2242void sock_enable_timestamp(struct sock *sk, int flag); 2243int sock_get_timestamp(struct sock *, struct timeval __user *); 2244int sock_get_timestampns(struct sock *, struct timespec __user *); 2245int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2246 int type); 2247 2248bool sk_ns_capable(const struct sock *sk, 2249 struct user_namespace *user_ns, int cap); 2250bool sk_capable(const struct sock *sk, int cap); 2251bool sk_net_capable(const struct sock *sk, int cap); 2252 2253extern __u32 sysctl_wmem_max; 2254extern __u32 sysctl_rmem_max; 2255 2256extern int sysctl_tstamp_allow_data; 2257extern int sysctl_optmem_max; 2258 2259extern __u32 sysctl_wmem_default; 2260extern __u32 sysctl_rmem_default; 2261 2262#endif /* _SOCK_H */ 2263