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 #include <net/smc.h> 74 75 /* 76 * This structure really needs to be cleaned up. 77 * Most of it is for TCP, and not used by any of 78 * the other protocols. 79 */ 80 81 /* Define this to get the SOCK_DBG debugging facility. */ 82 #define SOCK_DEBUGGING 83 #ifdef SOCK_DEBUGGING 84 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 85 printk(KERN_DEBUG msg); } while (0) 86 #else 87 /* Validate arguments and do nothing */ 88 static inline __printf(2, 3) 89 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 90 { 91 } 92 #endif 93 94 /* This is the per-socket lock. The spinlock provides a synchronization 95 * between user contexts and software interrupt processing, whereas the 96 * mini-semaphore synchronizes multiple users amongst themselves. 97 */ 98 typedef struct { 99 spinlock_t slock; 100 int owned; 101 wait_queue_head_t wq; 102 /* 103 * We express the mutex-alike socket_lock semantics 104 * to the lock validator by explicitly managing 105 * the slock as a lock variant (in addition to 106 * the slock itself): 107 */ 108 #ifdef CONFIG_DEBUG_LOCK_ALLOC 109 struct lockdep_map dep_map; 110 #endif 111 } socket_lock_t; 112 113 struct sock; 114 struct proto; 115 struct net; 116 117 typedef __u32 __bitwise __portpair; 118 typedef __u64 __bitwise __addrpair; 119 120 /** 121 * struct sock_common - minimal network layer representation of sockets 122 * @skc_daddr: Foreign IPv4 addr 123 * @skc_rcv_saddr: Bound local IPv4 addr 124 * @skc_hash: hash value used with various protocol lookup tables 125 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 126 * @skc_dport: placeholder for inet_dport/tw_dport 127 * @skc_num: placeholder for inet_num/tw_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_bound_dev_if: bound device index if != 0 133 * @skc_bind_node: bind hash linkage for various protocol lookup tables 134 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 135 * @skc_prot: protocol handlers inside a network family 136 * @skc_net: reference to the network namespace of this socket 137 * @skc_node: main hash linkage for various protocol lookup tables 138 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 139 * @skc_tx_queue_mapping: tx queue number for this connection 140 * @skc_flags: place holder for sk_flags 141 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 142 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 143 * @skc_incoming_cpu: record/match cpu processing incoming packets 144 * @skc_refcnt: reference count 145 * 146 * This is the minimal network layer representation of sockets, the header 147 * for struct sock and struct inet_timewait_sock. 148 */ 149 struct sock_common { 150 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 151 * address on 64bit arches : cf INET_MATCH() 152 */ 153 union { 154 __addrpair skc_addrpair; 155 struct { 156 __be32 skc_daddr; 157 __be32 skc_rcv_saddr; 158 }; 159 }; 160 union { 161 unsigned int skc_hash; 162 __u16 skc_u16hashes[2]; 163 }; 164 /* skc_dport && skc_num must be grouped as well */ 165 union { 166 __portpair skc_portpair; 167 struct { 168 __be16 skc_dport; 169 __u16 skc_num; 170 }; 171 }; 172 173 unsigned short skc_family; 174 volatile unsigned char skc_state; 175 unsigned char skc_reuse:4; 176 unsigned char skc_reuseport:1; 177 unsigned char skc_ipv6only:1; 178 unsigned char skc_net_refcnt:1; 179 int skc_bound_dev_if; 180 union { 181 struct hlist_node skc_bind_node; 182 struct hlist_node skc_portaddr_node; 183 }; 184 struct proto *skc_prot; 185 possible_net_t skc_net; 186 187 #if IS_ENABLED(CONFIG_IPV6) 188 struct in6_addr skc_v6_daddr; 189 struct in6_addr skc_v6_rcv_saddr; 190 #endif 191 192 atomic64_t skc_cookie; 193 194 /* following fields are padding to force 195 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 196 * assuming IPV6 is enabled. We use this padding differently 197 * for different kind of 'sockets' 198 */ 199 union { 200 unsigned long skc_flags; 201 struct sock *skc_listener; /* request_sock */ 202 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 203 }; 204 /* 205 * fields between dontcopy_begin/dontcopy_end 206 * are not copied in sock_copy() 207 */ 208 /* private: */ 209 int skc_dontcopy_begin[0]; 210 /* public: */ 211 union { 212 struct hlist_node skc_node; 213 struct hlist_nulls_node skc_nulls_node; 214 }; 215 int skc_tx_queue_mapping; 216 union { 217 int skc_incoming_cpu; 218 u32 skc_rcv_wnd; 219 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 220 }; 221 222 atomic_t skc_refcnt; 223 /* private: */ 224 int skc_dontcopy_end[0]; 225 union { 226 u32 skc_rxhash; 227 u32 skc_window_clamp; 228 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 229 }; 230 /* public: */ 231 }; 232 233 /** 234 * struct sock - network layer representation of sockets 235 * @__sk_common: shared layout with inet_timewait_sock 236 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 237 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 238 * @sk_lock: synchronizer 239 * @sk_rcvbuf: size of receive buffer in bytes 240 * @sk_wq: sock wait queue and async head 241 * @sk_rx_dst: receive input route used by early demux 242 * @sk_dst_cache: destination cache 243 * @sk_policy: flow policy 244 * @sk_receive_queue: incoming packets 245 * @sk_wmem_alloc: transmit queue bytes committed 246 * @sk_write_queue: Packet sending queue 247 * @sk_omem_alloc: "o" is "option" or "other" 248 * @sk_wmem_queued: persistent queue size 249 * @sk_forward_alloc: space allocated forward 250 * @sk_napi_id: id of the last napi context to receive data for sk 251 * @sk_ll_usec: usecs to busypoll when there is no data 252 * @sk_allocation: allocation mode 253 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 254 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 255 * @sk_sndbuf: size of send buffer in bytes 256 * @sk_padding: unused element for alignment 257 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 258 * @sk_no_check_rx: allow zero checksum in RX packets 259 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 260 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 261 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 262 * @sk_gso_max_size: Maximum GSO segment size to build 263 * @sk_gso_max_segs: Maximum number of GSO segments 264 * @sk_lingertime: %SO_LINGER l_linger setting 265 * @sk_backlog: always used with the per-socket spinlock held 266 * @sk_callback_lock: used with the callbacks in the end of this struct 267 * @sk_error_queue: rarely used 268 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 269 * IPV6_ADDRFORM for instance) 270 * @sk_err: last error 271 * @sk_err_soft: errors that don't cause failure but are the cause of a 272 * persistent failure not just 'timed out' 273 * @sk_drops: raw/udp drops counter 274 * @sk_ack_backlog: current listen backlog 275 * @sk_max_ack_backlog: listen backlog set in listen() 276 * @sk_priority: %SO_PRIORITY setting 277 * @sk_type: socket type (%SOCK_STREAM, etc) 278 * @sk_protocol: which protocol this socket belongs in this network family 279 * @sk_peer_pid: &struct pid for this socket's peer 280 * @sk_peer_cred: %SO_PEERCRED setting 281 * @sk_rcvlowat: %SO_RCVLOWAT setting 282 * @sk_rcvtimeo: %SO_RCVTIMEO setting 283 * @sk_sndtimeo: %SO_SNDTIMEO setting 284 * @sk_txhash: computed flow hash for use on transmit 285 * @sk_filter: socket filtering instructions 286 * @sk_timer: sock cleanup timer 287 * @sk_stamp: time stamp of last packet received 288 * @sk_tsflags: SO_TIMESTAMPING socket options 289 * @sk_tskey: counter to disambiguate concurrent tstamp requests 290 * @sk_socket: Identd and reporting IO signals 291 * @sk_user_data: RPC layer private data 292 * @sk_frag: cached page frag 293 * @sk_peek_off: current peek_offset value 294 * @sk_send_head: front of stuff to transmit 295 * @sk_security: used by security modules 296 * @sk_mark: generic packet mark 297 * @sk_cgrp_data: cgroup data for this cgroup 298 * @sk_memcg: this socket's memory cgroup association 299 * @sk_write_pending: a write to stream socket waits to start 300 * @sk_state_change: callback to indicate change in the state of the sock 301 * @sk_data_ready: callback to indicate there is data to be processed 302 * @sk_write_space: callback to indicate there is bf sending space available 303 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 304 * @sk_backlog_rcv: callback to process the backlog 305 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 306 * @sk_reuseport_cb: reuseport group container 307 * @sk_rcu: used during RCU grace period 308 */ 309 struct sock { 310 /* 311 * Now struct inet_timewait_sock also uses sock_common, so please just 312 * don't add nothing before this first member (__sk_common) --acme 313 */ 314 struct sock_common __sk_common; 315 #define sk_node __sk_common.skc_node 316 #define sk_nulls_node __sk_common.skc_nulls_node 317 #define sk_refcnt __sk_common.skc_refcnt 318 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 319 320 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 321 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 322 #define sk_hash __sk_common.skc_hash 323 #define sk_portpair __sk_common.skc_portpair 324 #define sk_num __sk_common.skc_num 325 #define sk_dport __sk_common.skc_dport 326 #define sk_addrpair __sk_common.skc_addrpair 327 #define sk_daddr __sk_common.skc_daddr 328 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 329 #define sk_family __sk_common.skc_family 330 #define sk_state __sk_common.skc_state 331 #define sk_reuse __sk_common.skc_reuse 332 #define sk_reuseport __sk_common.skc_reuseport 333 #define sk_ipv6only __sk_common.skc_ipv6only 334 #define sk_net_refcnt __sk_common.skc_net_refcnt 335 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 336 #define sk_bind_node __sk_common.skc_bind_node 337 #define sk_prot __sk_common.skc_prot 338 #define sk_net __sk_common.skc_net 339 #define sk_v6_daddr __sk_common.skc_v6_daddr 340 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 341 #define sk_cookie __sk_common.skc_cookie 342 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 343 #define sk_flags __sk_common.skc_flags 344 #define sk_rxhash __sk_common.skc_rxhash 345 346 socket_lock_t sk_lock; 347 atomic_t sk_drops; 348 int sk_rcvlowat; 349 struct sk_buff_head sk_error_queue; 350 struct sk_buff_head sk_receive_queue; 351 /* 352 * The backlog queue is special, it is always used with 353 * the per-socket spinlock held and requires low latency 354 * access. Therefore we special case it's implementation. 355 * Note : rmem_alloc is in this structure to fill a hole 356 * on 64bit arches, not because its logically part of 357 * backlog. 358 */ 359 struct { 360 atomic_t rmem_alloc; 361 int len; 362 struct sk_buff *head; 363 struct sk_buff *tail; 364 } sk_backlog; 365 #define sk_rmem_alloc sk_backlog.rmem_alloc 366 367 int sk_forward_alloc; 368 #ifdef CONFIG_NET_RX_BUSY_POLL 369 unsigned int sk_ll_usec; 370 /* ===== mostly read cache line ===== */ 371 unsigned int sk_napi_id; 372 #endif 373 int sk_rcvbuf; 374 375 struct sk_filter __rcu *sk_filter; 376 union { 377 struct socket_wq __rcu *sk_wq; 378 struct socket_wq *sk_wq_raw; 379 }; 380 #ifdef CONFIG_XFRM 381 struct xfrm_policy __rcu *sk_policy[2]; 382 #endif 383 struct dst_entry *sk_rx_dst; 384 struct dst_entry __rcu *sk_dst_cache; 385 atomic_t sk_omem_alloc; 386 int sk_sndbuf; 387 388 /* ===== cache line for TX ===== */ 389 int sk_wmem_queued; 390 atomic_t sk_wmem_alloc; 391 unsigned long sk_tsq_flags; 392 struct sk_buff *sk_send_head; 393 struct sk_buff_head sk_write_queue; 394 __s32 sk_peek_off; 395 int sk_write_pending; 396 long sk_sndtimeo; 397 struct timer_list sk_timer; 398 __u32 sk_priority; 399 __u32 sk_mark; 400 u32 sk_pacing_rate; /* bytes per second */ 401 u32 sk_max_pacing_rate; 402 struct page_frag sk_frag; 403 netdev_features_t sk_route_caps; 404 netdev_features_t sk_route_nocaps; 405 int sk_gso_type; 406 unsigned int sk_gso_max_size; 407 gfp_t sk_allocation; 408 __u32 sk_txhash; 409 410 /* 411 * Because of non atomicity rules, all 412 * changes are protected by socket lock. 413 */ 414 unsigned int __sk_flags_offset[0]; 415 #ifdef __BIG_ENDIAN_BITFIELD 416 #define SK_FL_PROTO_SHIFT 16 417 #define SK_FL_PROTO_MASK 0x00ff0000 418 419 #define SK_FL_TYPE_SHIFT 0 420 #define SK_FL_TYPE_MASK 0x0000ffff 421 #else 422 #define SK_FL_PROTO_SHIFT 8 423 #define SK_FL_PROTO_MASK 0x0000ff00 424 425 #define SK_FL_TYPE_SHIFT 16 426 #define SK_FL_TYPE_MASK 0xffff0000 427 #endif 428 429 kmemcheck_bitfield_begin(flags); 430 unsigned int sk_padding : 2, 431 sk_no_check_tx : 1, 432 sk_no_check_rx : 1, 433 sk_userlocks : 4, 434 sk_protocol : 8, 435 sk_type : 16; 436 #define SK_PROTOCOL_MAX U8_MAX 437 kmemcheck_bitfield_end(flags); 438 439 u16 sk_gso_max_segs; 440 unsigned long sk_lingertime; 441 struct proto *sk_prot_creator; 442 rwlock_t sk_callback_lock; 443 int sk_err, 444 sk_err_soft; 445 u32 sk_ack_backlog; 446 u32 sk_max_ack_backlog; 447 kuid_t sk_uid; 448 struct pid *sk_peer_pid; 449 const struct cred *sk_peer_cred; 450 long sk_rcvtimeo; 451 ktime_t sk_stamp; 452 u16 sk_tsflags; 453 u8 sk_shutdown; 454 u32 sk_tskey; 455 struct socket *sk_socket; 456 void *sk_user_data; 457 #ifdef CONFIG_SECURITY 458 void *sk_security; 459 #endif 460 struct sock_cgroup_data sk_cgrp_data; 461 struct mem_cgroup *sk_memcg; 462 void (*sk_state_change)(struct sock *sk); 463 void (*sk_data_ready)(struct sock *sk); 464 void (*sk_write_space)(struct sock *sk); 465 void (*sk_error_report)(struct sock *sk); 466 int (*sk_backlog_rcv)(struct sock *sk, 467 struct sk_buff *skb); 468 void (*sk_destruct)(struct sock *sk); 469 struct sock_reuseport __rcu *sk_reuseport_cb; 470 struct rcu_head sk_rcu; 471 }; 472 473 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 474 475 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 476 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 477 478 /* 479 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 480 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 481 * on a socket means that the socket will reuse everybody else's port 482 * without looking at the other's sk_reuse value. 483 */ 484 485 #define SK_NO_REUSE 0 486 #define SK_CAN_REUSE 1 487 #define SK_FORCE_REUSE 2 488 489 int sk_set_peek_off(struct sock *sk, int val); 490 491 static inline int sk_peek_offset(struct sock *sk, int flags) 492 { 493 if (unlikely(flags & MSG_PEEK)) { 494 s32 off = READ_ONCE(sk->sk_peek_off); 495 if (off >= 0) 496 return off; 497 } 498 499 return 0; 500 } 501 502 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 503 { 504 s32 off = READ_ONCE(sk->sk_peek_off); 505 506 if (unlikely(off >= 0)) { 507 off = max_t(s32, off - val, 0); 508 WRITE_ONCE(sk->sk_peek_off, off); 509 } 510 } 511 512 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 513 { 514 sk_peek_offset_bwd(sk, -val); 515 } 516 517 /* 518 * Hashed lists helper routines 519 */ 520 static inline struct sock *sk_entry(const struct hlist_node *node) 521 { 522 return hlist_entry(node, struct sock, sk_node); 523 } 524 525 static inline struct sock *__sk_head(const struct hlist_head *head) 526 { 527 return hlist_entry(head->first, struct sock, sk_node); 528 } 529 530 static inline struct sock *sk_head(const struct hlist_head *head) 531 { 532 return hlist_empty(head) ? NULL : __sk_head(head); 533 } 534 535 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 536 { 537 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 538 } 539 540 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 541 { 542 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 543 } 544 545 static inline struct sock *sk_next(const struct sock *sk) 546 { 547 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 548 } 549 550 static inline struct sock *sk_nulls_next(const struct sock *sk) 551 { 552 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 553 hlist_nulls_entry(sk->sk_nulls_node.next, 554 struct sock, sk_nulls_node) : 555 NULL; 556 } 557 558 static inline bool sk_unhashed(const struct sock *sk) 559 { 560 return hlist_unhashed(&sk->sk_node); 561 } 562 563 static inline bool sk_hashed(const struct sock *sk) 564 { 565 return !sk_unhashed(sk); 566 } 567 568 static inline void sk_node_init(struct hlist_node *node) 569 { 570 node->pprev = NULL; 571 } 572 573 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 574 { 575 node->pprev = NULL; 576 } 577 578 static inline void __sk_del_node(struct sock *sk) 579 { 580 __hlist_del(&sk->sk_node); 581 } 582 583 /* NB: equivalent to hlist_del_init_rcu */ 584 static inline bool __sk_del_node_init(struct sock *sk) 585 { 586 if (sk_hashed(sk)) { 587 __sk_del_node(sk); 588 sk_node_init(&sk->sk_node); 589 return true; 590 } 591 return false; 592 } 593 594 /* Grab socket reference count. This operation is valid only 595 when sk is ALREADY grabbed f.e. it is found in hash table 596 or a list and the lookup is made under lock preventing hash table 597 modifications. 598 */ 599 600 static __always_inline void sock_hold(struct sock *sk) 601 { 602 atomic_inc(&sk->sk_refcnt); 603 } 604 605 /* Ungrab socket in the context, which assumes that socket refcnt 606 cannot hit zero, f.e. it is true in context of any socketcall. 607 */ 608 static __always_inline void __sock_put(struct sock *sk) 609 { 610 atomic_dec(&sk->sk_refcnt); 611 } 612 613 static inline bool sk_del_node_init(struct sock *sk) 614 { 615 bool rc = __sk_del_node_init(sk); 616 617 if (rc) { 618 /* paranoid for a while -acme */ 619 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 620 __sock_put(sk); 621 } 622 return rc; 623 } 624 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 625 626 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 627 { 628 if (sk_hashed(sk)) { 629 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 630 return true; 631 } 632 return false; 633 } 634 635 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 636 { 637 bool rc = __sk_nulls_del_node_init_rcu(sk); 638 639 if (rc) { 640 /* paranoid for a while -acme */ 641 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 642 __sock_put(sk); 643 } 644 return rc; 645 } 646 647 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 648 { 649 hlist_add_head(&sk->sk_node, list); 650 } 651 652 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 653 { 654 sock_hold(sk); 655 __sk_add_node(sk, list); 656 } 657 658 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 659 { 660 sock_hold(sk); 661 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 662 sk->sk_family == AF_INET6) 663 hlist_add_tail_rcu(&sk->sk_node, list); 664 else 665 hlist_add_head_rcu(&sk->sk_node, list); 666 } 667 668 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 669 { 670 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 671 sk->sk_family == AF_INET6) 672 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 673 else 674 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 675 } 676 677 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 678 { 679 sock_hold(sk); 680 __sk_nulls_add_node_rcu(sk, list); 681 } 682 683 static inline void __sk_del_bind_node(struct sock *sk) 684 { 685 __hlist_del(&sk->sk_bind_node); 686 } 687 688 static inline void sk_add_bind_node(struct sock *sk, 689 struct hlist_head *list) 690 { 691 hlist_add_head(&sk->sk_bind_node, list); 692 } 693 694 #define sk_for_each(__sk, list) \ 695 hlist_for_each_entry(__sk, list, sk_node) 696 #define sk_for_each_rcu(__sk, list) \ 697 hlist_for_each_entry_rcu(__sk, list, sk_node) 698 #define sk_nulls_for_each(__sk, node, list) \ 699 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 700 #define sk_nulls_for_each_rcu(__sk, node, list) \ 701 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 702 #define sk_for_each_from(__sk) \ 703 hlist_for_each_entry_from(__sk, sk_node) 704 #define sk_nulls_for_each_from(__sk, node) \ 705 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 706 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 707 #define sk_for_each_safe(__sk, tmp, list) \ 708 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 709 #define sk_for_each_bound(__sk, list) \ 710 hlist_for_each_entry(__sk, list, sk_bind_node) 711 712 /** 713 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 714 * @tpos: the type * to use as a loop cursor. 715 * @pos: the &struct hlist_node to use as a loop cursor. 716 * @head: the head for your list. 717 * @offset: offset of hlist_node within the struct. 718 * 719 */ 720 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 721 for (pos = rcu_dereference((head)->first); \ 722 pos != NULL && \ 723 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 724 pos = rcu_dereference(pos->next)) 725 726 static inline struct user_namespace *sk_user_ns(struct sock *sk) 727 { 728 /* Careful only use this in a context where these parameters 729 * can not change and must all be valid, such as recvmsg from 730 * userspace. 731 */ 732 return sk->sk_socket->file->f_cred->user_ns; 733 } 734 735 /* Sock flags */ 736 enum sock_flags { 737 SOCK_DEAD, 738 SOCK_DONE, 739 SOCK_URGINLINE, 740 SOCK_KEEPOPEN, 741 SOCK_LINGER, 742 SOCK_DESTROY, 743 SOCK_BROADCAST, 744 SOCK_TIMESTAMP, 745 SOCK_ZAPPED, 746 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 747 SOCK_DBG, /* %SO_DEBUG setting */ 748 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 749 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 750 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 751 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 752 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 753 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 754 SOCK_FASYNC, /* fasync() active */ 755 SOCK_RXQ_OVFL, 756 SOCK_ZEROCOPY, /* buffers from userspace */ 757 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 758 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 759 * Will use last 4 bytes of packet sent from 760 * user-space instead. 761 */ 762 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 763 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 764 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 765 }; 766 767 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 768 769 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 770 { 771 nsk->sk_flags = osk->sk_flags; 772 } 773 774 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 775 { 776 __set_bit(flag, &sk->sk_flags); 777 } 778 779 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 780 { 781 __clear_bit(flag, &sk->sk_flags); 782 } 783 784 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 785 { 786 return test_bit(flag, &sk->sk_flags); 787 } 788 789 #ifdef CONFIG_NET 790 extern struct static_key memalloc_socks; 791 static inline int sk_memalloc_socks(void) 792 { 793 return static_key_false(&memalloc_socks); 794 } 795 #else 796 797 static inline int sk_memalloc_socks(void) 798 { 799 return 0; 800 } 801 802 #endif 803 804 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 805 { 806 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 807 } 808 809 static inline void sk_acceptq_removed(struct sock *sk) 810 { 811 sk->sk_ack_backlog--; 812 } 813 814 static inline void sk_acceptq_added(struct sock *sk) 815 { 816 sk->sk_ack_backlog++; 817 } 818 819 static inline bool sk_acceptq_is_full(const struct sock *sk) 820 { 821 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 822 } 823 824 /* 825 * Compute minimal free write space needed to queue new packets. 826 */ 827 static inline int sk_stream_min_wspace(const struct sock *sk) 828 { 829 return sk->sk_wmem_queued >> 1; 830 } 831 832 static inline int sk_stream_wspace(const struct sock *sk) 833 { 834 return sk->sk_sndbuf - sk->sk_wmem_queued; 835 } 836 837 void sk_stream_write_space(struct sock *sk); 838 839 /* OOB backlog add */ 840 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 841 { 842 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 843 skb_dst_force_safe(skb); 844 845 if (!sk->sk_backlog.tail) 846 sk->sk_backlog.head = skb; 847 else 848 sk->sk_backlog.tail->next = skb; 849 850 sk->sk_backlog.tail = skb; 851 skb->next = NULL; 852 } 853 854 /* 855 * Take into account size of receive queue and backlog queue 856 * Do not take into account this skb truesize, 857 * to allow even a single big packet to come. 858 */ 859 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 860 { 861 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 862 863 return qsize > limit; 864 } 865 866 /* The per-socket spinlock must be held here. */ 867 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 868 unsigned int limit) 869 { 870 if (sk_rcvqueues_full(sk, limit)) 871 return -ENOBUFS; 872 873 /* 874 * If the skb was allocated from pfmemalloc reserves, only 875 * allow SOCK_MEMALLOC sockets to use it as this socket is 876 * helping free memory 877 */ 878 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 879 return -ENOMEM; 880 881 __sk_add_backlog(sk, skb); 882 sk->sk_backlog.len += skb->truesize; 883 return 0; 884 } 885 886 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 887 888 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 889 { 890 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 891 return __sk_backlog_rcv(sk, skb); 892 893 return sk->sk_backlog_rcv(sk, skb); 894 } 895 896 static inline void sk_incoming_cpu_update(struct sock *sk) 897 { 898 sk->sk_incoming_cpu = raw_smp_processor_id(); 899 } 900 901 static inline void sock_rps_record_flow_hash(__u32 hash) 902 { 903 #ifdef CONFIG_RPS 904 struct rps_sock_flow_table *sock_flow_table; 905 906 rcu_read_lock(); 907 sock_flow_table = rcu_dereference(rps_sock_flow_table); 908 rps_record_sock_flow(sock_flow_table, hash); 909 rcu_read_unlock(); 910 #endif 911 } 912 913 static inline void sock_rps_record_flow(const struct sock *sk) 914 { 915 #ifdef CONFIG_RPS 916 if (static_key_false(&rfs_needed)) { 917 /* Reading sk->sk_rxhash might incur an expensive cache line 918 * miss. 919 * 920 * TCP_ESTABLISHED does cover almost all states where RFS 921 * might be useful, and is cheaper [1] than testing : 922 * IPv4: inet_sk(sk)->inet_daddr 923 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 924 * OR an additional socket flag 925 * [1] : sk_state and sk_prot are in the same cache line. 926 */ 927 if (sk->sk_state == TCP_ESTABLISHED) 928 sock_rps_record_flow_hash(sk->sk_rxhash); 929 } 930 #endif 931 } 932 933 static inline void sock_rps_save_rxhash(struct sock *sk, 934 const struct sk_buff *skb) 935 { 936 #ifdef CONFIG_RPS 937 if (unlikely(sk->sk_rxhash != skb->hash)) 938 sk->sk_rxhash = skb->hash; 939 #endif 940 } 941 942 static inline void sock_rps_reset_rxhash(struct sock *sk) 943 { 944 #ifdef CONFIG_RPS 945 sk->sk_rxhash = 0; 946 #endif 947 } 948 949 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 950 ({ int __rc; \ 951 release_sock(__sk); \ 952 __rc = __condition; \ 953 if (!__rc) { \ 954 *(__timeo) = wait_woken(__wait, \ 955 TASK_INTERRUPTIBLE, \ 956 *(__timeo)); \ 957 } \ 958 sched_annotate_sleep(); \ 959 lock_sock(__sk); \ 960 __rc = __condition; \ 961 __rc; \ 962 }) 963 964 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 965 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 966 void sk_stream_wait_close(struct sock *sk, long timeo_p); 967 int sk_stream_error(struct sock *sk, int flags, int err); 968 void sk_stream_kill_queues(struct sock *sk); 969 void sk_set_memalloc(struct sock *sk); 970 void sk_clear_memalloc(struct sock *sk); 971 972 void __sk_flush_backlog(struct sock *sk); 973 974 static inline bool sk_flush_backlog(struct sock *sk) 975 { 976 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 977 __sk_flush_backlog(sk); 978 return true; 979 } 980 return false; 981 } 982 983 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 984 985 struct request_sock_ops; 986 struct timewait_sock_ops; 987 struct inet_hashinfo; 988 struct raw_hashinfo; 989 struct smc_hashinfo; 990 struct module; 991 992 /* 993 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 994 * un-modified. Special care is taken when initializing object to zero. 995 */ 996 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 997 { 998 if (offsetof(struct sock, sk_node.next) != 0) 999 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1000 memset(&sk->sk_node.pprev, 0, 1001 size - offsetof(struct sock, sk_node.pprev)); 1002 } 1003 1004 /* Networking protocol blocks we attach to sockets. 1005 * socket layer -> transport layer interface 1006 */ 1007 struct proto { 1008 void (*close)(struct sock *sk, 1009 long timeout); 1010 int (*connect)(struct sock *sk, 1011 struct sockaddr *uaddr, 1012 int addr_len); 1013 int (*disconnect)(struct sock *sk, int flags); 1014 1015 struct sock * (*accept)(struct sock *sk, int flags, int *err); 1016 1017 int (*ioctl)(struct sock *sk, int cmd, 1018 unsigned long arg); 1019 int (*init)(struct sock *sk); 1020 void (*destroy)(struct sock *sk); 1021 void (*shutdown)(struct sock *sk, int how); 1022 int (*setsockopt)(struct sock *sk, int level, 1023 int optname, char __user *optval, 1024 unsigned int optlen); 1025 int (*getsockopt)(struct sock *sk, int level, 1026 int optname, char __user *optval, 1027 int __user *option); 1028 void (*keepalive)(struct sock *sk, int valbool); 1029 #ifdef CONFIG_COMPAT 1030 int (*compat_setsockopt)(struct sock *sk, 1031 int level, 1032 int optname, char __user *optval, 1033 unsigned int optlen); 1034 int (*compat_getsockopt)(struct sock *sk, 1035 int level, 1036 int optname, char __user *optval, 1037 int __user *option); 1038 int (*compat_ioctl)(struct sock *sk, 1039 unsigned int cmd, unsigned long arg); 1040 #endif 1041 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1042 size_t len); 1043 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1044 size_t len, int noblock, int flags, 1045 int *addr_len); 1046 int (*sendpage)(struct sock *sk, struct page *page, 1047 int offset, size_t size, int flags); 1048 int (*bind)(struct sock *sk, 1049 struct sockaddr *uaddr, int addr_len); 1050 1051 int (*backlog_rcv) (struct sock *sk, 1052 struct sk_buff *skb); 1053 1054 void (*release_cb)(struct sock *sk); 1055 1056 /* Keeping track of sk's, looking them up, and port selection methods. */ 1057 int (*hash)(struct sock *sk); 1058 void (*unhash)(struct sock *sk); 1059 void (*rehash)(struct sock *sk); 1060 int (*get_port)(struct sock *sk, unsigned short snum); 1061 1062 /* Keeping track of sockets in use */ 1063 #ifdef CONFIG_PROC_FS 1064 unsigned int inuse_idx; 1065 #endif 1066 1067 bool (*stream_memory_free)(const struct sock *sk); 1068 /* Memory pressure */ 1069 void (*enter_memory_pressure)(struct sock *sk); 1070 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1071 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1072 /* 1073 * Pressure flag: try to collapse. 1074 * Technical note: it is used by multiple contexts non atomically. 1075 * All the __sk_mem_schedule() is of this nature: accounting 1076 * is strict, actions are advisory and have some latency. 1077 */ 1078 int *memory_pressure; 1079 long *sysctl_mem; 1080 int *sysctl_wmem; 1081 int *sysctl_rmem; 1082 int max_header; 1083 bool no_autobind; 1084 1085 struct kmem_cache *slab; 1086 unsigned int obj_size; 1087 int slab_flags; 1088 1089 struct percpu_counter *orphan_count; 1090 1091 struct request_sock_ops *rsk_prot; 1092 struct timewait_sock_ops *twsk_prot; 1093 1094 union { 1095 struct inet_hashinfo *hashinfo; 1096 struct udp_table *udp_table; 1097 struct raw_hashinfo *raw_hash; 1098 struct smc_hashinfo *smc_hash; 1099 } h; 1100 1101 struct module *owner; 1102 1103 char name[32]; 1104 1105 struct list_head node; 1106 #ifdef SOCK_REFCNT_DEBUG 1107 atomic_t socks; 1108 #endif 1109 int (*diag_destroy)(struct sock *sk, int err); 1110 }; 1111 1112 int proto_register(struct proto *prot, int alloc_slab); 1113 void proto_unregister(struct proto *prot); 1114 1115 #ifdef SOCK_REFCNT_DEBUG 1116 static inline void sk_refcnt_debug_inc(struct sock *sk) 1117 { 1118 atomic_inc(&sk->sk_prot->socks); 1119 } 1120 1121 static inline void sk_refcnt_debug_dec(struct sock *sk) 1122 { 1123 atomic_dec(&sk->sk_prot->socks); 1124 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1125 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1126 } 1127 1128 static inline void sk_refcnt_debug_release(const struct sock *sk) 1129 { 1130 if (atomic_read(&sk->sk_refcnt) != 1) 1131 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1132 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1133 } 1134 #else /* SOCK_REFCNT_DEBUG */ 1135 #define sk_refcnt_debug_inc(sk) do { } while (0) 1136 #define sk_refcnt_debug_dec(sk) do { } while (0) 1137 #define sk_refcnt_debug_release(sk) do { } while (0) 1138 #endif /* SOCK_REFCNT_DEBUG */ 1139 1140 static inline bool sk_stream_memory_free(const struct sock *sk) 1141 { 1142 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1143 return false; 1144 1145 return sk->sk_prot->stream_memory_free ? 1146 sk->sk_prot->stream_memory_free(sk) : true; 1147 } 1148 1149 static inline bool sk_stream_is_writeable(const struct sock *sk) 1150 { 1151 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1152 sk_stream_memory_free(sk); 1153 } 1154 1155 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1156 struct cgroup *ancestor) 1157 { 1158 #ifdef CONFIG_SOCK_CGROUP_DATA 1159 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1160 ancestor); 1161 #else 1162 return -ENOTSUPP; 1163 #endif 1164 } 1165 1166 static inline bool sk_has_memory_pressure(const struct sock *sk) 1167 { 1168 return sk->sk_prot->memory_pressure != NULL; 1169 } 1170 1171 static inline bool sk_under_memory_pressure(const struct sock *sk) 1172 { 1173 if (!sk->sk_prot->memory_pressure) 1174 return false; 1175 1176 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1177 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1178 return true; 1179 1180 return !!*sk->sk_prot->memory_pressure; 1181 } 1182 1183 static inline void sk_leave_memory_pressure(struct sock *sk) 1184 { 1185 int *memory_pressure = sk->sk_prot->memory_pressure; 1186 1187 if (!memory_pressure) 1188 return; 1189 1190 if (*memory_pressure) 1191 *memory_pressure = 0; 1192 } 1193 1194 static inline void sk_enter_memory_pressure(struct sock *sk) 1195 { 1196 if (!sk->sk_prot->enter_memory_pressure) 1197 return; 1198 1199 sk->sk_prot->enter_memory_pressure(sk); 1200 } 1201 1202 static inline long 1203 sk_memory_allocated(const struct sock *sk) 1204 { 1205 return atomic_long_read(sk->sk_prot->memory_allocated); 1206 } 1207 1208 static inline long 1209 sk_memory_allocated_add(struct sock *sk, int amt) 1210 { 1211 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1212 } 1213 1214 static inline void 1215 sk_memory_allocated_sub(struct sock *sk, int amt) 1216 { 1217 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1218 } 1219 1220 static inline void sk_sockets_allocated_dec(struct sock *sk) 1221 { 1222 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1223 } 1224 1225 static inline void sk_sockets_allocated_inc(struct sock *sk) 1226 { 1227 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1228 } 1229 1230 static inline int 1231 sk_sockets_allocated_read_positive(struct sock *sk) 1232 { 1233 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1234 } 1235 1236 static inline int 1237 proto_sockets_allocated_sum_positive(struct proto *prot) 1238 { 1239 return percpu_counter_sum_positive(prot->sockets_allocated); 1240 } 1241 1242 static inline long 1243 proto_memory_allocated(struct proto *prot) 1244 { 1245 return atomic_long_read(prot->memory_allocated); 1246 } 1247 1248 static inline bool 1249 proto_memory_pressure(struct proto *prot) 1250 { 1251 if (!prot->memory_pressure) 1252 return false; 1253 return !!*prot->memory_pressure; 1254 } 1255 1256 1257 #ifdef CONFIG_PROC_FS 1258 /* Called with local bh disabled */ 1259 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1260 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1261 #else 1262 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1263 int inc) 1264 { 1265 } 1266 #endif 1267 1268 1269 /* With per-bucket locks this operation is not-atomic, so that 1270 * this version is not worse. 1271 */ 1272 static inline int __sk_prot_rehash(struct sock *sk) 1273 { 1274 sk->sk_prot->unhash(sk); 1275 return sk->sk_prot->hash(sk); 1276 } 1277 1278 /* About 10 seconds */ 1279 #define SOCK_DESTROY_TIME (10*HZ) 1280 1281 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1282 #define PROT_SOCK 1024 1283 1284 #define SHUTDOWN_MASK 3 1285 #define RCV_SHUTDOWN 1 1286 #define SEND_SHUTDOWN 2 1287 1288 #define SOCK_SNDBUF_LOCK 1 1289 #define SOCK_RCVBUF_LOCK 2 1290 #define SOCK_BINDADDR_LOCK 4 1291 #define SOCK_BINDPORT_LOCK 8 1292 1293 struct socket_alloc { 1294 struct socket socket; 1295 struct inode vfs_inode; 1296 }; 1297 1298 static inline struct socket *SOCKET_I(struct inode *inode) 1299 { 1300 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1301 } 1302 1303 static inline struct inode *SOCK_INODE(struct socket *socket) 1304 { 1305 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1306 } 1307 1308 /* 1309 * Functions for memory accounting 1310 */ 1311 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1312 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1313 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1314 void __sk_mem_reclaim(struct sock *sk, int amount); 1315 1316 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1317 * do not necessarily have 16x time more memory than 4KB ones. 1318 */ 1319 #define SK_MEM_QUANTUM 4096 1320 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1321 #define SK_MEM_SEND 0 1322 #define SK_MEM_RECV 1 1323 1324 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1325 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1326 { 1327 long val = sk->sk_prot->sysctl_mem[index]; 1328 1329 #if PAGE_SIZE > SK_MEM_QUANTUM 1330 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1331 #elif PAGE_SIZE < SK_MEM_QUANTUM 1332 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1333 #endif 1334 return val; 1335 } 1336 1337 static inline int sk_mem_pages(int amt) 1338 { 1339 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1340 } 1341 1342 static inline bool sk_has_account(struct sock *sk) 1343 { 1344 /* return true if protocol supports memory accounting */ 1345 return !!sk->sk_prot->memory_allocated; 1346 } 1347 1348 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1349 { 1350 if (!sk_has_account(sk)) 1351 return true; 1352 return size <= sk->sk_forward_alloc || 1353 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1354 } 1355 1356 static inline bool 1357 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1358 { 1359 if (!sk_has_account(sk)) 1360 return true; 1361 return size<= sk->sk_forward_alloc || 1362 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1363 skb_pfmemalloc(skb); 1364 } 1365 1366 static inline void sk_mem_reclaim(struct sock *sk) 1367 { 1368 if (!sk_has_account(sk)) 1369 return; 1370 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1371 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1372 } 1373 1374 static inline void sk_mem_reclaim_partial(struct sock *sk) 1375 { 1376 if (!sk_has_account(sk)) 1377 return; 1378 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1379 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1380 } 1381 1382 static inline void sk_mem_charge(struct sock *sk, int size) 1383 { 1384 if (!sk_has_account(sk)) 1385 return; 1386 sk->sk_forward_alloc -= size; 1387 } 1388 1389 static inline void sk_mem_uncharge(struct sock *sk, int size) 1390 { 1391 if (!sk_has_account(sk)) 1392 return; 1393 sk->sk_forward_alloc += size; 1394 1395 /* Avoid a possible overflow. 1396 * TCP send queues can make this happen, if sk_mem_reclaim() 1397 * is not called and more than 2 GBytes are released at once. 1398 * 1399 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1400 * no need to hold that much forward allocation anyway. 1401 */ 1402 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1403 __sk_mem_reclaim(sk, 1 << 20); 1404 } 1405 1406 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1407 { 1408 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1409 sk->sk_wmem_queued -= skb->truesize; 1410 sk_mem_uncharge(sk, skb->truesize); 1411 __kfree_skb(skb); 1412 } 1413 1414 static inline void sock_release_ownership(struct sock *sk) 1415 { 1416 if (sk->sk_lock.owned) { 1417 sk->sk_lock.owned = 0; 1418 1419 /* The sk_lock has mutex_unlock() semantics: */ 1420 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1421 } 1422 } 1423 1424 /* 1425 * Macro so as to not evaluate some arguments when 1426 * lockdep is not enabled. 1427 * 1428 * Mark both the sk_lock and the sk_lock.slock as a 1429 * per-address-family lock class. 1430 */ 1431 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1432 do { \ 1433 sk->sk_lock.owned = 0; \ 1434 init_waitqueue_head(&sk->sk_lock.wq); \ 1435 spin_lock_init(&(sk)->sk_lock.slock); \ 1436 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1437 sizeof((sk)->sk_lock)); \ 1438 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1439 (skey), (sname)); \ 1440 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1441 } while (0) 1442 1443 #ifdef CONFIG_LOCKDEP 1444 static inline bool lockdep_sock_is_held(const struct sock *csk) 1445 { 1446 struct sock *sk = (struct sock *)csk; 1447 1448 return lockdep_is_held(&sk->sk_lock) || 1449 lockdep_is_held(&sk->sk_lock.slock); 1450 } 1451 #endif 1452 1453 void lock_sock_nested(struct sock *sk, int subclass); 1454 1455 static inline void lock_sock(struct sock *sk) 1456 { 1457 lock_sock_nested(sk, 0); 1458 } 1459 1460 void release_sock(struct sock *sk); 1461 1462 /* BH context may only use the following locking interface. */ 1463 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1464 #define bh_lock_sock_nested(__sk) \ 1465 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1466 SINGLE_DEPTH_NESTING) 1467 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1468 1469 bool lock_sock_fast(struct sock *sk); 1470 /** 1471 * unlock_sock_fast - complement of lock_sock_fast 1472 * @sk: socket 1473 * @slow: slow mode 1474 * 1475 * fast unlock socket for user context. 1476 * If slow mode is on, we call regular release_sock() 1477 */ 1478 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1479 { 1480 if (slow) 1481 release_sock(sk); 1482 else 1483 spin_unlock_bh(&sk->sk_lock.slock); 1484 } 1485 1486 /* Used by processes to "lock" a socket state, so that 1487 * interrupts and bottom half handlers won't change it 1488 * from under us. It essentially blocks any incoming 1489 * packets, so that we won't get any new data or any 1490 * packets that change the state of the socket. 1491 * 1492 * While locked, BH processing will add new packets to 1493 * the backlog queue. This queue is processed by the 1494 * owner of the socket lock right before it is released. 1495 * 1496 * Since ~2.3.5 it is also exclusive sleep lock serializing 1497 * accesses from user process context. 1498 */ 1499 1500 static inline void sock_owned_by_me(const struct sock *sk) 1501 { 1502 #ifdef CONFIG_LOCKDEP 1503 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1504 #endif 1505 } 1506 1507 static inline bool sock_owned_by_user(const struct sock *sk) 1508 { 1509 sock_owned_by_me(sk); 1510 return sk->sk_lock.owned; 1511 } 1512 1513 /* no reclassification while locks are held */ 1514 static inline bool sock_allow_reclassification(const struct sock *csk) 1515 { 1516 struct sock *sk = (struct sock *)csk; 1517 1518 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1519 } 1520 1521 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1522 struct proto *prot, int kern); 1523 void sk_free(struct sock *sk); 1524 void sk_destruct(struct sock *sk); 1525 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1526 1527 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1528 gfp_t priority); 1529 void __sock_wfree(struct sk_buff *skb); 1530 void sock_wfree(struct sk_buff *skb); 1531 void skb_orphan_partial(struct sk_buff *skb); 1532 void sock_rfree(struct sk_buff *skb); 1533 void sock_efree(struct sk_buff *skb); 1534 #ifdef CONFIG_INET 1535 void sock_edemux(struct sk_buff *skb); 1536 #else 1537 #define sock_edemux sock_efree 1538 #endif 1539 1540 int sock_setsockopt(struct socket *sock, int level, int op, 1541 char __user *optval, unsigned int optlen); 1542 1543 int sock_getsockopt(struct socket *sock, int level, int op, 1544 char __user *optval, int __user *optlen); 1545 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1546 int noblock, int *errcode); 1547 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1548 unsigned long data_len, int noblock, 1549 int *errcode, int max_page_order); 1550 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1551 void sock_kfree_s(struct sock *sk, void *mem, int size); 1552 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1553 void sk_send_sigurg(struct sock *sk); 1554 1555 struct sockcm_cookie { 1556 u32 mark; 1557 u16 tsflags; 1558 }; 1559 1560 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1561 struct sockcm_cookie *sockc); 1562 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1563 struct sockcm_cookie *sockc); 1564 1565 /* 1566 * Functions to fill in entries in struct proto_ops when a protocol 1567 * does not implement a particular function. 1568 */ 1569 int sock_no_bind(struct socket *, struct sockaddr *, int); 1570 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1571 int sock_no_socketpair(struct socket *, struct socket *); 1572 int sock_no_accept(struct socket *, struct socket *, int); 1573 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1574 unsigned int sock_no_poll(struct file *, struct socket *, 1575 struct poll_table_struct *); 1576 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1577 int sock_no_listen(struct socket *, int); 1578 int sock_no_shutdown(struct socket *, int); 1579 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1580 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1581 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1582 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1583 int sock_no_mmap(struct file *file, struct socket *sock, 1584 struct vm_area_struct *vma); 1585 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1586 size_t size, int flags); 1587 1588 /* 1589 * Functions to fill in entries in struct proto_ops when a protocol 1590 * uses the inet style. 1591 */ 1592 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1593 char __user *optval, int __user *optlen); 1594 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1595 int flags); 1596 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1597 char __user *optval, unsigned int optlen); 1598 int compat_sock_common_getsockopt(struct socket *sock, int level, 1599 int optname, char __user *optval, int __user *optlen); 1600 int compat_sock_common_setsockopt(struct socket *sock, int level, 1601 int optname, char __user *optval, unsigned int optlen); 1602 1603 void sk_common_release(struct sock *sk); 1604 1605 /* 1606 * Default socket callbacks and setup code 1607 */ 1608 1609 /* Initialise core socket variables */ 1610 void sock_init_data(struct socket *sock, struct sock *sk); 1611 1612 /* 1613 * Socket reference counting postulates. 1614 * 1615 * * Each user of socket SHOULD hold a reference count. 1616 * * Each access point to socket (an hash table bucket, reference from a list, 1617 * running timer, skb in flight MUST hold a reference count. 1618 * * When reference count hits 0, it means it will never increase back. 1619 * * When reference count hits 0, it means that no references from 1620 * outside exist to this socket and current process on current CPU 1621 * is last user and may/should destroy this socket. 1622 * * sk_free is called from any context: process, BH, IRQ. When 1623 * it is called, socket has no references from outside -> sk_free 1624 * may release descendant resources allocated by the socket, but 1625 * to the time when it is called, socket is NOT referenced by any 1626 * hash tables, lists etc. 1627 * * Packets, delivered from outside (from network or from another process) 1628 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1629 * when they sit in queue. Otherwise, packets will leak to hole, when 1630 * socket is looked up by one cpu and unhasing is made by another CPU. 1631 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1632 * (leak to backlog). Packet socket does all the processing inside 1633 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1634 * use separate SMP lock, so that they are prone too. 1635 */ 1636 1637 /* Ungrab socket and destroy it, if it was the last reference. */ 1638 static inline void sock_put(struct sock *sk) 1639 { 1640 if (atomic_dec_and_test(&sk->sk_refcnt)) 1641 sk_free(sk); 1642 } 1643 /* Generic version of sock_put(), dealing with all sockets 1644 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1645 */ 1646 void sock_gen_put(struct sock *sk); 1647 1648 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1649 unsigned int trim_cap, bool refcounted); 1650 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1651 const int nested) 1652 { 1653 return __sk_receive_skb(sk, skb, nested, 1, true); 1654 } 1655 1656 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1657 { 1658 sk->sk_tx_queue_mapping = tx_queue; 1659 } 1660 1661 static inline void sk_tx_queue_clear(struct sock *sk) 1662 { 1663 sk->sk_tx_queue_mapping = -1; 1664 } 1665 1666 static inline int sk_tx_queue_get(const struct sock *sk) 1667 { 1668 return sk ? sk->sk_tx_queue_mapping : -1; 1669 } 1670 1671 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1672 { 1673 sk_tx_queue_clear(sk); 1674 sk->sk_socket = sock; 1675 } 1676 1677 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1678 { 1679 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1680 return &rcu_dereference_raw(sk->sk_wq)->wait; 1681 } 1682 /* Detach socket from process context. 1683 * Announce socket dead, detach it from wait queue and inode. 1684 * Note that parent inode held reference count on this struct sock, 1685 * we do not release it in this function, because protocol 1686 * probably wants some additional cleanups or even continuing 1687 * to work with this socket (TCP). 1688 */ 1689 static inline void sock_orphan(struct sock *sk) 1690 { 1691 write_lock_bh(&sk->sk_callback_lock); 1692 sock_set_flag(sk, SOCK_DEAD); 1693 sk_set_socket(sk, NULL); 1694 sk->sk_wq = NULL; 1695 write_unlock_bh(&sk->sk_callback_lock); 1696 } 1697 1698 static inline void sock_graft(struct sock *sk, struct socket *parent) 1699 { 1700 write_lock_bh(&sk->sk_callback_lock); 1701 sk->sk_wq = parent->wq; 1702 parent->sk = sk; 1703 sk_set_socket(sk, parent); 1704 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1705 security_sock_graft(sk, parent); 1706 write_unlock_bh(&sk->sk_callback_lock); 1707 } 1708 1709 kuid_t sock_i_uid(struct sock *sk); 1710 unsigned long sock_i_ino(struct sock *sk); 1711 1712 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1713 { 1714 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1715 } 1716 1717 static inline u32 net_tx_rndhash(void) 1718 { 1719 u32 v = prandom_u32(); 1720 1721 return v ?: 1; 1722 } 1723 1724 static inline void sk_set_txhash(struct sock *sk) 1725 { 1726 sk->sk_txhash = net_tx_rndhash(); 1727 } 1728 1729 static inline void sk_rethink_txhash(struct sock *sk) 1730 { 1731 if (sk->sk_txhash) 1732 sk_set_txhash(sk); 1733 } 1734 1735 static inline struct dst_entry * 1736 __sk_dst_get(struct sock *sk) 1737 { 1738 return rcu_dereference_check(sk->sk_dst_cache, 1739 lockdep_sock_is_held(sk)); 1740 } 1741 1742 static inline struct dst_entry * 1743 sk_dst_get(struct sock *sk) 1744 { 1745 struct dst_entry *dst; 1746 1747 rcu_read_lock(); 1748 dst = rcu_dereference(sk->sk_dst_cache); 1749 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1750 dst = NULL; 1751 rcu_read_unlock(); 1752 return dst; 1753 } 1754 1755 static inline void dst_negative_advice(struct sock *sk) 1756 { 1757 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1758 1759 sk_rethink_txhash(sk); 1760 1761 if (dst && dst->ops->negative_advice) { 1762 ndst = dst->ops->negative_advice(dst); 1763 1764 if (ndst != dst) { 1765 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1766 sk_tx_queue_clear(sk); 1767 } 1768 } 1769 } 1770 1771 static inline void 1772 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1773 { 1774 struct dst_entry *old_dst; 1775 1776 sk_tx_queue_clear(sk); 1777 /* 1778 * This can be called while sk is owned by the caller only, 1779 * with no state that can be checked in a rcu_dereference_check() cond 1780 */ 1781 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1782 rcu_assign_pointer(sk->sk_dst_cache, dst); 1783 dst_release(old_dst); 1784 } 1785 1786 static inline void 1787 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1788 { 1789 struct dst_entry *old_dst; 1790 1791 sk_tx_queue_clear(sk); 1792 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1793 dst_release(old_dst); 1794 } 1795 1796 static inline void 1797 __sk_dst_reset(struct sock *sk) 1798 { 1799 __sk_dst_set(sk, NULL); 1800 } 1801 1802 static inline void 1803 sk_dst_reset(struct sock *sk) 1804 { 1805 sk_dst_set(sk, NULL); 1806 } 1807 1808 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1809 1810 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1811 1812 bool sk_mc_loop(struct sock *sk); 1813 1814 static inline bool sk_can_gso(const struct sock *sk) 1815 { 1816 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1817 } 1818 1819 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1820 1821 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1822 { 1823 sk->sk_route_nocaps |= flags; 1824 sk->sk_route_caps &= ~flags; 1825 } 1826 1827 static inline bool sk_check_csum_caps(struct sock *sk) 1828 { 1829 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1830 (sk->sk_family == PF_INET && 1831 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1832 (sk->sk_family == PF_INET6 && 1833 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1834 } 1835 1836 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1837 struct iov_iter *from, char *to, 1838 int copy, int offset) 1839 { 1840 if (skb->ip_summed == CHECKSUM_NONE) { 1841 __wsum csum = 0; 1842 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1843 return -EFAULT; 1844 skb->csum = csum_block_add(skb->csum, csum, offset); 1845 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1846 if (!copy_from_iter_full_nocache(to, copy, from)) 1847 return -EFAULT; 1848 } else if (!copy_from_iter_full(to, copy, from)) 1849 return -EFAULT; 1850 1851 return 0; 1852 } 1853 1854 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1855 struct iov_iter *from, int copy) 1856 { 1857 int err, offset = skb->len; 1858 1859 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1860 copy, offset); 1861 if (err) 1862 __skb_trim(skb, offset); 1863 1864 return err; 1865 } 1866 1867 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1868 struct sk_buff *skb, 1869 struct page *page, 1870 int off, int copy) 1871 { 1872 int err; 1873 1874 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1875 copy, skb->len); 1876 if (err) 1877 return err; 1878 1879 skb->len += copy; 1880 skb->data_len += copy; 1881 skb->truesize += copy; 1882 sk->sk_wmem_queued += copy; 1883 sk_mem_charge(sk, copy); 1884 return 0; 1885 } 1886 1887 /** 1888 * sk_wmem_alloc_get - returns write allocations 1889 * @sk: socket 1890 * 1891 * Returns sk_wmem_alloc minus initial offset of one 1892 */ 1893 static inline int sk_wmem_alloc_get(const struct sock *sk) 1894 { 1895 return atomic_read(&sk->sk_wmem_alloc) - 1; 1896 } 1897 1898 /** 1899 * sk_rmem_alloc_get - returns read allocations 1900 * @sk: socket 1901 * 1902 * Returns sk_rmem_alloc 1903 */ 1904 static inline int sk_rmem_alloc_get(const struct sock *sk) 1905 { 1906 return atomic_read(&sk->sk_rmem_alloc); 1907 } 1908 1909 /** 1910 * sk_has_allocations - check if allocations are outstanding 1911 * @sk: socket 1912 * 1913 * Returns true if socket has write or read allocations 1914 */ 1915 static inline bool sk_has_allocations(const struct sock *sk) 1916 { 1917 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1918 } 1919 1920 /** 1921 * skwq_has_sleeper - check if there are any waiting processes 1922 * @wq: struct socket_wq 1923 * 1924 * Returns true if socket_wq has waiting processes 1925 * 1926 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1927 * barrier call. They were added due to the race found within the tcp code. 1928 * 1929 * Consider following tcp code paths: 1930 * 1931 * CPU1 CPU2 1932 * 1933 * sys_select receive packet 1934 * ... ... 1935 * __add_wait_queue update tp->rcv_nxt 1936 * ... ... 1937 * tp->rcv_nxt check sock_def_readable 1938 * ... { 1939 * schedule rcu_read_lock(); 1940 * wq = rcu_dereference(sk->sk_wq); 1941 * if (wq && waitqueue_active(&wq->wait)) 1942 * wake_up_interruptible(&wq->wait) 1943 * ... 1944 * } 1945 * 1946 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1947 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1948 * could then endup calling schedule and sleep forever if there are no more 1949 * data on the socket. 1950 * 1951 */ 1952 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1953 { 1954 return wq && wq_has_sleeper(&wq->wait); 1955 } 1956 1957 /** 1958 * sock_poll_wait - place memory barrier behind the poll_wait call. 1959 * @filp: file 1960 * @wait_address: socket wait queue 1961 * @p: poll_table 1962 * 1963 * See the comments in the wq_has_sleeper function. 1964 */ 1965 static inline void sock_poll_wait(struct file *filp, 1966 wait_queue_head_t *wait_address, poll_table *p) 1967 { 1968 if (!poll_does_not_wait(p) && wait_address) { 1969 poll_wait(filp, wait_address, p); 1970 /* We need to be sure we are in sync with the 1971 * socket flags modification. 1972 * 1973 * This memory barrier is paired in the wq_has_sleeper. 1974 */ 1975 smp_mb(); 1976 } 1977 } 1978 1979 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1980 { 1981 if (sk->sk_txhash) { 1982 skb->l4_hash = 1; 1983 skb->hash = sk->sk_txhash; 1984 } 1985 } 1986 1987 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 1988 1989 /* 1990 * Queue a received datagram if it will fit. Stream and sequenced 1991 * protocols can't normally use this as they need to fit buffers in 1992 * and play with them. 1993 * 1994 * Inlined as it's very short and called for pretty much every 1995 * packet ever received. 1996 */ 1997 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1998 { 1999 skb_orphan(skb); 2000 skb->sk = sk; 2001 skb->destructor = sock_rfree; 2002 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2003 sk_mem_charge(sk, skb->truesize); 2004 } 2005 2006 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2007 unsigned long expires); 2008 2009 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2010 2011 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff *skb, 2012 unsigned int flags); 2013 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2014 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2015 2016 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2017 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2018 2019 /* 2020 * Recover an error report and clear atomically 2021 */ 2022 2023 static inline int sock_error(struct sock *sk) 2024 { 2025 int err; 2026 if (likely(!sk->sk_err)) 2027 return 0; 2028 err = xchg(&sk->sk_err, 0); 2029 return -err; 2030 } 2031 2032 static inline unsigned long sock_wspace(struct sock *sk) 2033 { 2034 int amt = 0; 2035 2036 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2037 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2038 if (amt < 0) 2039 amt = 0; 2040 } 2041 return amt; 2042 } 2043 2044 /* Note: 2045 * We use sk->sk_wq_raw, from contexts knowing this 2046 * pointer is not NULL and cannot disappear/change. 2047 */ 2048 static inline void sk_set_bit(int nr, struct sock *sk) 2049 { 2050 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2051 !sock_flag(sk, SOCK_FASYNC)) 2052 return; 2053 2054 set_bit(nr, &sk->sk_wq_raw->flags); 2055 } 2056 2057 static inline void sk_clear_bit(int nr, struct sock *sk) 2058 { 2059 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2060 !sock_flag(sk, SOCK_FASYNC)) 2061 return; 2062 2063 clear_bit(nr, &sk->sk_wq_raw->flags); 2064 } 2065 2066 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2067 { 2068 if (sock_flag(sk, SOCK_FASYNC)) { 2069 rcu_read_lock(); 2070 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2071 rcu_read_unlock(); 2072 } 2073 } 2074 2075 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2076 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2077 * Note: for send buffers, TCP works better if we can build two skbs at 2078 * minimum. 2079 */ 2080 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2081 2082 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2083 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2084 2085 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2086 { 2087 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2088 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2089 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2090 } 2091 } 2092 2093 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2094 bool force_schedule); 2095 2096 /** 2097 * sk_page_frag - return an appropriate page_frag 2098 * @sk: socket 2099 * 2100 * If socket allocation mode allows current thread to sleep, it means its 2101 * safe to use the per task page_frag instead of the per socket one. 2102 */ 2103 static inline struct page_frag *sk_page_frag(struct sock *sk) 2104 { 2105 if (gfpflags_allow_blocking(sk->sk_allocation)) 2106 return ¤t->task_frag; 2107 2108 return &sk->sk_frag; 2109 } 2110 2111 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2112 2113 /* 2114 * Default write policy as shown to user space via poll/select/SIGIO 2115 */ 2116 static inline bool sock_writeable(const struct sock *sk) 2117 { 2118 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2119 } 2120 2121 static inline gfp_t gfp_any(void) 2122 { 2123 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2124 } 2125 2126 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2127 { 2128 return noblock ? 0 : sk->sk_rcvtimeo; 2129 } 2130 2131 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2132 { 2133 return noblock ? 0 : sk->sk_sndtimeo; 2134 } 2135 2136 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2137 { 2138 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2139 } 2140 2141 /* Alas, with timeout socket operations are not restartable. 2142 * Compare this to poll(). 2143 */ 2144 static inline int sock_intr_errno(long timeo) 2145 { 2146 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2147 } 2148 2149 struct sock_skb_cb { 2150 u32 dropcount; 2151 }; 2152 2153 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2154 * using skb->cb[] would keep using it directly and utilize its 2155 * alignement guarantee. 2156 */ 2157 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2158 sizeof(struct sock_skb_cb))) 2159 2160 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2161 SOCK_SKB_CB_OFFSET)) 2162 2163 #define sock_skb_cb_check_size(size) \ 2164 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2165 2166 static inline void 2167 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2168 { 2169 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2170 atomic_read(&sk->sk_drops) : 0; 2171 } 2172 2173 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2174 { 2175 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2176 2177 atomic_add(segs, &sk->sk_drops); 2178 } 2179 2180 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2181 struct sk_buff *skb); 2182 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2183 struct sk_buff *skb); 2184 2185 static inline void 2186 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2187 { 2188 ktime_t kt = skb->tstamp; 2189 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2190 2191 /* 2192 * generate control messages if 2193 * - receive time stamping in software requested 2194 * - software time stamp available and wanted 2195 * - hardware time stamps available and wanted 2196 */ 2197 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2198 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2199 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2200 (hwtstamps->hwtstamp && 2201 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2202 __sock_recv_timestamp(msg, sk, skb); 2203 else 2204 sk->sk_stamp = kt; 2205 2206 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2207 __sock_recv_wifi_status(msg, sk, skb); 2208 } 2209 2210 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2211 struct sk_buff *skb); 2212 2213 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2214 struct sk_buff *skb) 2215 { 2216 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2217 (1UL << SOCK_RCVTSTAMP)) 2218 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2219 SOF_TIMESTAMPING_RAW_HARDWARE) 2220 2221 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2222 __sock_recv_ts_and_drops(msg, sk, skb); 2223 else 2224 sk->sk_stamp = skb->tstamp; 2225 } 2226 2227 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2228 2229 /** 2230 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2231 * @sk: socket sending this packet 2232 * @tsflags: timestamping flags to use 2233 * @tx_flags: completed with instructions for time stamping 2234 * 2235 * Note : callers should take care of initial *tx_flags value (usually 0) 2236 */ 2237 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2238 __u8 *tx_flags) 2239 { 2240 if (unlikely(tsflags)) 2241 __sock_tx_timestamp(tsflags, tx_flags); 2242 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2243 *tx_flags |= SKBTX_WIFI_STATUS; 2244 } 2245 2246 /** 2247 * sk_eat_skb - Release a skb if it is no longer needed 2248 * @sk: socket to eat this skb from 2249 * @skb: socket buffer to eat 2250 * 2251 * This routine must be called with interrupts disabled or with the socket 2252 * locked so that the sk_buff queue operation is ok. 2253 */ 2254 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2255 { 2256 __skb_unlink(skb, &sk->sk_receive_queue); 2257 __kfree_skb(skb); 2258 } 2259 2260 static inline 2261 struct net *sock_net(const struct sock *sk) 2262 { 2263 return read_pnet(&sk->sk_net); 2264 } 2265 2266 static inline 2267 void sock_net_set(struct sock *sk, struct net *net) 2268 { 2269 write_pnet(&sk->sk_net, net); 2270 } 2271 2272 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2273 { 2274 if (skb->sk) { 2275 struct sock *sk = skb->sk; 2276 2277 skb->destructor = NULL; 2278 skb->sk = NULL; 2279 return sk; 2280 } 2281 return NULL; 2282 } 2283 2284 /* This helper checks if a socket is a full socket, 2285 * ie _not_ a timewait or request socket. 2286 */ 2287 static inline bool sk_fullsock(const struct sock *sk) 2288 { 2289 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2290 } 2291 2292 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2293 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2294 */ 2295 static inline bool sk_listener(const struct sock *sk) 2296 { 2297 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2298 } 2299 2300 /** 2301 * sk_state_load - read sk->sk_state for lockless contexts 2302 * @sk: socket pointer 2303 * 2304 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2305 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2306 */ 2307 static inline int sk_state_load(const struct sock *sk) 2308 { 2309 return smp_load_acquire(&sk->sk_state); 2310 } 2311 2312 /** 2313 * sk_state_store - update sk->sk_state 2314 * @sk: socket pointer 2315 * @newstate: new state 2316 * 2317 * Paired with sk_state_load(). Should be used in contexts where 2318 * state change might impact lockless readers. 2319 */ 2320 static inline void sk_state_store(struct sock *sk, int newstate) 2321 { 2322 smp_store_release(&sk->sk_state, newstate); 2323 } 2324 2325 void sock_enable_timestamp(struct sock *sk, int flag); 2326 int sock_get_timestamp(struct sock *, struct timeval __user *); 2327 int sock_get_timestampns(struct sock *, struct timespec __user *); 2328 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2329 int type); 2330 2331 bool sk_ns_capable(const struct sock *sk, 2332 struct user_namespace *user_ns, int cap); 2333 bool sk_capable(const struct sock *sk, int cap); 2334 bool sk_net_capable(const struct sock *sk, int cap); 2335 2336 extern __u32 sysctl_wmem_max; 2337 extern __u32 sysctl_rmem_max; 2338 2339 extern int sysctl_tstamp_allow_data; 2340 extern int sysctl_optmem_max; 2341 2342 extern __u32 sysctl_wmem_default; 2343 extern __u32 sysctl_rmem_default; 2344 2345 #endif /* _SOCK_H */ 2346