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_kern_sock: True if sock is using kernel lock classes 240 * @sk_rcvbuf: size of receive buffer in bytes 241 * @sk_wq: sock wait queue and async head 242 * @sk_rx_dst: receive input route used by early demux 243 * @sk_dst_cache: destination cache 244 * @sk_dst_pending_confirm: need to confirm neighbour 245 * @sk_policy: flow policy 246 * @sk_receive_queue: incoming packets 247 * @sk_wmem_alloc: transmit queue bytes committed 248 * @sk_write_queue: Packet sending queue 249 * @sk_omem_alloc: "o" is "option" or "other" 250 * @sk_wmem_queued: persistent queue size 251 * @sk_forward_alloc: space allocated forward 252 * @sk_napi_id: id of the last napi context to receive data for sk 253 * @sk_ll_usec: usecs to busypoll when there is no data 254 * @sk_allocation: allocation mode 255 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 256 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 257 * @sk_sndbuf: size of send buffer in bytes 258 * @sk_padding: unused element for alignment 259 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 260 * @sk_no_check_rx: allow zero checksum in RX packets 261 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 262 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 263 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 264 * @sk_gso_max_size: Maximum GSO segment size to build 265 * @sk_gso_max_segs: Maximum number of GSO segments 266 * @sk_lingertime: %SO_LINGER l_linger setting 267 * @sk_backlog: always used with the per-socket spinlock held 268 * @sk_callback_lock: used with the callbacks in the end of this struct 269 * @sk_error_queue: rarely used 270 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 271 * IPV6_ADDRFORM for instance) 272 * @sk_err: last error 273 * @sk_err_soft: errors that don't cause failure but are the cause of a 274 * persistent failure not just 'timed out' 275 * @sk_drops: raw/udp drops counter 276 * @sk_ack_backlog: current listen backlog 277 * @sk_max_ack_backlog: listen backlog set in listen() 278 * @sk_priority: %SO_PRIORITY setting 279 * @sk_type: socket type (%SOCK_STREAM, etc) 280 * @sk_protocol: which protocol this socket belongs in this network family 281 * @sk_peer_pid: &struct pid for this socket's peer 282 * @sk_peer_cred: %SO_PEERCRED setting 283 * @sk_rcvlowat: %SO_RCVLOWAT setting 284 * @sk_rcvtimeo: %SO_RCVTIMEO setting 285 * @sk_sndtimeo: %SO_SNDTIMEO setting 286 * @sk_txhash: computed flow hash for use on transmit 287 * @sk_filter: socket filtering instructions 288 * @sk_timer: sock cleanup timer 289 * @sk_stamp: time stamp of last packet received 290 * @sk_tsflags: SO_TIMESTAMPING socket options 291 * @sk_tskey: counter to disambiguate concurrent tstamp requests 292 * @sk_socket: Identd and reporting IO signals 293 * @sk_user_data: RPC layer private data 294 * @sk_frag: cached page frag 295 * @sk_peek_off: current peek_offset value 296 * @sk_send_head: front of stuff to transmit 297 * @sk_security: used by security modules 298 * @sk_mark: generic packet mark 299 * @sk_cgrp_data: cgroup data for this cgroup 300 * @sk_memcg: this socket's memory cgroup association 301 * @sk_write_pending: a write to stream socket waits to start 302 * @sk_state_change: callback to indicate change in the state of the sock 303 * @sk_data_ready: callback to indicate there is data to be processed 304 * @sk_write_space: callback to indicate there is bf sending space available 305 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 306 * @sk_backlog_rcv: callback to process the backlog 307 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 308 * @sk_reuseport_cb: reuseport group container 309 * @sk_rcu: used during RCU grace period 310 */ 311 struct sock { 312 /* 313 * Now struct inet_timewait_sock also uses sock_common, so please just 314 * don't add nothing before this first member (__sk_common) --acme 315 */ 316 struct sock_common __sk_common; 317 #define sk_node __sk_common.skc_node 318 #define sk_nulls_node __sk_common.skc_nulls_node 319 #define sk_refcnt __sk_common.skc_refcnt 320 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 321 322 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 323 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 324 #define sk_hash __sk_common.skc_hash 325 #define sk_portpair __sk_common.skc_portpair 326 #define sk_num __sk_common.skc_num 327 #define sk_dport __sk_common.skc_dport 328 #define sk_addrpair __sk_common.skc_addrpair 329 #define sk_daddr __sk_common.skc_daddr 330 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 331 #define sk_family __sk_common.skc_family 332 #define sk_state __sk_common.skc_state 333 #define sk_reuse __sk_common.skc_reuse 334 #define sk_reuseport __sk_common.skc_reuseport 335 #define sk_ipv6only __sk_common.skc_ipv6only 336 #define sk_net_refcnt __sk_common.skc_net_refcnt 337 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 338 #define sk_bind_node __sk_common.skc_bind_node 339 #define sk_prot __sk_common.skc_prot 340 #define sk_net __sk_common.skc_net 341 #define sk_v6_daddr __sk_common.skc_v6_daddr 342 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 343 #define sk_cookie __sk_common.skc_cookie 344 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 345 #define sk_flags __sk_common.skc_flags 346 #define sk_rxhash __sk_common.skc_rxhash 347 348 socket_lock_t sk_lock; 349 atomic_t sk_drops; 350 int sk_rcvlowat; 351 struct sk_buff_head sk_error_queue; 352 struct sk_buff_head sk_receive_queue; 353 /* 354 * The backlog queue is special, it is always used with 355 * the per-socket spinlock held and requires low latency 356 * access. Therefore we special case it's implementation. 357 * Note : rmem_alloc is in this structure to fill a hole 358 * on 64bit arches, not because its logically part of 359 * backlog. 360 */ 361 struct { 362 atomic_t rmem_alloc; 363 int len; 364 struct sk_buff *head; 365 struct sk_buff *tail; 366 } sk_backlog; 367 #define sk_rmem_alloc sk_backlog.rmem_alloc 368 369 int sk_forward_alloc; 370 #ifdef CONFIG_NET_RX_BUSY_POLL 371 unsigned int sk_ll_usec; 372 /* ===== mostly read cache line ===== */ 373 unsigned int sk_napi_id; 374 #endif 375 int sk_rcvbuf; 376 377 struct sk_filter __rcu *sk_filter; 378 union { 379 struct socket_wq __rcu *sk_wq; 380 struct socket_wq *sk_wq_raw; 381 }; 382 #ifdef CONFIG_XFRM 383 struct xfrm_policy __rcu *sk_policy[2]; 384 #endif 385 struct dst_entry *sk_rx_dst; 386 struct dst_entry __rcu *sk_dst_cache; 387 atomic_t sk_omem_alloc; 388 int sk_sndbuf; 389 390 /* ===== cache line for TX ===== */ 391 int sk_wmem_queued; 392 atomic_t sk_wmem_alloc; 393 unsigned long sk_tsq_flags; 394 struct sk_buff *sk_send_head; 395 struct sk_buff_head sk_write_queue; 396 __s32 sk_peek_off; 397 int sk_write_pending; 398 __u32 sk_dst_pending_confirm; 399 /* Note: 32bit hole on 64bit arches */ 400 long sk_sndtimeo; 401 struct timer_list sk_timer; 402 __u32 sk_priority; 403 __u32 sk_mark; 404 u32 sk_pacing_rate; /* bytes per second */ 405 u32 sk_max_pacing_rate; 406 struct page_frag sk_frag; 407 netdev_features_t sk_route_caps; 408 netdev_features_t sk_route_nocaps; 409 int sk_gso_type; 410 unsigned int sk_gso_max_size; 411 gfp_t sk_allocation; 412 __u32 sk_txhash; 413 414 /* 415 * Because of non atomicity rules, all 416 * changes are protected by socket lock. 417 */ 418 unsigned int __sk_flags_offset[0]; 419 #ifdef __BIG_ENDIAN_BITFIELD 420 #define SK_FL_PROTO_SHIFT 16 421 #define SK_FL_PROTO_MASK 0x00ff0000 422 423 #define SK_FL_TYPE_SHIFT 0 424 #define SK_FL_TYPE_MASK 0x0000ffff 425 #else 426 #define SK_FL_PROTO_SHIFT 8 427 #define SK_FL_PROTO_MASK 0x0000ff00 428 429 #define SK_FL_TYPE_SHIFT 16 430 #define SK_FL_TYPE_MASK 0xffff0000 431 #endif 432 433 kmemcheck_bitfield_begin(flags); 434 unsigned int sk_padding : 1, 435 sk_kern_sock : 1, 436 sk_no_check_tx : 1, 437 sk_no_check_rx : 1, 438 sk_userlocks : 4, 439 sk_protocol : 8, 440 sk_type : 16; 441 #define SK_PROTOCOL_MAX U8_MAX 442 kmemcheck_bitfield_end(flags); 443 444 u16 sk_gso_max_segs; 445 unsigned long sk_lingertime; 446 struct proto *sk_prot_creator; 447 rwlock_t sk_callback_lock; 448 int sk_err, 449 sk_err_soft; 450 u32 sk_ack_backlog; 451 u32 sk_max_ack_backlog; 452 kuid_t sk_uid; 453 struct pid *sk_peer_pid; 454 const struct cred *sk_peer_cred; 455 long sk_rcvtimeo; 456 ktime_t sk_stamp; 457 u16 sk_tsflags; 458 u8 sk_shutdown; 459 u32 sk_tskey; 460 struct socket *sk_socket; 461 void *sk_user_data; 462 #ifdef CONFIG_SECURITY 463 void *sk_security; 464 #endif 465 struct sock_cgroup_data sk_cgrp_data; 466 struct mem_cgroup *sk_memcg; 467 void (*sk_state_change)(struct sock *sk); 468 void (*sk_data_ready)(struct sock *sk); 469 void (*sk_write_space)(struct sock *sk); 470 void (*sk_error_report)(struct sock *sk); 471 int (*sk_backlog_rcv)(struct sock *sk, 472 struct sk_buff *skb); 473 void (*sk_destruct)(struct sock *sk); 474 struct sock_reuseport __rcu *sk_reuseport_cb; 475 struct rcu_head sk_rcu; 476 }; 477 478 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 479 480 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 481 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 482 483 /* 484 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 485 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 486 * on a socket means that the socket will reuse everybody else's port 487 * without looking at the other's sk_reuse value. 488 */ 489 490 #define SK_NO_REUSE 0 491 #define SK_CAN_REUSE 1 492 #define SK_FORCE_REUSE 2 493 494 int sk_set_peek_off(struct sock *sk, int val); 495 496 static inline int sk_peek_offset(struct sock *sk, int flags) 497 { 498 if (unlikely(flags & MSG_PEEK)) { 499 s32 off = READ_ONCE(sk->sk_peek_off); 500 if (off >= 0) 501 return off; 502 } 503 504 return 0; 505 } 506 507 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 508 { 509 s32 off = READ_ONCE(sk->sk_peek_off); 510 511 if (unlikely(off >= 0)) { 512 off = max_t(s32, off - val, 0); 513 WRITE_ONCE(sk->sk_peek_off, off); 514 } 515 } 516 517 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 518 { 519 sk_peek_offset_bwd(sk, -val); 520 } 521 522 /* 523 * Hashed lists helper routines 524 */ 525 static inline struct sock *sk_entry(const struct hlist_node *node) 526 { 527 return hlist_entry(node, struct sock, sk_node); 528 } 529 530 static inline struct sock *__sk_head(const struct hlist_head *head) 531 { 532 return hlist_entry(head->first, struct sock, sk_node); 533 } 534 535 static inline struct sock *sk_head(const struct hlist_head *head) 536 { 537 return hlist_empty(head) ? NULL : __sk_head(head); 538 } 539 540 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 541 { 542 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 543 } 544 545 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 546 { 547 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 548 } 549 550 static inline struct sock *sk_next(const struct sock *sk) 551 { 552 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 553 } 554 555 static inline struct sock *sk_nulls_next(const struct sock *sk) 556 { 557 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 558 hlist_nulls_entry(sk->sk_nulls_node.next, 559 struct sock, sk_nulls_node) : 560 NULL; 561 } 562 563 static inline bool sk_unhashed(const struct sock *sk) 564 { 565 return hlist_unhashed(&sk->sk_node); 566 } 567 568 static inline bool sk_hashed(const struct sock *sk) 569 { 570 return !sk_unhashed(sk); 571 } 572 573 static inline void sk_node_init(struct hlist_node *node) 574 { 575 node->pprev = NULL; 576 } 577 578 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 579 { 580 node->pprev = NULL; 581 } 582 583 static inline void __sk_del_node(struct sock *sk) 584 { 585 __hlist_del(&sk->sk_node); 586 } 587 588 /* NB: equivalent to hlist_del_init_rcu */ 589 static inline bool __sk_del_node_init(struct sock *sk) 590 { 591 if (sk_hashed(sk)) { 592 __sk_del_node(sk); 593 sk_node_init(&sk->sk_node); 594 return true; 595 } 596 return false; 597 } 598 599 /* Grab socket reference count. This operation is valid only 600 when sk is ALREADY grabbed f.e. it is found in hash table 601 or a list and the lookup is made under lock preventing hash table 602 modifications. 603 */ 604 605 static __always_inline void sock_hold(struct sock *sk) 606 { 607 atomic_inc(&sk->sk_refcnt); 608 } 609 610 /* Ungrab socket in the context, which assumes that socket refcnt 611 cannot hit zero, f.e. it is true in context of any socketcall. 612 */ 613 static __always_inline void __sock_put(struct sock *sk) 614 { 615 atomic_dec(&sk->sk_refcnt); 616 } 617 618 static inline bool sk_del_node_init(struct sock *sk) 619 { 620 bool rc = __sk_del_node_init(sk); 621 622 if (rc) { 623 /* paranoid for a while -acme */ 624 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 625 __sock_put(sk); 626 } 627 return rc; 628 } 629 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 630 631 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 632 { 633 if (sk_hashed(sk)) { 634 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 635 return true; 636 } 637 return false; 638 } 639 640 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 641 { 642 bool rc = __sk_nulls_del_node_init_rcu(sk); 643 644 if (rc) { 645 /* paranoid for a while -acme */ 646 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 647 __sock_put(sk); 648 } 649 return rc; 650 } 651 652 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 653 { 654 hlist_add_head(&sk->sk_node, list); 655 } 656 657 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 658 { 659 sock_hold(sk); 660 __sk_add_node(sk, list); 661 } 662 663 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 664 { 665 sock_hold(sk); 666 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 667 sk->sk_family == AF_INET6) 668 hlist_add_tail_rcu(&sk->sk_node, list); 669 else 670 hlist_add_head_rcu(&sk->sk_node, list); 671 } 672 673 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 674 { 675 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 676 sk->sk_family == AF_INET6) 677 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 678 else 679 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 680 } 681 682 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 683 { 684 sock_hold(sk); 685 __sk_nulls_add_node_rcu(sk, list); 686 } 687 688 static inline void __sk_del_bind_node(struct sock *sk) 689 { 690 __hlist_del(&sk->sk_bind_node); 691 } 692 693 static inline void sk_add_bind_node(struct sock *sk, 694 struct hlist_head *list) 695 { 696 hlist_add_head(&sk->sk_bind_node, list); 697 } 698 699 #define sk_for_each(__sk, list) \ 700 hlist_for_each_entry(__sk, list, sk_node) 701 #define sk_for_each_rcu(__sk, list) \ 702 hlist_for_each_entry_rcu(__sk, list, sk_node) 703 #define sk_nulls_for_each(__sk, node, list) \ 704 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 705 #define sk_nulls_for_each_rcu(__sk, node, list) \ 706 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 707 #define sk_for_each_from(__sk) \ 708 hlist_for_each_entry_from(__sk, sk_node) 709 #define sk_nulls_for_each_from(__sk, node) \ 710 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 711 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 712 #define sk_for_each_safe(__sk, tmp, list) \ 713 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 714 #define sk_for_each_bound(__sk, list) \ 715 hlist_for_each_entry(__sk, list, sk_bind_node) 716 717 /** 718 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 719 * @tpos: the type * to use as a loop cursor. 720 * @pos: the &struct hlist_node to use as a loop cursor. 721 * @head: the head for your list. 722 * @offset: offset of hlist_node within the struct. 723 * 724 */ 725 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 726 for (pos = rcu_dereference((head)->first); \ 727 pos != NULL && \ 728 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 729 pos = rcu_dereference(pos->next)) 730 731 static inline struct user_namespace *sk_user_ns(struct sock *sk) 732 { 733 /* Careful only use this in a context where these parameters 734 * can not change and must all be valid, such as recvmsg from 735 * userspace. 736 */ 737 return sk->sk_socket->file->f_cred->user_ns; 738 } 739 740 /* Sock flags */ 741 enum sock_flags { 742 SOCK_DEAD, 743 SOCK_DONE, 744 SOCK_URGINLINE, 745 SOCK_KEEPOPEN, 746 SOCK_LINGER, 747 SOCK_DESTROY, 748 SOCK_BROADCAST, 749 SOCK_TIMESTAMP, 750 SOCK_ZAPPED, 751 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 752 SOCK_DBG, /* %SO_DEBUG setting */ 753 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 754 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 755 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 756 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 757 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 758 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 759 SOCK_FASYNC, /* fasync() active */ 760 SOCK_RXQ_OVFL, 761 SOCK_ZEROCOPY, /* buffers from userspace */ 762 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 763 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 764 * Will use last 4 bytes of packet sent from 765 * user-space instead. 766 */ 767 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 768 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 769 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 770 }; 771 772 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 773 774 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 775 { 776 nsk->sk_flags = osk->sk_flags; 777 } 778 779 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 780 { 781 __set_bit(flag, &sk->sk_flags); 782 } 783 784 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 785 { 786 __clear_bit(flag, &sk->sk_flags); 787 } 788 789 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 790 { 791 return test_bit(flag, &sk->sk_flags); 792 } 793 794 #ifdef CONFIG_NET 795 extern struct static_key memalloc_socks; 796 static inline int sk_memalloc_socks(void) 797 { 798 return static_key_false(&memalloc_socks); 799 } 800 #else 801 802 static inline int sk_memalloc_socks(void) 803 { 804 return 0; 805 } 806 807 #endif 808 809 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 810 { 811 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 812 } 813 814 static inline void sk_acceptq_removed(struct sock *sk) 815 { 816 sk->sk_ack_backlog--; 817 } 818 819 static inline void sk_acceptq_added(struct sock *sk) 820 { 821 sk->sk_ack_backlog++; 822 } 823 824 static inline bool sk_acceptq_is_full(const struct sock *sk) 825 { 826 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 827 } 828 829 /* 830 * Compute minimal free write space needed to queue new packets. 831 */ 832 static inline int sk_stream_min_wspace(const struct sock *sk) 833 { 834 return sk->sk_wmem_queued >> 1; 835 } 836 837 static inline int sk_stream_wspace(const struct sock *sk) 838 { 839 return sk->sk_sndbuf - sk->sk_wmem_queued; 840 } 841 842 void sk_stream_write_space(struct sock *sk); 843 844 /* OOB backlog add */ 845 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 846 { 847 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 848 skb_dst_force_safe(skb); 849 850 if (!sk->sk_backlog.tail) 851 sk->sk_backlog.head = skb; 852 else 853 sk->sk_backlog.tail->next = skb; 854 855 sk->sk_backlog.tail = skb; 856 skb->next = NULL; 857 } 858 859 /* 860 * Take into account size of receive queue and backlog queue 861 * Do not take into account this skb truesize, 862 * to allow even a single big packet to come. 863 */ 864 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 865 { 866 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 867 868 return qsize > limit; 869 } 870 871 /* The per-socket spinlock must be held here. */ 872 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 873 unsigned int limit) 874 { 875 if (sk_rcvqueues_full(sk, limit)) 876 return -ENOBUFS; 877 878 /* 879 * If the skb was allocated from pfmemalloc reserves, only 880 * allow SOCK_MEMALLOC sockets to use it as this socket is 881 * helping free memory 882 */ 883 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 884 return -ENOMEM; 885 886 __sk_add_backlog(sk, skb); 887 sk->sk_backlog.len += skb->truesize; 888 return 0; 889 } 890 891 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 892 893 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 894 { 895 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 896 return __sk_backlog_rcv(sk, skb); 897 898 return sk->sk_backlog_rcv(sk, skb); 899 } 900 901 static inline void sk_incoming_cpu_update(struct sock *sk) 902 { 903 sk->sk_incoming_cpu = raw_smp_processor_id(); 904 } 905 906 static inline void sock_rps_record_flow_hash(__u32 hash) 907 { 908 #ifdef CONFIG_RPS 909 struct rps_sock_flow_table *sock_flow_table; 910 911 rcu_read_lock(); 912 sock_flow_table = rcu_dereference(rps_sock_flow_table); 913 rps_record_sock_flow(sock_flow_table, hash); 914 rcu_read_unlock(); 915 #endif 916 } 917 918 static inline void sock_rps_record_flow(const struct sock *sk) 919 { 920 #ifdef CONFIG_RPS 921 if (static_key_false(&rfs_needed)) { 922 /* Reading sk->sk_rxhash might incur an expensive cache line 923 * miss. 924 * 925 * TCP_ESTABLISHED does cover almost all states where RFS 926 * might be useful, and is cheaper [1] than testing : 927 * IPv4: inet_sk(sk)->inet_daddr 928 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 929 * OR an additional socket flag 930 * [1] : sk_state and sk_prot are in the same cache line. 931 */ 932 if (sk->sk_state == TCP_ESTABLISHED) 933 sock_rps_record_flow_hash(sk->sk_rxhash); 934 } 935 #endif 936 } 937 938 static inline void sock_rps_save_rxhash(struct sock *sk, 939 const struct sk_buff *skb) 940 { 941 #ifdef CONFIG_RPS 942 if (unlikely(sk->sk_rxhash != skb->hash)) 943 sk->sk_rxhash = skb->hash; 944 #endif 945 } 946 947 static inline void sock_rps_reset_rxhash(struct sock *sk) 948 { 949 #ifdef CONFIG_RPS 950 sk->sk_rxhash = 0; 951 #endif 952 } 953 954 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 955 ({ int __rc; \ 956 release_sock(__sk); \ 957 __rc = __condition; \ 958 if (!__rc) { \ 959 *(__timeo) = wait_woken(__wait, \ 960 TASK_INTERRUPTIBLE, \ 961 *(__timeo)); \ 962 } \ 963 sched_annotate_sleep(); \ 964 lock_sock(__sk); \ 965 __rc = __condition; \ 966 __rc; \ 967 }) 968 969 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 970 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 971 void sk_stream_wait_close(struct sock *sk, long timeo_p); 972 int sk_stream_error(struct sock *sk, int flags, int err); 973 void sk_stream_kill_queues(struct sock *sk); 974 void sk_set_memalloc(struct sock *sk); 975 void sk_clear_memalloc(struct sock *sk); 976 977 void __sk_flush_backlog(struct sock *sk); 978 979 static inline bool sk_flush_backlog(struct sock *sk) 980 { 981 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 982 __sk_flush_backlog(sk); 983 return true; 984 } 985 return false; 986 } 987 988 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 989 990 struct request_sock_ops; 991 struct timewait_sock_ops; 992 struct inet_hashinfo; 993 struct raw_hashinfo; 994 struct smc_hashinfo; 995 struct module; 996 997 /* 998 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 999 * un-modified. Special care is taken when initializing object to zero. 1000 */ 1001 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1002 { 1003 if (offsetof(struct sock, sk_node.next) != 0) 1004 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1005 memset(&sk->sk_node.pprev, 0, 1006 size - offsetof(struct sock, sk_node.pprev)); 1007 } 1008 1009 /* Networking protocol blocks we attach to sockets. 1010 * socket layer -> transport layer interface 1011 */ 1012 struct proto { 1013 void (*close)(struct sock *sk, 1014 long timeout); 1015 int (*connect)(struct sock *sk, 1016 struct sockaddr *uaddr, 1017 int addr_len); 1018 int (*disconnect)(struct sock *sk, int flags); 1019 1020 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1021 bool kern); 1022 1023 int (*ioctl)(struct sock *sk, int cmd, 1024 unsigned long arg); 1025 int (*init)(struct sock *sk); 1026 void (*destroy)(struct sock *sk); 1027 void (*shutdown)(struct sock *sk, int how); 1028 int (*setsockopt)(struct sock *sk, int level, 1029 int optname, char __user *optval, 1030 unsigned int optlen); 1031 int (*getsockopt)(struct sock *sk, int level, 1032 int optname, char __user *optval, 1033 int __user *option); 1034 void (*keepalive)(struct sock *sk, int valbool); 1035 #ifdef CONFIG_COMPAT 1036 int (*compat_setsockopt)(struct sock *sk, 1037 int level, 1038 int optname, char __user *optval, 1039 unsigned int optlen); 1040 int (*compat_getsockopt)(struct sock *sk, 1041 int level, 1042 int optname, char __user *optval, 1043 int __user *option); 1044 int (*compat_ioctl)(struct sock *sk, 1045 unsigned int cmd, unsigned long arg); 1046 #endif 1047 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1048 size_t len); 1049 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1050 size_t len, int noblock, int flags, 1051 int *addr_len); 1052 int (*sendpage)(struct sock *sk, struct page *page, 1053 int offset, size_t size, int flags); 1054 int (*bind)(struct sock *sk, 1055 struct sockaddr *uaddr, int addr_len); 1056 1057 int (*backlog_rcv) (struct sock *sk, 1058 struct sk_buff *skb); 1059 1060 void (*release_cb)(struct sock *sk); 1061 1062 /* Keeping track of sk's, looking them up, and port selection methods. */ 1063 int (*hash)(struct sock *sk); 1064 void (*unhash)(struct sock *sk); 1065 void (*rehash)(struct sock *sk); 1066 int (*get_port)(struct sock *sk, unsigned short snum); 1067 1068 /* Keeping track of sockets in use */ 1069 #ifdef CONFIG_PROC_FS 1070 unsigned int inuse_idx; 1071 #endif 1072 1073 bool (*stream_memory_free)(const struct sock *sk); 1074 /* Memory pressure */ 1075 void (*enter_memory_pressure)(struct sock *sk); 1076 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1077 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1078 /* 1079 * Pressure flag: try to collapse. 1080 * Technical note: it is used by multiple contexts non atomically. 1081 * All the __sk_mem_schedule() is of this nature: accounting 1082 * is strict, actions are advisory and have some latency. 1083 */ 1084 int *memory_pressure; 1085 long *sysctl_mem; 1086 int *sysctl_wmem; 1087 int *sysctl_rmem; 1088 int max_header; 1089 bool no_autobind; 1090 1091 struct kmem_cache *slab; 1092 unsigned int obj_size; 1093 int slab_flags; 1094 1095 struct percpu_counter *orphan_count; 1096 1097 struct request_sock_ops *rsk_prot; 1098 struct timewait_sock_ops *twsk_prot; 1099 1100 union { 1101 struct inet_hashinfo *hashinfo; 1102 struct udp_table *udp_table; 1103 struct raw_hashinfo *raw_hash; 1104 struct smc_hashinfo *smc_hash; 1105 } h; 1106 1107 struct module *owner; 1108 1109 char name[32]; 1110 1111 struct list_head node; 1112 #ifdef SOCK_REFCNT_DEBUG 1113 atomic_t socks; 1114 #endif 1115 int (*diag_destroy)(struct sock *sk, int err); 1116 }; 1117 1118 int proto_register(struct proto *prot, int alloc_slab); 1119 void proto_unregister(struct proto *prot); 1120 1121 #ifdef SOCK_REFCNT_DEBUG 1122 static inline void sk_refcnt_debug_inc(struct sock *sk) 1123 { 1124 atomic_inc(&sk->sk_prot->socks); 1125 } 1126 1127 static inline void sk_refcnt_debug_dec(struct sock *sk) 1128 { 1129 atomic_dec(&sk->sk_prot->socks); 1130 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1131 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1132 } 1133 1134 static inline void sk_refcnt_debug_release(const struct sock *sk) 1135 { 1136 if (atomic_read(&sk->sk_refcnt) != 1) 1137 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1138 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1139 } 1140 #else /* SOCK_REFCNT_DEBUG */ 1141 #define sk_refcnt_debug_inc(sk) do { } while (0) 1142 #define sk_refcnt_debug_dec(sk) do { } while (0) 1143 #define sk_refcnt_debug_release(sk) do { } while (0) 1144 #endif /* SOCK_REFCNT_DEBUG */ 1145 1146 static inline bool sk_stream_memory_free(const struct sock *sk) 1147 { 1148 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1149 return false; 1150 1151 return sk->sk_prot->stream_memory_free ? 1152 sk->sk_prot->stream_memory_free(sk) : true; 1153 } 1154 1155 static inline bool sk_stream_is_writeable(const struct sock *sk) 1156 { 1157 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1158 sk_stream_memory_free(sk); 1159 } 1160 1161 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1162 struct cgroup *ancestor) 1163 { 1164 #ifdef CONFIG_SOCK_CGROUP_DATA 1165 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1166 ancestor); 1167 #else 1168 return -ENOTSUPP; 1169 #endif 1170 } 1171 1172 static inline bool sk_has_memory_pressure(const struct sock *sk) 1173 { 1174 return sk->sk_prot->memory_pressure != NULL; 1175 } 1176 1177 static inline bool sk_under_memory_pressure(const struct sock *sk) 1178 { 1179 if (!sk->sk_prot->memory_pressure) 1180 return false; 1181 1182 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1183 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1184 return true; 1185 1186 return !!*sk->sk_prot->memory_pressure; 1187 } 1188 1189 static inline void sk_leave_memory_pressure(struct sock *sk) 1190 { 1191 int *memory_pressure = sk->sk_prot->memory_pressure; 1192 1193 if (!memory_pressure) 1194 return; 1195 1196 if (*memory_pressure) 1197 *memory_pressure = 0; 1198 } 1199 1200 static inline void sk_enter_memory_pressure(struct sock *sk) 1201 { 1202 if (!sk->sk_prot->enter_memory_pressure) 1203 return; 1204 1205 sk->sk_prot->enter_memory_pressure(sk); 1206 } 1207 1208 static inline long 1209 sk_memory_allocated(const struct sock *sk) 1210 { 1211 return atomic_long_read(sk->sk_prot->memory_allocated); 1212 } 1213 1214 static inline long 1215 sk_memory_allocated_add(struct sock *sk, int amt) 1216 { 1217 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1218 } 1219 1220 static inline void 1221 sk_memory_allocated_sub(struct sock *sk, int amt) 1222 { 1223 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1224 } 1225 1226 static inline void sk_sockets_allocated_dec(struct sock *sk) 1227 { 1228 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1229 } 1230 1231 static inline void sk_sockets_allocated_inc(struct sock *sk) 1232 { 1233 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1234 } 1235 1236 static inline int 1237 sk_sockets_allocated_read_positive(struct sock *sk) 1238 { 1239 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1240 } 1241 1242 static inline int 1243 proto_sockets_allocated_sum_positive(struct proto *prot) 1244 { 1245 return percpu_counter_sum_positive(prot->sockets_allocated); 1246 } 1247 1248 static inline long 1249 proto_memory_allocated(struct proto *prot) 1250 { 1251 return atomic_long_read(prot->memory_allocated); 1252 } 1253 1254 static inline bool 1255 proto_memory_pressure(struct proto *prot) 1256 { 1257 if (!prot->memory_pressure) 1258 return false; 1259 return !!*prot->memory_pressure; 1260 } 1261 1262 1263 #ifdef CONFIG_PROC_FS 1264 /* Called with local bh disabled */ 1265 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1266 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1267 #else 1268 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1269 int inc) 1270 { 1271 } 1272 #endif 1273 1274 1275 /* With per-bucket locks this operation is not-atomic, so that 1276 * this version is not worse. 1277 */ 1278 static inline int __sk_prot_rehash(struct sock *sk) 1279 { 1280 sk->sk_prot->unhash(sk); 1281 return sk->sk_prot->hash(sk); 1282 } 1283 1284 /* About 10 seconds */ 1285 #define SOCK_DESTROY_TIME (10*HZ) 1286 1287 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1288 #define PROT_SOCK 1024 1289 1290 #define SHUTDOWN_MASK 3 1291 #define RCV_SHUTDOWN 1 1292 #define SEND_SHUTDOWN 2 1293 1294 #define SOCK_SNDBUF_LOCK 1 1295 #define SOCK_RCVBUF_LOCK 2 1296 #define SOCK_BINDADDR_LOCK 4 1297 #define SOCK_BINDPORT_LOCK 8 1298 1299 struct socket_alloc { 1300 struct socket socket; 1301 struct inode vfs_inode; 1302 }; 1303 1304 static inline struct socket *SOCKET_I(struct inode *inode) 1305 { 1306 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1307 } 1308 1309 static inline struct inode *SOCK_INODE(struct socket *socket) 1310 { 1311 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1312 } 1313 1314 /* 1315 * Functions for memory accounting 1316 */ 1317 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1318 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1319 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1320 void __sk_mem_reclaim(struct sock *sk, int amount); 1321 1322 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1323 * do not necessarily have 16x time more memory than 4KB ones. 1324 */ 1325 #define SK_MEM_QUANTUM 4096 1326 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1327 #define SK_MEM_SEND 0 1328 #define SK_MEM_RECV 1 1329 1330 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1331 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1332 { 1333 long val = sk->sk_prot->sysctl_mem[index]; 1334 1335 #if PAGE_SIZE > SK_MEM_QUANTUM 1336 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1337 #elif PAGE_SIZE < SK_MEM_QUANTUM 1338 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1339 #endif 1340 return val; 1341 } 1342 1343 static inline int sk_mem_pages(int amt) 1344 { 1345 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1346 } 1347 1348 static inline bool sk_has_account(struct sock *sk) 1349 { 1350 /* return true if protocol supports memory accounting */ 1351 return !!sk->sk_prot->memory_allocated; 1352 } 1353 1354 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1355 { 1356 if (!sk_has_account(sk)) 1357 return true; 1358 return size <= sk->sk_forward_alloc || 1359 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1360 } 1361 1362 static inline bool 1363 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1364 { 1365 if (!sk_has_account(sk)) 1366 return true; 1367 return size<= sk->sk_forward_alloc || 1368 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1369 skb_pfmemalloc(skb); 1370 } 1371 1372 static inline void sk_mem_reclaim(struct sock *sk) 1373 { 1374 if (!sk_has_account(sk)) 1375 return; 1376 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1377 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1378 } 1379 1380 static inline void sk_mem_reclaim_partial(struct sock *sk) 1381 { 1382 if (!sk_has_account(sk)) 1383 return; 1384 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1385 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1386 } 1387 1388 static inline void sk_mem_charge(struct sock *sk, int size) 1389 { 1390 if (!sk_has_account(sk)) 1391 return; 1392 sk->sk_forward_alloc -= size; 1393 } 1394 1395 static inline void sk_mem_uncharge(struct sock *sk, int size) 1396 { 1397 if (!sk_has_account(sk)) 1398 return; 1399 sk->sk_forward_alloc += size; 1400 1401 /* Avoid a possible overflow. 1402 * TCP send queues can make this happen, if sk_mem_reclaim() 1403 * is not called and more than 2 GBytes are released at once. 1404 * 1405 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1406 * no need to hold that much forward allocation anyway. 1407 */ 1408 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1409 __sk_mem_reclaim(sk, 1 << 20); 1410 } 1411 1412 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1413 { 1414 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1415 sk->sk_wmem_queued -= skb->truesize; 1416 sk_mem_uncharge(sk, skb->truesize); 1417 __kfree_skb(skb); 1418 } 1419 1420 static inline void sock_release_ownership(struct sock *sk) 1421 { 1422 if (sk->sk_lock.owned) { 1423 sk->sk_lock.owned = 0; 1424 1425 /* The sk_lock has mutex_unlock() semantics: */ 1426 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1427 } 1428 } 1429 1430 /* 1431 * Macro so as to not evaluate some arguments when 1432 * lockdep is not enabled. 1433 * 1434 * Mark both the sk_lock and the sk_lock.slock as a 1435 * per-address-family lock class. 1436 */ 1437 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1438 do { \ 1439 sk->sk_lock.owned = 0; \ 1440 init_waitqueue_head(&sk->sk_lock.wq); \ 1441 spin_lock_init(&(sk)->sk_lock.slock); \ 1442 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1443 sizeof((sk)->sk_lock)); \ 1444 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1445 (skey), (sname)); \ 1446 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1447 } while (0) 1448 1449 #ifdef CONFIG_LOCKDEP 1450 static inline bool lockdep_sock_is_held(const struct sock *csk) 1451 { 1452 struct sock *sk = (struct sock *)csk; 1453 1454 return lockdep_is_held(&sk->sk_lock) || 1455 lockdep_is_held(&sk->sk_lock.slock); 1456 } 1457 #endif 1458 1459 void lock_sock_nested(struct sock *sk, int subclass); 1460 1461 static inline void lock_sock(struct sock *sk) 1462 { 1463 lock_sock_nested(sk, 0); 1464 } 1465 1466 void release_sock(struct sock *sk); 1467 1468 /* BH context may only use the following locking interface. */ 1469 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1470 #define bh_lock_sock_nested(__sk) \ 1471 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1472 SINGLE_DEPTH_NESTING) 1473 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1474 1475 bool lock_sock_fast(struct sock *sk); 1476 /** 1477 * unlock_sock_fast - complement of lock_sock_fast 1478 * @sk: socket 1479 * @slow: slow mode 1480 * 1481 * fast unlock socket for user context. 1482 * If slow mode is on, we call regular release_sock() 1483 */ 1484 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1485 { 1486 if (slow) 1487 release_sock(sk); 1488 else 1489 spin_unlock_bh(&sk->sk_lock.slock); 1490 } 1491 1492 /* Used by processes to "lock" a socket state, so that 1493 * interrupts and bottom half handlers won't change it 1494 * from under us. It essentially blocks any incoming 1495 * packets, so that we won't get any new data or any 1496 * packets that change the state of the socket. 1497 * 1498 * While locked, BH processing will add new packets to 1499 * the backlog queue. This queue is processed by the 1500 * owner of the socket lock right before it is released. 1501 * 1502 * Since ~2.3.5 it is also exclusive sleep lock serializing 1503 * accesses from user process context. 1504 */ 1505 1506 static inline void sock_owned_by_me(const struct sock *sk) 1507 { 1508 #ifdef CONFIG_LOCKDEP 1509 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1510 #endif 1511 } 1512 1513 static inline bool sock_owned_by_user(const struct sock *sk) 1514 { 1515 sock_owned_by_me(sk); 1516 return sk->sk_lock.owned; 1517 } 1518 1519 /* no reclassification while locks are held */ 1520 static inline bool sock_allow_reclassification(const struct sock *csk) 1521 { 1522 struct sock *sk = (struct sock *)csk; 1523 1524 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1525 } 1526 1527 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1528 struct proto *prot, int kern); 1529 void sk_free(struct sock *sk); 1530 void sk_destruct(struct sock *sk); 1531 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1532 void sk_free_unlock_clone(struct sock *sk); 1533 1534 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1535 gfp_t priority); 1536 void __sock_wfree(struct sk_buff *skb); 1537 void sock_wfree(struct sk_buff *skb); 1538 void skb_orphan_partial(struct sk_buff *skb); 1539 void sock_rfree(struct sk_buff *skb); 1540 void sock_efree(struct sk_buff *skb); 1541 #ifdef CONFIG_INET 1542 void sock_edemux(struct sk_buff *skb); 1543 #else 1544 #define sock_edemux sock_efree 1545 #endif 1546 1547 int sock_setsockopt(struct socket *sock, int level, int op, 1548 char __user *optval, unsigned int optlen); 1549 1550 int sock_getsockopt(struct socket *sock, int level, int op, 1551 char __user *optval, int __user *optlen); 1552 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1553 int noblock, int *errcode); 1554 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1555 unsigned long data_len, int noblock, 1556 int *errcode, int max_page_order); 1557 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1558 void sock_kfree_s(struct sock *sk, void *mem, int size); 1559 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1560 void sk_send_sigurg(struct sock *sk); 1561 1562 struct sockcm_cookie { 1563 u32 mark; 1564 u16 tsflags; 1565 }; 1566 1567 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1568 struct sockcm_cookie *sockc); 1569 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1570 struct sockcm_cookie *sockc); 1571 1572 /* 1573 * Functions to fill in entries in struct proto_ops when a protocol 1574 * does not implement a particular function. 1575 */ 1576 int sock_no_bind(struct socket *, struct sockaddr *, int); 1577 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1578 int sock_no_socketpair(struct socket *, struct socket *); 1579 int sock_no_accept(struct socket *, struct socket *, int, bool); 1580 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1581 unsigned int sock_no_poll(struct file *, struct socket *, 1582 struct poll_table_struct *); 1583 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1584 int sock_no_listen(struct socket *, int); 1585 int sock_no_shutdown(struct socket *, int); 1586 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1587 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1588 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1589 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1590 int sock_no_mmap(struct file *file, struct socket *sock, 1591 struct vm_area_struct *vma); 1592 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1593 size_t size, int flags); 1594 1595 /* 1596 * Functions to fill in entries in struct proto_ops when a protocol 1597 * uses the inet style. 1598 */ 1599 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1600 char __user *optval, int __user *optlen); 1601 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1602 int flags); 1603 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1604 char __user *optval, unsigned int optlen); 1605 int compat_sock_common_getsockopt(struct socket *sock, int level, 1606 int optname, char __user *optval, int __user *optlen); 1607 int compat_sock_common_setsockopt(struct socket *sock, int level, 1608 int optname, char __user *optval, unsigned int optlen); 1609 1610 void sk_common_release(struct sock *sk); 1611 1612 /* 1613 * Default socket callbacks and setup code 1614 */ 1615 1616 /* Initialise core socket variables */ 1617 void sock_init_data(struct socket *sock, struct sock *sk); 1618 1619 /* 1620 * Socket reference counting postulates. 1621 * 1622 * * Each user of socket SHOULD hold a reference count. 1623 * * Each access point to socket (an hash table bucket, reference from a list, 1624 * running timer, skb in flight MUST hold a reference count. 1625 * * When reference count hits 0, it means it will never increase back. 1626 * * When reference count hits 0, it means that no references from 1627 * outside exist to this socket and current process on current CPU 1628 * is last user and may/should destroy this socket. 1629 * * sk_free is called from any context: process, BH, IRQ. When 1630 * it is called, socket has no references from outside -> sk_free 1631 * may release descendant resources allocated by the socket, but 1632 * to the time when it is called, socket is NOT referenced by any 1633 * hash tables, lists etc. 1634 * * Packets, delivered from outside (from network or from another process) 1635 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1636 * when they sit in queue. Otherwise, packets will leak to hole, when 1637 * socket is looked up by one cpu and unhasing is made by another CPU. 1638 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1639 * (leak to backlog). Packet socket does all the processing inside 1640 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1641 * use separate SMP lock, so that they are prone too. 1642 */ 1643 1644 /* Ungrab socket and destroy it, if it was the last reference. */ 1645 static inline void sock_put(struct sock *sk) 1646 { 1647 if (atomic_dec_and_test(&sk->sk_refcnt)) 1648 sk_free(sk); 1649 } 1650 /* Generic version of sock_put(), dealing with all sockets 1651 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1652 */ 1653 void sock_gen_put(struct sock *sk); 1654 1655 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1656 unsigned int trim_cap, bool refcounted); 1657 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1658 const int nested) 1659 { 1660 return __sk_receive_skb(sk, skb, nested, 1, true); 1661 } 1662 1663 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1664 { 1665 sk->sk_tx_queue_mapping = tx_queue; 1666 } 1667 1668 static inline void sk_tx_queue_clear(struct sock *sk) 1669 { 1670 sk->sk_tx_queue_mapping = -1; 1671 } 1672 1673 static inline int sk_tx_queue_get(const struct sock *sk) 1674 { 1675 return sk ? sk->sk_tx_queue_mapping : -1; 1676 } 1677 1678 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1679 { 1680 sk_tx_queue_clear(sk); 1681 sk->sk_socket = sock; 1682 } 1683 1684 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1685 { 1686 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1687 return &rcu_dereference_raw(sk->sk_wq)->wait; 1688 } 1689 /* Detach socket from process context. 1690 * Announce socket dead, detach it from wait queue and inode. 1691 * Note that parent inode held reference count on this struct sock, 1692 * we do not release it in this function, because protocol 1693 * probably wants some additional cleanups or even continuing 1694 * to work with this socket (TCP). 1695 */ 1696 static inline void sock_orphan(struct sock *sk) 1697 { 1698 write_lock_bh(&sk->sk_callback_lock); 1699 sock_set_flag(sk, SOCK_DEAD); 1700 sk_set_socket(sk, NULL); 1701 sk->sk_wq = NULL; 1702 write_unlock_bh(&sk->sk_callback_lock); 1703 } 1704 1705 static inline void sock_graft(struct sock *sk, struct socket *parent) 1706 { 1707 write_lock_bh(&sk->sk_callback_lock); 1708 sk->sk_wq = parent->wq; 1709 parent->sk = sk; 1710 sk_set_socket(sk, parent); 1711 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1712 security_sock_graft(sk, parent); 1713 write_unlock_bh(&sk->sk_callback_lock); 1714 } 1715 1716 kuid_t sock_i_uid(struct sock *sk); 1717 unsigned long sock_i_ino(struct sock *sk); 1718 1719 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1720 { 1721 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1722 } 1723 1724 static inline u32 net_tx_rndhash(void) 1725 { 1726 u32 v = prandom_u32(); 1727 1728 return v ?: 1; 1729 } 1730 1731 static inline void sk_set_txhash(struct sock *sk) 1732 { 1733 sk->sk_txhash = net_tx_rndhash(); 1734 } 1735 1736 static inline void sk_rethink_txhash(struct sock *sk) 1737 { 1738 if (sk->sk_txhash) 1739 sk_set_txhash(sk); 1740 } 1741 1742 static inline struct dst_entry * 1743 __sk_dst_get(struct sock *sk) 1744 { 1745 return rcu_dereference_check(sk->sk_dst_cache, 1746 lockdep_sock_is_held(sk)); 1747 } 1748 1749 static inline struct dst_entry * 1750 sk_dst_get(struct sock *sk) 1751 { 1752 struct dst_entry *dst; 1753 1754 rcu_read_lock(); 1755 dst = rcu_dereference(sk->sk_dst_cache); 1756 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1757 dst = NULL; 1758 rcu_read_unlock(); 1759 return dst; 1760 } 1761 1762 static inline void dst_negative_advice(struct sock *sk) 1763 { 1764 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1765 1766 sk_rethink_txhash(sk); 1767 1768 if (dst && dst->ops->negative_advice) { 1769 ndst = dst->ops->negative_advice(dst); 1770 1771 if (ndst != dst) { 1772 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1773 sk_tx_queue_clear(sk); 1774 sk->sk_dst_pending_confirm = 0; 1775 } 1776 } 1777 } 1778 1779 static inline void 1780 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1781 { 1782 struct dst_entry *old_dst; 1783 1784 sk_tx_queue_clear(sk); 1785 sk->sk_dst_pending_confirm = 0; 1786 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1787 lockdep_sock_is_held(sk)); 1788 rcu_assign_pointer(sk->sk_dst_cache, dst); 1789 dst_release(old_dst); 1790 } 1791 1792 static inline void 1793 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1794 { 1795 struct dst_entry *old_dst; 1796 1797 sk_tx_queue_clear(sk); 1798 sk->sk_dst_pending_confirm = 0; 1799 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1800 dst_release(old_dst); 1801 } 1802 1803 static inline void 1804 __sk_dst_reset(struct sock *sk) 1805 { 1806 __sk_dst_set(sk, NULL); 1807 } 1808 1809 static inline void 1810 sk_dst_reset(struct sock *sk) 1811 { 1812 sk_dst_set(sk, NULL); 1813 } 1814 1815 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1816 1817 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1818 1819 static inline void sk_dst_confirm(struct sock *sk) 1820 { 1821 if (!sk->sk_dst_pending_confirm) 1822 sk->sk_dst_pending_confirm = 1; 1823 } 1824 1825 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1826 { 1827 if (skb_get_dst_pending_confirm(skb)) { 1828 struct sock *sk = skb->sk; 1829 unsigned long now = jiffies; 1830 1831 /* avoid dirtying neighbour */ 1832 if (n->confirmed != now) 1833 n->confirmed = now; 1834 if (sk && sk->sk_dst_pending_confirm) 1835 sk->sk_dst_pending_confirm = 0; 1836 } 1837 } 1838 1839 bool sk_mc_loop(struct sock *sk); 1840 1841 static inline bool sk_can_gso(const struct sock *sk) 1842 { 1843 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1844 } 1845 1846 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1847 1848 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1849 { 1850 sk->sk_route_nocaps |= flags; 1851 sk->sk_route_caps &= ~flags; 1852 } 1853 1854 static inline bool sk_check_csum_caps(struct sock *sk) 1855 { 1856 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1857 (sk->sk_family == PF_INET && 1858 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1859 (sk->sk_family == PF_INET6 && 1860 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1861 } 1862 1863 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1864 struct iov_iter *from, char *to, 1865 int copy, int offset) 1866 { 1867 if (skb->ip_summed == CHECKSUM_NONE) { 1868 __wsum csum = 0; 1869 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1870 return -EFAULT; 1871 skb->csum = csum_block_add(skb->csum, csum, offset); 1872 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1873 if (!copy_from_iter_full_nocache(to, copy, from)) 1874 return -EFAULT; 1875 } else if (!copy_from_iter_full(to, copy, from)) 1876 return -EFAULT; 1877 1878 return 0; 1879 } 1880 1881 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1882 struct iov_iter *from, int copy) 1883 { 1884 int err, offset = skb->len; 1885 1886 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1887 copy, offset); 1888 if (err) 1889 __skb_trim(skb, offset); 1890 1891 return err; 1892 } 1893 1894 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1895 struct sk_buff *skb, 1896 struct page *page, 1897 int off, int copy) 1898 { 1899 int err; 1900 1901 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1902 copy, skb->len); 1903 if (err) 1904 return err; 1905 1906 skb->len += copy; 1907 skb->data_len += copy; 1908 skb->truesize += copy; 1909 sk->sk_wmem_queued += copy; 1910 sk_mem_charge(sk, copy); 1911 return 0; 1912 } 1913 1914 /** 1915 * sk_wmem_alloc_get - returns write allocations 1916 * @sk: socket 1917 * 1918 * Returns sk_wmem_alloc minus initial offset of one 1919 */ 1920 static inline int sk_wmem_alloc_get(const struct sock *sk) 1921 { 1922 return atomic_read(&sk->sk_wmem_alloc) - 1; 1923 } 1924 1925 /** 1926 * sk_rmem_alloc_get - returns read allocations 1927 * @sk: socket 1928 * 1929 * Returns sk_rmem_alloc 1930 */ 1931 static inline int sk_rmem_alloc_get(const struct sock *sk) 1932 { 1933 return atomic_read(&sk->sk_rmem_alloc); 1934 } 1935 1936 /** 1937 * sk_has_allocations - check if allocations are outstanding 1938 * @sk: socket 1939 * 1940 * Returns true if socket has write or read allocations 1941 */ 1942 static inline bool sk_has_allocations(const struct sock *sk) 1943 { 1944 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1945 } 1946 1947 /** 1948 * skwq_has_sleeper - check if there are any waiting processes 1949 * @wq: struct socket_wq 1950 * 1951 * Returns true if socket_wq has waiting processes 1952 * 1953 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1954 * barrier call. They were added due to the race found within the tcp code. 1955 * 1956 * Consider following tcp code paths: 1957 * 1958 * CPU1 CPU2 1959 * 1960 * sys_select receive packet 1961 * ... ... 1962 * __add_wait_queue update tp->rcv_nxt 1963 * ... ... 1964 * tp->rcv_nxt check sock_def_readable 1965 * ... { 1966 * schedule rcu_read_lock(); 1967 * wq = rcu_dereference(sk->sk_wq); 1968 * if (wq && waitqueue_active(&wq->wait)) 1969 * wake_up_interruptible(&wq->wait) 1970 * ... 1971 * } 1972 * 1973 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1974 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1975 * could then endup calling schedule and sleep forever if there are no more 1976 * data on the socket. 1977 * 1978 */ 1979 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1980 { 1981 return wq && wq_has_sleeper(&wq->wait); 1982 } 1983 1984 /** 1985 * sock_poll_wait - place memory barrier behind the poll_wait call. 1986 * @filp: file 1987 * @wait_address: socket wait queue 1988 * @p: poll_table 1989 * 1990 * See the comments in the wq_has_sleeper function. 1991 */ 1992 static inline void sock_poll_wait(struct file *filp, 1993 wait_queue_head_t *wait_address, poll_table *p) 1994 { 1995 if (!poll_does_not_wait(p) && wait_address) { 1996 poll_wait(filp, wait_address, p); 1997 /* We need to be sure we are in sync with the 1998 * socket flags modification. 1999 * 2000 * This memory barrier is paired in the wq_has_sleeper. 2001 */ 2002 smp_mb(); 2003 } 2004 } 2005 2006 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2007 { 2008 if (sk->sk_txhash) { 2009 skb->l4_hash = 1; 2010 skb->hash = sk->sk_txhash; 2011 } 2012 } 2013 2014 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2015 2016 /* 2017 * Queue a received datagram if it will fit. Stream and sequenced 2018 * protocols can't normally use this as they need to fit buffers in 2019 * and play with them. 2020 * 2021 * Inlined as it's very short and called for pretty much every 2022 * packet ever received. 2023 */ 2024 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2025 { 2026 skb_orphan(skb); 2027 skb->sk = sk; 2028 skb->destructor = sock_rfree; 2029 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2030 sk_mem_charge(sk, skb->truesize); 2031 } 2032 2033 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2034 unsigned long expires); 2035 2036 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2037 2038 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff *skb, 2039 unsigned int flags, 2040 void (*destructor)(struct sock *sk, 2041 struct sk_buff *skb)); 2042 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2043 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2044 2045 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2046 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2047 2048 /* 2049 * Recover an error report and clear atomically 2050 */ 2051 2052 static inline int sock_error(struct sock *sk) 2053 { 2054 int err; 2055 if (likely(!sk->sk_err)) 2056 return 0; 2057 err = xchg(&sk->sk_err, 0); 2058 return -err; 2059 } 2060 2061 static inline unsigned long sock_wspace(struct sock *sk) 2062 { 2063 int amt = 0; 2064 2065 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2066 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2067 if (amt < 0) 2068 amt = 0; 2069 } 2070 return amt; 2071 } 2072 2073 /* Note: 2074 * We use sk->sk_wq_raw, from contexts knowing this 2075 * pointer is not NULL and cannot disappear/change. 2076 */ 2077 static inline void sk_set_bit(int nr, struct sock *sk) 2078 { 2079 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2080 !sock_flag(sk, SOCK_FASYNC)) 2081 return; 2082 2083 set_bit(nr, &sk->sk_wq_raw->flags); 2084 } 2085 2086 static inline void sk_clear_bit(int nr, struct sock *sk) 2087 { 2088 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2089 !sock_flag(sk, SOCK_FASYNC)) 2090 return; 2091 2092 clear_bit(nr, &sk->sk_wq_raw->flags); 2093 } 2094 2095 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2096 { 2097 if (sock_flag(sk, SOCK_FASYNC)) { 2098 rcu_read_lock(); 2099 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2100 rcu_read_unlock(); 2101 } 2102 } 2103 2104 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2105 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2106 * Note: for send buffers, TCP works better if we can build two skbs at 2107 * minimum. 2108 */ 2109 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2110 2111 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2112 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2113 2114 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2115 { 2116 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2117 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2118 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2119 } 2120 } 2121 2122 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2123 bool force_schedule); 2124 2125 /** 2126 * sk_page_frag - return an appropriate page_frag 2127 * @sk: socket 2128 * 2129 * If socket allocation mode allows current thread to sleep, it means its 2130 * safe to use the per task page_frag instead of the per socket one. 2131 */ 2132 static inline struct page_frag *sk_page_frag(struct sock *sk) 2133 { 2134 if (gfpflags_allow_blocking(sk->sk_allocation)) 2135 return ¤t->task_frag; 2136 2137 return &sk->sk_frag; 2138 } 2139 2140 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2141 2142 /* 2143 * Default write policy as shown to user space via poll/select/SIGIO 2144 */ 2145 static inline bool sock_writeable(const struct sock *sk) 2146 { 2147 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2148 } 2149 2150 static inline gfp_t gfp_any(void) 2151 { 2152 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2153 } 2154 2155 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2156 { 2157 return noblock ? 0 : sk->sk_rcvtimeo; 2158 } 2159 2160 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2161 { 2162 return noblock ? 0 : sk->sk_sndtimeo; 2163 } 2164 2165 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2166 { 2167 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2168 } 2169 2170 /* Alas, with timeout socket operations are not restartable. 2171 * Compare this to poll(). 2172 */ 2173 static inline int sock_intr_errno(long timeo) 2174 { 2175 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2176 } 2177 2178 struct sock_skb_cb { 2179 u32 dropcount; 2180 }; 2181 2182 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2183 * using skb->cb[] would keep using it directly and utilize its 2184 * alignement guarantee. 2185 */ 2186 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2187 sizeof(struct sock_skb_cb))) 2188 2189 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2190 SOCK_SKB_CB_OFFSET)) 2191 2192 #define sock_skb_cb_check_size(size) \ 2193 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2194 2195 static inline void 2196 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2197 { 2198 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2199 atomic_read(&sk->sk_drops) : 0; 2200 } 2201 2202 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2203 { 2204 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2205 2206 atomic_add(segs, &sk->sk_drops); 2207 } 2208 2209 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2210 struct sk_buff *skb); 2211 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2212 struct sk_buff *skb); 2213 2214 static inline void 2215 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2216 { 2217 ktime_t kt = skb->tstamp; 2218 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2219 2220 /* 2221 * generate control messages if 2222 * - receive time stamping in software requested 2223 * - software time stamp available and wanted 2224 * - hardware time stamps available and wanted 2225 */ 2226 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2227 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2228 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2229 (hwtstamps->hwtstamp && 2230 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2231 __sock_recv_timestamp(msg, sk, skb); 2232 else 2233 sk->sk_stamp = kt; 2234 2235 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2236 __sock_recv_wifi_status(msg, sk, skb); 2237 } 2238 2239 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2240 struct sk_buff *skb); 2241 2242 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2243 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2244 struct sk_buff *skb) 2245 { 2246 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2247 (1UL << SOCK_RCVTSTAMP)) 2248 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2249 SOF_TIMESTAMPING_RAW_HARDWARE) 2250 2251 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2252 __sock_recv_ts_and_drops(msg, sk, skb); 2253 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2254 sk->sk_stamp = skb->tstamp; 2255 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2256 sk->sk_stamp = 0; 2257 } 2258 2259 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2260 2261 /** 2262 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2263 * @sk: socket sending this packet 2264 * @tsflags: timestamping flags to use 2265 * @tx_flags: completed with instructions for time stamping 2266 * 2267 * Note : callers should take care of initial *tx_flags value (usually 0) 2268 */ 2269 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2270 __u8 *tx_flags) 2271 { 2272 if (unlikely(tsflags)) 2273 __sock_tx_timestamp(tsflags, tx_flags); 2274 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2275 *tx_flags |= SKBTX_WIFI_STATUS; 2276 } 2277 2278 /** 2279 * sk_eat_skb - Release a skb if it is no longer needed 2280 * @sk: socket to eat this skb from 2281 * @skb: socket buffer to eat 2282 * 2283 * This routine must be called with interrupts disabled or with the socket 2284 * locked so that the sk_buff queue operation is ok. 2285 */ 2286 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2287 { 2288 __skb_unlink(skb, &sk->sk_receive_queue); 2289 __kfree_skb(skb); 2290 } 2291 2292 static inline 2293 struct net *sock_net(const struct sock *sk) 2294 { 2295 return read_pnet(&sk->sk_net); 2296 } 2297 2298 static inline 2299 void sock_net_set(struct sock *sk, struct net *net) 2300 { 2301 write_pnet(&sk->sk_net, net); 2302 } 2303 2304 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2305 { 2306 if (skb->sk) { 2307 struct sock *sk = skb->sk; 2308 2309 skb->destructor = NULL; 2310 skb->sk = NULL; 2311 return sk; 2312 } 2313 return NULL; 2314 } 2315 2316 /* This helper checks if a socket is a full socket, 2317 * ie _not_ a timewait or request socket. 2318 */ 2319 static inline bool sk_fullsock(const struct sock *sk) 2320 { 2321 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2322 } 2323 2324 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2325 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2326 */ 2327 static inline bool sk_listener(const struct sock *sk) 2328 { 2329 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2330 } 2331 2332 /** 2333 * sk_state_load - read sk->sk_state for lockless contexts 2334 * @sk: socket pointer 2335 * 2336 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2337 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2338 */ 2339 static inline int sk_state_load(const struct sock *sk) 2340 { 2341 return smp_load_acquire(&sk->sk_state); 2342 } 2343 2344 /** 2345 * sk_state_store - update sk->sk_state 2346 * @sk: socket pointer 2347 * @newstate: new state 2348 * 2349 * Paired with sk_state_load(). Should be used in contexts where 2350 * state change might impact lockless readers. 2351 */ 2352 static inline void sk_state_store(struct sock *sk, int newstate) 2353 { 2354 smp_store_release(&sk->sk_state, newstate); 2355 } 2356 2357 void sock_enable_timestamp(struct sock *sk, int flag); 2358 int sock_get_timestamp(struct sock *, struct timeval __user *); 2359 int sock_get_timestampns(struct sock *, struct timespec __user *); 2360 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2361 int type); 2362 2363 bool sk_ns_capable(const struct sock *sk, 2364 struct user_namespace *user_ns, int cap); 2365 bool sk_capable(const struct sock *sk, int cap); 2366 bool sk_net_capable(const struct sock *sk, int cap); 2367 2368 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2369 2370 extern __u32 sysctl_wmem_max; 2371 extern __u32 sysctl_rmem_max; 2372 2373 extern int sysctl_tstamp_allow_data; 2374 extern int sysctl_optmem_max; 2375 2376 extern __u32 sysctl_wmem_default; 2377 extern __u32 sysctl_rmem_default; 2378 2379 #endif /* _SOCK_H */ 2380