1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the AF_INET socket handler. 7 * 8 * Version: @(#)sock.h 1.0.4 05/13/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Corey Minyard <wf-rch!minyard@relay.EU.net> 13 * Florian La Roche <flla@stud.uni-sb.de> 14 * 15 * Fixes: 16 * Alan Cox : Volatiles in skbuff pointers. See 17 * skbuff comments. May be overdone, 18 * better to prove they can be removed 19 * than the reverse. 20 * Alan Cox : Added a zapped field for tcp to note 21 * a socket is reset and must stay shut up 22 * Alan Cox : New fields for options 23 * Pauline Middelink : identd support 24 * Alan Cox : Eliminate low level recv/recvfrom 25 * David S. Miller : New socket lookup architecture. 26 * Steve Whitehouse: Default routines for sock_ops 27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 28 * protinfo be just a void pointer, as the 29 * protocol specific parts were moved to 30 * respective headers and ipv4/v6, etc now 31 * use private slabcaches for its socks 32 * Pedro Hortas : New flags field for socket options 33 * 34 * 35 * This program is free software; you can redistribute it and/or 36 * modify it under the terms of the GNU General Public License 37 * as published by the Free Software Foundation; either version 38 * 2 of the License, or (at your option) any later version. 39 */ 40 #ifndef _SOCK_H 41 #define _SOCK_H 42 43 #include <linux/hardirq.h> 44 #include <linux/kernel.h> 45 #include <linux/list.h> 46 #include <linux/list_nulls.h> 47 #include <linux/timer.h> 48 #include <linux/cache.h> 49 #include <linux/bitops.h> 50 #include <linux/lockdep.h> 51 #include <linux/netdevice.h> 52 #include <linux/skbuff.h> /* struct sk_buff */ 53 #include <linux/mm.h> 54 #include <linux/security.h> 55 #include <linux/slab.h> 56 #include <linux/uaccess.h> 57 #include <linux/page_counter.h> 58 #include <linux/memcontrol.h> 59 #include <linux/static_key.h> 60 #include <linux/sched.h> 61 #include <linux/wait.h> 62 #include <linux/cgroup-defs.h> 63 64 #include <linux/filter.h> 65 #include <linux/rculist_nulls.h> 66 #include <linux/poll.h> 67 68 #include <linux/atomic.h> 69 #include <net/dst.h> 70 #include <net/checksum.h> 71 #include <net/tcp_states.h> 72 #include <linux/net_tstamp.h> 73 74 /* 75 * This structure really needs to be cleaned up. 76 * Most of it is for TCP, and not used by any of 77 * the other protocols. 78 */ 79 80 /* Define this to get the SOCK_DBG debugging facility. */ 81 #define SOCK_DEBUGGING 82 #ifdef SOCK_DEBUGGING 83 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 84 printk(KERN_DEBUG msg); } while (0) 85 #else 86 /* Validate arguments and do nothing */ 87 static inline __printf(2, 3) 88 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 89 { 90 } 91 #endif 92 93 /* This is the per-socket lock. The spinlock provides a synchronization 94 * between user contexts and software interrupt processing, whereas the 95 * mini-semaphore synchronizes multiple users amongst themselves. 96 */ 97 typedef struct { 98 spinlock_t slock; 99 int owned; 100 wait_queue_head_t wq; 101 /* 102 * We express the mutex-alike socket_lock semantics 103 * to the lock validator by explicitly managing 104 * the slock as a lock variant (in addition to 105 * the slock itself): 106 */ 107 #ifdef CONFIG_DEBUG_LOCK_ALLOC 108 struct lockdep_map dep_map; 109 #endif 110 } socket_lock_t; 111 112 struct sock; 113 struct proto; 114 struct net; 115 116 typedef __u32 __bitwise __portpair; 117 typedef __u64 __bitwise __addrpair; 118 119 /** 120 * struct sock_common - minimal network layer representation of sockets 121 * @skc_daddr: Foreign IPv4 addr 122 * @skc_rcv_saddr: Bound local IPv4 addr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_family: network address family 128 * @skc_state: Connection state 129 * @skc_reuse: %SO_REUSEADDR setting 130 * @skc_reuseport: %SO_REUSEPORT setting 131 * @skc_bound_dev_if: bound device index if != 0 132 * @skc_bind_node: bind hash linkage for various protocol lookup tables 133 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 134 * @skc_prot: protocol handlers inside a network family 135 * @skc_net: reference to the network namespace of this socket 136 * @skc_node: main hash linkage for various protocol lookup tables 137 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 138 * @skc_tx_queue_mapping: tx queue number for this connection 139 * @skc_flags: place holder for sk_flags 140 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 141 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 142 * @skc_incoming_cpu: record/match cpu processing incoming packets 143 * @skc_refcnt: reference count 144 * 145 * This is the minimal network layer representation of sockets, the header 146 * for struct sock and struct inet_timewait_sock. 147 */ 148 struct sock_common { 149 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 150 * address on 64bit arches : cf INET_MATCH() 151 */ 152 union { 153 __addrpair skc_addrpair; 154 struct { 155 __be32 skc_daddr; 156 __be32 skc_rcv_saddr; 157 }; 158 }; 159 union { 160 unsigned int skc_hash; 161 __u16 skc_u16hashes[2]; 162 }; 163 /* skc_dport && skc_num must be grouped as well */ 164 union { 165 __portpair skc_portpair; 166 struct { 167 __be16 skc_dport; 168 __u16 skc_num; 169 }; 170 }; 171 172 unsigned short skc_family; 173 volatile unsigned char skc_state; 174 unsigned char skc_reuse:4; 175 unsigned char skc_reuseport:1; 176 unsigned char skc_ipv6only:1; 177 unsigned char skc_net_refcnt:1; 178 int skc_bound_dev_if; 179 union { 180 struct hlist_node skc_bind_node; 181 struct hlist_node skc_portaddr_node; 182 }; 183 struct proto *skc_prot; 184 possible_net_t skc_net; 185 186 #if IS_ENABLED(CONFIG_IPV6) 187 struct in6_addr skc_v6_daddr; 188 struct in6_addr skc_v6_rcv_saddr; 189 #endif 190 191 atomic64_t skc_cookie; 192 193 /* following fields are padding to force 194 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 195 * assuming IPV6 is enabled. We use this padding differently 196 * for different kind of 'sockets' 197 */ 198 union { 199 unsigned long skc_flags; 200 struct sock *skc_listener; /* request_sock */ 201 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 202 }; 203 /* 204 * fields between dontcopy_begin/dontcopy_end 205 * are not copied in sock_copy() 206 */ 207 /* private: */ 208 int skc_dontcopy_begin[0]; 209 /* public: */ 210 union { 211 struct hlist_node skc_node; 212 struct hlist_nulls_node skc_nulls_node; 213 }; 214 int skc_tx_queue_mapping; 215 union { 216 int skc_incoming_cpu; 217 u32 skc_rcv_wnd; 218 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 219 }; 220 221 atomic_t skc_refcnt; 222 /* private: */ 223 int skc_dontcopy_end[0]; 224 union { 225 u32 skc_rxhash; 226 u32 skc_window_clamp; 227 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 228 }; 229 /* public: */ 230 }; 231 232 /** 233 * struct sock - network layer representation of sockets 234 * @__sk_common: shared layout with inet_timewait_sock 235 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 236 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 237 * @sk_lock: synchronizer 238 * @sk_rcvbuf: size of receive buffer in bytes 239 * @sk_wq: sock wait queue and async head 240 * @sk_rx_dst: receive input route used by early demux 241 * @sk_dst_cache: destination cache 242 * @sk_policy: flow policy 243 * @sk_receive_queue: incoming packets 244 * @sk_wmem_alloc: transmit queue bytes committed 245 * @sk_write_queue: Packet sending queue 246 * @sk_omem_alloc: "o" is "option" or "other" 247 * @sk_wmem_queued: persistent queue size 248 * @sk_forward_alloc: space allocated forward 249 * @sk_napi_id: id of the last napi context to receive data for sk 250 * @sk_ll_usec: usecs to busypoll when there is no data 251 * @sk_allocation: allocation mode 252 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 253 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 254 * @sk_sndbuf: size of send buffer in bytes 255 * @sk_padding: unused element for alignment 256 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 257 * @sk_no_check_rx: allow zero checksum in RX packets 258 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 259 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 260 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 261 * @sk_gso_max_size: Maximum GSO segment size to build 262 * @sk_gso_max_segs: Maximum number of GSO segments 263 * @sk_lingertime: %SO_LINGER l_linger setting 264 * @sk_backlog: always used with the per-socket spinlock held 265 * @sk_callback_lock: used with the callbacks in the end of this struct 266 * @sk_error_queue: rarely used 267 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 268 * IPV6_ADDRFORM for instance) 269 * @sk_err: last error 270 * @sk_err_soft: errors that don't cause failure but are the cause of a 271 * persistent failure not just 'timed out' 272 * @sk_drops: raw/udp drops counter 273 * @sk_ack_backlog: current listen backlog 274 * @sk_max_ack_backlog: listen backlog set in listen() 275 * @sk_priority: %SO_PRIORITY setting 276 * @sk_type: socket type (%SOCK_STREAM, etc) 277 * @sk_protocol: which protocol this socket belongs in this network family 278 * @sk_peer_pid: &struct pid for this socket's peer 279 * @sk_peer_cred: %SO_PEERCRED setting 280 * @sk_rcvlowat: %SO_RCVLOWAT setting 281 * @sk_rcvtimeo: %SO_RCVTIMEO setting 282 * @sk_sndtimeo: %SO_SNDTIMEO setting 283 * @sk_txhash: computed flow hash for use on transmit 284 * @sk_filter: socket filtering instructions 285 * @sk_timer: sock cleanup timer 286 * @sk_stamp: time stamp of last packet received 287 * @sk_tsflags: SO_TIMESTAMPING socket options 288 * @sk_tskey: counter to disambiguate concurrent tstamp requests 289 * @sk_socket: Identd and reporting IO signals 290 * @sk_user_data: RPC layer private data 291 * @sk_frag: cached page frag 292 * @sk_peek_off: current peek_offset value 293 * @sk_send_head: front of stuff to transmit 294 * @sk_security: used by security modules 295 * @sk_mark: generic packet mark 296 * @sk_cgrp_data: cgroup data for this cgroup 297 * @sk_memcg: this socket's memory cgroup association 298 * @sk_write_pending: a write to stream socket waits to start 299 * @sk_state_change: callback to indicate change in the state of the sock 300 * @sk_data_ready: callback to indicate there is data to be processed 301 * @sk_write_space: callback to indicate there is bf sending space available 302 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 303 * @sk_backlog_rcv: callback to process the backlog 304 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 305 * @sk_reuseport_cb: reuseport group container 306 * @sk_rcu: used during RCU grace period 307 */ 308 struct sock { 309 /* 310 * Now struct inet_timewait_sock also uses sock_common, so please just 311 * don't add nothing before this first member (__sk_common) --acme 312 */ 313 struct sock_common __sk_common; 314 #define sk_node __sk_common.skc_node 315 #define sk_nulls_node __sk_common.skc_nulls_node 316 #define sk_refcnt __sk_common.skc_refcnt 317 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 318 319 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 320 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 321 #define sk_hash __sk_common.skc_hash 322 #define sk_portpair __sk_common.skc_portpair 323 #define sk_num __sk_common.skc_num 324 #define sk_dport __sk_common.skc_dport 325 #define sk_addrpair __sk_common.skc_addrpair 326 #define sk_daddr __sk_common.skc_daddr 327 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 328 #define sk_family __sk_common.skc_family 329 #define sk_state __sk_common.skc_state 330 #define sk_reuse __sk_common.skc_reuse 331 #define sk_reuseport __sk_common.skc_reuseport 332 #define sk_ipv6only __sk_common.skc_ipv6only 333 #define sk_net_refcnt __sk_common.skc_net_refcnt 334 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 335 #define sk_bind_node __sk_common.skc_bind_node 336 #define sk_prot __sk_common.skc_prot 337 #define sk_net __sk_common.skc_net 338 #define sk_v6_daddr __sk_common.skc_v6_daddr 339 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 340 #define sk_cookie __sk_common.skc_cookie 341 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 342 #define sk_flags __sk_common.skc_flags 343 #define sk_rxhash __sk_common.skc_rxhash 344 345 socket_lock_t sk_lock; 346 struct sk_buff_head sk_receive_queue; 347 /* 348 * The backlog queue is special, it is always used with 349 * the per-socket spinlock held and requires low latency 350 * access. Therefore we special case it's implementation. 351 * Note : rmem_alloc is in this structure to fill a hole 352 * on 64bit arches, not because its logically part of 353 * backlog. 354 */ 355 struct { 356 atomic_t rmem_alloc; 357 int len; 358 struct sk_buff *head; 359 struct sk_buff *tail; 360 } sk_backlog; 361 #define sk_rmem_alloc sk_backlog.rmem_alloc 362 int sk_forward_alloc; 363 364 __u32 sk_txhash; 365 #ifdef CONFIG_NET_RX_BUSY_POLL 366 unsigned int sk_napi_id; 367 unsigned int sk_ll_usec; 368 #endif 369 atomic_t sk_drops; 370 int sk_rcvbuf; 371 372 struct sk_filter __rcu *sk_filter; 373 union { 374 struct socket_wq __rcu *sk_wq; 375 struct socket_wq *sk_wq_raw; 376 }; 377 #ifdef CONFIG_XFRM 378 struct xfrm_policy __rcu *sk_policy[2]; 379 #endif 380 struct dst_entry *sk_rx_dst; 381 struct dst_entry __rcu *sk_dst_cache; 382 /* Note: 32bit hole on 64bit arches */ 383 atomic_t sk_wmem_alloc; 384 atomic_t sk_omem_alloc; 385 int sk_sndbuf; 386 struct sk_buff_head sk_write_queue; 387 388 /* 389 * Because of non atomicity rules, all 390 * changes are protected by socket lock. 391 */ 392 kmemcheck_bitfield_begin(flags); 393 unsigned int sk_padding : 2, 394 sk_no_check_tx : 1, 395 sk_no_check_rx : 1, 396 sk_userlocks : 4, 397 sk_protocol : 8, 398 sk_type : 16; 399 #define SK_PROTOCOL_MAX U8_MAX 400 kmemcheck_bitfield_end(flags); 401 402 int sk_wmem_queued; 403 gfp_t sk_allocation; 404 u32 sk_pacing_rate; /* bytes per second */ 405 u32 sk_max_pacing_rate; 406 netdev_features_t sk_route_caps; 407 netdev_features_t sk_route_nocaps; 408 int sk_gso_type; 409 unsigned int sk_gso_max_size; 410 u16 sk_gso_max_segs; 411 int sk_rcvlowat; 412 unsigned long sk_lingertime; 413 struct sk_buff_head sk_error_queue; 414 struct proto *sk_prot_creator; 415 rwlock_t sk_callback_lock; 416 int sk_err, 417 sk_err_soft; 418 u32 sk_ack_backlog; 419 u32 sk_max_ack_backlog; 420 __u32 sk_priority; 421 __u32 sk_mark; 422 kuid_t sk_uid; 423 struct pid *sk_peer_pid; 424 const struct cred *sk_peer_cred; 425 long sk_rcvtimeo; 426 long sk_sndtimeo; 427 struct timer_list sk_timer; 428 ktime_t sk_stamp; 429 u16 sk_tsflags; 430 u8 sk_shutdown; 431 u32 sk_tskey; 432 struct socket *sk_socket; 433 void *sk_user_data; 434 struct page_frag sk_frag; 435 struct sk_buff *sk_send_head; 436 __s32 sk_peek_off; 437 int sk_write_pending; 438 #ifdef CONFIG_SECURITY 439 void *sk_security; 440 #endif 441 struct sock_cgroup_data sk_cgrp_data; 442 struct mem_cgroup *sk_memcg; 443 void (*sk_state_change)(struct sock *sk); 444 void (*sk_data_ready)(struct sock *sk); 445 void (*sk_write_space)(struct sock *sk); 446 void (*sk_error_report)(struct sock *sk); 447 int (*sk_backlog_rcv)(struct sock *sk, 448 struct sk_buff *skb); 449 void (*sk_destruct)(struct sock *sk); 450 struct sock_reuseport __rcu *sk_reuseport_cb; 451 struct rcu_head sk_rcu; 452 }; 453 454 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 455 456 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 457 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 458 459 /* 460 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 461 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 462 * on a socket means that the socket will reuse everybody else's port 463 * without looking at the other's sk_reuse value. 464 */ 465 466 #define SK_NO_REUSE 0 467 #define SK_CAN_REUSE 1 468 #define SK_FORCE_REUSE 2 469 470 int sk_set_peek_off(struct sock *sk, int val); 471 472 static inline int sk_peek_offset(struct sock *sk, int flags) 473 { 474 if (unlikely(flags & MSG_PEEK)) { 475 s32 off = READ_ONCE(sk->sk_peek_off); 476 if (off >= 0) 477 return off; 478 } 479 480 return 0; 481 } 482 483 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 484 { 485 s32 off = READ_ONCE(sk->sk_peek_off); 486 487 if (unlikely(off >= 0)) { 488 off = max_t(s32, off - val, 0); 489 WRITE_ONCE(sk->sk_peek_off, off); 490 } 491 } 492 493 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 494 { 495 sk_peek_offset_bwd(sk, -val); 496 } 497 498 /* 499 * Hashed lists helper routines 500 */ 501 static inline struct sock *sk_entry(const struct hlist_node *node) 502 { 503 return hlist_entry(node, struct sock, sk_node); 504 } 505 506 static inline struct sock *__sk_head(const struct hlist_head *head) 507 { 508 return hlist_entry(head->first, struct sock, sk_node); 509 } 510 511 static inline struct sock *sk_head(const struct hlist_head *head) 512 { 513 return hlist_empty(head) ? NULL : __sk_head(head); 514 } 515 516 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 517 { 518 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 519 } 520 521 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 522 { 523 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 524 } 525 526 static inline struct sock *sk_next(const struct sock *sk) 527 { 528 return sk->sk_node.next ? 529 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 530 } 531 532 static inline struct sock *sk_nulls_next(const struct sock *sk) 533 { 534 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 535 hlist_nulls_entry(sk->sk_nulls_node.next, 536 struct sock, sk_nulls_node) : 537 NULL; 538 } 539 540 static inline bool sk_unhashed(const struct sock *sk) 541 { 542 return hlist_unhashed(&sk->sk_node); 543 } 544 545 static inline bool sk_hashed(const struct sock *sk) 546 { 547 return !sk_unhashed(sk); 548 } 549 550 static inline void sk_node_init(struct hlist_node *node) 551 { 552 node->pprev = NULL; 553 } 554 555 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 556 { 557 node->pprev = NULL; 558 } 559 560 static inline void __sk_del_node(struct sock *sk) 561 { 562 __hlist_del(&sk->sk_node); 563 } 564 565 /* NB: equivalent to hlist_del_init_rcu */ 566 static inline bool __sk_del_node_init(struct sock *sk) 567 { 568 if (sk_hashed(sk)) { 569 __sk_del_node(sk); 570 sk_node_init(&sk->sk_node); 571 return true; 572 } 573 return false; 574 } 575 576 /* Grab socket reference count. This operation is valid only 577 when sk is ALREADY grabbed f.e. it is found in hash table 578 or a list and the lookup is made under lock preventing hash table 579 modifications. 580 */ 581 582 static __always_inline void sock_hold(struct sock *sk) 583 { 584 atomic_inc(&sk->sk_refcnt); 585 } 586 587 /* Ungrab socket in the context, which assumes that socket refcnt 588 cannot hit zero, f.e. it is true in context of any socketcall. 589 */ 590 static __always_inline void __sock_put(struct sock *sk) 591 { 592 atomic_dec(&sk->sk_refcnt); 593 } 594 595 static inline bool sk_del_node_init(struct sock *sk) 596 { 597 bool rc = __sk_del_node_init(sk); 598 599 if (rc) { 600 /* paranoid for a while -acme */ 601 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 602 __sock_put(sk); 603 } 604 return rc; 605 } 606 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 607 608 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 609 { 610 if (sk_hashed(sk)) { 611 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 612 return true; 613 } 614 return false; 615 } 616 617 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 618 { 619 bool rc = __sk_nulls_del_node_init_rcu(sk); 620 621 if (rc) { 622 /* paranoid for a while -acme */ 623 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 624 __sock_put(sk); 625 } 626 return rc; 627 } 628 629 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 630 { 631 hlist_add_head(&sk->sk_node, list); 632 } 633 634 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 635 { 636 sock_hold(sk); 637 __sk_add_node(sk, list); 638 } 639 640 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 641 { 642 sock_hold(sk); 643 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 644 sk->sk_family == AF_INET6) 645 hlist_add_tail_rcu(&sk->sk_node, list); 646 else 647 hlist_add_head_rcu(&sk->sk_node, list); 648 } 649 650 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 651 { 652 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 653 sk->sk_family == AF_INET6) 654 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 655 else 656 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 657 } 658 659 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 660 { 661 sock_hold(sk); 662 __sk_nulls_add_node_rcu(sk, list); 663 } 664 665 static inline void __sk_del_bind_node(struct sock *sk) 666 { 667 __hlist_del(&sk->sk_bind_node); 668 } 669 670 static inline void sk_add_bind_node(struct sock *sk, 671 struct hlist_head *list) 672 { 673 hlist_add_head(&sk->sk_bind_node, list); 674 } 675 676 #define sk_for_each(__sk, list) \ 677 hlist_for_each_entry(__sk, list, sk_node) 678 #define sk_for_each_rcu(__sk, list) \ 679 hlist_for_each_entry_rcu(__sk, list, sk_node) 680 #define sk_nulls_for_each(__sk, node, list) \ 681 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 682 #define sk_nulls_for_each_rcu(__sk, node, list) \ 683 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 684 #define sk_for_each_from(__sk) \ 685 hlist_for_each_entry_from(__sk, sk_node) 686 #define sk_nulls_for_each_from(__sk, node) \ 687 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 688 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 689 #define sk_for_each_safe(__sk, tmp, list) \ 690 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 691 #define sk_for_each_bound(__sk, list) \ 692 hlist_for_each_entry(__sk, list, sk_bind_node) 693 694 /** 695 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 696 * @tpos: the type * to use as a loop cursor. 697 * @pos: the &struct hlist_node to use as a loop cursor. 698 * @head: the head for your list. 699 * @offset: offset of hlist_node within the struct. 700 * 701 */ 702 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 703 for (pos = rcu_dereference((head)->first); \ 704 pos != NULL && \ 705 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 706 pos = rcu_dereference(pos->next)) 707 708 static inline struct user_namespace *sk_user_ns(struct sock *sk) 709 { 710 /* Careful only use this in a context where these parameters 711 * can not change and must all be valid, such as recvmsg from 712 * userspace. 713 */ 714 return sk->sk_socket->file->f_cred->user_ns; 715 } 716 717 /* Sock flags */ 718 enum sock_flags { 719 SOCK_DEAD, 720 SOCK_DONE, 721 SOCK_URGINLINE, 722 SOCK_KEEPOPEN, 723 SOCK_LINGER, 724 SOCK_DESTROY, 725 SOCK_BROADCAST, 726 SOCK_TIMESTAMP, 727 SOCK_ZAPPED, 728 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 729 SOCK_DBG, /* %SO_DEBUG setting */ 730 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 731 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 732 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 733 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 734 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 735 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 736 SOCK_FASYNC, /* fasync() active */ 737 SOCK_RXQ_OVFL, 738 SOCK_ZEROCOPY, /* buffers from userspace */ 739 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 740 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 741 * Will use last 4 bytes of packet sent from 742 * user-space instead. 743 */ 744 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 745 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 746 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 747 }; 748 749 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 750 751 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 752 { 753 nsk->sk_flags = osk->sk_flags; 754 } 755 756 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 757 { 758 __set_bit(flag, &sk->sk_flags); 759 } 760 761 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 762 { 763 __clear_bit(flag, &sk->sk_flags); 764 } 765 766 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 767 { 768 return test_bit(flag, &sk->sk_flags); 769 } 770 771 #ifdef CONFIG_NET 772 extern struct static_key memalloc_socks; 773 static inline int sk_memalloc_socks(void) 774 { 775 return static_key_false(&memalloc_socks); 776 } 777 #else 778 779 static inline int sk_memalloc_socks(void) 780 { 781 return 0; 782 } 783 784 #endif 785 786 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 787 { 788 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 789 } 790 791 static inline void sk_acceptq_removed(struct sock *sk) 792 { 793 sk->sk_ack_backlog--; 794 } 795 796 static inline void sk_acceptq_added(struct sock *sk) 797 { 798 sk->sk_ack_backlog++; 799 } 800 801 static inline bool sk_acceptq_is_full(const struct sock *sk) 802 { 803 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 804 } 805 806 /* 807 * Compute minimal free write space needed to queue new packets. 808 */ 809 static inline int sk_stream_min_wspace(const struct sock *sk) 810 { 811 return sk->sk_wmem_queued >> 1; 812 } 813 814 static inline int sk_stream_wspace(const struct sock *sk) 815 { 816 return sk->sk_sndbuf - sk->sk_wmem_queued; 817 } 818 819 void sk_stream_write_space(struct sock *sk); 820 821 /* OOB backlog add */ 822 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 823 { 824 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 825 skb_dst_force_safe(skb); 826 827 if (!sk->sk_backlog.tail) 828 sk->sk_backlog.head = skb; 829 else 830 sk->sk_backlog.tail->next = skb; 831 832 sk->sk_backlog.tail = skb; 833 skb->next = NULL; 834 } 835 836 /* 837 * Take into account size of receive queue and backlog queue 838 * Do not take into account this skb truesize, 839 * to allow even a single big packet to come. 840 */ 841 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 842 { 843 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 844 845 return qsize > limit; 846 } 847 848 /* The per-socket spinlock must be held here. */ 849 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 850 unsigned int limit) 851 { 852 if (sk_rcvqueues_full(sk, limit)) 853 return -ENOBUFS; 854 855 /* 856 * If the skb was allocated from pfmemalloc reserves, only 857 * allow SOCK_MEMALLOC sockets to use it as this socket is 858 * helping free memory 859 */ 860 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 861 return -ENOMEM; 862 863 __sk_add_backlog(sk, skb); 864 sk->sk_backlog.len += skb->truesize; 865 return 0; 866 } 867 868 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 869 870 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 871 { 872 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 873 return __sk_backlog_rcv(sk, skb); 874 875 return sk->sk_backlog_rcv(sk, skb); 876 } 877 878 static inline void sk_incoming_cpu_update(struct sock *sk) 879 { 880 sk->sk_incoming_cpu = raw_smp_processor_id(); 881 } 882 883 static inline void sock_rps_record_flow_hash(__u32 hash) 884 { 885 #ifdef CONFIG_RPS 886 struct rps_sock_flow_table *sock_flow_table; 887 888 rcu_read_lock(); 889 sock_flow_table = rcu_dereference(rps_sock_flow_table); 890 rps_record_sock_flow(sock_flow_table, hash); 891 rcu_read_unlock(); 892 #endif 893 } 894 895 static inline void sock_rps_record_flow(const struct sock *sk) 896 { 897 #ifdef CONFIG_RPS 898 sock_rps_record_flow_hash(sk->sk_rxhash); 899 #endif 900 } 901 902 static inline void sock_rps_save_rxhash(struct sock *sk, 903 const struct sk_buff *skb) 904 { 905 #ifdef CONFIG_RPS 906 if (unlikely(sk->sk_rxhash != skb->hash)) 907 sk->sk_rxhash = skb->hash; 908 #endif 909 } 910 911 static inline void sock_rps_reset_rxhash(struct sock *sk) 912 { 913 #ifdef CONFIG_RPS 914 sk->sk_rxhash = 0; 915 #endif 916 } 917 918 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 919 ({ int __rc; \ 920 release_sock(__sk); \ 921 __rc = __condition; \ 922 if (!__rc) { \ 923 *(__timeo) = wait_woken(__wait, \ 924 TASK_INTERRUPTIBLE, \ 925 *(__timeo)); \ 926 } \ 927 sched_annotate_sleep(); \ 928 lock_sock(__sk); \ 929 __rc = __condition; \ 930 __rc; \ 931 }) 932 933 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 934 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 935 void sk_stream_wait_close(struct sock *sk, long timeo_p); 936 int sk_stream_error(struct sock *sk, int flags, int err); 937 void sk_stream_kill_queues(struct sock *sk); 938 void sk_set_memalloc(struct sock *sk); 939 void sk_clear_memalloc(struct sock *sk); 940 941 void __sk_flush_backlog(struct sock *sk); 942 943 static inline bool sk_flush_backlog(struct sock *sk) 944 { 945 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 946 __sk_flush_backlog(sk); 947 return true; 948 } 949 return false; 950 } 951 952 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 953 954 struct request_sock_ops; 955 struct timewait_sock_ops; 956 struct inet_hashinfo; 957 struct raw_hashinfo; 958 struct module; 959 960 /* 961 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 962 * un-modified. Special care is taken when initializing object to zero. 963 */ 964 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 965 { 966 if (offsetof(struct sock, sk_node.next) != 0) 967 memset(sk, 0, offsetof(struct sock, sk_node.next)); 968 memset(&sk->sk_node.pprev, 0, 969 size - offsetof(struct sock, sk_node.pprev)); 970 } 971 972 /* Networking protocol blocks we attach to sockets. 973 * socket layer -> transport layer interface 974 */ 975 struct proto { 976 void (*close)(struct sock *sk, 977 long timeout); 978 int (*connect)(struct sock *sk, 979 struct sockaddr *uaddr, 980 int addr_len); 981 int (*disconnect)(struct sock *sk, int flags); 982 983 struct sock * (*accept)(struct sock *sk, int flags, int *err); 984 985 int (*ioctl)(struct sock *sk, int cmd, 986 unsigned long arg); 987 int (*init)(struct sock *sk); 988 void (*destroy)(struct sock *sk); 989 void (*shutdown)(struct sock *sk, int how); 990 int (*setsockopt)(struct sock *sk, int level, 991 int optname, char __user *optval, 992 unsigned int optlen); 993 int (*getsockopt)(struct sock *sk, int level, 994 int optname, char __user *optval, 995 int __user *option); 996 #ifdef CONFIG_COMPAT 997 int (*compat_setsockopt)(struct sock *sk, 998 int level, 999 int optname, char __user *optval, 1000 unsigned int optlen); 1001 int (*compat_getsockopt)(struct sock *sk, 1002 int level, 1003 int optname, char __user *optval, 1004 int __user *option); 1005 int (*compat_ioctl)(struct sock *sk, 1006 unsigned int cmd, unsigned long arg); 1007 #endif 1008 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1009 size_t len); 1010 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1011 size_t len, int noblock, int flags, 1012 int *addr_len); 1013 int (*sendpage)(struct sock *sk, struct page *page, 1014 int offset, size_t size, int flags); 1015 int (*bind)(struct sock *sk, 1016 struct sockaddr *uaddr, int addr_len); 1017 1018 int (*backlog_rcv) (struct sock *sk, 1019 struct sk_buff *skb); 1020 1021 void (*release_cb)(struct sock *sk); 1022 1023 /* Keeping track of sk's, looking them up, and port selection methods. */ 1024 int (*hash)(struct sock *sk); 1025 void (*unhash)(struct sock *sk); 1026 void (*rehash)(struct sock *sk); 1027 int (*get_port)(struct sock *sk, unsigned short snum); 1028 1029 /* Keeping track of sockets in use */ 1030 #ifdef CONFIG_PROC_FS 1031 unsigned int inuse_idx; 1032 #endif 1033 1034 bool (*stream_memory_free)(const struct sock *sk); 1035 /* Memory pressure */ 1036 void (*enter_memory_pressure)(struct sock *sk); 1037 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1038 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1039 /* 1040 * Pressure flag: try to collapse. 1041 * Technical note: it is used by multiple contexts non atomically. 1042 * All the __sk_mem_schedule() is of this nature: accounting 1043 * is strict, actions are advisory and have some latency. 1044 */ 1045 int *memory_pressure; 1046 long *sysctl_mem; 1047 int *sysctl_wmem; 1048 int *sysctl_rmem; 1049 int max_header; 1050 bool no_autobind; 1051 1052 struct kmem_cache *slab; 1053 unsigned int obj_size; 1054 int slab_flags; 1055 1056 struct percpu_counter *orphan_count; 1057 1058 struct request_sock_ops *rsk_prot; 1059 struct timewait_sock_ops *twsk_prot; 1060 1061 union { 1062 struct inet_hashinfo *hashinfo; 1063 struct udp_table *udp_table; 1064 struct raw_hashinfo *raw_hash; 1065 } h; 1066 1067 struct module *owner; 1068 1069 char name[32]; 1070 1071 struct list_head node; 1072 #ifdef SOCK_REFCNT_DEBUG 1073 atomic_t socks; 1074 #endif 1075 int (*diag_destroy)(struct sock *sk, int err); 1076 }; 1077 1078 int proto_register(struct proto *prot, int alloc_slab); 1079 void proto_unregister(struct proto *prot); 1080 1081 #ifdef SOCK_REFCNT_DEBUG 1082 static inline void sk_refcnt_debug_inc(struct sock *sk) 1083 { 1084 atomic_inc(&sk->sk_prot->socks); 1085 } 1086 1087 static inline void sk_refcnt_debug_dec(struct sock *sk) 1088 { 1089 atomic_dec(&sk->sk_prot->socks); 1090 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1091 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1092 } 1093 1094 static inline void sk_refcnt_debug_release(const struct sock *sk) 1095 { 1096 if (atomic_read(&sk->sk_refcnt) != 1) 1097 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1098 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1099 } 1100 #else /* SOCK_REFCNT_DEBUG */ 1101 #define sk_refcnt_debug_inc(sk) do { } while (0) 1102 #define sk_refcnt_debug_dec(sk) do { } while (0) 1103 #define sk_refcnt_debug_release(sk) do { } while (0) 1104 #endif /* SOCK_REFCNT_DEBUG */ 1105 1106 static inline bool sk_stream_memory_free(const struct sock *sk) 1107 { 1108 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1109 return false; 1110 1111 return sk->sk_prot->stream_memory_free ? 1112 sk->sk_prot->stream_memory_free(sk) : true; 1113 } 1114 1115 static inline bool sk_stream_is_writeable(const struct sock *sk) 1116 { 1117 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1118 sk_stream_memory_free(sk); 1119 } 1120 1121 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1122 struct cgroup *ancestor) 1123 { 1124 #ifdef CONFIG_SOCK_CGROUP_DATA 1125 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1126 ancestor); 1127 #else 1128 return -ENOTSUPP; 1129 #endif 1130 } 1131 1132 static inline bool sk_has_memory_pressure(const struct sock *sk) 1133 { 1134 return sk->sk_prot->memory_pressure != NULL; 1135 } 1136 1137 static inline bool sk_under_memory_pressure(const struct sock *sk) 1138 { 1139 if (!sk->sk_prot->memory_pressure) 1140 return false; 1141 1142 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1143 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1144 return true; 1145 1146 return !!*sk->sk_prot->memory_pressure; 1147 } 1148 1149 static inline void sk_leave_memory_pressure(struct sock *sk) 1150 { 1151 int *memory_pressure = sk->sk_prot->memory_pressure; 1152 1153 if (!memory_pressure) 1154 return; 1155 1156 if (*memory_pressure) 1157 *memory_pressure = 0; 1158 } 1159 1160 static inline void sk_enter_memory_pressure(struct sock *sk) 1161 { 1162 if (!sk->sk_prot->enter_memory_pressure) 1163 return; 1164 1165 sk->sk_prot->enter_memory_pressure(sk); 1166 } 1167 1168 static inline long 1169 sk_memory_allocated(const struct sock *sk) 1170 { 1171 return atomic_long_read(sk->sk_prot->memory_allocated); 1172 } 1173 1174 static inline long 1175 sk_memory_allocated_add(struct sock *sk, int amt) 1176 { 1177 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1178 } 1179 1180 static inline void 1181 sk_memory_allocated_sub(struct sock *sk, int amt) 1182 { 1183 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1184 } 1185 1186 static inline void sk_sockets_allocated_dec(struct sock *sk) 1187 { 1188 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1189 } 1190 1191 static inline void sk_sockets_allocated_inc(struct sock *sk) 1192 { 1193 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1194 } 1195 1196 static inline int 1197 sk_sockets_allocated_read_positive(struct sock *sk) 1198 { 1199 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1200 } 1201 1202 static inline int 1203 proto_sockets_allocated_sum_positive(struct proto *prot) 1204 { 1205 return percpu_counter_sum_positive(prot->sockets_allocated); 1206 } 1207 1208 static inline long 1209 proto_memory_allocated(struct proto *prot) 1210 { 1211 return atomic_long_read(prot->memory_allocated); 1212 } 1213 1214 static inline bool 1215 proto_memory_pressure(struct proto *prot) 1216 { 1217 if (!prot->memory_pressure) 1218 return false; 1219 return !!*prot->memory_pressure; 1220 } 1221 1222 1223 #ifdef CONFIG_PROC_FS 1224 /* Called with local bh disabled */ 1225 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1226 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1227 #else 1228 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1229 int inc) 1230 { 1231 } 1232 #endif 1233 1234 1235 /* With per-bucket locks this operation is not-atomic, so that 1236 * this version is not worse. 1237 */ 1238 static inline int __sk_prot_rehash(struct sock *sk) 1239 { 1240 sk->sk_prot->unhash(sk); 1241 return sk->sk_prot->hash(sk); 1242 } 1243 1244 /* About 10 seconds */ 1245 #define SOCK_DESTROY_TIME (10*HZ) 1246 1247 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1248 #define PROT_SOCK 1024 1249 1250 #define SHUTDOWN_MASK 3 1251 #define RCV_SHUTDOWN 1 1252 #define SEND_SHUTDOWN 2 1253 1254 #define SOCK_SNDBUF_LOCK 1 1255 #define SOCK_RCVBUF_LOCK 2 1256 #define SOCK_BINDADDR_LOCK 4 1257 #define SOCK_BINDPORT_LOCK 8 1258 1259 struct socket_alloc { 1260 struct socket socket; 1261 struct inode vfs_inode; 1262 }; 1263 1264 static inline struct socket *SOCKET_I(struct inode *inode) 1265 { 1266 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1267 } 1268 1269 static inline struct inode *SOCK_INODE(struct socket *socket) 1270 { 1271 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1272 } 1273 1274 /* 1275 * Functions for memory accounting 1276 */ 1277 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1278 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1279 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1280 void __sk_mem_reclaim(struct sock *sk, int amount); 1281 1282 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1283 * do not necessarily have 16x time more memory than 4KB ones. 1284 */ 1285 #define SK_MEM_QUANTUM 4096 1286 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1287 #define SK_MEM_SEND 0 1288 #define SK_MEM_RECV 1 1289 1290 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1291 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1292 { 1293 long val = sk->sk_prot->sysctl_mem[index]; 1294 1295 #if PAGE_SIZE > SK_MEM_QUANTUM 1296 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1297 #elif PAGE_SIZE < SK_MEM_QUANTUM 1298 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1299 #endif 1300 return val; 1301 } 1302 1303 static inline int sk_mem_pages(int amt) 1304 { 1305 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1306 } 1307 1308 static inline bool sk_has_account(struct sock *sk) 1309 { 1310 /* return true if protocol supports memory accounting */ 1311 return !!sk->sk_prot->memory_allocated; 1312 } 1313 1314 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1315 { 1316 if (!sk_has_account(sk)) 1317 return true; 1318 return size <= sk->sk_forward_alloc || 1319 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1320 } 1321 1322 static inline bool 1323 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1324 { 1325 if (!sk_has_account(sk)) 1326 return true; 1327 return size<= sk->sk_forward_alloc || 1328 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1329 skb_pfmemalloc(skb); 1330 } 1331 1332 static inline void sk_mem_reclaim(struct sock *sk) 1333 { 1334 if (!sk_has_account(sk)) 1335 return; 1336 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1337 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1338 } 1339 1340 static inline void sk_mem_reclaim_partial(struct sock *sk) 1341 { 1342 if (!sk_has_account(sk)) 1343 return; 1344 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1345 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1346 } 1347 1348 static inline void sk_mem_charge(struct sock *sk, int size) 1349 { 1350 if (!sk_has_account(sk)) 1351 return; 1352 sk->sk_forward_alloc -= size; 1353 } 1354 1355 static inline void sk_mem_uncharge(struct sock *sk, int size) 1356 { 1357 if (!sk_has_account(sk)) 1358 return; 1359 sk->sk_forward_alloc += size; 1360 1361 /* Avoid a possible overflow. 1362 * TCP send queues can make this happen, if sk_mem_reclaim() 1363 * is not called and more than 2 GBytes are released at once. 1364 * 1365 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1366 * no need to hold that much forward allocation anyway. 1367 */ 1368 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1369 __sk_mem_reclaim(sk, 1 << 20); 1370 } 1371 1372 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1373 { 1374 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1375 sk->sk_wmem_queued -= skb->truesize; 1376 sk_mem_uncharge(sk, skb->truesize); 1377 __kfree_skb(skb); 1378 } 1379 1380 static inline void sock_release_ownership(struct sock *sk) 1381 { 1382 if (sk->sk_lock.owned) { 1383 sk->sk_lock.owned = 0; 1384 1385 /* The sk_lock has mutex_unlock() semantics: */ 1386 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1387 } 1388 } 1389 1390 /* 1391 * Macro so as to not evaluate some arguments when 1392 * lockdep is not enabled. 1393 * 1394 * Mark both the sk_lock and the sk_lock.slock as a 1395 * per-address-family lock class. 1396 */ 1397 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1398 do { \ 1399 sk->sk_lock.owned = 0; \ 1400 init_waitqueue_head(&sk->sk_lock.wq); \ 1401 spin_lock_init(&(sk)->sk_lock.slock); \ 1402 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1403 sizeof((sk)->sk_lock)); \ 1404 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1405 (skey), (sname)); \ 1406 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1407 } while (0) 1408 1409 #ifdef CONFIG_LOCKDEP 1410 static inline bool lockdep_sock_is_held(const struct sock *csk) 1411 { 1412 struct sock *sk = (struct sock *)csk; 1413 1414 return lockdep_is_held(&sk->sk_lock) || 1415 lockdep_is_held(&sk->sk_lock.slock); 1416 } 1417 #endif 1418 1419 void lock_sock_nested(struct sock *sk, int subclass); 1420 1421 static inline void lock_sock(struct sock *sk) 1422 { 1423 lock_sock_nested(sk, 0); 1424 } 1425 1426 void release_sock(struct sock *sk); 1427 1428 /* BH context may only use the following locking interface. */ 1429 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1430 #define bh_lock_sock_nested(__sk) \ 1431 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1432 SINGLE_DEPTH_NESTING) 1433 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1434 1435 bool lock_sock_fast(struct sock *sk); 1436 /** 1437 * unlock_sock_fast - complement of lock_sock_fast 1438 * @sk: socket 1439 * @slow: slow mode 1440 * 1441 * fast unlock socket for user context. 1442 * If slow mode is on, we call regular release_sock() 1443 */ 1444 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1445 { 1446 if (slow) 1447 release_sock(sk); 1448 else 1449 spin_unlock_bh(&sk->sk_lock.slock); 1450 } 1451 1452 /* Used by processes to "lock" a socket state, so that 1453 * interrupts and bottom half handlers won't change it 1454 * from under us. It essentially blocks any incoming 1455 * packets, so that we won't get any new data or any 1456 * packets that change the state of the socket. 1457 * 1458 * While locked, BH processing will add new packets to 1459 * the backlog queue. This queue is processed by the 1460 * owner of the socket lock right before it is released. 1461 * 1462 * Since ~2.3.5 it is also exclusive sleep lock serializing 1463 * accesses from user process context. 1464 */ 1465 1466 static inline void sock_owned_by_me(const struct sock *sk) 1467 { 1468 #ifdef CONFIG_LOCKDEP 1469 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1470 #endif 1471 } 1472 1473 static inline bool sock_owned_by_user(const struct sock *sk) 1474 { 1475 sock_owned_by_me(sk); 1476 return sk->sk_lock.owned; 1477 } 1478 1479 /* no reclassification while locks are held */ 1480 static inline bool sock_allow_reclassification(const struct sock *csk) 1481 { 1482 struct sock *sk = (struct sock *)csk; 1483 1484 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1485 } 1486 1487 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1488 struct proto *prot, int kern); 1489 void sk_free(struct sock *sk); 1490 void sk_destruct(struct sock *sk); 1491 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1492 1493 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1494 gfp_t priority); 1495 void __sock_wfree(struct sk_buff *skb); 1496 void sock_wfree(struct sk_buff *skb); 1497 void skb_orphan_partial(struct sk_buff *skb); 1498 void sock_rfree(struct sk_buff *skb); 1499 void sock_efree(struct sk_buff *skb); 1500 #ifdef CONFIG_INET 1501 void sock_edemux(struct sk_buff *skb); 1502 #else 1503 #define sock_edemux(skb) sock_efree(skb) 1504 #endif 1505 1506 int sock_setsockopt(struct socket *sock, int level, int op, 1507 char __user *optval, unsigned int optlen); 1508 1509 int sock_getsockopt(struct socket *sock, int level, int op, 1510 char __user *optval, int __user *optlen); 1511 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1512 int noblock, int *errcode); 1513 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1514 unsigned long data_len, int noblock, 1515 int *errcode, int max_page_order); 1516 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1517 void sock_kfree_s(struct sock *sk, void *mem, int size); 1518 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1519 void sk_send_sigurg(struct sock *sk); 1520 1521 struct sockcm_cookie { 1522 u32 mark; 1523 u16 tsflags; 1524 }; 1525 1526 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1527 struct sockcm_cookie *sockc); 1528 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1529 struct sockcm_cookie *sockc); 1530 1531 /* 1532 * Functions to fill in entries in struct proto_ops when a protocol 1533 * does not implement a particular function. 1534 */ 1535 int sock_no_bind(struct socket *, struct sockaddr *, int); 1536 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1537 int sock_no_socketpair(struct socket *, struct socket *); 1538 int sock_no_accept(struct socket *, struct socket *, int); 1539 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1540 unsigned int sock_no_poll(struct file *, struct socket *, 1541 struct poll_table_struct *); 1542 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1543 int sock_no_listen(struct socket *, int); 1544 int sock_no_shutdown(struct socket *, int); 1545 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1546 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1547 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1548 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1549 int sock_no_mmap(struct file *file, struct socket *sock, 1550 struct vm_area_struct *vma); 1551 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1552 size_t size, int flags); 1553 1554 /* 1555 * Functions to fill in entries in struct proto_ops when a protocol 1556 * uses the inet style. 1557 */ 1558 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1559 char __user *optval, int __user *optlen); 1560 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1561 int flags); 1562 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1563 char __user *optval, unsigned int optlen); 1564 int compat_sock_common_getsockopt(struct socket *sock, int level, 1565 int optname, char __user *optval, int __user *optlen); 1566 int compat_sock_common_setsockopt(struct socket *sock, int level, 1567 int optname, char __user *optval, unsigned int optlen); 1568 1569 void sk_common_release(struct sock *sk); 1570 1571 /* 1572 * Default socket callbacks and setup code 1573 */ 1574 1575 /* Initialise core socket variables */ 1576 void sock_init_data(struct socket *sock, struct sock *sk); 1577 1578 /* 1579 * Socket reference counting postulates. 1580 * 1581 * * Each user of socket SHOULD hold a reference count. 1582 * * Each access point to socket (an hash table bucket, reference from a list, 1583 * running timer, skb in flight MUST hold a reference count. 1584 * * When reference count hits 0, it means it will never increase back. 1585 * * When reference count hits 0, it means that no references from 1586 * outside exist to this socket and current process on current CPU 1587 * is last user and may/should destroy this socket. 1588 * * sk_free is called from any context: process, BH, IRQ. When 1589 * it is called, socket has no references from outside -> sk_free 1590 * may release descendant resources allocated by the socket, but 1591 * to the time when it is called, socket is NOT referenced by any 1592 * hash tables, lists etc. 1593 * * Packets, delivered from outside (from network or from another process) 1594 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1595 * when they sit in queue. Otherwise, packets will leak to hole, when 1596 * socket is looked up by one cpu and unhasing is made by another CPU. 1597 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1598 * (leak to backlog). Packet socket does all the processing inside 1599 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1600 * use separate SMP lock, so that they are prone too. 1601 */ 1602 1603 /* Ungrab socket and destroy it, if it was the last reference. */ 1604 static inline void sock_put(struct sock *sk) 1605 { 1606 if (atomic_dec_and_test(&sk->sk_refcnt)) 1607 sk_free(sk); 1608 } 1609 /* Generic version of sock_put(), dealing with all sockets 1610 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1611 */ 1612 void sock_gen_put(struct sock *sk); 1613 1614 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1615 unsigned int trim_cap, bool refcounted); 1616 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1617 const int nested) 1618 { 1619 return __sk_receive_skb(sk, skb, nested, 1, true); 1620 } 1621 1622 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1623 { 1624 sk->sk_tx_queue_mapping = tx_queue; 1625 } 1626 1627 static inline void sk_tx_queue_clear(struct sock *sk) 1628 { 1629 sk->sk_tx_queue_mapping = -1; 1630 } 1631 1632 static inline int sk_tx_queue_get(const struct sock *sk) 1633 { 1634 return sk ? sk->sk_tx_queue_mapping : -1; 1635 } 1636 1637 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1638 { 1639 sk_tx_queue_clear(sk); 1640 sk->sk_socket = sock; 1641 } 1642 1643 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1644 { 1645 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1646 return &rcu_dereference_raw(sk->sk_wq)->wait; 1647 } 1648 /* Detach socket from process context. 1649 * Announce socket dead, detach it from wait queue and inode. 1650 * Note that parent inode held reference count on this struct sock, 1651 * we do not release it in this function, because protocol 1652 * probably wants some additional cleanups or even continuing 1653 * to work with this socket (TCP). 1654 */ 1655 static inline void sock_orphan(struct sock *sk) 1656 { 1657 write_lock_bh(&sk->sk_callback_lock); 1658 sock_set_flag(sk, SOCK_DEAD); 1659 sk_set_socket(sk, NULL); 1660 sk->sk_wq = NULL; 1661 write_unlock_bh(&sk->sk_callback_lock); 1662 } 1663 1664 static inline void sock_graft(struct sock *sk, struct socket *parent) 1665 { 1666 write_lock_bh(&sk->sk_callback_lock); 1667 sk->sk_wq = parent->wq; 1668 parent->sk = sk; 1669 sk_set_socket(sk, parent); 1670 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1671 security_sock_graft(sk, parent); 1672 write_unlock_bh(&sk->sk_callback_lock); 1673 } 1674 1675 kuid_t sock_i_uid(struct sock *sk); 1676 unsigned long sock_i_ino(struct sock *sk); 1677 1678 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1679 { 1680 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1681 } 1682 1683 static inline u32 net_tx_rndhash(void) 1684 { 1685 u32 v = prandom_u32(); 1686 1687 return v ?: 1; 1688 } 1689 1690 static inline void sk_set_txhash(struct sock *sk) 1691 { 1692 sk->sk_txhash = net_tx_rndhash(); 1693 } 1694 1695 static inline void sk_rethink_txhash(struct sock *sk) 1696 { 1697 if (sk->sk_txhash) 1698 sk_set_txhash(sk); 1699 } 1700 1701 static inline struct dst_entry * 1702 __sk_dst_get(struct sock *sk) 1703 { 1704 return rcu_dereference_check(sk->sk_dst_cache, 1705 lockdep_sock_is_held(sk)); 1706 } 1707 1708 static inline struct dst_entry * 1709 sk_dst_get(struct sock *sk) 1710 { 1711 struct dst_entry *dst; 1712 1713 rcu_read_lock(); 1714 dst = rcu_dereference(sk->sk_dst_cache); 1715 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1716 dst = NULL; 1717 rcu_read_unlock(); 1718 return dst; 1719 } 1720 1721 static inline void dst_negative_advice(struct sock *sk) 1722 { 1723 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1724 1725 sk_rethink_txhash(sk); 1726 1727 if (dst && dst->ops->negative_advice) { 1728 ndst = dst->ops->negative_advice(dst); 1729 1730 if (ndst != dst) { 1731 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1732 sk_tx_queue_clear(sk); 1733 } 1734 } 1735 } 1736 1737 static inline void 1738 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1739 { 1740 struct dst_entry *old_dst; 1741 1742 sk_tx_queue_clear(sk); 1743 /* 1744 * This can be called while sk is owned by the caller only, 1745 * with no state that can be checked in a rcu_dereference_check() cond 1746 */ 1747 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1748 rcu_assign_pointer(sk->sk_dst_cache, dst); 1749 dst_release(old_dst); 1750 } 1751 1752 static inline void 1753 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1754 { 1755 struct dst_entry *old_dst; 1756 1757 sk_tx_queue_clear(sk); 1758 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1759 dst_release(old_dst); 1760 } 1761 1762 static inline void 1763 __sk_dst_reset(struct sock *sk) 1764 { 1765 __sk_dst_set(sk, NULL); 1766 } 1767 1768 static inline void 1769 sk_dst_reset(struct sock *sk) 1770 { 1771 sk_dst_set(sk, NULL); 1772 } 1773 1774 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1775 1776 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1777 1778 bool sk_mc_loop(struct sock *sk); 1779 1780 static inline bool sk_can_gso(const struct sock *sk) 1781 { 1782 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1783 } 1784 1785 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1786 1787 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1788 { 1789 sk->sk_route_nocaps |= flags; 1790 sk->sk_route_caps &= ~flags; 1791 } 1792 1793 static inline bool sk_check_csum_caps(struct sock *sk) 1794 { 1795 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1796 (sk->sk_family == PF_INET && 1797 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1798 (sk->sk_family == PF_INET6 && 1799 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1800 } 1801 1802 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1803 struct iov_iter *from, char *to, 1804 int copy, int offset) 1805 { 1806 if (skb->ip_summed == CHECKSUM_NONE) { 1807 __wsum csum = 0; 1808 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy) 1809 return -EFAULT; 1810 skb->csum = csum_block_add(skb->csum, csum, offset); 1811 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1812 if (copy_from_iter_nocache(to, copy, from) != copy) 1813 return -EFAULT; 1814 } else if (copy_from_iter(to, copy, from) != copy) 1815 return -EFAULT; 1816 1817 return 0; 1818 } 1819 1820 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1821 struct iov_iter *from, int copy) 1822 { 1823 int err, offset = skb->len; 1824 1825 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1826 copy, offset); 1827 if (err) 1828 __skb_trim(skb, offset); 1829 1830 return err; 1831 } 1832 1833 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1834 struct sk_buff *skb, 1835 struct page *page, 1836 int off, int copy) 1837 { 1838 int err; 1839 1840 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1841 copy, skb->len); 1842 if (err) 1843 return err; 1844 1845 skb->len += copy; 1846 skb->data_len += copy; 1847 skb->truesize += copy; 1848 sk->sk_wmem_queued += copy; 1849 sk_mem_charge(sk, copy); 1850 return 0; 1851 } 1852 1853 /** 1854 * sk_wmem_alloc_get - returns write allocations 1855 * @sk: socket 1856 * 1857 * Returns sk_wmem_alloc minus initial offset of one 1858 */ 1859 static inline int sk_wmem_alloc_get(const struct sock *sk) 1860 { 1861 return atomic_read(&sk->sk_wmem_alloc) - 1; 1862 } 1863 1864 /** 1865 * sk_rmem_alloc_get - returns read allocations 1866 * @sk: socket 1867 * 1868 * Returns sk_rmem_alloc 1869 */ 1870 static inline int sk_rmem_alloc_get(const struct sock *sk) 1871 { 1872 return atomic_read(&sk->sk_rmem_alloc); 1873 } 1874 1875 /** 1876 * sk_has_allocations - check if allocations are outstanding 1877 * @sk: socket 1878 * 1879 * Returns true if socket has write or read allocations 1880 */ 1881 static inline bool sk_has_allocations(const struct sock *sk) 1882 { 1883 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1884 } 1885 1886 /** 1887 * skwq_has_sleeper - check if there are any waiting processes 1888 * @wq: struct socket_wq 1889 * 1890 * Returns true if socket_wq has waiting processes 1891 * 1892 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1893 * barrier call. They were added due to the race found within the tcp code. 1894 * 1895 * Consider following tcp code paths: 1896 * 1897 * CPU1 CPU2 1898 * 1899 * sys_select receive packet 1900 * ... ... 1901 * __add_wait_queue update tp->rcv_nxt 1902 * ... ... 1903 * tp->rcv_nxt check sock_def_readable 1904 * ... { 1905 * schedule rcu_read_lock(); 1906 * wq = rcu_dereference(sk->sk_wq); 1907 * if (wq && waitqueue_active(&wq->wait)) 1908 * wake_up_interruptible(&wq->wait) 1909 * ... 1910 * } 1911 * 1912 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1913 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1914 * could then endup calling schedule and sleep forever if there are no more 1915 * data on the socket. 1916 * 1917 */ 1918 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1919 { 1920 return wq && wq_has_sleeper(&wq->wait); 1921 } 1922 1923 /** 1924 * sock_poll_wait - place memory barrier behind the poll_wait call. 1925 * @filp: file 1926 * @wait_address: socket wait queue 1927 * @p: poll_table 1928 * 1929 * See the comments in the wq_has_sleeper function. 1930 */ 1931 static inline void sock_poll_wait(struct file *filp, 1932 wait_queue_head_t *wait_address, poll_table *p) 1933 { 1934 if (!poll_does_not_wait(p) && wait_address) { 1935 poll_wait(filp, wait_address, p); 1936 /* We need to be sure we are in sync with the 1937 * socket flags modification. 1938 * 1939 * This memory barrier is paired in the wq_has_sleeper. 1940 */ 1941 smp_mb(); 1942 } 1943 } 1944 1945 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1946 { 1947 if (sk->sk_txhash) { 1948 skb->l4_hash = 1; 1949 skb->hash = sk->sk_txhash; 1950 } 1951 } 1952 1953 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 1954 1955 /* 1956 * Queue a received datagram if it will fit. Stream and sequenced 1957 * protocols can't normally use this as they need to fit buffers in 1958 * and play with them. 1959 * 1960 * Inlined as it's very short and called for pretty much every 1961 * packet ever received. 1962 */ 1963 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1964 { 1965 skb_orphan(skb); 1966 skb->sk = sk; 1967 skb->destructor = sock_rfree; 1968 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1969 sk_mem_charge(sk, skb->truesize); 1970 } 1971 1972 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1973 unsigned long expires); 1974 1975 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1976 1977 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff *skb, 1978 unsigned int flags); 1979 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1980 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1981 1982 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1983 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 1984 1985 /* 1986 * Recover an error report and clear atomically 1987 */ 1988 1989 static inline int sock_error(struct sock *sk) 1990 { 1991 int err; 1992 if (likely(!sk->sk_err)) 1993 return 0; 1994 err = xchg(&sk->sk_err, 0); 1995 return -err; 1996 } 1997 1998 static inline unsigned long sock_wspace(struct sock *sk) 1999 { 2000 int amt = 0; 2001 2002 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2003 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2004 if (amt < 0) 2005 amt = 0; 2006 } 2007 return amt; 2008 } 2009 2010 /* Note: 2011 * We use sk->sk_wq_raw, from contexts knowing this 2012 * pointer is not NULL and cannot disappear/change. 2013 */ 2014 static inline void sk_set_bit(int nr, struct sock *sk) 2015 { 2016 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2017 !sock_flag(sk, SOCK_FASYNC)) 2018 return; 2019 2020 set_bit(nr, &sk->sk_wq_raw->flags); 2021 } 2022 2023 static inline void sk_clear_bit(int nr, struct sock *sk) 2024 { 2025 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2026 !sock_flag(sk, SOCK_FASYNC)) 2027 return; 2028 2029 clear_bit(nr, &sk->sk_wq_raw->flags); 2030 } 2031 2032 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2033 { 2034 if (sock_flag(sk, SOCK_FASYNC)) { 2035 rcu_read_lock(); 2036 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2037 rcu_read_unlock(); 2038 } 2039 } 2040 2041 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2042 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2043 * Note: for send buffers, TCP works better if we can build two skbs at 2044 * minimum. 2045 */ 2046 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2047 2048 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2049 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2050 2051 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2052 { 2053 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2054 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2055 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2056 } 2057 } 2058 2059 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2060 bool force_schedule); 2061 2062 /** 2063 * sk_page_frag - return an appropriate page_frag 2064 * @sk: socket 2065 * 2066 * If socket allocation mode allows current thread to sleep, it means its 2067 * safe to use the per task page_frag instead of the per socket one. 2068 */ 2069 static inline struct page_frag *sk_page_frag(struct sock *sk) 2070 { 2071 if (gfpflags_allow_blocking(sk->sk_allocation)) 2072 return ¤t->task_frag; 2073 2074 return &sk->sk_frag; 2075 } 2076 2077 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2078 2079 /* 2080 * Default write policy as shown to user space via poll/select/SIGIO 2081 */ 2082 static inline bool sock_writeable(const struct sock *sk) 2083 { 2084 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2085 } 2086 2087 static inline gfp_t gfp_any(void) 2088 { 2089 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2090 } 2091 2092 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2093 { 2094 return noblock ? 0 : sk->sk_rcvtimeo; 2095 } 2096 2097 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2098 { 2099 return noblock ? 0 : sk->sk_sndtimeo; 2100 } 2101 2102 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2103 { 2104 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2105 } 2106 2107 /* Alas, with timeout socket operations are not restartable. 2108 * Compare this to poll(). 2109 */ 2110 static inline int sock_intr_errno(long timeo) 2111 { 2112 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2113 } 2114 2115 struct sock_skb_cb { 2116 u32 dropcount; 2117 }; 2118 2119 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2120 * using skb->cb[] would keep using it directly and utilize its 2121 * alignement guarantee. 2122 */ 2123 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2124 sizeof(struct sock_skb_cb))) 2125 2126 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2127 SOCK_SKB_CB_OFFSET)) 2128 2129 #define sock_skb_cb_check_size(size) \ 2130 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2131 2132 static inline void 2133 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2134 { 2135 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops); 2136 } 2137 2138 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2139 { 2140 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2141 2142 atomic_add(segs, &sk->sk_drops); 2143 } 2144 2145 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2146 struct sk_buff *skb); 2147 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2148 struct sk_buff *skb); 2149 2150 static inline void 2151 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2152 { 2153 ktime_t kt = skb->tstamp; 2154 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2155 2156 /* 2157 * generate control messages if 2158 * - receive time stamping in software requested 2159 * - software time stamp available and wanted 2160 * - hardware time stamps available and wanted 2161 */ 2162 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2163 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2164 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2165 (hwtstamps->hwtstamp.tv64 && 2166 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2167 __sock_recv_timestamp(msg, sk, skb); 2168 else 2169 sk->sk_stamp = kt; 2170 2171 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2172 __sock_recv_wifi_status(msg, sk, skb); 2173 } 2174 2175 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2176 struct sk_buff *skb); 2177 2178 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2179 struct sk_buff *skb) 2180 { 2181 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2182 (1UL << SOCK_RCVTSTAMP)) 2183 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2184 SOF_TIMESTAMPING_RAW_HARDWARE) 2185 2186 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2187 __sock_recv_ts_and_drops(msg, sk, skb); 2188 else 2189 sk->sk_stamp = skb->tstamp; 2190 } 2191 2192 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2193 2194 /** 2195 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2196 * @sk: socket sending this packet 2197 * @tsflags: timestamping flags to use 2198 * @tx_flags: completed with instructions for time stamping 2199 * 2200 * Note : callers should take care of initial *tx_flags value (usually 0) 2201 */ 2202 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2203 __u8 *tx_flags) 2204 { 2205 if (unlikely(tsflags)) 2206 __sock_tx_timestamp(tsflags, tx_flags); 2207 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2208 *tx_flags |= SKBTX_WIFI_STATUS; 2209 } 2210 2211 /** 2212 * sk_eat_skb - Release a skb if it is no longer needed 2213 * @sk: socket to eat this skb from 2214 * @skb: socket buffer to eat 2215 * 2216 * This routine must be called with interrupts disabled or with the socket 2217 * locked so that the sk_buff queue operation is ok. 2218 */ 2219 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2220 { 2221 __skb_unlink(skb, &sk->sk_receive_queue); 2222 __kfree_skb(skb); 2223 } 2224 2225 static inline 2226 struct net *sock_net(const struct sock *sk) 2227 { 2228 return read_pnet(&sk->sk_net); 2229 } 2230 2231 static inline 2232 void sock_net_set(struct sock *sk, struct net *net) 2233 { 2234 write_pnet(&sk->sk_net, net); 2235 } 2236 2237 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2238 { 2239 if (skb->sk) { 2240 struct sock *sk = skb->sk; 2241 2242 skb->destructor = NULL; 2243 skb->sk = NULL; 2244 return sk; 2245 } 2246 return NULL; 2247 } 2248 2249 /* This helper checks if a socket is a full socket, 2250 * ie _not_ a timewait or request socket. 2251 */ 2252 static inline bool sk_fullsock(const struct sock *sk) 2253 { 2254 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2255 } 2256 2257 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2258 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2259 */ 2260 static inline bool sk_listener(const struct sock *sk) 2261 { 2262 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2263 } 2264 2265 /** 2266 * sk_state_load - read sk->sk_state for lockless contexts 2267 * @sk: socket pointer 2268 * 2269 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2270 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2271 */ 2272 static inline int sk_state_load(const struct sock *sk) 2273 { 2274 return smp_load_acquire(&sk->sk_state); 2275 } 2276 2277 /** 2278 * sk_state_store - update sk->sk_state 2279 * @sk: socket pointer 2280 * @newstate: new state 2281 * 2282 * Paired with sk_state_load(). Should be used in contexts where 2283 * state change might impact lockless readers. 2284 */ 2285 static inline void sk_state_store(struct sock *sk, int newstate) 2286 { 2287 smp_store_release(&sk->sk_state, newstate); 2288 } 2289 2290 void sock_enable_timestamp(struct sock *sk, int flag); 2291 int sock_get_timestamp(struct sock *, struct timeval __user *); 2292 int sock_get_timestampns(struct sock *, struct timespec __user *); 2293 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2294 int type); 2295 2296 bool sk_ns_capable(const struct sock *sk, 2297 struct user_namespace *user_ns, int cap); 2298 bool sk_capable(const struct sock *sk, int cap); 2299 bool sk_net_capable(const struct sock *sk, int cap); 2300 2301 extern __u32 sysctl_wmem_max; 2302 extern __u32 sysctl_rmem_max; 2303 2304 extern int sysctl_tstamp_allow_data; 2305 extern int sysctl_optmem_max; 2306 2307 extern __u32 sysctl_wmem_default; 2308 extern __u32 sysctl_rmem_default; 2309 2310 #endif /* _SOCK_H */ 2311