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 <linux/refcount.h> 70 #include <net/dst.h> 71 #include <net/checksum.h> 72 #include <net/tcp_states.h> 73 #include <linux/net_tstamp.h> 74 #include <net/smc.h> 75 76 /* 77 * This structure really needs to be cleaned up. 78 * Most of it is for TCP, and not used by any of 79 * the other protocols. 80 */ 81 82 /* Define this to get the SOCK_DBG debugging facility. */ 83 #define SOCK_DEBUGGING 84 #ifdef SOCK_DEBUGGING 85 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 86 printk(KERN_DEBUG msg); } while (0) 87 #else 88 /* Validate arguments and do nothing */ 89 static inline __printf(2, 3) 90 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 91 { 92 } 93 #endif 94 95 /* This is the per-socket lock. The spinlock provides a synchronization 96 * between user contexts and software interrupt processing, whereas the 97 * mini-semaphore synchronizes multiple users amongst themselves. 98 */ 99 typedef struct { 100 spinlock_t slock; 101 int owned; 102 wait_queue_head_t wq; 103 /* 104 * We express the mutex-alike socket_lock semantics 105 * to the lock validator by explicitly managing 106 * the slock as a lock variant (in addition to 107 * the slock itself): 108 */ 109 #ifdef CONFIG_DEBUG_LOCK_ALLOC 110 struct lockdep_map dep_map; 111 #endif 112 } socket_lock_t; 113 114 struct sock; 115 struct proto; 116 struct net; 117 118 typedef __u32 __bitwise __portpair; 119 typedef __u64 __bitwise __addrpair; 120 121 /** 122 * struct sock_common - minimal network layer representation of sockets 123 * @skc_daddr: Foreign IPv4 addr 124 * @skc_rcv_saddr: Bound local IPv4 addr 125 * @skc_hash: hash value used with various protocol lookup tables 126 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 127 * @skc_dport: placeholder for inet_dport/tw_dport 128 * @skc_num: placeholder for inet_num/tw_num 129 * @skc_family: network address family 130 * @skc_state: Connection state 131 * @skc_reuse: %SO_REUSEADDR setting 132 * @skc_reuseport: %SO_REUSEPORT setting 133 * @skc_bound_dev_if: bound device index if != 0 134 * @skc_bind_node: bind hash linkage for various protocol lookup tables 135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 136 * @skc_prot: protocol handlers inside a network family 137 * @skc_net: reference to the network namespace of this socket 138 * @skc_node: main hash linkage for various protocol lookup tables 139 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 140 * @skc_tx_queue_mapping: tx queue number for this connection 141 * @skc_flags: place holder for sk_flags 142 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 143 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 144 * @skc_incoming_cpu: record/match cpu processing incoming packets 145 * @skc_refcnt: reference count 146 * 147 * This is the minimal network layer representation of sockets, the header 148 * for struct sock and struct inet_timewait_sock. 149 */ 150 struct sock_common { 151 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 152 * address on 64bit arches : cf INET_MATCH() 153 */ 154 union { 155 __addrpair skc_addrpair; 156 struct { 157 __be32 skc_daddr; 158 __be32 skc_rcv_saddr; 159 }; 160 }; 161 union { 162 unsigned int skc_hash; 163 __u16 skc_u16hashes[2]; 164 }; 165 /* skc_dport && skc_num must be grouped as well */ 166 union { 167 __portpair skc_portpair; 168 struct { 169 __be16 skc_dport; 170 __u16 skc_num; 171 }; 172 }; 173 174 unsigned short skc_family; 175 volatile unsigned char skc_state; 176 unsigned char skc_reuse:4; 177 unsigned char skc_reuseport:1; 178 unsigned char skc_ipv6only:1; 179 unsigned char skc_net_refcnt:1; 180 int skc_bound_dev_if; 181 union { 182 struct hlist_node skc_bind_node; 183 struct hlist_node skc_portaddr_node; 184 }; 185 struct proto *skc_prot; 186 possible_net_t skc_net; 187 188 #if IS_ENABLED(CONFIG_IPV6) 189 struct in6_addr skc_v6_daddr; 190 struct in6_addr skc_v6_rcv_saddr; 191 #endif 192 193 atomic64_t skc_cookie; 194 195 /* following fields are padding to force 196 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 197 * assuming IPV6 is enabled. We use this padding differently 198 * for different kind of 'sockets' 199 */ 200 union { 201 unsigned long skc_flags; 202 struct sock *skc_listener; /* request_sock */ 203 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 204 }; 205 /* 206 * fields between dontcopy_begin/dontcopy_end 207 * are not copied in sock_copy() 208 */ 209 /* private: */ 210 int skc_dontcopy_begin[0]; 211 /* public: */ 212 union { 213 struct hlist_node skc_node; 214 struct hlist_nulls_node skc_nulls_node; 215 }; 216 int skc_tx_queue_mapping; 217 union { 218 int skc_incoming_cpu; 219 u32 skc_rcv_wnd; 220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 221 }; 222 223 refcount_t skc_refcnt; 224 /* private: */ 225 int skc_dontcopy_end[0]; 226 union { 227 u32 skc_rxhash; 228 u32 skc_window_clamp; 229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 230 }; 231 /* public: */ 232 }; 233 234 /** 235 * struct sock - network layer representation of sockets 236 * @__sk_common: shared layout with inet_timewait_sock 237 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 238 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 239 * @sk_lock: synchronizer 240 * @sk_kern_sock: True if sock is using kernel lock classes 241 * @sk_rcvbuf: size of receive buffer in bytes 242 * @sk_wq: sock wait queue and async head 243 * @sk_rx_dst: receive input route used by early demux 244 * @sk_dst_cache: destination cache 245 * @sk_dst_pending_confirm: need to confirm neighbour 246 * @sk_policy: flow policy 247 * @sk_receive_queue: incoming packets 248 * @sk_wmem_alloc: transmit queue bytes committed 249 * @sk_write_queue: Packet sending queue 250 * @sk_omem_alloc: "o" is "option" or "other" 251 * @sk_wmem_queued: persistent queue size 252 * @sk_forward_alloc: space allocated forward 253 * @sk_napi_id: id of the last napi context to receive data for sk 254 * @sk_ll_usec: usecs to busypoll when there is no data 255 * @sk_allocation: allocation mode 256 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 257 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 258 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 259 * @sk_sndbuf: size of send buffer in bytes 260 * @sk_padding: unused element for alignment 261 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 262 * @sk_no_check_rx: allow zero checksum in RX packets 263 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 264 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 265 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 266 * @sk_gso_max_size: Maximum GSO segment size to build 267 * @sk_gso_max_segs: Maximum number of GSO segments 268 * @sk_lingertime: %SO_LINGER l_linger setting 269 * @sk_backlog: always used with the per-socket spinlock held 270 * @sk_callback_lock: used with the callbacks in the end of this struct 271 * @sk_error_queue: rarely used 272 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 273 * IPV6_ADDRFORM for instance) 274 * @sk_err: last error 275 * @sk_err_soft: errors that don't cause failure but are the cause of a 276 * persistent failure not just 'timed out' 277 * @sk_drops: raw/udp drops counter 278 * @sk_ack_backlog: current listen backlog 279 * @sk_max_ack_backlog: listen backlog set in listen() 280 * @sk_priority: %SO_PRIORITY setting 281 * @sk_type: socket type (%SOCK_STREAM, etc) 282 * @sk_protocol: which protocol this socket belongs in this network family 283 * @sk_peer_pid: &struct pid for this socket's peer 284 * @sk_peer_cred: %SO_PEERCRED setting 285 * @sk_rcvlowat: %SO_RCVLOWAT setting 286 * @sk_rcvtimeo: %SO_RCVTIMEO setting 287 * @sk_sndtimeo: %SO_SNDTIMEO setting 288 * @sk_txhash: computed flow hash for use on transmit 289 * @sk_filter: socket filtering instructions 290 * @sk_timer: sock cleanup timer 291 * @sk_stamp: time stamp of last packet received 292 * @sk_tsflags: SO_TIMESTAMPING socket options 293 * @sk_tskey: counter to disambiguate concurrent tstamp requests 294 * @sk_socket: Identd and reporting IO signals 295 * @sk_user_data: RPC layer private data 296 * @sk_frag: cached page frag 297 * @sk_peek_off: current peek_offset value 298 * @sk_send_head: front of stuff to transmit 299 * @sk_security: used by security modules 300 * @sk_mark: generic packet mark 301 * @sk_cgrp_data: cgroup data for this cgroup 302 * @sk_memcg: this socket's memory cgroup association 303 * @sk_write_pending: a write to stream socket waits to start 304 * @sk_state_change: callback to indicate change in the state of the sock 305 * @sk_data_ready: callback to indicate there is data to be processed 306 * @sk_write_space: callback to indicate there is bf sending space available 307 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 308 * @sk_backlog_rcv: callback to process the backlog 309 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 310 * @sk_reuseport_cb: reuseport group container 311 * @sk_rcu: used during RCU grace period 312 */ 313 struct sock { 314 /* 315 * Now struct inet_timewait_sock also uses sock_common, so please just 316 * don't add nothing before this first member (__sk_common) --acme 317 */ 318 struct sock_common __sk_common; 319 #define sk_node __sk_common.skc_node 320 #define sk_nulls_node __sk_common.skc_nulls_node 321 #define sk_refcnt __sk_common.skc_refcnt 322 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 323 324 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 325 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 326 #define sk_hash __sk_common.skc_hash 327 #define sk_portpair __sk_common.skc_portpair 328 #define sk_num __sk_common.skc_num 329 #define sk_dport __sk_common.skc_dport 330 #define sk_addrpair __sk_common.skc_addrpair 331 #define sk_daddr __sk_common.skc_daddr 332 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 333 #define sk_family __sk_common.skc_family 334 #define sk_state __sk_common.skc_state 335 #define sk_reuse __sk_common.skc_reuse 336 #define sk_reuseport __sk_common.skc_reuseport 337 #define sk_ipv6only __sk_common.skc_ipv6only 338 #define sk_net_refcnt __sk_common.skc_net_refcnt 339 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 340 #define sk_bind_node __sk_common.skc_bind_node 341 #define sk_prot __sk_common.skc_prot 342 #define sk_net __sk_common.skc_net 343 #define sk_v6_daddr __sk_common.skc_v6_daddr 344 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 345 #define sk_cookie __sk_common.skc_cookie 346 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 347 #define sk_flags __sk_common.skc_flags 348 #define sk_rxhash __sk_common.skc_rxhash 349 350 socket_lock_t sk_lock; 351 atomic_t sk_drops; 352 int sk_rcvlowat; 353 struct sk_buff_head sk_error_queue; 354 struct sk_buff_head sk_receive_queue; 355 /* 356 * The backlog queue is special, it is always used with 357 * the per-socket spinlock held and requires low latency 358 * access. Therefore we special case it's implementation. 359 * Note : rmem_alloc is in this structure to fill a hole 360 * on 64bit arches, not because its logically part of 361 * backlog. 362 */ 363 struct { 364 atomic_t rmem_alloc; 365 int len; 366 struct sk_buff *head; 367 struct sk_buff *tail; 368 } sk_backlog; 369 #define sk_rmem_alloc sk_backlog.rmem_alloc 370 371 int sk_forward_alloc; 372 #ifdef CONFIG_NET_RX_BUSY_POLL 373 unsigned int sk_ll_usec; 374 /* ===== mostly read cache line ===== */ 375 unsigned int sk_napi_id; 376 #endif 377 int sk_rcvbuf; 378 379 struct sk_filter __rcu *sk_filter; 380 union { 381 struct socket_wq __rcu *sk_wq; 382 struct socket_wq *sk_wq_raw; 383 }; 384 #ifdef CONFIG_XFRM 385 struct xfrm_policy __rcu *sk_policy[2]; 386 #endif 387 struct dst_entry *sk_rx_dst; 388 struct dst_entry __rcu *sk_dst_cache; 389 atomic_t sk_omem_alloc; 390 int sk_sndbuf; 391 392 /* ===== cache line for TX ===== */ 393 int sk_wmem_queued; 394 refcount_t sk_wmem_alloc; 395 unsigned long sk_tsq_flags; 396 struct sk_buff *sk_send_head; 397 struct sk_buff_head sk_write_queue; 398 __s32 sk_peek_off; 399 int sk_write_pending; 400 __u32 sk_dst_pending_confirm; 401 u32 sk_pacing_status; /* see enum sk_pacing */ 402 long sk_sndtimeo; 403 struct timer_list sk_timer; 404 __u32 sk_priority; 405 __u32 sk_mark; 406 u32 sk_pacing_rate; /* bytes per second */ 407 u32 sk_max_pacing_rate; 408 struct page_frag sk_frag; 409 netdev_features_t sk_route_caps; 410 netdev_features_t sk_route_nocaps; 411 int sk_gso_type; 412 unsigned int sk_gso_max_size; 413 gfp_t sk_allocation; 414 __u32 sk_txhash; 415 416 /* 417 * Because of non atomicity rules, all 418 * changes are protected by socket lock. 419 */ 420 unsigned int __sk_flags_offset[0]; 421 #ifdef __BIG_ENDIAN_BITFIELD 422 #define SK_FL_PROTO_SHIFT 16 423 #define SK_FL_PROTO_MASK 0x00ff0000 424 425 #define SK_FL_TYPE_SHIFT 0 426 #define SK_FL_TYPE_MASK 0x0000ffff 427 #else 428 #define SK_FL_PROTO_SHIFT 8 429 #define SK_FL_PROTO_MASK 0x0000ff00 430 431 #define SK_FL_TYPE_SHIFT 16 432 #define SK_FL_TYPE_MASK 0xffff0000 433 #endif 434 435 kmemcheck_bitfield_begin(flags); 436 unsigned int sk_padding : 1, 437 sk_kern_sock : 1, 438 sk_no_check_tx : 1, 439 sk_no_check_rx : 1, 440 sk_userlocks : 4, 441 sk_protocol : 8, 442 sk_type : 16; 443 #define SK_PROTOCOL_MAX U8_MAX 444 kmemcheck_bitfield_end(flags); 445 446 u16 sk_gso_max_segs; 447 unsigned long sk_lingertime; 448 struct proto *sk_prot_creator; 449 rwlock_t sk_callback_lock; 450 int sk_err, 451 sk_err_soft; 452 u32 sk_ack_backlog; 453 u32 sk_max_ack_backlog; 454 kuid_t sk_uid; 455 struct pid *sk_peer_pid; 456 const struct cred *sk_peer_cred; 457 long sk_rcvtimeo; 458 ktime_t sk_stamp; 459 u16 sk_tsflags; 460 u8 sk_shutdown; 461 u32 sk_tskey; 462 struct socket *sk_socket; 463 void *sk_user_data; 464 #ifdef CONFIG_SECURITY 465 void *sk_security; 466 #endif 467 struct sock_cgroup_data sk_cgrp_data; 468 struct mem_cgroup *sk_memcg; 469 void (*sk_state_change)(struct sock *sk); 470 void (*sk_data_ready)(struct sock *sk); 471 void (*sk_write_space)(struct sock *sk); 472 void (*sk_error_report)(struct sock *sk); 473 int (*sk_backlog_rcv)(struct sock *sk, 474 struct sk_buff *skb); 475 void (*sk_destruct)(struct sock *sk); 476 struct sock_reuseport __rcu *sk_reuseport_cb; 477 struct rcu_head sk_rcu; 478 }; 479 480 enum sk_pacing { 481 SK_PACING_NONE = 0, 482 SK_PACING_NEEDED = 1, 483 SK_PACING_FQ = 2, 484 }; 485 486 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 487 488 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 489 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 490 491 /* 492 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 493 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 494 * on a socket means that the socket will reuse everybody else's port 495 * without looking at the other's sk_reuse value. 496 */ 497 498 #define SK_NO_REUSE 0 499 #define SK_CAN_REUSE 1 500 #define SK_FORCE_REUSE 2 501 502 int sk_set_peek_off(struct sock *sk, int val); 503 504 static inline int sk_peek_offset(struct sock *sk, int flags) 505 { 506 if (unlikely(flags & MSG_PEEK)) { 507 s32 off = READ_ONCE(sk->sk_peek_off); 508 if (off >= 0) 509 return off; 510 } 511 512 return 0; 513 } 514 515 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 516 { 517 s32 off = READ_ONCE(sk->sk_peek_off); 518 519 if (unlikely(off >= 0)) { 520 off = max_t(s32, off - val, 0); 521 WRITE_ONCE(sk->sk_peek_off, off); 522 } 523 } 524 525 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 526 { 527 sk_peek_offset_bwd(sk, -val); 528 } 529 530 /* 531 * Hashed lists helper routines 532 */ 533 static inline struct sock *sk_entry(const struct hlist_node *node) 534 { 535 return hlist_entry(node, struct sock, sk_node); 536 } 537 538 static inline struct sock *__sk_head(const struct hlist_head *head) 539 { 540 return hlist_entry(head->first, struct sock, sk_node); 541 } 542 543 static inline struct sock *sk_head(const struct hlist_head *head) 544 { 545 return hlist_empty(head) ? NULL : __sk_head(head); 546 } 547 548 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 549 { 550 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 551 } 552 553 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 554 { 555 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 556 } 557 558 static inline struct sock *sk_next(const struct sock *sk) 559 { 560 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 561 } 562 563 static inline struct sock *sk_nulls_next(const struct sock *sk) 564 { 565 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 566 hlist_nulls_entry(sk->sk_nulls_node.next, 567 struct sock, sk_nulls_node) : 568 NULL; 569 } 570 571 static inline bool sk_unhashed(const struct sock *sk) 572 { 573 return hlist_unhashed(&sk->sk_node); 574 } 575 576 static inline bool sk_hashed(const struct sock *sk) 577 { 578 return !sk_unhashed(sk); 579 } 580 581 static inline void sk_node_init(struct hlist_node *node) 582 { 583 node->pprev = NULL; 584 } 585 586 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 587 { 588 node->pprev = NULL; 589 } 590 591 static inline void __sk_del_node(struct sock *sk) 592 { 593 __hlist_del(&sk->sk_node); 594 } 595 596 /* NB: equivalent to hlist_del_init_rcu */ 597 static inline bool __sk_del_node_init(struct sock *sk) 598 { 599 if (sk_hashed(sk)) { 600 __sk_del_node(sk); 601 sk_node_init(&sk->sk_node); 602 return true; 603 } 604 return false; 605 } 606 607 /* Grab socket reference count. This operation is valid only 608 when sk is ALREADY grabbed f.e. it is found in hash table 609 or a list and the lookup is made under lock preventing hash table 610 modifications. 611 */ 612 613 static __always_inline void sock_hold(struct sock *sk) 614 { 615 refcount_inc(&sk->sk_refcnt); 616 } 617 618 /* Ungrab socket in the context, which assumes that socket refcnt 619 cannot hit zero, f.e. it is true in context of any socketcall. 620 */ 621 static __always_inline void __sock_put(struct sock *sk) 622 { 623 refcount_dec(&sk->sk_refcnt); 624 } 625 626 static inline bool sk_del_node_init(struct sock *sk) 627 { 628 bool rc = __sk_del_node_init(sk); 629 630 if (rc) { 631 /* paranoid for a while -acme */ 632 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 633 __sock_put(sk); 634 } 635 return rc; 636 } 637 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 638 639 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 640 { 641 if (sk_hashed(sk)) { 642 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 643 return true; 644 } 645 return false; 646 } 647 648 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 649 { 650 bool rc = __sk_nulls_del_node_init_rcu(sk); 651 652 if (rc) { 653 /* paranoid for a while -acme */ 654 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 655 __sock_put(sk); 656 } 657 return rc; 658 } 659 660 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 661 { 662 hlist_add_head(&sk->sk_node, list); 663 } 664 665 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 666 { 667 sock_hold(sk); 668 __sk_add_node(sk, list); 669 } 670 671 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 672 { 673 sock_hold(sk); 674 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 675 sk->sk_family == AF_INET6) 676 hlist_add_tail_rcu(&sk->sk_node, list); 677 else 678 hlist_add_head_rcu(&sk->sk_node, list); 679 } 680 681 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 682 { 683 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 684 sk->sk_family == AF_INET6) 685 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 686 else 687 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 688 } 689 690 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 691 { 692 sock_hold(sk); 693 __sk_nulls_add_node_rcu(sk, list); 694 } 695 696 static inline void __sk_del_bind_node(struct sock *sk) 697 { 698 __hlist_del(&sk->sk_bind_node); 699 } 700 701 static inline void sk_add_bind_node(struct sock *sk, 702 struct hlist_head *list) 703 { 704 hlist_add_head(&sk->sk_bind_node, list); 705 } 706 707 #define sk_for_each(__sk, list) \ 708 hlist_for_each_entry(__sk, list, sk_node) 709 #define sk_for_each_rcu(__sk, list) \ 710 hlist_for_each_entry_rcu(__sk, list, sk_node) 711 #define sk_nulls_for_each(__sk, node, list) \ 712 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 713 #define sk_nulls_for_each_rcu(__sk, node, list) \ 714 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 715 #define sk_for_each_from(__sk) \ 716 hlist_for_each_entry_from(__sk, sk_node) 717 #define sk_nulls_for_each_from(__sk, node) \ 718 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 719 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 720 #define sk_for_each_safe(__sk, tmp, list) \ 721 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 722 #define sk_for_each_bound(__sk, list) \ 723 hlist_for_each_entry(__sk, list, sk_bind_node) 724 725 /** 726 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 727 * @tpos: the type * to use as a loop cursor. 728 * @pos: the &struct hlist_node to use as a loop cursor. 729 * @head: the head for your list. 730 * @offset: offset of hlist_node within the struct. 731 * 732 */ 733 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 734 for (pos = rcu_dereference((head)->first); \ 735 pos != NULL && \ 736 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 737 pos = rcu_dereference(pos->next)) 738 739 static inline struct user_namespace *sk_user_ns(struct sock *sk) 740 { 741 /* Careful only use this in a context where these parameters 742 * can not change and must all be valid, such as recvmsg from 743 * userspace. 744 */ 745 return sk->sk_socket->file->f_cred->user_ns; 746 } 747 748 /* Sock flags */ 749 enum sock_flags { 750 SOCK_DEAD, 751 SOCK_DONE, 752 SOCK_URGINLINE, 753 SOCK_KEEPOPEN, 754 SOCK_LINGER, 755 SOCK_DESTROY, 756 SOCK_BROADCAST, 757 SOCK_TIMESTAMP, 758 SOCK_ZAPPED, 759 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 760 SOCK_DBG, /* %SO_DEBUG setting */ 761 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 762 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 763 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 764 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 765 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 766 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 767 SOCK_FASYNC, /* fasync() active */ 768 SOCK_RXQ_OVFL, 769 SOCK_ZEROCOPY, /* buffers from userspace */ 770 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 771 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 772 * Will use last 4 bytes of packet sent from 773 * user-space instead. 774 */ 775 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 776 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 777 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 778 }; 779 780 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 781 782 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 783 { 784 nsk->sk_flags = osk->sk_flags; 785 } 786 787 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 788 { 789 __set_bit(flag, &sk->sk_flags); 790 } 791 792 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 793 { 794 __clear_bit(flag, &sk->sk_flags); 795 } 796 797 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 798 { 799 return test_bit(flag, &sk->sk_flags); 800 } 801 802 #ifdef CONFIG_NET 803 extern struct static_key memalloc_socks; 804 static inline int sk_memalloc_socks(void) 805 { 806 return static_key_false(&memalloc_socks); 807 } 808 #else 809 810 static inline int sk_memalloc_socks(void) 811 { 812 return 0; 813 } 814 815 #endif 816 817 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 818 { 819 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 820 } 821 822 static inline void sk_acceptq_removed(struct sock *sk) 823 { 824 sk->sk_ack_backlog--; 825 } 826 827 static inline void sk_acceptq_added(struct sock *sk) 828 { 829 sk->sk_ack_backlog++; 830 } 831 832 static inline bool sk_acceptq_is_full(const struct sock *sk) 833 { 834 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 835 } 836 837 /* 838 * Compute minimal free write space needed to queue new packets. 839 */ 840 static inline int sk_stream_min_wspace(const struct sock *sk) 841 { 842 return sk->sk_wmem_queued >> 1; 843 } 844 845 static inline int sk_stream_wspace(const struct sock *sk) 846 { 847 return sk->sk_sndbuf - sk->sk_wmem_queued; 848 } 849 850 void sk_stream_write_space(struct sock *sk); 851 852 /* OOB backlog add */ 853 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 854 { 855 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 856 skb_dst_force_safe(skb); 857 858 if (!sk->sk_backlog.tail) 859 sk->sk_backlog.head = skb; 860 else 861 sk->sk_backlog.tail->next = skb; 862 863 sk->sk_backlog.tail = skb; 864 skb->next = NULL; 865 } 866 867 /* 868 * Take into account size of receive queue and backlog queue 869 * Do not take into account this skb truesize, 870 * to allow even a single big packet to come. 871 */ 872 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 873 { 874 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 875 876 return qsize > limit; 877 } 878 879 /* The per-socket spinlock must be held here. */ 880 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 881 unsigned int limit) 882 { 883 if (sk_rcvqueues_full(sk, limit)) 884 return -ENOBUFS; 885 886 /* 887 * If the skb was allocated from pfmemalloc reserves, only 888 * allow SOCK_MEMALLOC sockets to use it as this socket is 889 * helping free memory 890 */ 891 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 892 return -ENOMEM; 893 894 __sk_add_backlog(sk, skb); 895 sk->sk_backlog.len += skb->truesize; 896 return 0; 897 } 898 899 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 900 901 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 902 { 903 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 904 return __sk_backlog_rcv(sk, skb); 905 906 return sk->sk_backlog_rcv(sk, skb); 907 } 908 909 static inline void sk_incoming_cpu_update(struct sock *sk) 910 { 911 int cpu = raw_smp_processor_id(); 912 913 if (unlikely(sk->sk_incoming_cpu != cpu)) 914 sk->sk_incoming_cpu = cpu; 915 } 916 917 static inline void sock_rps_record_flow_hash(__u32 hash) 918 { 919 #ifdef CONFIG_RPS 920 struct rps_sock_flow_table *sock_flow_table; 921 922 rcu_read_lock(); 923 sock_flow_table = rcu_dereference(rps_sock_flow_table); 924 rps_record_sock_flow(sock_flow_table, hash); 925 rcu_read_unlock(); 926 #endif 927 } 928 929 static inline void sock_rps_record_flow(const struct sock *sk) 930 { 931 #ifdef CONFIG_RPS 932 if (static_key_false(&rfs_needed)) { 933 /* Reading sk->sk_rxhash might incur an expensive cache line 934 * miss. 935 * 936 * TCP_ESTABLISHED does cover almost all states where RFS 937 * might be useful, and is cheaper [1] than testing : 938 * IPv4: inet_sk(sk)->inet_daddr 939 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 940 * OR an additional socket flag 941 * [1] : sk_state and sk_prot are in the same cache line. 942 */ 943 if (sk->sk_state == TCP_ESTABLISHED) 944 sock_rps_record_flow_hash(sk->sk_rxhash); 945 } 946 #endif 947 } 948 949 static inline void sock_rps_save_rxhash(struct sock *sk, 950 const struct sk_buff *skb) 951 { 952 #ifdef CONFIG_RPS 953 if (unlikely(sk->sk_rxhash != skb->hash)) 954 sk->sk_rxhash = skb->hash; 955 #endif 956 } 957 958 static inline void sock_rps_reset_rxhash(struct sock *sk) 959 { 960 #ifdef CONFIG_RPS 961 sk->sk_rxhash = 0; 962 #endif 963 } 964 965 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 966 ({ int __rc; \ 967 release_sock(__sk); \ 968 __rc = __condition; \ 969 if (!__rc) { \ 970 *(__timeo) = wait_woken(__wait, \ 971 TASK_INTERRUPTIBLE, \ 972 *(__timeo)); \ 973 } \ 974 sched_annotate_sleep(); \ 975 lock_sock(__sk); \ 976 __rc = __condition; \ 977 __rc; \ 978 }) 979 980 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 981 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 982 void sk_stream_wait_close(struct sock *sk, long timeo_p); 983 int sk_stream_error(struct sock *sk, int flags, int err); 984 void sk_stream_kill_queues(struct sock *sk); 985 void sk_set_memalloc(struct sock *sk); 986 void sk_clear_memalloc(struct sock *sk); 987 988 void __sk_flush_backlog(struct sock *sk); 989 990 static inline bool sk_flush_backlog(struct sock *sk) 991 { 992 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 993 __sk_flush_backlog(sk); 994 return true; 995 } 996 return false; 997 } 998 999 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1000 1001 struct request_sock_ops; 1002 struct timewait_sock_ops; 1003 struct inet_hashinfo; 1004 struct raw_hashinfo; 1005 struct smc_hashinfo; 1006 struct module; 1007 1008 /* 1009 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1010 * un-modified. Special care is taken when initializing object to zero. 1011 */ 1012 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1013 { 1014 if (offsetof(struct sock, sk_node.next) != 0) 1015 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1016 memset(&sk->sk_node.pprev, 0, 1017 size - offsetof(struct sock, sk_node.pprev)); 1018 } 1019 1020 /* Networking protocol blocks we attach to sockets. 1021 * socket layer -> transport layer interface 1022 */ 1023 struct proto { 1024 void (*close)(struct sock *sk, 1025 long timeout); 1026 int (*connect)(struct sock *sk, 1027 struct sockaddr *uaddr, 1028 int addr_len); 1029 int (*disconnect)(struct sock *sk, int flags); 1030 1031 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1032 bool kern); 1033 1034 int (*ioctl)(struct sock *sk, int cmd, 1035 unsigned long arg); 1036 int (*init)(struct sock *sk); 1037 void (*destroy)(struct sock *sk); 1038 void (*shutdown)(struct sock *sk, int how); 1039 int (*setsockopt)(struct sock *sk, int level, 1040 int optname, char __user *optval, 1041 unsigned int optlen); 1042 int (*getsockopt)(struct sock *sk, int level, 1043 int optname, char __user *optval, 1044 int __user *option); 1045 void (*keepalive)(struct sock *sk, int valbool); 1046 #ifdef CONFIG_COMPAT 1047 int (*compat_setsockopt)(struct sock *sk, 1048 int level, 1049 int optname, char __user *optval, 1050 unsigned int optlen); 1051 int (*compat_getsockopt)(struct sock *sk, 1052 int level, 1053 int optname, char __user *optval, 1054 int __user *option); 1055 int (*compat_ioctl)(struct sock *sk, 1056 unsigned int cmd, unsigned long arg); 1057 #endif 1058 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1059 size_t len); 1060 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1061 size_t len, int noblock, int flags, 1062 int *addr_len); 1063 int (*sendpage)(struct sock *sk, struct page *page, 1064 int offset, size_t size, int flags); 1065 int (*bind)(struct sock *sk, 1066 struct sockaddr *uaddr, int addr_len); 1067 1068 int (*backlog_rcv) (struct sock *sk, 1069 struct sk_buff *skb); 1070 1071 void (*release_cb)(struct sock *sk); 1072 1073 /* Keeping track of sk's, looking them up, and port selection methods. */ 1074 int (*hash)(struct sock *sk); 1075 void (*unhash)(struct sock *sk); 1076 void (*rehash)(struct sock *sk); 1077 int (*get_port)(struct sock *sk, unsigned short snum); 1078 1079 /* Keeping track of sockets in use */ 1080 #ifdef CONFIG_PROC_FS 1081 unsigned int inuse_idx; 1082 #endif 1083 1084 bool (*stream_memory_free)(const struct sock *sk); 1085 /* Memory pressure */ 1086 void (*enter_memory_pressure)(struct sock *sk); 1087 void (*leave_memory_pressure)(struct sock *sk); 1088 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1089 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1090 /* 1091 * Pressure flag: try to collapse. 1092 * Technical note: it is used by multiple contexts non atomically. 1093 * All the __sk_mem_schedule() is of this nature: accounting 1094 * is strict, actions are advisory and have some latency. 1095 */ 1096 unsigned long *memory_pressure; 1097 long *sysctl_mem; 1098 int *sysctl_wmem; 1099 int *sysctl_rmem; 1100 int max_header; 1101 bool no_autobind; 1102 1103 struct kmem_cache *slab; 1104 unsigned int obj_size; 1105 int slab_flags; 1106 1107 struct percpu_counter *orphan_count; 1108 1109 struct request_sock_ops *rsk_prot; 1110 struct timewait_sock_ops *twsk_prot; 1111 1112 union { 1113 struct inet_hashinfo *hashinfo; 1114 struct udp_table *udp_table; 1115 struct raw_hashinfo *raw_hash; 1116 struct smc_hashinfo *smc_hash; 1117 } h; 1118 1119 struct module *owner; 1120 1121 char name[32]; 1122 1123 struct list_head node; 1124 #ifdef SOCK_REFCNT_DEBUG 1125 atomic_t socks; 1126 #endif 1127 int (*diag_destroy)(struct sock *sk, int err); 1128 }; 1129 1130 int proto_register(struct proto *prot, int alloc_slab); 1131 void proto_unregister(struct proto *prot); 1132 1133 #ifdef SOCK_REFCNT_DEBUG 1134 static inline void sk_refcnt_debug_inc(struct sock *sk) 1135 { 1136 atomic_inc(&sk->sk_prot->socks); 1137 } 1138 1139 static inline void sk_refcnt_debug_dec(struct sock *sk) 1140 { 1141 atomic_dec(&sk->sk_prot->socks); 1142 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1143 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1144 } 1145 1146 static inline void sk_refcnt_debug_release(const struct sock *sk) 1147 { 1148 if (refcount_read(&sk->sk_refcnt) != 1) 1149 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1150 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1151 } 1152 #else /* SOCK_REFCNT_DEBUG */ 1153 #define sk_refcnt_debug_inc(sk) do { } while (0) 1154 #define sk_refcnt_debug_dec(sk) do { } while (0) 1155 #define sk_refcnt_debug_release(sk) do { } while (0) 1156 #endif /* SOCK_REFCNT_DEBUG */ 1157 1158 static inline bool sk_stream_memory_free(const struct sock *sk) 1159 { 1160 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1161 return false; 1162 1163 return sk->sk_prot->stream_memory_free ? 1164 sk->sk_prot->stream_memory_free(sk) : true; 1165 } 1166 1167 static inline bool sk_stream_is_writeable(const struct sock *sk) 1168 { 1169 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1170 sk_stream_memory_free(sk); 1171 } 1172 1173 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1174 struct cgroup *ancestor) 1175 { 1176 #ifdef CONFIG_SOCK_CGROUP_DATA 1177 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1178 ancestor); 1179 #else 1180 return -ENOTSUPP; 1181 #endif 1182 } 1183 1184 static inline bool sk_has_memory_pressure(const struct sock *sk) 1185 { 1186 return sk->sk_prot->memory_pressure != NULL; 1187 } 1188 1189 static inline bool sk_under_memory_pressure(const struct sock *sk) 1190 { 1191 if (!sk->sk_prot->memory_pressure) 1192 return false; 1193 1194 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1195 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1196 return true; 1197 1198 return !!*sk->sk_prot->memory_pressure; 1199 } 1200 1201 static inline long 1202 sk_memory_allocated(const struct sock *sk) 1203 { 1204 return atomic_long_read(sk->sk_prot->memory_allocated); 1205 } 1206 1207 static inline long 1208 sk_memory_allocated_add(struct sock *sk, int amt) 1209 { 1210 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1211 } 1212 1213 static inline void 1214 sk_memory_allocated_sub(struct sock *sk, int amt) 1215 { 1216 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1217 } 1218 1219 static inline void sk_sockets_allocated_dec(struct sock *sk) 1220 { 1221 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1222 } 1223 1224 static inline void sk_sockets_allocated_inc(struct sock *sk) 1225 { 1226 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1227 } 1228 1229 static inline int 1230 sk_sockets_allocated_read_positive(struct sock *sk) 1231 { 1232 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1233 } 1234 1235 static inline int 1236 proto_sockets_allocated_sum_positive(struct proto *prot) 1237 { 1238 return percpu_counter_sum_positive(prot->sockets_allocated); 1239 } 1240 1241 static inline long 1242 proto_memory_allocated(struct proto *prot) 1243 { 1244 return atomic_long_read(prot->memory_allocated); 1245 } 1246 1247 static inline bool 1248 proto_memory_pressure(struct proto *prot) 1249 { 1250 if (!prot->memory_pressure) 1251 return false; 1252 return !!*prot->memory_pressure; 1253 } 1254 1255 1256 #ifdef CONFIG_PROC_FS 1257 /* Called with local bh disabled */ 1258 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1259 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1260 #else 1261 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1262 int inc) 1263 { 1264 } 1265 #endif 1266 1267 1268 /* With per-bucket locks this operation is not-atomic, so that 1269 * this version is not worse. 1270 */ 1271 static inline int __sk_prot_rehash(struct sock *sk) 1272 { 1273 sk->sk_prot->unhash(sk); 1274 return sk->sk_prot->hash(sk); 1275 } 1276 1277 /* About 10 seconds */ 1278 #define SOCK_DESTROY_TIME (10*HZ) 1279 1280 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1281 #define PROT_SOCK 1024 1282 1283 #define SHUTDOWN_MASK 3 1284 #define RCV_SHUTDOWN 1 1285 #define SEND_SHUTDOWN 2 1286 1287 #define SOCK_SNDBUF_LOCK 1 1288 #define SOCK_RCVBUF_LOCK 2 1289 #define SOCK_BINDADDR_LOCK 4 1290 #define SOCK_BINDPORT_LOCK 8 1291 1292 struct socket_alloc { 1293 struct socket socket; 1294 struct inode vfs_inode; 1295 }; 1296 1297 static inline struct socket *SOCKET_I(struct inode *inode) 1298 { 1299 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1300 } 1301 1302 static inline struct inode *SOCK_INODE(struct socket *socket) 1303 { 1304 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1305 } 1306 1307 /* 1308 * Functions for memory accounting 1309 */ 1310 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1311 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1312 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1313 void __sk_mem_reclaim(struct sock *sk, int amount); 1314 1315 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1316 * do not necessarily have 16x time more memory than 4KB ones. 1317 */ 1318 #define SK_MEM_QUANTUM 4096 1319 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1320 #define SK_MEM_SEND 0 1321 #define SK_MEM_RECV 1 1322 1323 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1324 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1325 { 1326 long val = sk->sk_prot->sysctl_mem[index]; 1327 1328 #if PAGE_SIZE > SK_MEM_QUANTUM 1329 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1330 #elif PAGE_SIZE < SK_MEM_QUANTUM 1331 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1332 #endif 1333 return val; 1334 } 1335 1336 static inline int sk_mem_pages(int amt) 1337 { 1338 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1339 } 1340 1341 static inline bool sk_has_account(struct sock *sk) 1342 { 1343 /* return true if protocol supports memory accounting */ 1344 return !!sk->sk_prot->memory_allocated; 1345 } 1346 1347 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1348 { 1349 if (!sk_has_account(sk)) 1350 return true; 1351 return size <= sk->sk_forward_alloc || 1352 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1353 } 1354 1355 static inline bool 1356 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1357 { 1358 if (!sk_has_account(sk)) 1359 return true; 1360 return size<= sk->sk_forward_alloc || 1361 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1362 skb_pfmemalloc(skb); 1363 } 1364 1365 static inline void sk_mem_reclaim(struct sock *sk) 1366 { 1367 if (!sk_has_account(sk)) 1368 return; 1369 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1370 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1371 } 1372 1373 static inline void sk_mem_reclaim_partial(struct sock *sk) 1374 { 1375 if (!sk_has_account(sk)) 1376 return; 1377 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1378 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1379 } 1380 1381 static inline void sk_mem_charge(struct sock *sk, int size) 1382 { 1383 if (!sk_has_account(sk)) 1384 return; 1385 sk->sk_forward_alloc -= size; 1386 } 1387 1388 static inline void sk_mem_uncharge(struct sock *sk, int size) 1389 { 1390 if (!sk_has_account(sk)) 1391 return; 1392 sk->sk_forward_alloc += size; 1393 1394 /* Avoid a possible overflow. 1395 * TCP send queues can make this happen, if sk_mem_reclaim() 1396 * is not called and more than 2 GBytes are released at once. 1397 * 1398 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1399 * no need to hold that much forward allocation anyway. 1400 */ 1401 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1402 __sk_mem_reclaim(sk, 1 << 20); 1403 } 1404 1405 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1406 { 1407 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1408 sk->sk_wmem_queued -= skb->truesize; 1409 sk_mem_uncharge(sk, skb->truesize); 1410 __kfree_skb(skb); 1411 } 1412 1413 static inline void sock_release_ownership(struct sock *sk) 1414 { 1415 if (sk->sk_lock.owned) { 1416 sk->sk_lock.owned = 0; 1417 1418 /* The sk_lock has mutex_unlock() semantics: */ 1419 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1420 } 1421 } 1422 1423 /* 1424 * Macro so as to not evaluate some arguments when 1425 * lockdep is not enabled. 1426 * 1427 * Mark both the sk_lock and the sk_lock.slock as a 1428 * per-address-family lock class. 1429 */ 1430 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1431 do { \ 1432 sk->sk_lock.owned = 0; \ 1433 init_waitqueue_head(&sk->sk_lock.wq); \ 1434 spin_lock_init(&(sk)->sk_lock.slock); \ 1435 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1436 sizeof((sk)->sk_lock)); \ 1437 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1438 (skey), (sname)); \ 1439 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1440 } while (0) 1441 1442 #ifdef CONFIG_LOCKDEP 1443 static inline bool lockdep_sock_is_held(const struct sock *csk) 1444 { 1445 struct sock *sk = (struct sock *)csk; 1446 1447 return lockdep_is_held(&sk->sk_lock) || 1448 lockdep_is_held(&sk->sk_lock.slock); 1449 } 1450 #endif 1451 1452 void lock_sock_nested(struct sock *sk, int subclass); 1453 1454 static inline void lock_sock(struct sock *sk) 1455 { 1456 lock_sock_nested(sk, 0); 1457 } 1458 1459 void release_sock(struct sock *sk); 1460 1461 /* BH context may only use the following locking interface. */ 1462 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1463 #define bh_lock_sock_nested(__sk) \ 1464 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1465 SINGLE_DEPTH_NESTING) 1466 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1467 1468 bool lock_sock_fast(struct sock *sk); 1469 /** 1470 * unlock_sock_fast - complement of lock_sock_fast 1471 * @sk: socket 1472 * @slow: slow mode 1473 * 1474 * fast unlock socket for user context. 1475 * If slow mode is on, we call regular release_sock() 1476 */ 1477 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1478 { 1479 if (slow) 1480 release_sock(sk); 1481 else 1482 spin_unlock_bh(&sk->sk_lock.slock); 1483 } 1484 1485 /* Used by processes to "lock" a socket state, so that 1486 * interrupts and bottom half handlers won't change it 1487 * from under us. It essentially blocks any incoming 1488 * packets, so that we won't get any new data or any 1489 * packets that change the state of the socket. 1490 * 1491 * While locked, BH processing will add new packets to 1492 * the backlog queue. This queue is processed by the 1493 * owner of the socket lock right before it is released. 1494 * 1495 * Since ~2.3.5 it is also exclusive sleep lock serializing 1496 * accesses from user process context. 1497 */ 1498 1499 static inline void sock_owned_by_me(const struct sock *sk) 1500 { 1501 #ifdef CONFIG_LOCKDEP 1502 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1503 #endif 1504 } 1505 1506 static inline bool sock_owned_by_user(const struct sock *sk) 1507 { 1508 sock_owned_by_me(sk); 1509 return sk->sk_lock.owned; 1510 } 1511 1512 /* no reclassification while locks are held */ 1513 static inline bool sock_allow_reclassification(const struct sock *csk) 1514 { 1515 struct sock *sk = (struct sock *)csk; 1516 1517 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1518 } 1519 1520 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1521 struct proto *prot, int kern); 1522 void sk_free(struct sock *sk); 1523 void sk_destruct(struct sock *sk); 1524 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1525 void sk_free_unlock_clone(struct sock *sk); 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, bool); 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 (refcount_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 WARN_ON(parent->sk); 1701 write_lock_bh(&sk->sk_callback_lock); 1702 sk->sk_wq = parent->wq; 1703 parent->sk = sk; 1704 sk_set_socket(sk, parent); 1705 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1706 security_sock_graft(sk, parent); 1707 write_unlock_bh(&sk->sk_callback_lock); 1708 } 1709 1710 kuid_t sock_i_uid(struct sock *sk); 1711 unsigned long sock_i_ino(struct sock *sk); 1712 1713 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1714 { 1715 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1716 } 1717 1718 static inline u32 net_tx_rndhash(void) 1719 { 1720 u32 v = prandom_u32(); 1721 1722 return v ?: 1; 1723 } 1724 1725 static inline void sk_set_txhash(struct sock *sk) 1726 { 1727 sk->sk_txhash = net_tx_rndhash(); 1728 } 1729 1730 static inline void sk_rethink_txhash(struct sock *sk) 1731 { 1732 if (sk->sk_txhash) 1733 sk_set_txhash(sk); 1734 } 1735 1736 static inline struct dst_entry * 1737 __sk_dst_get(struct sock *sk) 1738 { 1739 return rcu_dereference_check(sk->sk_dst_cache, 1740 lockdep_sock_is_held(sk)); 1741 } 1742 1743 static inline struct dst_entry * 1744 sk_dst_get(struct sock *sk) 1745 { 1746 struct dst_entry *dst; 1747 1748 rcu_read_lock(); 1749 dst = rcu_dereference(sk->sk_dst_cache); 1750 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1751 dst = NULL; 1752 rcu_read_unlock(); 1753 return dst; 1754 } 1755 1756 static inline void dst_negative_advice(struct sock *sk) 1757 { 1758 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1759 1760 sk_rethink_txhash(sk); 1761 1762 if (dst && dst->ops->negative_advice) { 1763 ndst = dst->ops->negative_advice(dst); 1764 1765 if (ndst != dst) { 1766 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1767 sk_tx_queue_clear(sk); 1768 sk->sk_dst_pending_confirm = 0; 1769 } 1770 } 1771 } 1772 1773 static inline void 1774 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1775 { 1776 struct dst_entry *old_dst; 1777 1778 sk_tx_queue_clear(sk); 1779 sk->sk_dst_pending_confirm = 0; 1780 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1781 lockdep_sock_is_held(sk)); 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 sk->sk_dst_pending_confirm = 0; 1793 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1794 dst_release(old_dst); 1795 } 1796 1797 static inline void 1798 __sk_dst_reset(struct sock *sk) 1799 { 1800 __sk_dst_set(sk, NULL); 1801 } 1802 1803 static inline void 1804 sk_dst_reset(struct sock *sk) 1805 { 1806 sk_dst_set(sk, NULL); 1807 } 1808 1809 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1810 1811 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1812 1813 static inline void sk_dst_confirm(struct sock *sk) 1814 { 1815 if (!sk->sk_dst_pending_confirm) 1816 sk->sk_dst_pending_confirm = 1; 1817 } 1818 1819 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1820 { 1821 if (skb_get_dst_pending_confirm(skb)) { 1822 struct sock *sk = skb->sk; 1823 unsigned long now = jiffies; 1824 1825 /* avoid dirtying neighbour */ 1826 if (n->confirmed != now) 1827 n->confirmed = now; 1828 if (sk && sk->sk_dst_pending_confirm) 1829 sk->sk_dst_pending_confirm = 0; 1830 } 1831 } 1832 1833 bool sk_mc_loop(struct sock *sk); 1834 1835 static inline bool sk_can_gso(const struct sock *sk) 1836 { 1837 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1838 } 1839 1840 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1841 1842 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1843 { 1844 sk->sk_route_nocaps |= flags; 1845 sk->sk_route_caps &= ~flags; 1846 } 1847 1848 static inline bool sk_check_csum_caps(struct sock *sk) 1849 { 1850 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1851 (sk->sk_family == PF_INET && 1852 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1853 (sk->sk_family == PF_INET6 && 1854 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1855 } 1856 1857 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1858 struct iov_iter *from, char *to, 1859 int copy, int offset) 1860 { 1861 if (skb->ip_summed == CHECKSUM_NONE) { 1862 __wsum csum = 0; 1863 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1864 return -EFAULT; 1865 skb->csum = csum_block_add(skb->csum, csum, offset); 1866 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1867 if (!copy_from_iter_full_nocache(to, copy, from)) 1868 return -EFAULT; 1869 } else if (!copy_from_iter_full(to, copy, from)) 1870 return -EFAULT; 1871 1872 return 0; 1873 } 1874 1875 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1876 struct iov_iter *from, int copy) 1877 { 1878 int err, offset = skb->len; 1879 1880 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1881 copy, offset); 1882 if (err) 1883 __skb_trim(skb, offset); 1884 1885 return err; 1886 } 1887 1888 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1889 struct sk_buff *skb, 1890 struct page *page, 1891 int off, int copy) 1892 { 1893 int err; 1894 1895 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1896 copy, skb->len); 1897 if (err) 1898 return err; 1899 1900 skb->len += copy; 1901 skb->data_len += copy; 1902 skb->truesize += copy; 1903 sk->sk_wmem_queued += copy; 1904 sk_mem_charge(sk, copy); 1905 return 0; 1906 } 1907 1908 /** 1909 * sk_wmem_alloc_get - returns write allocations 1910 * @sk: socket 1911 * 1912 * Returns sk_wmem_alloc minus initial offset of one 1913 */ 1914 static inline int sk_wmem_alloc_get(const struct sock *sk) 1915 { 1916 return refcount_read(&sk->sk_wmem_alloc) - 1; 1917 } 1918 1919 /** 1920 * sk_rmem_alloc_get - returns read allocations 1921 * @sk: socket 1922 * 1923 * Returns sk_rmem_alloc 1924 */ 1925 static inline int sk_rmem_alloc_get(const struct sock *sk) 1926 { 1927 return atomic_read(&sk->sk_rmem_alloc); 1928 } 1929 1930 /** 1931 * sk_has_allocations - check if allocations are outstanding 1932 * @sk: socket 1933 * 1934 * Returns true if socket has write or read allocations 1935 */ 1936 static inline bool sk_has_allocations(const struct sock *sk) 1937 { 1938 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1939 } 1940 1941 /** 1942 * skwq_has_sleeper - check if there are any waiting processes 1943 * @wq: struct socket_wq 1944 * 1945 * Returns true if socket_wq has waiting processes 1946 * 1947 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1948 * barrier call. They were added due to the race found within the tcp code. 1949 * 1950 * Consider following tcp code paths:: 1951 * 1952 * CPU1 CPU2 1953 * sys_select receive packet 1954 * ... ... 1955 * __add_wait_queue update tp->rcv_nxt 1956 * ... ... 1957 * tp->rcv_nxt check sock_def_readable 1958 * ... { 1959 * schedule rcu_read_lock(); 1960 * wq = rcu_dereference(sk->sk_wq); 1961 * if (wq && waitqueue_active(&wq->wait)) 1962 * wake_up_interruptible(&wq->wait) 1963 * ... 1964 * } 1965 * 1966 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1967 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1968 * could then endup calling schedule and sleep forever if there are no more 1969 * data on the socket. 1970 * 1971 */ 1972 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1973 { 1974 return wq && wq_has_sleeper(&wq->wait); 1975 } 1976 1977 /** 1978 * sock_poll_wait - place memory barrier behind the poll_wait call. 1979 * @filp: file 1980 * @wait_address: socket wait queue 1981 * @p: poll_table 1982 * 1983 * See the comments in the wq_has_sleeper function. 1984 */ 1985 static inline void sock_poll_wait(struct file *filp, 1986 wait_queue_head_t *wait_address, poll_table *p) 1987 { 1988 if (!poll_does_not_wait(p) && wait_address) { 1989 poll_wait(filp, wait_address, p); 1990 /* We need to be sure we are in sync with the 1991 * socket flags modification. 1992 * 1993 * This memory barrier is paired in the wq_has_sleeper. 1994 */ 1995 smp_mb(); 1996 } 1997 } 1998 1999 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2000 { 2001 if (sk->sk_txhash) { 2002 skb->l4_hash = 1; 2003 skb->hash = sk->sk_txhash; 2004 } 2005 } 2006 2007 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2008 2009 /* 2010 * Queue a received datagram if it will fit. Stream and sequenced 2011 * protocols can't normally use this as they need to fit buffers in 2012 * and play with them. 2013 * 2014 * Inlined as it's very short and called for pretty much every 2015 * packet ever received. 2016 */ 2017 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2018 { 2019 skb_orphan(skb); 2020 skb->sk = sk; 2021 skb->destructor = sock_rfree; 2022 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2023 sk_mem_charge(sk, skb->truesize); 2024 } 2025 2026 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2027 unsigned long expires); 2028 2029 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2030 2031 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2032 struct sk_buff *skb, unsigned int flags, 2033 void (*destructor)(struct sock *sk, 2034 struct sk_buff *skb)); 2035 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2036 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2037 2038 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2039 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2040 2041 /* 2042 * Recover an error report and clear atomically 2043 */ 2044 2045 static inline int sock_error(struct sock *sk) 2046 { 2047 int err; 2048 if (likely(!sk->sk_err)) 2049 return 0; 2050 err = xchg(&sk->sk_err, 0); 2051 return -err; 2052 } 2053 2054 static inline unsigned long sock_wspace(struct sock *sk) 2055 { 2056 int amt = 0; 2057 2058 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2059 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2060 if (amt < 0) 2061 amt = 0; 2062 } 2063 return amt; 2064 } 2065 2066 /* Note: 2067 * We use sk->sk_wq_raw, from contexts knowing this 2068 * pointer is not NULL and cannot disappear/change. 2069 */ 2070 static inline void sk_set_bit(int nr, struct sock *sk) 2071 { 2072 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2073 !sock_flag(sk, SOCK_FASYNC)) 2074 return; 2075 2076 set_bit(nr, &sk->sk_wq_raw->flags); 2077 } 2078 2079 static inline void sk_clear_bit(int nr, struct sock *sk) 2080 { 2081 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2082 !sock_flag(sk, SOCK_FASYNC)) 2083 return; 2084 2085 clear_bit(nr, &sk->sk_wq_raw->flags); 2086 } 2087 2088 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2089 { 2090 if (sock_flag(sk, SOCK_FASYNC)) { 2091 rcu_read_lock(); 2092 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2093 rcu_read_unlock(); 2094 } 2095 } 2096 2097 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2098 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2099 * Note: for send buffers, TCP works better if we can build two skbs at 2100 * minimum. 2101 */ 2102 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2103 2104 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2105 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2106 2107 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2108 { 2109 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2110 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2111 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2112 } 2113 } 2114 2115 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2116 bool force_schedule); 2117 2118 /** 2119 * sk_page_frag - return an appropriate page_frag 2120 * @sk: socket 2121 * 2122 * If socket allocation mode allows current thread to sleep, it means its 2123 * safe to use the per task page_frag instead of the per socket one. 2124 */ 2125 static inline struct page_frag *sk_page_frag(struct sock *sk) 2126 { 2127 if (gfpflags_allow_blocking(sk->sk_allocation)) 2128 return ¤t->task_frag; 2129 2130 return &sk->sk_frag; 2131 } 2132 2133 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2134 2135 /* 2136 * Default write policy as shown to user space via poll/select/SIGIO 2137 */ 2138 static inline bool sock_writeable(const struct sock *sk) 2139 { 2140 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2141 } 2142 2143 static inline gfp_t gfp_any(void) 2144 { 2145 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2146 } 2147 2148 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2149 { 2150 return noblock ? 0 : sk->sk_rcvtimeo; 2151 } 2152 2153 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2154 { 2155 return noblock ? 0 : sk->sk_sndtimeo; 2156 } 2157 2158 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2159 { 2160 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2161 } 2162 2163 /* Alas, with timeout socket operations are not restartable. 2164 * Compare this to poll(). 2165 */ 2166 static inline int sock_intr_errno(long timeo) 2167 { 2168 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2169 } 2170 2171 struct sock_skb_cb { 2172 u32 dropcount; 2173 }; 2174 2175 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2176 * using skb->cb[] would keep using it directly and utilize its 2177 * alignement guarantee. 2178 */ 2179 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2180 sizeof(struct sock_skb_cb))) 2181 2182 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2183 SOCK_SKB_CB_OFFSET)) 2184 2185 #define sock_skb_cb_check_size(size) \ 2186 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2187 2188 static inline void 2189 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2190 { 2191 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2192 atomic_read(&sk->sk_drops) : 0; 2193 } 2194 2195 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2196 { 2197 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2198 2199 atomic_add(segs, &sk->sk_drops); 2200 } 2201 2202 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2203 struct sk_buff *skb); 2204 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2205 struct sk_buff *skb); 2206 2207 static inline void 2208 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2209 { 2210 ktime_t kt = skb->tstamp; 2211 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2212 2213 /* 2214 * generate control messages if 2215 * - receive time stamping in software requested 2216 * - software time stamp available and wanted 2217 * - hardware time stamps available and wanted 2218 */ 2219 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2220 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2221 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2222 (hwtstamps->hwtstamp && 2223 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2224 __sock_recv_timestamp(msg, sk, skb); 2225 else 2226 sk->sk_stamp = kt; 2227 2228 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2229 __sock_recv_wifi_status(msg, sk, skb); 2230 } 2231 2232 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2233 struct sk_buff *skb); 2234 2235 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2236 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2237 struct sk_buff *skb) 2238 { 2239 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2240 (1UL << SOCK_RCVTSTAMP)) 2241 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2242 SOF_TIMESTAMPING_RAW_HARDWARE) 2243 2244 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2245 __sock_recv_ts_and_drops(msg, sk, skb); 2246 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2247 sk->sk_stamp = skb->tstamp; 2248 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2249 sk->sk_stamp = 0; 2250 } 2251 2252 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2253 2254 /** 2255 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2256 * @sk: socket sending this packet 2257 * @tsflags: timestamping flags to use 2258 * @tx_flags: completed with instructions for time stamping 2259 * 2260 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2261 */ 2262 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2263 __u8 *tx_flags) 2264 { 2265 if (unlikely(tsflags)) 2266 __sock_tx_timestamp(tsflags, tx_flags); 2267 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2268 *tx_flags |= SKBTX_WIFI_STATUS; 2269 } 2270 2271 /** 2272 * sk_eat_skb - Release a skb if it is no longer needed 2273 * @sk: socket to eat this skb from 2274 * @skb: socket buffer to eat 2275 * 2276 * This routine must be called with interrupts disabled or with the socket 2277 * locked so that the sk_buff queue operation is ok. 2278 */ 2279 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2280 { 2281 __skb_unlink(skb, &sk->sk_receive_queue); 2282 __kfree_skb(skb); 2283 } 2284 2285 static inline 2286 struct net *sock_net(const struct sock *sk) 2287 { 2288 return read_pnet(&sk->sk_net); 2289 } 2290 2291 static inline 2292 void sock_net_set(struct sock *sk, struct net *net) 2293 { 2294 write_pnet(&sk->sk_net, net); 2295 } 2296 2297 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2298 { 2299 if (skb->sk) { 2300 struct sock *sk = skb->sk; 2301 2302 skb->destructor = NULL; 2303 skb->sk = NULL; 2304 return sk; 2305 } 2306 return NULL; 2307 } 2308 2309 /* This helper checks if a socket is a full socket, 2310 * ie _not_ a timewait or request socket. 2311 */ 2312 static inline bool sk_fullsock(const struct sock *sk) 2313 { 2314 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2315 } 2316 2317 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2318 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2319 */ 2320 static inline bool sk_listener(const struct sock *sk) 2321 { 2322 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2323 } 2324 2325 /** 2326 * sk_state_load - read sk->sk_state for lockless contexts 2327 * @sk: socket pointer 2328 * 2329 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2330 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2331 */ 2332 static inline int sk_state_load(const struct sock *sk) 2333 { 2334 return smp_load_acquire(&sk->sk_state); 2335 } 2336 2337 /** 2338 * sk_state_store - update sk->sk_state 2339 * @sk: socket pointer 2340 * @newstate: new state 2341 * 2342 * Paired with sk_state_load(). Should be used in contexts where 2343 * state change might impact lockless readers. 2344 */ 2345 static inline void sk_state_store(struct sock *sk, int newstate) 2346 { 2347 smp_store_release(&sk->sk_state, newstate); 2348 } 2349 2350 void sock_enable_timestamp(struct sock *sk, int flag); 2351 int sock_get_timestamp(struct sock *, struct timeval __user *); 2352 int sock_get_timestampns(struct sock *, struct timespec __user *); 2353 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2354 int type); 2355 2356 bool sk_ns_capable(const struct sock *sk, 2357 struct user_namespace *user_ns, int cap); 2358 bool sk_capable(const struct sock *sk, int cap); 2359 bool sk_net_capable(const struct sock *sk, int cap); 2360 2361 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2362 2363 extern __u32 sysctl_wmem_max; 2364 extern __u32 sysctl_rmem_max; 2365 2366 extern int sysctl_tstamp_allow_data; 2367 extern int sysctl_optmem_max; 2368 2369 extern __u32 sysctl_wmem_default; 2370 extern __u32 sysctl_rmem_default; 2371 2372 #endif /* _SOCK_H */ 2373