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 #include <linux/wait.h> 62 #include <linux/cgroup-defs.h> 63 64 #include <linux/filter.h> 65 #include <linux/rculist_nulls.h> 66 #include <linux/poll.h> 67 68 #include <linux/atomic.h> 69 #include <net/dst.h> 70 #include <net/checksum.h> 71 #include <net/tcp_states.h> 72 #include <linux/net_tstamp.h> 73 74 /* 75 * This structure really needs to be cleaned up. 76 * Most of it is for TCP, and not used by any of 77 * the other protocols. 78 */ 79 80 /* Define this to get the SOCK_DBG debugging facility. */ 81 #define SOCK_DEBUGGING 82 #ifdef SOCK_DEBUGGING 83 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 84 printk(KERN_DEBUG msg); } while (0) 85 #else 86 /* Validate arguments and do nothing */ 87 static inline __printf(2, 3) 88 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 89 { 90 } 91 #endif 92 93 /* This is the per-socket lock. The spinlock provides a synchronization 94 * between user contexts and software interrupt processing, whereas the 95 * mini-semaphore synchronizes multiple users amongst themselves. 96 */ 97 typedef struct { 98 spinlock_t slock; 99 int owned; 100 wait_queue_head_t wq; 101 /* 102 * We express the mutex-alike socket_lock semantics 103 * to the lock validator by explicitly managing 104 * the slock as a lock variant (in addition to 105 * the slock itself): 106 */ 107 #ifdef CONFIG_DEBUG_LOCK_ALLOC 108 struct lockdep_map dep_map; 109 #endif 110 } socket_lock_t; 111 112 struct sock; 113 struct proto; 114 struct net; 115 116 typedef __u32 __bitwise __portpair; 117 typedef __u64 __bitwise __addrpair; 118 119 /** 120 * struct sock_common - minimal network layer representation of sockets 121 * @skc_daddr: Foreign IPv4 addr 122 * @skc_rcv_saddr: Bound local IPv4 addr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_family: network address family 128 * @skc_state: Connection state 129 * @skc_reuse: %SO_REUSEADDR setting 130 * @skc_reuseport: %SO_REUSEPORT setting 131 * @skc_bound_dev_if: bound device index if != 0 132 * @skc_bind_node: bind hash linkage for various protocol lookup tables 133 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 134 * @skc_prot: protocol handlers inside a network family 135 * @skc_net: reference to the network namespace of this socket 136 * @skc_node: main hash linkage for various protocol lookup tables 137 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 138 * @skc_tx_queue_mapping: tx queue number for this connection 139 * @skc_flags: place holder for sk_flags 140 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 141 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 142 * @skc_incoming_cpu: record/match cpu processing incoming packets 143 * @skc_refcnt: reference count 144 * 145 * This is the minimal network layer representation of sockets, the header 146 * for struct sock and struct inet_timewait_sock. 147 */ 148 struct sock_common { 149 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 150 * address on 64bit arches : cf INET_MATCH() 151 */ 152 union { 153 __addrpair skc_addrpair; 154 struct { 155 __be32 skc_daddr; 156 __be32 skc_rcv_saddr; 157 }; 158 }; 159 union { 160 unsigned int skc_hash; 161 __u16 skc_u16hashes[2]; 162 }; 163 /* skc_dport && skc_num must be grouped as well */ 164 union { 165 __portpair skc_portpair; 166 struct { 167 __be16 skc_dport; 168 __u16 skc_num; 169 }; 170 }; 171 172 unsigned short skc_family; 173 volatile unsigned char skc_state; 174 unsigned char skc_reuse:4; 175 unsigned char skc_reuseport:1; 176 unsigned char skc_ipv6only:1; 177 unsigned char skc_net_refcnt:1; 178 int skc_bound_dev_if; 179 union { 180 struct hlist_node skc_bind_node; 181 struct hlist_node skc_portaddr_node; 182 }; 183 struct proto *skc_prot; 184 possible_net_t skc_net; 185 186 #if IS_ENABLED(CONFIG_IPV6) 187 struct in6_addr skc_v6_daddr; 188 struct in6_addr skc_v6_rcv_saddr; 189 #endif 190 191 atomic64_t skc_cookie; 192 193 /* following fields are padding to force 194 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 195 * assuming IPV6 is enabled. We use this padding differently 196 * for different kind of 'sockets' 197 */ 198 union { 199 unsigned long skc_flags; 200 struct sock *skc_listener; /* request_sock */ 201 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 202 }; 203 /* 204 * fields between dontcopy_begin/dontcopy_end 205 * are not copied in sock_copy() 206 */ 207 /* private: */ 208 int skc_dontcopy_begin[0]; 209 /* public: */ 210 union { 211 struct hlist_node skc_node; 212 struct hlist_nulls_node skc_nulls_node; 213 }; 214 int skc_tx_queue_mapping; 215 union { 216 int skc_incoming_cpu; 217 u32 skc_rcv_wnd; 218 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 219 }; 220 221 atomic_t skc_refcnt; 222 /* private: */ 223 int skc_dontcopy_end[0]; 224 union { 225 u32 skc_rxhash; 226 u32 skc_window_clamp; 227 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 228 }; 229 /* public: */ 230 }; 231 232 /** 233 * struct sock - network layer representation of sockets 234 * @__sk_common: shared layout with inet_timewait_sock 235 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 236 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 237 * @sk_lock: synchronizer 238 * @sk_rcvbuf: size of receive buffer in bytes 239 * @sk_wq: sock wait queue and async head 240 * @sk_rx_dst: receive input route used by early demux 241 * @sk_dst_cache: destination cache 242 * @sk_policy: flow policy 243 * @sk_receive_queue: incoming packets 244 * @sk_wmem_alloc: transmit queue bytes committed 245 * @sk_write_queue: Packet sending queue 246 * @sk_omem_alloc: "o" is "option" or "other" 247 * @sk_wmem_queued: persistent queue size 248 * @sk_forward_alloc: space allocated forward 249 * @sk_napi_id: id of the last napi context to receive data for sk 250 * @sk_ll_usec: usecs to busypoll when there is no data 251 * @sk_allocation: allocation mode 252 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 253 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 254 * @sk_sndbuf: size of send buffer in bytes 255 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 256 * @sk_no_check_rx: allow zero checksum in RX packets 257 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 258 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 259 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 260 * @sk_gso_max_size: Maximum GSO segment size to build 261 * @sk_gso_max_segs: Maximum number of GSO segments 262 * @sk_lingertime: %SO_LINGER l_linger setting 263 * @sk_backlog: always used with the per-socket spinlock held 264 * @sk_callback_lock: used with the callbacks in the end of this struct 265 * @sk_error_queue: rarely used 266 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 267 * IPV6_ADDRFORM for instance) 268 * @sk_err: last error 269 * @sk_err_soft: errors that don't cause failure but are the cause of a 270 * persistent failure not just 'timed out' 271 * @sk_drops: raw/udp drops counter 272 * @sk_ack_backlog: current listen backlog 273 * @sk_max_ack_backlog: listen backlog set in listen() 274 * @sk_priority: %SO_PRIORITY setting 275 * @sk_type: socket type (%SOCK_STREAM, etc) 276 * @sk_protocol: which protocol this socket belongs in this network family 277 * @sk_peer_pid: &struct pid for this socket's peer 278 * @sk_peer_cred: %SO_PEERCRED setting 279 * @sk_rcvlowat: %SO_RCVLOWAT setting 280 * @sk_rcvtimeo: %SO_RCVTIMEO setting 281 * @sk_sndtimeo: %SO_SNDTIMEO setting 282 * @sk_txhash: computed flow hash for use on transmit 283 * @sk_filter: socket filtering instructions 284 * @sk_timer: sock cleanup timer 285 * @sk_stamp: time stamp of last packet received 286 * @sk_tsflags: SO_TIMESTAMPING socket options 287 * @sk_tskey: counter to disambiguate concurrent tstamp requests 288 * @sk_socket: Identd and reporting IO signals 289 * @sk_user_data: RPC layer private data 290 * @sk_frag: cached page frag 291 * @sk_peek_off: current peek_offset value 292 * @sk_send_head: front of stuff to transmit 293 * @sk_security: used by security modules 294 * @sk_mark: generic packet mark 295 * @sk_cgrp_data: cgroup data for this cgroup 296 * @sk_memcg: this socket's memory cgroup association 297 * @sk_write_pending: a write to stream socket waits to start 298 * @sk_state_change: callback to indicate change in the state of the sock 299 * @sk_data_ready: callback to indicate there is data to be processed 300 * @sk_write_space: callback to indicate there is bf sending space available 301 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 302 * @sk_backlog_rcv: callback to process the backlog 303 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 304 * @sk_reuseport_cb: reuseport group container 305 */ 306 struct sock { 307 /* 308 * Now struct inet_timewait_sock also uses sock_common, so please just 309 * don't add nothing before this first member (__sk_common) --acme 310 */ 311 struct sock_common __sk_common; 312 #define sk_node __sk_common.skc_node 313 #define sk_nulls_node __sk_common.skc_nulls_node 314 #define sk_refcnt __sk_common.skc_refcnt 315 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 316 317 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 318 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 319 #define sk_hash __sk_common.skc_hash 320 #define sk_portpair __sk_common.skc_portpair 321 #define sk_num __sk_common.skc_num 322 #define sk_dport __sk_common.skc_dport 323 #define sk_addrpair __sk_common.skc_addrpair 324 #define sk_daddr __sk_common.skc_daddr 325 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 326 #define sk_family __sk_common.skc_family 327 #define sk_state __sk_common.skc_state 328 #define sk_reuse __sk_common.skc_reuse 329 #define sk_reuseport __sk_common.skc_reuseport 330 #define sk_ipv6only __sk_common.skc_ipv6only 331 #define sk_net_refcnt __sk_common.skc_net_refcnt 332 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 333 #define sk_bind_node __sk_common.skc_bind_node 334 #define sk_prot __sk_common.skc_prot 335 #define sk_net __sk_common.skc_net 336 #define sk_v6_daddr __sk_common.skc_v6_daddr 337 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 338 #define sk_cookie __sk_common.skc_cookie 339 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 340 #define sk_flags __sk_common.skc_flags 341 #define sk_rxhash __sk_common.skc_rxhash 342 343 socket_lock_t sk_lock; 344 struct sk_buff_head sk_receive_queue; 345 /* 346 * The backlog queue is special, it is always used with 347 * the per-socket spinlock held and requires low latency 348 * access. Therefore we special case it's implementation. 349 * Note : rmem_alloc is in this structure to fill a hole 350 * on 64bit arches, not because its logically part of 351 * backlog. 352 */ 353 struct { 354 atomic_t rmem_alloc; 355 int len; 356 struct sk_buff *head; 357 struct sk_buff *tail; 358 } sk_backlog; 359 #define sk_rmem_alloc sk_backlog.rmem_alloc 360 int sk_forward_alloc; 361 362 __u32 sk_txhash; 363 #ifdef CONFIG_NET_RX_BUSY_POLL 364 unsigned int sk_napi_id; 365 unsigned int sk_ll_usec; 366 #endif 367 atomic_t sk_drops; 368 int sk_rcvbuf; 369 370 struct sk_filter __rcu *sk_filter; 371 union { 372 struct socket_wq __rcu *sk_wq; 373 struct socket_wq *sk_wq_raw; 374 }; 375 #ifdef CONFIG_XFRM 376 struct xfrm_policy __rcu *sk_policy[2]; 377 #endif 378 struct dst_entry *sk_rx_dst; 379 struct dst_entry __rcu *sk_dst_cache; 380 /* Note: 32bit hole on 64bit arches */ 381 atomic_t sk_wmem_alloc; 382 atomic_t sk_omem_alloc; 383 int sk_sndbuf; 384 struct sk_buff_head sk_write_queue; 385 386 /* 387 * Because of non atomicity rules, all 388 * changes are protected by socket lock. 389 */ 390 kmemcheck_bitfield_begin(flags); 391 unsigned int sk_padding : 2, 392 sk_no_check_tx : 1, 393 sk_no_check_rx : 1, 394 sk_userlocks : 4, 395 sk_protocol : 8, 396 sk_type : 16; 397 #define SK_PROTOCOL_MAX U8_MAX 398 kmemcheck_bitfield_end(flags); 399 400 int sk_wmem_queued; 401 gfp_t sk_allocation; 402 u32 sk_pacing_rate; /* bytes per second */ 403 u32 sk_max_pacing_rate; 404 netdev_features_t sk_route_caps; 405 netdev_features_t sk_route_nocaps; 406 int sk_gso_type; 407 unsigned int sk_gso_max_size; 408 u16 sk_gso_max_segs; 409 int sk_rcvlowat; 410 unsigned long sk_lingertime; 411 struct sk_buff_head sk_error_queue; 412 struct proto *sk_prot_creator; 413 rwlock_t sk_callback_lock; 414 int sk_err, 415 sk_err_soft; 416 u32 sk_ack_backlog; 417 u32 sk_max_ack_backlog; 418 __u32 sk_priority; 419 __u32 sk_mark; 420 struct pid *sk_peer_pid; 421 const struct cred *sk_peer_cred; 422 long sk_rcvtimeo; 423 long sk_sndtimeo; 424 struct timer_list sk_timer; 425 ktime_t sk_stamp; 426 u16 sk_tsflags; 427 u8 sk_shutdown; 428 u32 sk_tskey; 429 struct socket *sk_socket; 430 void *sk_user_data; 431 struct page_frag sk_frag; 432 struct sk_buff *sk_send_head; 433 __s32 sk_peek_off; 434 int sk_write_pending; 435 #ifdef CONFIG_SECURITY 436 void *sk_security; 437 #endif 438 struct sock_cgroup_data sk_cgrp_data; 439 struct mem_cgroup *sk_memcg; 440 void (*sk_state_change)(struct sock *sk); 441 void (*sk_data_ready)(struct sock *sk); 442 void (*sk_write_space)(struct sock *sk); 443 void (*sk_error_report)(struct sock *sk); 444 int (*sk_backlog_rcv)(struct sock *sk, 445 struct sk_buff *skb); 446 void (*sk_destruct)(struct sock *sk); 447 struct sock_reuseport __rcu *sk_reuseport_cb; 448 struct rcu_head sk_rcu; 449 }; 450 451 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 452 453 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 454 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 455 456 /* 457 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 458 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 459 * on a socket means that the socket will reuse everybody else's port 460 * without looking at the other's sk_reuse value. 461 */ 462 463 #define SK_NO_REUSE 0 464 #define SK_CAN_REUSE 1 465 #define SK_FORCE_REUSE 2 466 467 int sk_set_peek_off(struct sock *sk, int val); 468 469 static inline int sk_peek_offset(struct sock *sk, int flags) 470 { 471 if (unlikely(flags & MSG_PEEK)) { 472 s32 off = READ_ONCE(sk->sk_peek_off); 473 if (off >= 0) 474 return off; 475 } 476 477 return 0; 478 } 479 480 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 481 { 482 s32 off = READ_ONCE(sk->sk_peek_off); 483 484 if (unlikely(off >= 0)) { 485 off = max_t(s32, off - val, 0); 486 WRITE_ONCE(sk->sk_peek_off, off); 487 } 488 } 489 490 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 491 { 492 sk_peek_offset_bwd(sk, -val); 493 } 494 495 /* 496 * Hashed lists helper routines 497 */ 498 static inline struct sock *sk_entry(const struct hlist_node *node) 499 { 500 return hlist_entry(node, struct sock, sk_node); 501 } 502 503 static inline struct sock *__sk_head(const struct hlist_head *head) 504 { 505 return hlist_entry(head->first, struct sock, sk_node); 506 } 507 508 static inline struct sock *sk_head(const struct hlist_head *head) 509 { 510 return hlist_empty(head) ? NULL : __sk_head(head); 511 } 512 513 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 514 { 515 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 516 } 517 518 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 519 { 520 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 521 } 522 523 static inline struct sock *sk_next(const struct sock *sk) 524 { 525 return sk->sk_node.next ? 526 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 527 } 528 529 static inline struct sock *sk_nulls_next(const struct sock *sk) 530 { 531 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 532 hlist_nulls_entry(sk->sk_nulls_node.next, 533 struct sock, sk_nulls_node) : 534 NULL; 535 } 536 537 static inline bool sk_unhashed(const struct sock *sk) 538 { 539 return hlist_unhashed(&sk->sk_node); 540 } 541 542 static inline bool sk_hashed(const struct sock *sk) 543 { 544 return !sk_unhashed(sk); 545 } 546 547 static inline void sk_node_init(struct hlist_node *node) 548 { 549 node->pprev = NULL; 550 } 551 552 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 553 { 554 node->pprev = NULL; 555 } 556 557 static inline void __sk_del_node(struct sock *sk) 558 { 559 __hlist_del(&sk->sk_node); 560 } 561 562 /* NB: equivalent to hlist_del_init_rcu */ 563 static inline bool __sk_del_node_init(struct sock *sk) 564 { 565 if (sk_hashed(sk)) { 566 __sk_del_node(sk); 567 sk_node_init(&sk->sk_node); 568 return true; 569 } 570 return false; 571 } 572 573 /* Grab socket reference count. This operation is valid only 574 when sk is ALREADY grabbed f.e. it is found in hash table 575 or a list and the lookup is made under lock preventing hash table 576 modifications. 577 */ 578 579 static __always_inline void sock_hold(struct sock *sk) 580 { 581 atomic_inc(&sk->sk_refcnt); 582 } 583 584 /* Ungrab socket in the context, which assumes that socket refcnt 585 cannot hit zero, f.e. it is true in context of any socketcall. 586 */ 587 static __always_inline void __sock_put(struct sock *sk) 588 { 589 atomic_dec(&sk->sk_refcnt); 590 } 591 592 static inline bool sk_del_node_init(struct sock *sk) 593 { 594 bool rc = __sk_del_node_init(sk); 595 596 if (rc) { 597 /* paranoid for a while -acme */ 598 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 599 __sock_put(sk); 600 } 601 return rc; 602 } 603 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 604 605 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 606 { 607 if (sk_hashed(sk)) { 608 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 609 return true; 610 } 611 return false; 612 } 613 614 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 615 { 616 bool rc = __sk_nulls_del_node_init_rcu(sk); 617 618 if (rc) { 619 /* paranoid for a while -acme */ 620 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 621 __sock_put(sk); 622 } 623 return rc; 624 } 625 626 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 627 { 628 hlist_add_head(&sk->sk_node, list); 629 } 630 631 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 632 { 633 sock_hold(sk); 634 __sk_add_node(sk, list); 635 } 636 637 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 638 { 639 sock_hold(sk); 640 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 641 sk->sk_family == AF_INET6) 642 hlist_add_tail_rcu(&sk->sk_node, list); 643 else 644 hlist_add_head_rcu(&sk->sk_node, list); 645 } 646 647 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 648 { 649 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 650 sk->sk_family == AF_INET6) 651 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 652 else 653 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 654 } 655 656 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 657 { 658 sock_hold(sk); 659 __sk_nulls_add_node_rcu(sk, list); 660 } 661 662 static inline void __sk_del_bind_node(struct sock *sk) 663 { 664 __hlist_del(&sk->sk_bind_node); 665 } 666 667 static inline void sk_add_bind_node(struct sock *sk, 668 struct hlist_head *list) 669 { 670 hlist_add_head(&sk->sk_bind_node, list); 671 } 672 673 #define sk_for_each(__sk, list) \ 674 hlist_for_each_entry(__sk, list, sk_node) 675 #define sk_for_each_rcu(__sk, list) \ 676 hlist_for_each_entry_rcu(__sk, list, sk_node) 677 #define sk_nulls_for_each(__sk, node, list) \ 678 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 679 #define sk_nulls_for_each_rcu(__sk, node, list) \ 680 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 681 #define sk_for_each_from(__sk) \ 682 hlist_for_each_entry_from(__sk, sk_node) 683 #define sk_nulls_for_each_from(__sk, node) \ 684 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 685 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 686 #define sk_for_each_safe(__sk, tmp, list) \ 687 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 688 #define sk_for_each_bound(__sk, list) \ 689 hlist_for_each_entry(__sk, list, sk_bind_node) 690 691 /** 692 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 693 * @tpos: the type * to use as a loop cursor. 694 * @pos: the &struct hlist_node to use as a loop cursor. 695 * @head: the head for your list. 696 * @offset: offset of hlist_node within the struct. 697 * 698 */ 699 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 700 for (pos = rcu_dereference((head)->first); \ 701 pos != NULL && \ 702 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 703 pos = rcu_dereference(pos->next)) 704 705 static inline struct user_namespace *sk_user_ns(struct sock *sk) 706 { 707 /* Careful only use this in a context where these parameters 708 * can not change and must all be valid, such as recvmsg from 709 * userspace. 710 */ 711 return sk->sk_socket->file->f_cred->user_ns; 712 } 713 714 /* Sock flags */ 715 enum sock_flags { 716 SOCK_DEAD, 717 SOCK_DONE, 718 SOCK_URGINLINE, 719 SOCK_KEEPOPEN, 720 SOCK_LINGER, 721 SOCK_DESTROY, 722 SOCK_BROADCAST, 723 SOCK_TIMESTAMP, 724 SOCK_ZAPPED, 725 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 726 SOCK_DBG, /* %SO_DEBUG setting */ 727 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 728 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 729 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 730 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 731 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 732 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 733 SOCK_FASYNC, /* fasync() active */ 734 SOCK_RXQ_OVFL, 735 SOCK_ZEROCOPY, /* buffers from userspace */ 736 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 737 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 738 * Will use last 4 bytes of packet sent from 739 * user-space instead. 740 */ 741 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 742 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 743 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 744 }; 745 746 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 747 748 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 749 { 750 nsk->sk_flags = osk->sk_flags; 751 } 752 753 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 754 { 755 __set_bit(flag, &sk->sk_flags); 756 } 757 758 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 759 { 760 __clear_bit(flag, &sk->sk_flags); 761 } 762 763 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 764 { 765 return test_bit(flag, &sk->sk_flags); 766 } 767 768 #ifdef CONFIG_NET 769 extern struct static_key memalloc_socks; 770 static inline int sk_memalloc_socks(void) 771 { 772 return static_key_false(&memalloc_socks); 773 } 774 #else 775 776 static inline int sk_memalloc_socks(void) 777 { 778 return 0; 779 } 780 781 #endif 782 783 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 784 { 785 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 786 } 787 788 static inline void sk_acceptq_removed(struct sock *sk) 789 { 790 sk->sk_ack_backlog--; 791 } 792 793 static inline void sk_acceptq_added(struct sock *sk) 794 { 795 sk->sk_ack_backlog++; 796 } 797 798 static inline bool sk_acceptq_is_full(const struct sock *sk) 799 { 800 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 801 } 802 803 /* 804 * Compute minimal free write space needed to queue new packets. 805 */ 806 static inline int sk_stream_min_wspace(const struct sock *sk) 807 { 808 return sk->sk_wmem_queued >> 1; 809 } 810 811 static inline int sk_stream_wspace(const struct sock *sk) 812 { 813 return sk->sk_sndbuf - sk->sk_wmem_queued; 814 } 815 816 void sk_stream_write_space(struct sock *sk); 817 818 /* OOB backlog add */ 819 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 820 { 821 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 822 skb_dst_force_safe(skb); 823 824 if (!sk->sk_backlog.tail) 825 sk->sk_backlog.head = skb; 826 else 827 sk->sk_backlog.tail->next = skb; 828 829 sk->sk_backlog.tail = skb; 830 skb->next = NULL; 831 } 832 833 /* 834 * Take into account size of receive queue and backlog queue 835 * Do not take into account this skb truesize, 836 * to allow even a single big packet to come. 837 */ 838 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 839 { 840 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 841 842 return qsize > limit; 843 } 844 845 /* The per-socket spinlock must be held here. */ 846 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 847 unsigned int limit) 848 { 849 if (sk_rcvqueues_full(sk, limit)) 850 return -ENOBUFS; 851 852 /* 853 * If the skb was allocated from pfmemalloc reserves, only 854 * allow SOCK_MEMALLOC sockets to use it as this socket is 855 * helping free memory 856 */ 857 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 858 return -ENOMEM; 859 860 __sk_add_backlog(sk, skb); 861 sk->sk_backlog.len += skb->truesize; 862 return 0; 863 } 864 865 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 866 867 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 868 { 869 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 870 return __sk_backlog_rcv(sk, skb); 871 872 return sk->sk_backlog_rcv(sk, skb); 873 } 874 875 static inline void sk_incoming_cpu_update(struct sock *sk) 876 { 877 sk->sk_incoming_cpu = raw_smp_processor_id(); 878 } 879 880 static inline void sock_rps_record_flow_hash(__u32 hash) 881 { 882 #ifdef CONFIG_RPS 883 struct rps_sock_flow_table *sock_flow_table; 884 885 rcu_read_lock(); 886 sock_flow_table = rcu_dereference(rps_sock_flow_table); 887 rps_record_sock_flow(sock_flow_table, hash); 888 rcu_read_unlock(); 889 #endif 890 } 891 892 static inline void sock_rps_record_flow(const struct sock *sk) 893 { 894 #ifdef CONFIG_RPS 895 sock_rps_record_flow_hash(sk->sk_rxhash); 896 #endif 897 } 898 899 static inline void sock_rps_save_rxhash(struct sock *sk, 900 const struct sk_buff *skb) 901 { 902 #ifdef CONFIG_RPS 903 if (unlikely(sk->sk_rxhash != skb->hash)) 904 sk->sk_rxhash = skb->hash; 905 #endif 906 } 907 908 static inline void sock_rps_reset_rxhash(struct sock *sk) 909 { 910 #ifdef CONFIG_RPS 911 sk->sk_rxhash = 0; 912 #endif 913 } 914 915 #define sk_wait_event(__sk, __timeo, __condition) \ 916 ({ int __rc; \ 917 release_sock(__sk); \ 918 __rc = __condition; \ 919 if (!__rc) { \ 920 *(__timeo) = schedule_timeout(*(__timeo)); \ 921 } \ 922 sched_annotate_sleep(); \ 923 lock_sock(__sk); \ 924 __rc = __condition; \ 925 __rc; \ 926 }) 927 928 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 929 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 930 void sk_stream_wait_close(struct sock *sk, long timeo_p); 931 int sk_stream_error(struct sock *sk, int flags, int err); 932 void sk_stream_kill_queues(struct sock *sk); 933 void sk_set_memalloc(struct sock *sk); 934 void sk_clear_memalloc(struct sock *sk); 935 936 void __sk_flush_backlog(struct sock *sk); 937 938 static inline bool sk_flush_backlog(struct sock *sk) 939 { 940 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 941 __sk_flush_backlog(sk); 942 return true; 943 } 944 return false; 945 } 946 947 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 948 949 struct request_sock_ops; 950 struct timewait_sock_ops; 951 struct inet_hashinfo; 952 struct raw_hashinfo; 953 struct module; 954 955 /* 956 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 957 * un-modified. Special care is taken when initializing object to zero. 958 */ 959 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 960 { 961 if (offsetof(struct sock, sk_node.next) != 0) 962 memset(sk, 0, offsetof(struct sock, sk_node.next)); 963 memset(&sk->sk_node.pprev, 0, 964 size - offsetof(struct sock, sk_node.pprev)); 965 } 966 967 /* Networking protocol blocks we attach to sockets. 968 * socket layer -> transport layer interface 969 */ 970 struct proto { 971 void (*close)(struct sock *sk, 972 long timeout); 973 int (*connect)(struct sock *sk, 974 struct sockaddr *uaddr, 975 int addr_len); 976 int (*disconnect)(struct sock *sk, int flags); 977 978 struct sock * (*accept)(struct sock *sk, int flags, int *err); 979 980 int (*ioctl)(struct sock *sk, int cmd, 981 unsigned long arg); 982 int (*init)(struct sock *sk); 983 void (*destroy)(struct sock *sk); 984 void (*shutdown)(struct sock *sk, int how); 985 int (*setsockopt)(struct sock *sk, int level, 986 int optname, char __user *optval, 987 unsigned int optlen); 988 int (*getsockopt)(struct sock *sk, int level, 989 int optname, char __user *optval, 990 int __user *option); 991 #ifdef CONFIG_COMPAT 992 int (*compat_setsockopt)(struct sock *sk, 993 int level, 994 int optname, char __user *optval, 995 unsigned int optlen); 996 int (*compat_getsockopt)(struct sock *sk, 997 int level, 998 int optname, char __user *optval, 999 int __user *option); 1000 int (*compat_ioctl)(struct sock *sk, 1001 unsigned int cmd, unsigned long arg); 1002 #endif 1003 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1004 size_t len); 1005 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1006 size_t len, int noblock, int flags, 1007 int *addr_len); 1008 int (*sendpage)(struct sock *sk, struct page *page, 1009 int offset, size_t size, int flags); 1010 int (*bind)(struct sock *sk, 1011 struct sockaddr *uaddr, int addr_len); 1012 1013 int (*backlog_rcv) (struct sock *sk, 1014 struct sk_buff *skb); 1015 1016 void (*release_cb)(struct sock *sk); 1017 1018 /* Keeping track of sk's, looking them up, and port selection methods. */ 1019 int (*hash)(struct sock *sk); 1020 void (*unhash)(struct sock *sk); 1021 void (*rehash)(struct sock *sk); 1022 int (*get_port)(struct sock *sk, unsigned short snum); 1023 1024 /* Keeping track of sockets in use */ 1025 #ifdef CONFIG_PROC_FS 1026 unsigned int inuse_idx; 1027 #endif 1028 1029 bool (*stream_memory_free)(const struct sock *sk); 1030 /* Memory pressure */ 1031 void (*enter_memory_pressure)(struct sock *sk); 1032 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1033 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1034 /* 1035 * Pressure flag: try to collapse. 1036 * Technical note: it is used by multiple contexts non atomically. 1037 * All the __sk_mem_schedule() is of this nature: accounting 1038 * is strict, actions are advisory and have some latency. 1039 */ 1040 int *memory_pressure; 1041 long *sysctl_mem; 1042 int *sysctl_wmem; 1043 int *sysctl_rmem; 1044 int max_header; 1045 bool no_autobind; 1046 1047 struct kmem_cache *slab; 1048 unsigned int obj_size; 1049 int slab_flags; 1050 1051 struct percpu_counter *orphan_count; 1052 1053 struct request_sock_ops *rsk_prot; 1054 struct timewait_sock_ops *twsk_prot; 1055 1056 union { 1057 struct inet_hashinfo *hashinfo; 1058 struct udp_table *udp_table; 1059 struct raw_hashinfo *raw_hash; 1060 } h; 1061 1062 struct module *owner; 1063 1064 char name[32]; 1065 1066 struct list_head node; 1067 #ifdef SOCK_REFCNT_DEBUG 1068 atomic_t socks; 1069 #endif 1070 int (*diag_destroy)(struct sock *sk, int err); 1071 }; 1072 1073 int proto_register(struct proto *prot, int alloc_slab); 1074 void proto_unregister(struct proto *prot); 1075 1076 #ifdef SOCK_REFCNT_DEBUG 1077 static inline void sk_refcnt_debug_inc(struct sock *sk) 1078 { 1079 atomic_inc(&sk->sk_prot->socks); 1080 } 1081 1082 static inline void sk_refcnt_debug_dec(struct sock *sk) 1083 { 1084 atomic_dec(&sk->sk_prot->socks); 1085 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1086 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1087 } 1088 1089 static inline void sk_refcnt_debug_release(const struct sock *sk) 1090 { 1091 if (atomic_read(&sk->sk_refcnt) != 1) 1092 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1093 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1094 } 1095 #else /* SOCK_REFCNT_DEBUG */ 1096 #define sk_refcnt_debug_inc(sk) do { } while (0) 1097 #define sk_refcnt_debug_dec(sk) do { } while (0) 1098 #define sk_refcnt_debug_release(sk) do { } while (0) 1099 #endif /* SOCK_REFCNT_DEBUG */ 1100 1101 static inline bool sk_stream_memory_free(const struct sock *sk) 1102 { 1103 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1104 return false; 1105 1106 return sk->sk_prot->stream_memory_free ? 1107 sk->sk_prot->stream_memory_free(sk) : true; 1108 } 1109 1110 static inline bool sk_stream_is_writeable(const struct sock *sk) 1111 { 1112 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1113 sk_stream_memory_free(sk); 1114 } 1115 1116 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1117 struct cgroup *ancestor) 1118 { 1119 #ifdef CONFIG_SOCK_CGROUP_DATA 1120 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1121 ancestor); 1122 #else 1123 return -ENOTSUPP; 1124 #endif 1125 } 1126 1127 static inline bool sk_has_memory_pressure(const struct sock *sk) 1128 { 1129 return sk->sk_prot->memory_pressure != NULL; 1130 } 1131 1132 static inline bool sk_under_memory_pressure(const struct sock *sk) 1133 { 1134 if (!sk->sk_prot->memory_pressure) 1135 return false; 1136 1137 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1138 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1139 return true; 1140 1141 return !!*sk->sk_prot->memory_pressure; 1142 } 1143 1144 static inline void sk_leave_memory_pressure(struct sock *sk) 1145 { 1146 int *memory_pressure = sk->sk_prot->memory_pressure; 1147 1148 if (!memory_pressure) 1149 return; 1150 1151 if (*memory_pressure) 1152 *memory_pressure = 0; 1153 } 1154 1155 static inline void sk_enter_memory_pressure(struct sock *sk) 1156 { 1157 if (!sk->sk_prot->enter_memory_pressure) 1158 return; 1159 1160 sk->sk_prot->enter_memory_pressure(sk); 1161 } 1162 1163 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1164 { 1165 return sk->sk_prot->sysctl_mem[index]; 1166 } 1167 1168 static inline long 1169 sk_memory_allocated(const struct sock *sk) 1170 { 1171 return atomic_long_read(sk->sk_prot->memory_allocated); 1172 } 1173 1174 static inline long 1175 sk_memory_allocated_add(struct sock *sk, int amt) 1176 { 1177 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1178 } 1179 1180 static inline void 1181 sk_memory_allocated_sub(struct sock *sk, int amt) 1182 { 1183 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1184 } 1185 1186 static inline void sk_sockets_allocated_dec(struct sock *sk) 1187 { 1188 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1189 } 1190 1191 static inline void sk_sockets_allocated_inc(struct sock *sk) 1192 { 1193 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1194 } 1195 1196 static inline int 1197 sk_sockets_allocated_read_positive(struct sock *sk) 1198 { 1199 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1200 } 1201 1202 static inline int 1203 proto_sockets_allocated_sum_positive(struct proto *prot) 1204 { 1205 return percpu_counter_sum_positive(prot->sockets_allocated); 1206 } 1207 1208 static inline long 1209 proto_memory_allocated(struct proto *prot) 1210 { 1211 return atomic_long_read(prot->memory_allocated); 1212 } 1213 1214 static inline bool 1215 proto_memory_pressure(struct proto *prot) 1216 { 1217 if (!prot->memory_pressure) 1218 return false; 1219 return !!*prot->memory_pressure; 1220 } 1221 1222 1223 #ifdef CONFIG_PROC_FS 1224 /* Called with local bh disabled */ 1225 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1226 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1227 #else 1228 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1229 int inc) 1230 { 1231 } 1232 #endif 1233 1234 1235 /* With per-bucket locks this operation is not-atomic, so that 1236 * this version is not worse. 1237 */ 1238 static inline int __sk_prot_rehash(struct sock *sk) 1239 { 1240 sk->sk_prot->unhash(sk); 1241 return sk->sk_prot->hash(sk); 1242 } 1243 1244 /* About 10 seconds */ 1245 #define SOCK_DESTROY_TIME (10*HZ) 1246 1247 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1248 #define PROT_SOCK 1024 1249 1250 #define SHUTDOWN_MASK 3 1251 #define RCV_SHUTDOWN 1 1252 #define SEND_SHUTDOWN 2 1253 1254 #define SOCK_SNDBUF_LOCK 1 1255 #define SOCK_RCVBUF_LOCK 2 1256 #define SOCK_BINDADDR_LOCK 4 1257 #define SOCK_BINDPORT_LOCK 8 1258 1259 struct socket_alloc { 1260 struct socket socket; 1261 struct inode vfs_inode; 1262 }; 1263 1264 static inline struct socket *SOCKET_I(struct inode *inode) 1265 { 1266 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1267 } 1268 1269 static inline struct inode *SOCK_INODE(struct socket *socket) 1270 { 1271 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1272 } 1273 1274 /* 1275 * Functions for memory accounting 1276 */ 1277 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1278 void __sk_mem_reclaim(struct sock *sk, int amount); 1279 1280 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1281 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1282 #define SK_MEM_SEND 0 1283 #define SK_MEM_RECV 1 1284 1285 static inline int sk_mem_pages(int amt) 1286 { 1287 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1288 } 1289 1290 static inline bool sk_has_account(struct sock *sk) 1291 { 1292 /* return true if protocol supports memory accounting */ 1293 return !!sk->sk_prot->memory_allocated; 1294 } 1295 1296 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1297 { 1298 if (!sk_has_account(sk)) 1299 return true; 1300 return size <= sk->sk_forward_alloc || 1301 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1302 } 1303 1304 static inline bool 1305 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1306 { 1307 if (!sk_has_account(sk)) 1308 return true; 1309 return size<= sk->sk_forward_alloc || 1310 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1311 skb_pfmemalloc(skb); 1312 } 1313 1314 static inline void sk_mem_reclaim(struct sock *sk) 1315 { 1316 if (!sk_has_account(sk)) 1317 return; 1318 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1319 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1320 } 1321 1322 static inline void sk_mem_reclaim_partial(struct sock *sk) 1323 { 1324 if (!sk_has_account(sk)) 1325 return; 1326 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1327 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1328 } 1329 1330 static inline void sk_mem_charge(struct sock *sk, int size) 1331 { 1332 if (!sk_has_account(sk)) 1333 return; 1334 sk->sk_forward_alloc -= size; 1335 } 1336 1337 static inline void sk_mem_uncharge(struct sock *sk, int size) 1338 { 1339 if (!sk_has_account(sk)) 1340 return; 1341 sk->sk_forward_alloc += size; 1342 } 1343 1344 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1345 { 1346 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1347 sk->sk_wmem_queued -= skb->truesize; 1348 sk_mem_uncharge(sk, skb->truesize); 1349 __kfree_skb(skb); 1350 } 1351 1352 static inline void sock_release_ownership(struct sock *sk) 1353 { 1354 if (sk->sk_lock.owned) { 1355 sk->sk_lock.owned = 0; 1356 1357 /* The sk_lock has mutex_unlock() semantics: */ 1358 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1359 } 1360 } 1361 1362 /* 1363 * Macro so as to not evaluate some arguments when 1364 * lockdep is not enabled. 1365 * 1366 * Mark both the sk_lock and the sk_lock.slock as a 1367 * per-address-family lock class. 1368 */ 1369 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1370 do { \ 1371 sk->sk_lock.owned = 0; \ 1372 init_waitqueue_head(&sk->sk_lock.wq); \ 1373 spin_lock_init(&(sk)->sk_lock.slock); \ 1374 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1375 sizeof((sk)->sk_lock)); \ 1376 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1377 (skey), (sname)); \ 1378 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1379 } while (0) 1380 1381 #ifdef CONFIG_LOCKDEP 1382 static inline bool lockdep_sock_is_held(const struct sock *csk) 1383 { 1384 struct sock *sk = (struct sock *)csk; 1385 1386 return lockdep_is_held(&sk->sk_lock) || 1387 lockdep_is_held(&sk->sk_lock.slock); 1388 } 1389 #endif 1390 1391 void lock_sock_nested(struct sock *sk, int subclass); 1392 1393 static inline void lock_sock(struct sock *sk) 1394 { 1395 lock_sock_nested(sk, 0); 1396 } 1397 1398 void release_sock(struct sock *sk); 1399 1400 /* BH context may only use the following locking interface. */ 1401 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1402 #define bh_lock_sock_nested(__sk) \ 1403 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1404 SINGLE_DEPTH_NESTING) 1405 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1406 1407 bool lock_sock_fast(struct sock *sk); 1408 /** 1409 * unlock_sock_fast - complement of lock_sock_fast 1410 * @sk: socket 1411 * @slow: slow mode 1412 * 1413 * fast unlock socket for user context. 1414 * If slow mode is on, we call regular release_sock() 1415 */ 1416 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1417 { 1418 if (slow) 1419 release_sock(sk); 1420 else 1421 spin_unlock_bh(&sk->sk_lock.slock); 1422 } 1423 1424 /* Used by processes to "lock" a socket state, so that 1425 * interrupts and bottom half handlers won't change it 1426 * from under us. It essentially blocks any incoming 1427 * packets, so that we won't get any new data or any 1428 * packets that change the state of the socket. 1429 * 1430 * While locked, BH processing will add new packets to 1431 * the backlog queue. This queue is processed by the 1432 * owner of the socket lock right before it is released. 1433 * 1434 * Since ~2.3.5 it is also exclusive sleep lock serializing 1435 * accesses from user process context. 1436 */ 1437 1438 static inline void sock_owned_by_me(const struct sock *sk) 1439 { 1440 #ifdef CONFIG_LOCKDEP 1441 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1442 #endif 1443 } 1444 1445 static inline bool sock_owned_by_user(const struct sock *sk) 1446 { 1447 sock_owned_by_me(sk); 1448 return sk->sk_lock.owned; 1449 } 1450 1451 /* no reclassification while locks are held */ 1452 static inline bool sock_allow_reclassification(const struct sock *csk) 1453 { 1454 struct sock *sk = (struct sock *)csk; 1455 1456 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1457 } 1458 1459 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1460 struct proto *prot, int kern); 1461 void sk_free(struct sock *sk); 1462 void sk_destruct(struct sock *sk); 1463 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1464 1465 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1466 gfp_t priority); 1467 void __sock_wfree(struct sk_buff *skb); 1468 void sock_wfree(struct sk_buff *skb); 1469 void skb_orphan_partial(struct sk_buff *skb); 1470 void sock_rfree(struct sk_buff *skb); 1471 void sock_efree(struct sk_buff *skb); 1472 #ifdef CONFIG_INET 1473 void sock_edemux(struct sk_buff *skb); 1474 #else 1475 #define sock_edemux(skb) sock_efree(skb) 1476 #endif 1477 1478 int sock_setsockopt(struct socket *sock, int level, int op, 1479 char __user *optval, unsigned int optlen); 1480 1481 int sock_getsockopt(struct socket *sock, int level, int op, 1482 char __user *optval, int __user *optlen); 1483 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1484 int noblock, int *errcode); 1485 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1486 unsigned long data_len, int noblock, 1487 int *errcode, int max_page_order); 1488 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1489 void sock_kfree_s(struct sock *sk, void *mem, int size); 1490 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1491 void sk_send_sigurg(struct sock *sk); 1492 1493 struct sockcm_cookie { 1494 u32 mark; 1495 u16 tsflags; 1496 }; 1497 1498 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1499 struct sockcm_cookie *sockc); 1500 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1501 struct sockcm_cookie *sockc); 1502 1503 /* 1504 * Functions to fill in entries in struct proto_ops when a protocol 1505 * does not implement a particular function. 1506 */ 1507 int sock_no_bind(struct socket *, struct sockaddr *, int); 1508 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1509 int sock_no_socketpair(struct socket *, struct socket *); 1510 int sock_no_accept(struct socket *, struct socket *, int); 1511 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1512 unsigned int sock_no_poll(struct file *, struct socket *, 1513 struct poll_table_struct *); 1514 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1515 int sock_no_listen(struct socket *, int); 1516 int sock_no_shutdown(struct socket *, int); 1517 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1518 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1519 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1520 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1521 int sock_no_mmap(struct file *file, struct socket *sock, 1522 struct vm_area_struct *vma); 1523 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1524 size_t size, int flags); 1525 1526 /* 1527 * Functions to fill in entries in struct proto_ops when a protocol 1528 * uses the inet style. 1529 */ 1530 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1531 char __user *optval, int __user *optlen); 1532 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1533 int flags); 1534 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1535 char __user *optval, unsigned int optlen); 1536 int compat_sock_common_getsockopt(struct socket *sock, int level, 1537 int optname, char __user *optval, int __user *optlen); 1538 int compat_sock_common_setsockopt(struct socket *sock, int level, 1539 int optname, char __user *optval, unsigned int optlen); 1540 1541 void sk_common_release(struct sock *sk); 1542 1543 /* 1544 * Default socket callbacks and setup code 1545 */ 1546 1547 /* Initialise core socket variables */ 1548 void sock_init_data(struct socket *sock, struct sock *sk); 1549 1550 /* 1551 * Socket reference counting postulates. 1552 * 1553 * * Each user of socket SHOULD hold a reference count. 1554 * * Each access point to socket (an hash table bucket, reference from a list, 1555 * running timer, skb in flight MUST hold a reference count. 1556 * * When reference count hits 0, it means it will never increase back. 1557 * * When reference count hits 0, it means that no references from 1558 * outside exist to this socket and current process on current CPU 1559 * is last user and may/should destroy this socket. 1560 * * sk_free is called from any context: process, BH, IRQ. When 1561 * it is called, socket has no references from outside -> sk_free 1562 * may release descendant resources allocated by the socket, but 1563 * to the time when it is called, socket is NOT referenced by any 1564 * hash tables, lists etc. 1565 * * Packets, delivered from outside (from network or from another process) 1566 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1567 * when they sit in queue. Otherwise, packets will leak to hole, when 1568 * socket is looked up by one cpu and unhasing is made by another CPU. 1569 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1570 * (leak to backlog). Packet socket does all the processing inside 1571 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1572 * use separate SMP lock, so that they are prone too. 1573 */ 1574 1575 /* Ungrab socket and destroy it, if it was the last reference. */ 1576 static inline void sock_put(struct sock *sk) 1577 { 1578 if (atomic_dec_and_test(&sk->sk_refcnt)) 1579 sk_free(sk); 1580 } 1581 /* Generic version of sock_put(), dealing with all sockets 1582 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1583 */ 1584 void sock_gen_put(struct sock *sk); 1585 1586 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1587 unsigned int trim_cap); 1588 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1589 const int nested) 1590 { 1591 return __sk_receive_skb(sk, skb, nested, 1); 1592 } 1593 1594 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1595 { 1596 sk->sk_tx_queue_mapping = tx_queue; 1597 } 1598 1599 static inline void sk_tx_queue_clear(struct sock *sk) 1600 { 1601 sk->sk_tx_queue_mapping = -1; 1602 } 1603 1604 static inline int sk_tx_queue_get(const struct sock *sk) 1605 { 1606 return sk ? sk->sk_tx_queue_mapping : -1; 1607 } 1608 1609 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1610 { 1611 sk_tx_queue_clear(sk); 1612 sk->sk_socket = sock; 1613 } 1614 1615 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1616 { 1617 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1618 return &rcu_dereference_raw(sk->sk_wq)->wait; 1619 } 1620 /* Detach socket from process context. 1621 * Announce socket dead, detach it from wait queue and inode. 1622 * Note that parent inode held reference count on this struct sock, 1623 * we do not release it in this function, because protocol 1624 * probably wants some additional cleanups or even continuing 1625 * to work with this socket (TCP). 1626 */ 1627 static inline void sock_orphan(struct sock *sk) 1628 { 1629 write_lock_bh(&sk->sk_callback_lock); 1630 sock_set_flag(sk, SOCK_DEAD); 1631 sk_set_socket(sk, NULL); 1632 sk->sk_wq = NULL; 1633 write_unlock_bh(&sk->sk_callback_lock); 1634 } 1635 1636 static inline void sock_graft(struct sock *sk, struct socket *parent) 1637 { 1638 write_lock_bh(&sk->sk_callback_lock); 1639 sk->sk_wq = parent->wq; 1640 parent->sk = sk; 1641 sk_set_socket(sk, parent); 1642 security_sock_graft(sk, parent); 1643 write_unlock_bh(&sk->sk_callback_lock); 1644 } 1645 1646 kuid_t sock_i_uid(struct sock *sk); 1647 unsigned long sock_i_ino(struct sock *sk); 1648 1649 static inline u32 net_tx_rndhash(void) 1650 { 1651 u32 v = prandom_u32(); 1652 1653 return v ?: 1; 1654 } 1655 1656 static inline void sk_set_txhash(struct sock *sk) 1657 { 1658 sk->sk_txhash = net_tx_rndhash(); 1659 } 1660 1661 static inline void sk_rethink_txhash(struct sock *sk) 1662 { 1663 if (sk->sk_txhash) 1664 sk_set_txhash(sk); 1665 } 1666 1667 static inline struct dst_entry * 1668 __sk_dst_get(struct sock *sk) 1669 { 1670 return rcu_dereference_check(sk->sk_dst_cache, 1671 lockdep_sock_is_held(sk)); 1672 } 1673 1674 static inline struct dst_entry * 1675 sk_dst_get(struct sock *sk) 1676 { 1677 struct dst_entry *dst; 1678 1679 rcu_read_lock(); 1680 dst = rcu_dereference(sk->sk_dst_cache); 1681 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1682 dst = NULL; 1683 rcu_read_unlock(); 1684 return dst; 1685 } 1686 1687 static inline void dst_negative_advice(struct sock *sk) 1688 { 1689 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1690 1691 sk_rethink_txhash(sk); 1692 1693 if (dst && dst->ops->negative_advice) { 1694 ndst = dst->ops->negative_advice(dst); 1695 1696 if (ndst != dst) { 1697 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1698 sk_tx_queue_clear(sk); 1699 } 1700 } 1701 } 1702 1703 static inline void 1704 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1705 { 1706 struct dst_entry *old_dst; 1707 1708 sk_tx_queue_clear(sk); 1709 /* 1710 * This can be called while sk is owned by the caller only, 1711 * with no state that can be checked in a rcu_dereference_check() cond 1712 */ 1713 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1714 rcu_assign_pointer(sk->sk_dst_cache, dst); 1715 dst_release(old_dst); 1716 } 1717 1718 static inline void 1719 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1720 { 1721 struct dst_entry *old_dst; 1722 1723 sk_tx_queue_clear(sk); 1724 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1725 dst_release(old_dst); 1726 } 1727 1728 static inline void 1729 __sk_dst_reset(struct sock *sk) 1730 { 1731 __sk_dst_set(sk, NULL); 1732 } 1733 1734 static inline void 1735 sk_dst_reset(struct sock *sk) 1736 { 1737 sk_dst_set(sk, NULL); 1738 } 1739 1740 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1741 1742 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1743 1744 bool sk_mc_loop(struct sock *sk); 1745 1746 static inline bool sk_can_gso(const struct sock *sk) 1747 { 1748 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1749 } 1750 1751 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1752 1753 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1754 { 1755 sk->sk_route_nocaps |= flags; 1756 sk->sk_route_caps &= ~flags; 1757 } 1758 1759 static inline bool sk_check_csum_caps(struct sock *sk) 1760 { 1761 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1762 (sk->sk_family == PF_INET && 1763 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1764 (sk->sk_family == PF_INET6 && 1765 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1766 } 1767 1768 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1769 struct iov_iter *from, char *to, 1770 int copy, int offset) 1771 { 1772 if (skb->ip_summed == CHECKSUM_NONE) { 1773 __wsum csum = 0; 1774 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy) 1775 return -EFAULT; 1776 skb->csum = csum_block_add(skb->csum, csum, offset); 1777 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1778 if (copy_from_iter_nocache(to, copy, from) != copy) 1779 return -EFAULT; 1780 } else if (copy_from_iter(to, copy, from) != copy) 1781 return -EFAULT; 1782 1783 return 0; 1784 } 1785 1786 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1787 struct iov_iter *from, int copy) 1788 { 1789 int err, offset = skb->len; 1790 1791 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1792 copy, offset); 1793 if (err) 1794 __skb_trim(skb, offset); 1795 1796 return err; 1797 } 1798 1799 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1800 struct sk_buff *skb, 1801 struct page *page, 1802 int off, int copy) 1803 { 1804 int err; 1805 1806 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1807 copy, skb->len); 1808 if (err) 1809 return err; 1810 1811 skb->len += copy; 1812 skb->data_len += copy; 1813 skb->truesize += copy; 1814 sk->sk_wmem_queued += copy; 1815 sk_mem_charge(sk, copy); 1816 return 0; 1817 } 1818 1819 /** 1820 * sk_wmem_alloc_get - returns write allocations 1821 * @sk: socket 1822 * 1823 * Returns sk_wmem_alloc minus initial offset of one 1824 */ 1825 static inline int sk_wmem_alloc_get(const struct sock *sk) 1826 { 1827 return atomic_read(&sk->sk_wmem_alloc) - 1; 1828 } 1829 1830 /** 1831 * sk_rmem_alloc_get - returns read allocations 1832 * @sk: socket 1833 * 1834 * Returns sk_rmem_alloc 1835 */ 1836 static inline int sk_rmem_alloc_get(const struct sock *sk) 1837 { 1838 return atomic_read(&sk->sk_rmem_alloc); 1839 } 1840 1841 /** 1842 * sk_has_allocations - check if allocations are outstanding 1843 * @sk: socket 1844 * 1845 * Returns true if socket has write or read allocations 1846 */ 1847 static inline bool sk_has_allocations(const struct sock *sk) 1848 { 1849 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1850 } 1851 1852 /** 1853 * skwq_has_sleeper - check if there are any waiting processes 1854 * @wq: struct socket_wq 1855 * 1856 * Returns true if socket_wq has waiting processes 1857 * 1858 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1859 * barrier call. They were added due to the race found within the tcp code. 1860 * 1861 * Consider following tcp code paths: 1862 * 1863 * CPU1 CPU2 1864 * 1865 * sys_select receive packet 1866 * ... ... 1867 * __add_wait_queue update tp->rcv_nxt 1868 * ... ... 1869 * tp->rcv_nxt check sock_def_readable 1870 * ... { 1871 * schedule rcu_read_lock(); 1872 * wq = rcu_dereference(sk->sk_wq); 1873 * if (wq && waitqueue_active(&wq->wait)) 1874 * wake_up_interruptible(&wq->wait) 1875 * ... 1876 * } 1877 * 1878 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1879 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1880 * could then endup calling schedule and sleep forever if there are no more 1881 * data on the socket. 1882 * 1883 */ 1884 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1885 { 1886 return wq && wq_has_sleeper(&wq->wait); 1887 } 1888 1889 /** 1890 * sock_poll_wait - place memory barrier behind the poll_wait call. 1891 * @filp: file 1892 * @wait_address: socket wait queue 1893 * @p: poll_table 1894 * 1895 * See the comments in the wq_has_sleeper function. 1896 */ 1897 static inline void sock_poll_wait(struct file *filp, 1898 wait_queue_head_t *wait_address, poll_table *p) 1899 { 1900 if (!poll_does_not_wait(p) && wait_address) { 1901 poll_wait(filp, wait_address, p); 1902 /* We need to be sure we are in sync with the 1903 * socket flags modification. 1904 * 1905 * This memory barrier is paired in the wq_has_sleeper. 1906 */ 1907 smp_mb(); 1908 } 1909 } 1910 1911 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1912 { 1913 if (sk->sk_txhash) { 1914 skb->l4_hash = 1; 1915 skb->hash = sk->sk_txhash; 1916 } 1917 } 1918 1919 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 1920 1921 /* 1922 * Queue a received datagram if it will fit. Stream and sequenced 1923 * protocols can't normally use this as they need to fit buffers in 1924 * and play with them. 1925 * 1926 * Inlined as it's very short and called for pretty much every 1927 * packet ever received. 1928 */ 1929 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1930 { 1931 skb_orphan(skb); 1932 skb->sk = sk; 1933 skb->destructor = sock_rfree; 1934 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1935 sk_mem_charge(sk, skb->truesize); 1936 } 1937 1938 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1939 unsigned long expires); 1940 1941 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1942 1943 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1944 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1945 1946 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1947 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 1948 1949 /* 1950 * Recover an error report and clear atomically 1951 */ 1952 1953 static inline int sock_error(struct sock *sk) 1954 { 1955 int err; 1956 if (likely(!sk->sk_err)) 1957 return 0; 1958 err = xchg(&sk->sk_err, 0); 1959 return -err; 1960 } 1961 1962 static inline unsigned long sock_wspace(struct sock *sk) 1963 { 1964 int amt = 0; 1965 1966 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1967 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 1968 if (amt < 0) 1969 amt = 0; 1970 } 1971 return amt; 1972 } 1973 1974 /* Note: 1975 * We use sk->sk_wq_raw, from contexts knowing this 1976 * pointer is not NULL and cannot disappear/change. 1977 */ 1978 static inline void sk_set_bit(int nr, struct sock *sk) 1979 { 1980 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 1981 !sock_flag(sk, SOCK_FASYNC)) 1982 return; 1983 1984 set_bit(nr, &sk->sk_wq_raw->flags); 1985 } 1986 1987 static inline void sk_clear_bit(int nr, struct sock *sk) 1988 { 1989 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 1990 !sock_flag(sk, SOCK_FASYNC)) 1991 return; 1992 1993 clear_bit(nr, &sk->sk_wq_raw->flags); 1994 } 1995 1996 static inline void sk_wake_async(const struct sock *sk, int how, int band) 1997 { 1998 if (sock_flag(sk, SOCK_FASYNC)) { 1999 rcu_read_lock(); 2000 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2001 rcu_read_unlock(); 2002 } 2003 } 2004 2005 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2006 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2007 * Note: for send buffers, TCP works better if we can build two skbs at 2008 * minimum. 2009 */ 2010 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2011 2012 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2013 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2014 2015 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2016 { 2017 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2018 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2019 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2020 } 2021 } 2022 2023 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2024 bool force_schedule); 2025 2026 /** 2027 * sk_page_frag - return an appropriate page_frag 2028 * @sk: socket 2029 * 2030 * If socket allocation mode allows current thread to sleep, it means its 2031 * safe to use the per task page_frag instead of the per socket one. 2032 */ 2033 static inline struct page_frag *sk_page_frag(struct sock *sk) 2034 { 2035 if (gfpflags_allow_blocking(sk->sk_allocation)) 2036 return ¤t->task_frag; 2037 2038 return &sk->sk_frag; 2039 } 2040 2041 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2042 2043 /* 2044 * Default write policy as shown to user space via poll/select/SIGIO 2045 */ 2046 static inline bool sock_writeable(const struct sock *sk) 2047 { 2048 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2049 } 2050 2051 static inline gfp_t gfp_any(void) 2052 { 2053 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2054 } 2055 2056 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2057 { 2058 return noblock ? 0 : sk->sk_rcvtimeo; 2059 } 2060 2061 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2062 { 2063 return noblock ? 0 : sk->sk_sndtimeo; 2064 } 2065 2066 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2067 { 2068 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2069 } 2070 2071 /* Alas, with timeout socket operations are not restartable. 2072 * Compare this to poll(). 2073 */ 2074 static inline int sock_intr_errno(long timeo) 2075 { 2076 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2077 } 2078 2079 struct sock_skb_cb { 2080 u32 dropcount; 2081 }; 2082 2083 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2084 * using skb->cb[] would keep using it directly and utilize its 2085 * alignement guarantee. 2086 */ 2087 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2088 sizeof(struct sock_skb_cb))) 2089 2090 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2091 SOCK_SKB_CB_OFFSET)) 2092 2093 #define sock_skb_cb_check_size(size) \ 2094 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2095 2096 static inline void 2097 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2098 { 2099 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops); 2100 } 2101 2102 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2103 { 2104 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2105 2106 atomic_add(segs, &sk->sk_drops); 2107 } 2108 2109 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2110 struct sk_buff *skb); 2111 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2112 struct sk_buff *skb); 2113 2114 static inline void 2115 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2116 { 2117 ktime_t kt = skb->tstamp; 2118 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2119 2120 /* 2121 * generate control messages if 2122 * - receive time stamping in software requested 2123 * - software time stamp available and wanted 2124 * - hardware time stamps available and wanted 2125 */ 2126 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2127 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2128 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2129 (hwtstamps->hwtstamp.tv64 && 2130 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2131 __sock_recv_timestamp(msg, sk, skb); 2132 else 2133 sk->sk_stamp = kt; 2134 2135 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2136 __sock_recv_wifi_status(msg, sk, skb); 2137 } 2138 2139 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2140 struct sk_buff *skb); 2141 2142 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2143 struct sk_buff *skb) 2144 { 2145 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2146 (1UL << SOCK_RCVTSTAMP)) 2147 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2148 SOF_TIMESTAMPING_RAW_HARDWARE) 2149 2150 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2151 __sock_recv_ts_and_drops(msg, sk, skb); 2152 else 2153 sk->sk_stamp = skb->tstamp; 2154 } 2155 2156 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2157 2158 /** 2159 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2160 * @sk: socket sending this packet 2161 * @tsflags: timestamping flags to use 2162 * @tx_flags: completed with instructions for time stamping 2163 * 2164 * Note : callers should take care of initial *tx_flags value (usually 0) 2165 */ 2166 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2167 __u8 *tx_flags) 2168 { 2169 if (unlikely(tsflags)) 2170 __sock_tx_timestamp(tsflags, tx_flags); 2171 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2172 *tx_flags |= SKBTX_WIFI_STATUS; 2173 } 2174 2175 /** 2176 * sk_eat_skb - Release a skb if it is no longer needed 2177 * @sk: socket to eat this skb from 2178 * @skb: socket buffer to eat 2179 * 2180 * This routine must be called with interrupts disabled or with the socket 2181 * locked so that the sk_buff queue operation is ok. 2182 */ 2183 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2184 { 2185 __skb_unlink(skb, &sk->sk_receive_queue); 2186 __kfree_skb(skb); 2187 } 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 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2202 { 2203 if (skb->sk) { 2204 struct sock *sk = skb->sk; 2205 2206 skb->destructor = NULL; 2207 skb->sk = NULL; 2208 return sk; 2209 } 2210 return NULL; 2211 } 2212 2213 /* This helper checks if a socket is a full socket, 2214 * ie _not_ a timewait or request socket. 2215 */ 2216 static inline bool sk_fullsock(const struct sock *sk) 2217 { 2218 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2219 } 2220 2221 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2222 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2223 */ 2224 static inline bool sk_listener(const struct sock *sk) 2225 { 2226 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2227 } 2228 2229 /** 2230 * sk_state_load - read sk->sk_state for lockless contexts 2231 * @sk: socket pointer 2232 * 2233 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2234 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2235 */ 2236 static inline int sk_state_load(const struct sock *sk) 2237 { 2238 return smp_load_acquire(&sk->sk_state); 2239 } 2240 2241 /** 2242 * sk_state_store - update sk->sk_state 2243 * @sk: socket pointer 2244 * @newstate: new state 2245 * 2246 * Paired with sk_state_load(). Should be used in contexts where 2247 * state change might impact lockless readers. 2248 */ 2249 static inline void sk_state_store(struct sock *sk, int newstate) 2250 { 2251 smp_store_release(&sk->sk_state, newstate); 2252 } 2253 2254 void sock_enable_timestamp(struct sock *sk, int flag); 2255 int sock_get_timestamp(struct sock *, struct timeval __user *); 2256 int sock_get_timestampns(struct sock *, struct timespec __user *); 2257 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2258 int type); 2259 2260 bool sk_ns_capable(const struct sock *sk, 2261 struct user_namespace *user_ns, int cap); 2262 bool sk_capable(const struct sock *sk, int cap); 2263 bool sk_net_capable(const struct sock *sk, int cap); 2264 2265 extern __u32 sysctl_wmem_max; 2266 extern __u32 sysctl_rmem_max; 2267 2268 extern int sysctl_tstamp_allow_data; 2269 extern int sysctl_optmem_max; 2270 2271 extern __u32 sysctl_wmem_default; 2272 extern __u32 sysctl_rmem_default; 2273 2274 #endif /* _SOCK_H */ 2275