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 unsigned int __sk_flags_offset[0]; 393 #ifdef __BIG_ENDIAN_BITFIELD 394 #define SK_FL_PROTO_SHIFT 16 395 #define SK_FL_PROTO_MASK 0x00ff0000 396 397 #define SK_FL_TYPE_SHIFT 0 398 #define SK_FL_TYPE_MASK 0x0000ffff 399 #else 400 #define SK_FL_PROTO_SHIFT 8 401 #define SK_FL_PROTO_MASK 0x0000ff00 402 403 #define SK_FL_TYPE_SHIFT 16 404 #define SK_FL_TYPE_MASK 0xffff0000 405 #endif 406 407 kmemcheck_bitfield_begin(flags); 408 unsigned int sk_padding : 2, 409 sk_no_check_tx : 1, 410 sk_no_check_rx : 1, 411 sk_userlocks : 4, 412 sk_protocol : 8, 413 sk_type : 16; 414 #define SK_PROTOCOL_MAX U8_MAX 415 kmemcheck_bitfield_end(flags); 416 417 int sk_wmem_queued; 418 gfp_t sk_allocation; 419 u32 sk_pacing_rate; /* bytes per second */ 420 u32 sk_max_pacing_rate; 421 netdev_features_t sk_route_caps; 422 netdev_features_t sk_route_nocaps; 423 int sk_gso_type; 424 unsigned int sk_gso_max_size; 425 u16 sk_gso_max_segs; 426 int sk_rcvlowat; 427 unsigned long sk_lingertime; 428 struct sk_buff_head sk_error_queue; 429 struct proto *sk_prot_creator; 430 rwlock_t sk_callback_lock; 431 int sk_err, 432 sk_err_soft; 433 u32 sk_ack_backlog; 434 u32 sk_max_ack_backlog; 435 __u32 sk_priority; 436 __u32 sk_mark; 437 kuid_t sk_uid; 438 struct pid *sk_peer_pid; 439 const struct cred *sk_peer_cred; 440 long sk_rcvtimeo; 441 long sk_sndtimeo; 442 struct timer_list sk_timer; 443 ktime_t sk_stamp; 444 u16 sk_tsflags; 445 u8 sk_shutdown; 446 u32 sk_tskey; 447 struct socket *sk_socket; 448 void *sk_user_data; 449 struct page_frag sk_frag; 450 struct sk_buff *sk_send_head; 451 __s32 sk_peek_off; 452 int sk_write_pending; 453 #ifdef CONFIG_SECURITY 454 void *sk_security; 455 #endif 456 struct sock_cgroup_data sk_cgrp_data; 457 struct mem_cgroup *sk_memcg; 458 void (*sk_state_change)(struct sock *sk); 459 void (*sk_data_ready)(struct sock *sk); 460 void (*sk_write_space)(struct sock *sk); 461 void (*sk_error_report)(struct sock *sk); 462 int (*sk_backlog_rcv)(struct sock *sk, 463 struct sk_buff *skb); 464 void (*sk_destruct)(struct sock *sk); 465 struct sock_reuseport __rcu *sk_reuseport_cb; 466 struct rcu_head sk_rcu; 467 }; 468 469 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 470 471 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 472 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 473 474 /* 475 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 476 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 477 * on a socket means that the socket will reuse everybody else's port 478 * without looking at the other's sk_reuse value. 479 */ 480 481 #define SK_NO_REUSE 0 482 #define SK_CAN_REUSE 1 483 #define SK_FORCE_REUSE 2 484 485 int sk_set_peek_off(struct sock *sk, int val); 486 487 static inline int sk_peek_offset(struct sock *sk, int flags) 488 { 489 if (unlikely(flags & MSG_PEEK)) { 490 s32 off = READ_ONCE(sk->sk_peek_off); 491 if (off >= 0) 492 return off; 493 } 494 495 return 0; 496 } 497 498 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 499 { 500 s32 off = READ_ONCE(sk->sk_peek_off); 501 502 if (unlikely(off >= 0)) { 503 off = max_t(s32, off - val, 0); 504 WRITE_ONCE(sk->sk_peek_off, off); 505 } 506 } 507 508 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 509 { 510 sk_peek_offset_bwd(sk, -val); 511 } 512 513 /* 514 * Hashed lists helper routines 515 */ 516 static inline struct sock *sk_entry(const struct hlist_node *node) 517 { 518 return hlist_entry(node, struct sock, sk_node); 519 } 520 521 static inline struct sock *__sk_head(const struct hlist_head *head) 522 { 523 return hlist_entry(head->first, struct sock, sk_node); 524 } 525 526 static inline struct sock *sk_head(const struct hlist_head *head) 527 { 528 return hlist_empty(head) ? NULL : __sk_head(head); 529 } 530 531 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 532 { 533 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 534 } 535 536 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 537 { 538 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 539 } 540 541 static inline struct sock *sk_next(const struct sock *sk) 542 { 543 return sk->sk_node.next ? 544 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 545 } 546 547 static inline struct sock *sk_nulls_next(const struct sock *sk) 548 { 549 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 550 hlist_nulls_entry(sk->sk_nulls_node.next, 551 struct sock, sk_nulls_node) : 552 NULL; 553 } 554 555 static inline bool sk_unhashed(const struct sock *sk) 556 { 557 return hlist_unhashed(&sk->sk_node); 558 } 559 560 static inline bool sk_hashed(const struct sock *sk) 561 { 562 return !sk_unhashed(sk); 563 } 564 565 static inline void sk_node_init(struct hlist_node *node) 566 { 567 node->pprev = NULL; 568 } 569 570 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 571 { 572 node->pprev = NULL; 573 } 574 575 static inline void __sk_del_node(struct sock *sk) 576 { 577 __hlist_del(&sk->sk_node); 578 } 579 580 /* NB: equivalent to hlist_del_init_rcu */ 581 static inline bool __sk_del_node_init(struct sock *sk) 582 { 583 if (sk_hashed(sk)) { 584 __sk_del_node(sk); 585 sk_node_init(&sk->sk_node); 586 return true; 587 } 588 return false; 589 } 590 591 /* Grab socket reference count. This operation is valid only 592 when sk is ALREADY grabbed f.e. it is found in hash table 593 or a list and the lookup is made under lock preventing hash table 594 modifications. 595 */ 596 597 static __always_inline void sock_hold(struct sock *sk) 598 { 599 atomic_inc(&sk->sk_refcnt); 600 } 601 602 /* Ungrab socket in the context, which assumes that socket refcnt 603 cannot hit zero, f.e. it is true in context of any socketcall. 604 */ 605 static __always_inline void __sock_put(struct sock *sk) 606 { 607 atomic_dec(&sk->sk_refcnt); 608 } 609 610 static inline bool sk_del_node_init(struct sock *sk) 611 { 612 bool rc = __sk_del_node_init(sk); 613 614 if (rc) { 615 /* paranoid for a while -acme */ 616 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 617 __sock_put(sk); 618 } 619 return rc; 620 } 621 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 622 623 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 624 { 625 if (sk_hashed(sk)) { 626 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 627 return true; 628 } 629 return false; 630 } 631 632 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 633 { 634 bool rc = __sk_nulls_del_node_init_rcu(sk); 635 636 if (rc) { 637 /* paranoid for a while -acme */ 638 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 639 __sock_put(sk); 640 } 641 return rc; 642 } 643 644 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 645 { 646 hlist_add_head(&sk->sk_node, list); 647 } 648 649 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 650 { 651 sock_hold(sk); 652 __sk_add_node(sk, list); 653 } 654 655 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 656 { 657 sock_hold(sk); 658 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 659 sk->sk_family == AF_INET6) 660 hlist_add_tail_rcu(&sk->sk_node, list); 661 else 662 hlist_add_head_rcu(&sk->sk_node, list); 663 } 664 665 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 666 { 667 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 668 sk->sk_family == AF_INET6) 669 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 670 else 671 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 672 } 673 674 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 675 { 676 sock_hold(sk); 677 __sk_nulls_add_node_rcu(sk, list); 678 } 679 680 static inline void __sk_del_bind_node(struct sock *sk) 681 { 682 __hlist_del(&sk->sk_bind_node); 683 } 684 685 static inline void sk_add_bind_node(struct sock *sk, 686 struct hlist_head *list) 687 { 688 hlist_add_head(&sk->sk_bind_node, list); 689 } 690 691 #define sk_for_each(__sk, list) \ 692 hlist_for_each_entry(__sk, list, sk_node) 693 #define sk_for_each_rcu(__sk, list) \ 694 hlist_for_each_entry_rcu(__sk, list, sk_node) 695 #define sk_nulls_for_each(__sk, node, list) \ 696 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 697 #define sk_nulls_for_each_rcu(__sk, node, list) \ 698 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 699 #define sk_for_each_from(__sk) \ 700 hlist_for_each_entry_from(__sk, sk_node) 701 #define sk_nulls_for_each_from(__sk, node) \ 702 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 703 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 704 #define sk_for_each_safe(__sk, tmp, list) \ 705 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 706 #define sk_for_each_bound(__sk, list) \ 707 hlist_for_each_entry(__sk, list, sk_bind_node) 708 709 /** 710 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 711 * @tpos: the type * to use as a loop cursor. 712 * @pos: the &struct hlist_node to use as a loop cursor. 713 * @head: the head for your list. 714 * @offset: offset of hlist_node within the struct. 715 * 716 */ 717 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 718 for (pos = rcu_dereference((head)->first); \ 719 pos != NULL && \ 720 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 721 pos = rcu_dereference(pos->next)) 722 723 static inline struct user_namespace *sk_user_ns(struct sock *sk) 724 { 725 /* Careful only use this in a context where these parameters 726 * can not change and must all be valid, such as recvmsg from 727 * userspace. 728 */ 729 return sk->sk_socket->file->f_cred->user_ns; 730 } 731 732 /* Sock flags */ 733 enum sock_flags { 734 SOCK_DEAD, 735 SOCK_DONE, 736 SOCK_URGINLINE, 737 SOCK_KEEPOPEN, 738 SOCK_LINGER, 739 SOCK_DESTROY, 740 SOCK_BROADCAST, 741 SOCK_TIMESTAMP, 742 SOCK_ZAPPED, 743 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 744 SOCK_DBG, /* %SO_DEBUG setting */ 745 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 746 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 747 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 748 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 749 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 750 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 751 SOCK_FASYNC, /* fasync() active */ 752 SOCK_RXQ_OVFL, 753 SOCK_ZEROCOPY, /* buffers from userspace */ 754 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 755 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 756 * Will use last 4 bytes of packet sent from 757 * user-space instead. 758 */ 759 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 760 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 761 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 762 }; 763 764 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 765 766 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 767 { 768 nsk->sk_flags = osk->sk_flags; 769 } 770 771 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 772 { 773 __set_bit(flag, &sk->sk_flags); 774 } 775 776 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 777 { 778 __clear_bit(flag, &sk->sk_flags); 779 } 780 781 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 782 { 783 return test_bit(flag, &sk->sk_flags); 784 } 785 786 #ifdef CONFIG_NET 787 extern struct static_key memalloc_socks; 788 static inline int sk_memalloc_socks(void) 789 { 790 return static_key_false(&memalloc_socks); 791 } 792 #else 793 794 static inline int sk_memalloc_socks(void) 795 { 796 return 0; 797 } 798 799 #endif 800 801 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 802 { 803 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 804 } 805 806 static inline void sk_acceptq_removed(struct sock *sk) 807 { 808 sk->sk_ack_backlog--; 809 } 810 811 static inline void sk_acceptq_added(struct sock *sk) 812 { 813 sk->sk_ack_backlog++; 814 } 815 816 static inline bool sk_acceptq_is_full(const struct sock *sk) 817 { 818 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 819 } 820 821 /* 822 * Compute minimal free write space needed to queue new packets. 823 */ 824 static inline int sk_stream_min_wspace(const struct sock *sk) 825 { 826 return sk->sk_wmem_queued >> 1; 827 } 828 829 static inline int sk_stream_wspace(const struct sock *sk) 830 { 831 return sk->sk_sndbuf - sk->sk_wmem_queued; 832 } 833 834 void sk_stream_write_space(struct sock *sk); 835 836 /* OOB backlog add */ 837 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 838 { 839 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 840 skb_dst_force_safe(skb); 841 842 if (!sk->sk_backlog.tail) 843 sk->sk_backlog.head = skb; 844 else 845 sk->sk_backlog.tail->next = skb; 846 847 sk->sk_backlog.tail = skb; 848 skb->next = NULL; 849 } 850 851 /* 852 * Take into account size of receive queue and backlog queue 853 * Do not take into account this skb truesize, 854 * to allow even a single big packet to come. 855 */ 856 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 857 { 858 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 859 860 return qsize > limit; 861 } 862 863 /* The per-socket spinlock must be held here. */ 864 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 865 unsigned int limit) 866 { 867 if (sk_rcvqueues_full(sk, limit)) 868 return -ENOBUFS; 869 870 /* 871 * If the skb was allocated from pfmemalloc reserves, only 872 * allow SOCK_MEMALLOC sockets to use it as this socket is 873 * helping free memory 874 */ 875 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 876 return -ENOMEM; 877 878 __sk_add_backlog(sk, skb); 879 sk->sk_backlog.len += skb->truesize; 880 return 0; 881 } 882 883 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 884 885 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 886 { 887 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 888 return __sk_backlog_rcv(sk, skb); 889 890 return sk->sk_backlog_rcv(sk, skb); 891 } 892 893 static inline void sk_incoming_cpu_update(struct sock *sk) 894 { 895 sk->sk_incoming_cpu = raw_smp_processor_id(); 896 } 897 898 static inline void sock_rps_record_flow_hash(__u32 hash) 899 { 900 #ifdef CONFIG_RPS 901 struct rps_sock_flow_table *sock_flow_table; 902 903 rcu_read_lock(); 904 sock_flow_table = rcu_dereference(rps_sock_flow_table); 905 rps_record_sock_flow(sock_flow_table, hash); 906 rcu_read_unlock(); 907 #endif 908 } 909 910 static inline void sock_rps_record_flow(const struct sock *sk) 911 { 912 #ifdef CONFIG_RPS 913 sock_rps_record_flow_hash(sk->sk_rxhash); 914 #endif 915 } 916 917 static inline void sock_rps_save_rxhash(struct sock *sk, 918 const struct sk_buff *skb) 919 { 920 #ifdef CONFIG_RPS 921 if (unlikely(sk->sk_rxhash != skb->hash)) 922 sk->sk_rxhash = skb->hash; 923 #endif 924 } 925 926 static inline void sock_rps_reset_rxhash(struct sock *sk) 927 { 928 #ifdef CONFIG_RPS 929 sk->sk_rxhash = 0; 930 #endif 931 } 932 933 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 934 ({ int __rc; \ 935 release_sock(__sk); \ 936 __rc = __condition; \ 937 if (!__rc) { \ 938 *(__timeo) = wait_woken(__wait, \ 939 TASK_INTERRUPTIBLE, \ 940 *(__timeo)); \ 941 } \ 942 sched_annotate_sleep(); \ 943 lock_sock(__sk); \ 944 __rc = __condition; \ 945 __rc; \ 946 }) 947 948 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 949 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 950 void sk_stream_wait_close(struct sock *sk, long timeo_p); 951 int sk_stream_error(struct sock *sk, int flags, int err); 952 void sk_stream_kill_queues(struct sock *sk); 953 void sk_set_memalloc(struct sock *sk); 954 void sk_clear_memalloc(struct sock *sk); 955 956 void __sk_flush_backlog(struct sock *sk); 957 958 static inline bool sk_flush_backlog(struct sock *sk) 959 { 960 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 961 __sk_flush_backlog(sk); 962 return true; 963 } 964 return false; 965 } 966 967 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 968 969 struct request_sock_ops; 970 struct timewait_sock_ops; 971 struct inet_hashinfo; 972 struct raw_hashinfo; 973 struct module; 974 975 /* 976 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 977 * un-modified. Special care is taken when initializing object to zero. 978 */ 979 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 980 { 981 if (offsetof(struct sock, sk_node.next) != 0) 982 memset(sk, 0, offsetof(struct sock, sk_node.next)); 983 memset(&sk->sk_node.pprev, 0, 984 size - offsetof(struct sock, sk_node.pprev)); 985 } 986 987 /* Networking protocol blocks we attach to sockets. 988 * socket layer -> transport layer interface 989 */ 990 struct proto { 991 void (*close)(struct sock *sk, 992 long timeout); 993 int (*connect)(struct sock *sk, 994 struct sockaddr *uaddr, 995 int addr_len); 996 int (*disconnect)(struct sock *sk, int flags); 997 998 struct sock * (*accept)(struct sock *sk, int flags, int *err); 999 1000 int (*ioctl)(struct sock *sk, int cmd, 1001 unsigned long arg); 1002 int (*init)(struct sock *sk); 1003 void (*destroy)(struct sock *sk); 1004 void (*shutdown)(struct sock *sk, int how); 1005 int (*setsockopt)(struct sock *sk, int level, 1006 int optname, char __user *optval, 1007 unsigned int optlen); 1008 int (*getsockopt)(struct sock *sk, int level, 1009 int optname, char __user *optval, 1010 int __user *option); 1011 #ifdef CONFIG_COMPAT 1012 int (*compat_setsockopt)(struct sock *sk, 1013 int level, 1014 int optname, char __user *optval, 1015 unsigned int optlen); 1016 int (*compat_getsockopt)(struct sock *sk, 1017 int level, 1018 int optname, char __user *optval, 1019 int __user *option); 1020 int (*compat_ioctl)(struct sock *sk, 1021 unsigned int cmd, unsigned long arg); 1022 #endif 1023 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1024 size_t len); 1025 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1026 size_t len, int noblock, int flags, 1027 int *addr_len); 1028 int (*sendpage)(struct sock *sk, struct page *page, 1029 int offset, size_t size, int flags); 1030 int (*bind)(struct sock *sk, 1031 struct sockaddr *uaddr, int addr_len); 1032 1033 int (*backlog_rcv) (struct sock *sk, 1034 struct sk_buff *skb); 1035 1036 void (*release_cb)(struct sock *sk); 1037 1038 /* Keeping track of sk's, looking them up, and port selection methods. */ 1039 int (*hash)(struct sock *sk); 1040 void (*unhash)(struct sock *sk); 1041 void (*rehash)(struct sock *sk); 1042 int (*get_port)(struct sock *sk, unsigned short snum); 1043 1044 /* Keeping track of sockets in use */ 1045 #ifdef CONFIG_PROC_FS 1046 unsigned int inuse_idx; 1047 #endif 1048 1049 bool (*stream_memory_free)(const struct sock *sk); 1050 /* Memory pressure */ 1051 void (*enter_memory_pressure)(struct sock *sk); 1052 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1053 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1054 /* 1055 * Pressure flag: try to collapse. 1056 * Technical note: it is used by multiple contexts non atomically. 1057 * All the __sk_mem_schedule() is of this nature: accounting 1058 * is strict, actions are advisory and have some latency. 1059 */ 1060 int *memory_pressure; 1061 long *sysctl_mem; 1062 int *sysctl_wmem; 1063 int *sysctl_rmem; 1064 int max_header; 1065 bool no_autobind; 1066 1067 struct kmem_cache *slab; 1068 unsigned int obj_size; 1069 int slab_flags; 1070 1071 struct percpu_counter *orphan_count; 1072 1073 struct request_sock_ops *rsk_prot; 1074 struct timewait_sock_ops *twsk_prot; 1075 1076 union { 1077 struct inet_hashinfo *hashinfo; 1078 struct udp_table *udp_table; 1079 struct raw_hashinfo *raw_hash; 1080 } h; 1081 1082 struct module *owner; 1083 1084 char name[32]; 1085 1086 struct list_head node; 1087 #ifdef SOCK_REFCNT_DEBUG 1088 atomic_t socks; 1089 #endif 1090 int (*diag_destroy)(struct sock *sk, int err); 1091 }; 1092 1093 int proto_register(struct proto *prot, int alloc_slab); 1094 void proto_unregister(struct proto *prot); 1095 1096 #ifdef SOCK_REFCNT_DEBUG 1097 static inline void sk_refcnt_debug_inc(struct sock *sk) 1098 { 1099 atomic_inc(&sk->sk_prot->socks); 1100 } 1101 1102 static inline void sk_refcnt_debug_dec(struct sock *sk) 1103 { 1104 atomic_dec(&sk->sk_prot->socks); 1105 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1106 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1107 } 1108 1109 static inline void sk_refcnt_debug_release(const struct sock *sk) 1110 { 1111 if (atomic_read(&sk->sk_refcnt) != 1) 1112 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1113 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1114 } 1115 #else /* SOCK_REFCNT_DEBUG */ 1116 #define sk_refcnt_debug_inc(sk) do { } while (0) 1117 #define sk_refcnt_debug_dec(sk) do { } while (0) 1118 #define sk_refcnt_debug_release(sk) do { } while (0) 1119 #endif /* SOCK_REFCNT_DEBUG */ 1120 1121 static inline bool sk_stream_memory_free(const struct sock *sk) 1122 { 1123 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1124 return false; 1125 1126 return sk->sk_prot->stream_memory_free ? 1127 sk->sk_prot->stream_memory_free(sk) : true; 1128 } 1129 1130 static inline bool sk_stream_is_writeable(const struct sock *sk) 1131 { 1132 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1133 sk_stream_memory_free(sk); 1134 } 1135 1136 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1137 struct cgroup *ancestor) 1138 { 1139 #ifdef CONFIG_SOCK_CGROUP_DATA 1140 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1141 ancestor); 1142 #else 1143 return -ENOTSUPP; 1144 #endif 1145 } 1146 1147 static inline bool sk_has_memory_pressure(const struct sock *sk) 1148 { 1149 return sk->sk_prot->memory_pressure != NULL; 1150 } 1151 1152 static inline bool sk_under_memory_pressure(const struct sock *sk) 1153 { 1154 if (!sk->sk_prot->memory_pressure) 1155 return false; 1156 1157 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1158 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1159 return true; 1160 1161 return !!*sk->sk_prot->memory_pressure; 1162 } 1163 1164 static inline void sk_leave_memory_pressure(struct sock *sk) 1165 { 1166 int *memory_pressure = sk->sk_prot->memory_pressure; 1167 1168 if (!memory_pressure) 1169 return; 1170 1171 if (*memory_pressure) 1172 *memory_pressure = 0; 1173 } 1174 1175 static inline void sk_enter_memory_pressure(struct sock *sk) 1176 { 1177 if (!sk->sk_prot->enter_memory_pressure) 1178 return; 1179 1180 sk->sk_prot->enter_memory_pressure(sk); 1181 } 1182 1183 static inline long 1184 sk_memory_allocated(const struct sock *sk) 1185 { 1186 return atomic_long_read(sk->sk_prot->memory_allocated); 1187 } 1188 1189 static inline long 1190 sk_memory_allocated_add(struct sock *sk, int amt) 1191 { 1192 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1193 } 1194 1195 static inline void 1196 sk_memory_allocated_sub(struct sock *sk, int amt) 1197 { 1198 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1199 } 1200 1201 static inline void sk_sockets_allocated_dec(struct sock *sk) 1202 { 1203 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1204 } 1205 1206 static inline void sk_sockets_allocated_inc(struct sock *sk) 1207 { 1208 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1209 } 1210 1211 static inline int 1212 sk_sockets_allocated_read_positive(struct sock *sk) 1213 { 1214 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1215 } 1216 1217 static inline int 1218 proto_sockets_allocated_sum_positive(struct proto *prot) 1219 { 1220 return percpu_counter_sum_positive(prot->sockets_allocated); 1221 } 1222 1223 static inline long 1224 proto_memory_allocated(struct proto *prot) 1225 { 1226 return atomic_long_read(prot->memory_allocated); 1227 } 1228 1229 static inline bool 1230 proto_memory_pressure(struct proto *prot) 1231 { 1232 if (!prot->memory_pressure) 1233 return false; 1234 return !!*prot->memory_pressure; 1235 } 1236 1237 1238 #ifdef CONFIG_PROC_FS 1239 /* Called with local bh disabled */ 1240 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1241 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1242 #else 1243 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1244 int inc) 1245 { 1246 } 1247 #endif 1248 1249 1250 /* With per-bucket locks this operation is not-atomic, so that 1251 * this version is not worse. 1252 */ 1253 static inline int __sk_prot_rehash(struct sock *sk) 1254 { 1255 sk->sk_prot->unhash(sk); 1256 return sk->sk_prot->hash(sk); 1257 } 1258 1259 /* About 10 seconds */ 1260 #define SOCK_DESTROY_TIME (10*HZ) 1261 1262 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1263 #define PROT_SOCK 1024 1264 1265 #define SHUTDOWN_MASK 3 1266 #define RCV_SHUTDOWN 1 1267 #define SEND_SHUTDOWN 2 1268 1269 #define SOCK_SNDBUF_LOCK 1 1270 #define SOCK_RCVBUF_LOCK 2 1271 #define SOCK_BINDADDR_LOCK 4 1272 #define SOCK_BINDPORT_LOCK 8 1273 1274 struct socket_alloc { 1275 struct socket socket; 1276 struct inode vfs_inode; 1277 }; 1278 1279 static inline struct socket *SOCKET_I(struct inode *inode) 1280 { 1281 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1282 } 1283 1284 static inline struct inode *SOCK_INODE(struct socket *socket) 1285 { 1286 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1287 } 1288 1289 /* 1290 * Functions for memory accounting 1291 */ 1292 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1293 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1294 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1295 void __sk_mem_reclaim(struct sock *sk, int amount); 1296 1297 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1298 * do not necessarily have 16x time more memory than 4KB ones. 1299 */ 1300 #define SK_MEM_QUANTUM 4096 1301 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1302 #define SK_MEM_SEND 0 1303 #define SK_MEM_RECV 1 1304 1305 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1306 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1307 { 1308 long val = sk->sk_prot->sysctl_mem[index]; 1309 1310 #if PAGE_SIZE > SK_MEM_QUANTUM 1311 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1312 #elif PAGE_SIZE < SK_MEM_QUANTUM 1313 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1314 #endif 1315 return val; 1316 } 1317 1318 static inline int sk_mem_pages(int amt) 1319 { 1320 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1321 } 1322 1323 static inline bool sk_has_account(struct sock *sk) 1324 { 1325 /* return true if protocol supports memory accounting */ 1326 return !!sk->sk_prot->memory_allocated; 1327 } 1328 1329 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1330 { 1331 if (!sk_has_account(sk)) 1332 return true; 1333 return size <= sk->sk_forward_alloc || 1334 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1335 } 1336 1337 static inline bool 1338 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1339 { 1340 if (!sk_has_account(sk)) 1341 return true; 1342 return size<= sk->sk_forward_alloc || 1343 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1344 skb_pfmemalloc(skb); 1345 } 1346 1347 static inline void sk_mem_reclaim(struct sock *sk) 1348 { 1349 if (!sk_has_account(sk)) 1350 return; 1351 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1352 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1353 } 1354 1355 static inline void sk_mem_reclaim_partial(struct sock *sk) 1356 { 1357 if (!sk_has_account(sk)) 1358 return; 1359 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1360 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1361 } 1362 1363 static inline void sk_mem_charge(struct sock *sk, int size) 1364 { 1365 if (!sk_has_account(sk)) 1366 return; 1367 sk->sk_forward_alloc -= size; 1368 } 1369 1370 static inline void sk_mem_uncharge(struct sock *sk, int size) 1371 { 1372 if (!sk_has_account(sk)) 1373 return; 1374 sk->sk_forward_alloc += size; 1375 1376 /* Avoid a possible overflow. 1377 * TCP send queues can make this happen, if sk_mem_reclaim() 1378 * is not called and more than 2 GBytes are released at once. 1379 * 1380 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1381 * no need to hold that much forward allocation anyway. 1382 */ 1383 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1384 __sk_mem_reclaim(sk, 1 << 20); 1385 } 1386 1387 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1388 { 1389 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1390 sk->sk_wmem_queued -= skb->truesize; 1391 sk_mem_uncharge(sk, skb->truesize); 1392 __kfree_skb(skb); 1393 } 1394 1395 static inline void sock_release_ownership(struct sock *sk) 1396 { 1397 if (sk->sk_lock.owned) { 1398 sk->sk_lock.owned = 0; 1399 1400 /* The sk_lock has mutex_unlock() semantics: */ 1401 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1402 } 1403 } 1404 1405 /* 1406 * Macro so as to not evaluate some arguments when 1407 * lockdep is not enabled. 1408 * 1409 * Mark both the sk_lock and the sk_lock.slock as a 1410 * per-address-family lock class. 1411 */ 1412 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1413 do { \ 1414 sk->sk_lock.owned = 0; \ 1415 init_waitqueue_head(&sk->sk_lock.wq); \ 1416 spin_lock_init(&(sk)->sk_lock.slock); \ 1417 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1418 sizeof((sk)->sk_lock)); \ 1419 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1420 (skey), (sname)); \ 1421 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1422 } while (0) 1423 1424 #ifdef CONFIG_LOCKDEP 1425 static inline bool lockdep_sock_is_held(const struct sock *csk) 1426 { 1427 struct sock *sk = (struct sock *)csk; 1428 1429 return lockdep_is_held(&sk->sk_lock) || 1430 lockdep_is_held(&sk->sk_lock.slock); 1431 } 1432 #endif 1433 1434 void lock_sock_nested(struct sock *sk, int subclass); 1435 1436 static inline void lock_sock(struct sock *sk) 1437 { 1438 lock_sock_nested(sk, 0); 1439 } 1440 1441 void release_sock(struct sock *sk); 1442 1443 /* BH context may only use the following locking interface. */ 1444 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1445 #define bh_lock_sock_nested(__sk) \ 1446 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1447 SINGLE_DEPTH_NESTING) 1448 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1449 1450 bool lock_sock_fast(struct sock *sk); 1451 /** 1452 * unlock_sock_fast - complement of lock_sock_fast 1453 * @sk: socket 1454 * @slow: slow mode 1455 * 1456 * fast unlock socket for user context. 1457 * If slow mode is on, we call regular release_sock() 1458 */ 1459 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1460 { 1461 if (slow) 1462 release_sock(sk); 1463 else 1464 spin_unlock_bh(&sk->sk_lock.slock); 1465 } 1466 1467 /* Used by processes to "lock" a socket state, so that 1468 * interrupts and bottom half handlers won't change it 1469 * from under us. It essentially blocks any incoming 1470 * packets, so that we won't get any new data or any 1471 * packets that change the state of the socket. 1472 * 1473 * While locked, BH processing will add new packets to 1474 * the backlog queue. This queue is processed by the 1475 * owner of the socket lock right before it is released. 1476 * 1477 * Since ~2.3.5 it is also exclusive sleep lock serializing 1478 * accesses from user process context. 1479 */ 1480 1481 static inline void sock_owned_by_me(const struct sock *sk) 1482 { 1483 #ifdef CONFIG_LOCKDEP 1484 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1485 #endif 1486 } 1487 1488 static inline bool sock_owned_by_user(const struct sock *sk) 1489 { 1490 sock_owned_by_me(sk); 1491 return sk->sk_lock.owned; 1492 } 1493 1494 /* no reclassification while locks are held */ 1495 static inline bool sock_allow_reclassification(const struct sock *csk) 1496 { 1497 struct sock *sk = (struct sock *)csk; 1498 1499 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1500 } 1501 1502 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1503 struct proto *prot, int kern); 1504 void sk_free(struct sock *sk); 1505 void sk_destruct(struct sock *sk); 1506 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1507 1508 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1509 gfp_t priority); 1510 void __sock_wfree(struct sk_buff *skb); 1511 void sock_wfree(struct sk_buff *skb); 1512 void skb_orphan_partial(struct sk_buff *skb); 1513 void sock_rfree(struct sk_buff *skb); 1514 void sock_efree(struct sk_buff *skb); 1515 #ifdef CONFIG_INET 1516 void sock_edemux(struct sk_buff *skb); 1517 #else 1518 #define sock_edemux(skb) sock_efree(skb) 1519 #endif 1520 1521 int sock_setsockopt(struct socket *sock, int level, int op, 1522 char __user *optval, unsigned int optlen); 1523 1524 int sock_getsockopt(struct socket *sock, int level, int op, 1525 char __user *optval, int __user *optlen); 1526 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1527 int noblock, int *errcode); 1528 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1529 unsigned long data_len, int noblock, 1530 int *errcode, int max_page_order); 1531 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1532 void sock_kfree_s(struct sock *sk, void *mem, int size); 1533 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1534 void sk_send_sigurg(struct sock *sk); 1535 1536 struct sockcm_cookie { 1537 u32 mark; 1538 u16 tsflags; 1539 }; 1540 1541 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1542 struct sockcm_cookie *sockc); 1543 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1544 struct sockcm_cookie *sockc); 1545 1546 /* 1547 * Functions to fill in entries in struct proto_ops when a protocol 1548 * does not implement a particular function. 1549 */ 1550 int sock_no_bind(struct socket *, struct sockaddr *, int); 1551 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1552 int sock_no_socketpair(struct socket *, struct socket *); 1553 int sock_no_accept(struct socket *, struct socket *, int); 1554 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1555 unsigned int sock_no_poll(struct file *, struct socket *, 1556 struct poll_table_struct *); 1557 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1558 int sock_no_listen(struct socket *, int); 1559 int sock_no_shutdown(struct socket *, int); 1560 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1561 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1562 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1563 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1564 int sock_no_mmap(struct file *file, struct socket *sock, 1565 struct vm_area_struct *vma); 1566 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1567 size_t size, int flags); 1568 1569 /* 1570 * Functions to fill in entries in struct proto_ops when a protocol 1571 * uses the inet style. 1572 */ 1573 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1574 char __user *optval, int __user *optlen); 1575 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1576 int flags); 1577 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1578 char __user *optval, unsigned int optlen); 1579 int compat_sock_common_getsockopt(struct socket *sock, int level, 1580 int optname, char __user *optval, int __user *optlen); 1581 int compat_sock_common_setsockopt(struct socket *sock, int level, 1582 int optname, char __user *optval, unsigned int optlen); 1583 1584 void sk_common_release(struct sock *sk); 1585 1586 /* 1587 * Default socket callbacks and setup code 1588 */ 1589 1590 /* Initialise core socket variables */ 1591 void sock_init_data(struct socket *sock, struct sock *sk); 1592 1593 /* 1594 * Socket reference counting postulates. 1595 * 1596 * * Each user of socket SHOULD hold a reference count. 1597 * * Each access point to socket (an hash table bucket, reference from a list, 1598 * running timer, skb in flight MUST hold a reference count. 1599 * * When reference count hits 0, it means it will never increase back. 1600 * * When reference count hits 0, it means that no references from 1601 * outside exist to this socket and current process on current CPU 1602 * is last user and may/should destroy this socket. 1603 * * sk_free is called from any context: process, BH, IRQ. When 1604 * it is called, socket has no references from outside -> sk_free 1605 * may release descendant resources allocated by the socket, but 1606 * to the time when it is called, socket is NOT referenced by any 1607 * hash tables, lists etc. 1608 * * Packets, delivered from outside (from network or from another process) 1609 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1610 * when they sit in queue. Otherwise, packets will leak to hole, when 1611 * socket is looked up by one cpu and unhasing is made by another CPU. 1612 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1613 * (leak to backlog). Packet socket does all the processing inside 1614 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1615 * use separate SMP lock, so that they are prone too. 1616 */ 1617 1618 /* Ungrab socket and destroy it, if it was the last reference. */ 1619 static inline void sock_put(struct sock *sk) 1620 { 1621 if (atomic_dec_and_test(&sk->sk_refcnt)) 1622 sk_free(sk); 1623 } 1624 /* Generic version of sock_put(), dealing with all sockets 1625 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1626 */ 1627 void sock_gen_put(struct sock *sk); 1628 1629 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1630 unsigned int trim_cap, bool refcounted); 1631 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1632 const int nested) 1633 { 1634 return __sk_receive_skb(sk, skb, nested, 1, true); 1635 } 1636 1637 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1638 { 1639 sk->sk_tx_queue_mapping = tx_queue; 1640 } 1641 1642 static inline void sk_tx_queue_clear(struct sock *sk) 1643 { 1644 sk->sk_tx_queue_mapping = -1; 1645 } 1646 1647 static inline int sk_tx_queue_get(const struct sock *sk) 1648 { 1649 return sk ? sk->sk_tx_queue_mapping : -1; 1650 } 1651 1652 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1653 { 1654 sk_tx_queue_clear(sk); 1655 sk->sk_socket = sock; 1656 } 1657 1658 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1659 { 1660 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1661 return &rcu_dereference_raw(sk->sk_wq)->wait; 1662 } 1663 /* Detach socket from process context. 1664 * Announce socket dead, detach it from wait queue and inode. 1665 * Note that parent inode held reference count on this struct sock, 1666 * we do not release it in this function, because protocol 1667 * probably wants some additional cleanups or even continuing 1668 * to work with this socket (TCP). 1669 */ 1670 static inline void sock_orphan(struct sock *sk) 1671 { 1672 write_lock_bh(&sk->sk_callback_lock); 1673 sock_set_flag(sk, SOCK_DEAD); 1674 sk_set_socket(sk, NULL); 1675 sk->sk_wq = NULL; 1676 write_unlock_bh(&sk->sk_callback_lock); 1677 } 1678 1679 static inline void sock_graft(struct sock *sk, struct socket *parent) 1680 { 1681 write_lock_bh(&sk->sk_callback_lock); 1682 sk->sk_wq = parent->wq; 1683 parent->sk = sk; 1684 sk_set_socket(sk, parent); 1685 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1686 security_sock_graft(sk, parent); 1687 write_unlock_bh(&sk->sk_callback_lock); 1688 } 1689 1690 kuid_t sock_i_uid(struct sock *sk); 1691 unsigned long sock_i_ino(struct sock *sk); 1692 1693 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1694 { 1695 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1696 } 1697 1698 static inline u32 net_tx_rndhash(void) 1699 { 1700 u32 v = prandom_u32(); 1701 1702 return v ?: 1; 1703 } 1704 1705 static inline void sk_set_txhash(struct sock *sk) 1706 { 1707 sk->sk_txhash = net_tx_rndhash(); 1708 } 1709 1710 static inline void sk_rethink_txhash(struct sock *sk) 1711 { 1712 if (sk->sk_txhash) 1713 sk_set_txhash(sk); 1714 } 1715 1716 static inline struct dst_entry * 1717 __sk_dst_get(struct sock *sk) 1718 { 1719 return rcu_dereference_check(sk->sk_dst_cache, 1720 lockdep_sock_is_held(sk)); 1721 } 1722 1723 static inline struct dst_entry * 1724 sk_dst_get(struct sock *sk) 1725 { 1726 struct dst_entry *dst; 1727 1728 rcu_read_lock(); 1729 dst = rcu_dereference(sk->sk_dst_cache); 1730 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1731 dst = NULL; 1732 rcu_read_unlock(); 1733 return dst; 1734 } 1735 1736 static inline void dst_negative_advice(struct sock *sk) 1737 { 1738 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1739 1740 sk_rethink_txhash(sk); 1741 1742 if (dst && dst->ops->negative_advice) { 1743 ndst = dst->ops->negative_advice(dst); 1744 1745 if (ndst != dst) { 1746 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1747 sk_tx_queue_clear(sk); 1748 } 1749 } 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 /* 1759 * This can be called while sk is owned by the caller only, 1760 * with no state that can be checked in a rcu_dereference_check() cond 1761 */ 1762 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1763 rcu_assign_pointer(sk->sk_dst_cache, dst); 1764 dst_release(old_dst); 1765 } 1766 1767 static inline void 1768 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1769 { 1770 struct dst_entry *old_dst; 1771 1772 sk_tx_queue_clear(sk); 1773 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1774 dst_release(old_dst); 1775 } 1776 1777 static inline void 1778 __sk_dst_reset(struct sock *sk) 1779 { 1780 __sk_dst_set(sk, NULL); 1781 } 1782 1783 static inline void 1784 sk_dst_reset(struct sock *sk) 1785 { 1786 sk_dst_set(sk, NULL); 1787 } 1788 1789 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1790 1791 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1792 1793 bool sk_mc_loop(struct sock *sk); 1794 1795 static inline bool sk_can_gso(const struct sock *sk) 1796 { 1797 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1798 } 1799 1800 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1801 1802 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1803 { 1804 sk->sk_route_nocaps |= flags; 1805 sk->sk_route_caps &= ~flags; 1806 } 1807 1808 static inline bool sk_check_csum_caps(struct sock *sk) 1809 { 1810 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1811 (sk->sk_family == PF_INET && 1812 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1813 (sk->sk_family == PF_INET6 && 1814 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1815 } 1816 1817 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1818 struct iov_iter *from, char *to, 1819 int copy, int offset) 1820 { 1821 if (skb->ip_summed == CHECKSUM_NONE) { 1822 __wsum csum = 0; 1823 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy) 1824 return -EFAULT; 1825 skb->csum = csum_block_add(skb->csum, csum, offset); 1826 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1827 if (copy_from_iter_nocache(to, copy, from) != copy) 1828 return -EFAULT; 1829 } else if (copy_from_iter(to, copy, from) != copy) 1830 return -EFAULT; 1831 1832 return 0; 1833 } 1834 1835 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1836 struct iov_iter *from, int copy) 1837 { 1838 int err, offset = skb->len; 1839 1840 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1841 copy, offset); 1842 if (err) 1843 __skb_trim(skb, offset); 1844 1845 return err; 1846 } 1847 1848 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1849 struct sk_buff *skb, 1850 struct page *page, 1851 int off, int copy) 1852 { 1853 int err; 1854 1855 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1856 copy, skb->len); 1857 if (err) 1858 return err; 1859 1860 skb->len += copy; 1861 skb->data_len += copy; 1862 skb->truesize += copy; 1863 sk->sk_wmem_queued += copy; 1864 sk_mem_charge(sk, copy); 1865 return 0; 1866 } 1867 1868 /** 1869 * sk_wmem_alloc_get - returns write allocations 1870 * @sk: socket 1871 * 1872 * Returns sk_wmem_alloc minus initial offset of one 1873 */ 1874 static inline int sk_wmem_alloc_get(const struct sock *sk) 1875 { 1876 return atomic_read(&sk->sk_wmem_alloc) - 1; 1877 } 1878 1879 /** 1880 * sk_rmem_alloc_get - returns read allocations 1881 * @sk: socket 1882 * 1883 * Returns sk_rmem_alloc 1884 */ 1885 static inline int sk_rmem_alloc_get(const struct sock *sk) 1886 { 1887 return atomic_read(&sk->sk_rmem_alloc); 1888 } 1889 1890 /** 1891 * sk_has_allocations - check if allocations are outstanding 1892 * @sk: socket 1893 * 1894 * Returns true if socket has write or read allocations 1895 */ 1896 static inline bool sk_has_allocations(const struct sock *sk) 1897 { 1898 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1899 } 1900 1901 /** 1902 * skwq_has_sleeper - check if there are any waiting processes 1903 * @wq: struct socket_wq 1904 * 1905 * Returns true if socket_wq has waiting processes 1906 * 1907 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1908 * barrier call. They were added due to the race found within the tcp code. 1909 * 1910 * Consider following tcp code paths: 1911 * 1912 * CPU1 CPU2 1913 * 1914 * sys_select receive packet 1915 * ... ... 1916 * __add_wait_queue update tp->rcv_nxt 1917 * ... ... 1918 * tp->rcv_nxt check sock_def_readable 1919 * ... { 1920 * schedule rcu_read_lock(); 1921 * wq = rcu_dereference(sk->sk_wq); 1922 * if (wq && waitqueue_active(&wq->wait)) 1923 * wake_up_interruptible(&wq->wait) 1924 * ... 1925 * } 1926 * 1927 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1928 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1929 * could then endup calling schedule and sleep forever if there are no more 1930 * data on the socket. 1931 * 1932 */ 1933 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1934 { 1935 return wq && wq_has_sleeper(&wq->wait); 1936 } 1937 1938 /** 1939 * sock_poll_wait - place memory barrier behind the poll_wait call. 1940 * @filp: file 1941 * @wait_address: socket wait queue 1942 * @p: poll_table 1943 * 1944 * See the comments in the wq_has_sleeper function. 1945 */ 1946 static inline void sock_poll_wait(struct file *filp, 1947 wait_queue_head_t *wait_address, poll_table *p) 1948 { 1949 if (!poll_does_not_wait(p) && wait_address) { 1950 poll_wait(filp, wait_address, p); 1951 /* We need to be sure we are in sync with the 1952 * socket flags modification. 1953 * 1954 * This memory barrier is paired in the wq_has_sleeper. 1955 */ 1956 smp_mb(); 1957 } 1958 } 1959 1960 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1961 { 1962 if (sk->sk_txhash) { 1963 skb->l4_hash = 1; 1964 skb->hash = sk->sk_txhash; 1965 } 1966 } 1967 1968 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 1969 1970 /* 1971 * Queue a received datagram if it will fit. Stream and sequenced 1972 * protocols can't normally use this as they need to fit buffers in 1973 * and play with them. 1974 * 1975 * Inlined as it's very short and called for pretty much every 1976 * packet ever received. 1977 */ 1978 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1979 { 1980 skb_orphan(skb); 1981 skb->sk = sk; 1982 skb->destructor = sock_rfree; 1983 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1984 sk_mem_charge(sk, skb->truesize); 1985 } 1986 1987 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1988 unsigned long expires); 1989 1990 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1991 1992 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff *skb, 1993 unsigned int flags); 1994 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1995 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1996 1997 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1998 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 1999 2000 /* 2001 * Recover an error report and clear atomically 2002 */ 2003 2004 static inline int sock_error(struct sock *sk) 2005 { 2006 int err; 2007 if (likely(!sk->sk_err)) 2008 return 0; 2009 err = xchg(&sk->sk_err, 0); 2010 return -err; 2011 } 2012 2013 static inline unsigned long sock_wspace(struct sock *sk) 2014 { 2015 int amt = 0; 2016 2017 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2018 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2019 if (amt < 0) 2020 amt = 0; 2021 } 2022 return amt; 2023 } 2024 2025 /* Note: 2026 * We use sk->sk_wq_raw, from contexts knowing this 2027 * pointer is not NULL and cannot disappear/change. 2028 */ 2029 static inline void sk_set_bit(int nr, struct sock *sk) 2030 { 2031 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2032 !sock_flag(sk, SOCK_FASYNC)) 2033 return; 2034 2035 set_bit(nr, &sk->sk_wq_raw->flags); 2036 } 2037 2038 static inline void sk_clear_bit(int nr, struct sock *sk) 2039 { 2040 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2041 !sock_flag(sk, SOCK_FASYNC)) 2042 return; 2043 2044 clear_bit(nr, &sk->sk_wq_raw->flags); 2045 } 2046 2047 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2048 { 2049 if (sock_flag(sk, SOCK_FASYNC)) { 2050 rcu_read_lock(); 2051 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2052 rcu_read_unlock(); 2053 } 2054 } 2055 2056 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2057 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2058 * Note: for send buffers, TCP works better if we can build two skbs at 2059 * minimum. 2060 */ 2061 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2062 2063 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2064 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2065 2066 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2067 { 2068 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2069 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2070 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2071 } 2072 } 2073 2074 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2075 bool force_schedule); 2076 2077 /** 2078 * sk_page_frag - return an appropriate page_frag 2079 * @sk: socket 2080 * 2081 * If socket allocation mode allows current thread to sleep, it means its 2082 * safe to use the per task page_frag instead of the per socket one. 2083 */ 2084 static inline struct page_frag *sk_page_frag(struct sock *sk) 2085 { 2086 if (gfpflags_allow_blocking(sk->sk_allocation)) 2087 return ¤t->task_frag; 2088 2089 return &sk->sk_frag; 2090 } 2091 2092 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2093 2094 /* 2095 * Default write policy as shown to user space via poll/select/SIGIO 2096 */ 2097 static inline bool sock_writeable(const struct sock *sk) 2098 { 2099 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2100 } 2101 2102 static inline gfp_t gfp_any(void) 2103 { 2104 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2105 } 2106 2107 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2108 { 2109 return noblock ? 0 : sk->sk_rcvtimeo; 2110 } 2111 2112 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2113 { 2114 return noblock ? 0 : sk->sk_sndtimeo; 2115 } 2116 2117 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2118 { 2119 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2120 } 2121 2122 /* Alas, with timeout socket operations are not restartable. 2123 * Compare this to poll(). 2124 */ 2125 static inline int sock_intr_errno(long timeo) 2126 { 2127 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2128 } 2129 2130 struct sock_skb_cb { 2131 u32 dropcount; 2132 }; 2133 2134 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2135 * using skb->cb[] would keep using it directly and utilize its 2136 * alignement guarantee. 2137 */ 2138 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2139 sizeof(struct sock_skb_cb))) 2140 2141 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2142 SOCK_SKB_CB_OFFSET)) 2143 2144 #define sock_skb_cb_check_size(size) \ 2145 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2146 2147 static inline void 2148 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2149 { 2150 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops); 2151 } 2152 2153 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2154 { 2155 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2156 2157 atomic_add(segs, &sk->sk_drops); 2158 } 2159 2160 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2161 struct sk_buff *skb); 2162 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2163 struct sk_buff *skb); 2164 2165 static inline void 2166 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2167 { 2168 ktime_t kt = skb->tstamp; 2169 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2170 2171 /* 2172 * generate control messages if 2173 * - receive time stamping in software requested 2174 * - software time stamp available and wanted 2175 * - hardware time stamps available and wanted 2176 */ 2177 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2178 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2179 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2180 (hwtstamps->hwtstamp.tv64 && 2181 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2182 __sock_recv_timestamp(msg, sk, skb); 2183 else 2184 sk->sk_stamp = kt; 2185 2186 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2187 __sock_recv_wifi_status(msg, sk, skb); 2188 } 2189 2190 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2191 struct sk_buff *skb); 2192 2193 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2194 struct sk_buff *skb) 2195 { 2196 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2197 (1UL << SOCK_RCVTSTAMP)) 2198 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2199 SOF_TIMESTAMPING_RAW_HARDWARE) 2200 2201 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2202 __sock_recv_ts_and_drops(msg, sk, skb); 2203 else 2204 sk->sk_stamp = skb->tstamp; 2205 } 2206 2207 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2208 2209 /** 2210 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2211 * @sk: socket sending this packet 2212 * @tsflags: timestamping flags to use 2213 * @tx_flags: completed with instructions for time stamping 2214 * 2215 * Note : callers should take care of initial *tx_flags value (usually 0) 2216 */ 2217 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2218 __u8 *tx_flags) 2219 { 2220 if (unlikely(tsflags)) 2221 __sock_tx_timestamp(tsflags, tx_flags); 2222 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2223 *tx_flags |= SKBTX_WIFI_STATUS; 2224 } 2225 2226 /** 2227 * sk_eat_skb - Release a skb if it is no longer needed 2228 * @sk: socket to eat this skb from 2229 * @skb: socket buffer to eat 2230 * 2231 * This routine must be called with interrupts disabled or with the socket 2232 * locked so that the sk_buff queue operation is ok. 2233 */ 2234 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2235 { 2236 __skb_unlink(skb, &sk->sk_receive_queue); 2237 __kfree_skb(skb); 2238 } 2239 2240 static inline 2241 struct net *sock_net(const struct sock *sk) 2242 { 2243 return read_pnet(&sk->sk_net); 2244 } 2245 2246 static inline 2247 void sock_net_set(struct sock *sk, struct net *net) 2248 { 2249 write_pnet(&sk->sk_net, net); 2250 } 2251 2252 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2253 { 2254 if (skb->sk) { 2255 struct sock *sk = skb->sk; 2256 2257 skb->destructor = NULL; 2258 skb->sk = NULL; 2259 return sk; 2260 } 2261 return NULL; 2262 } 2263 2264 /* This helper checks if a socket is a full socket, 2265 * ie _not_ a timewait or request socket. 2266 */ 2267 static inline bool sk_fullsock(const struct sock *sk) 2268 { 2269 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2270 } 2271 2272 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2273 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2274 */ 2275 static inline bool sk_listener(const struct sock *sk) 2276 { 2277 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2278 } 2279 2280 /** 2281 * sk_state_load - read sk->sk_state for lockless contexts 2282 * @sk: socket pointer 2283 * 2284 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2285 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2286 */ 2287 static inline int sk_state_load(const struct sock *sk) 2288 { 2289 return smp_load_acquire(&sk->sk_state); 2290 } 2291 2292 /** 2293 * sk_state_store - update sk->sk_state 2294 * @sk: socket pointer 2295 * @newstate: new state 2296 * 2297 * Paired with sk_state_load(). Should be used in contexts where 2298 * state change might impact lockless readers. 2299 */ 2300 static inline void sk_state_store(struct sock *sk, int newstate) 2301 { 2302 smp_store_release(&sk->sk_state, newstate); 2303 } 2304 2305 void sock_enable_timestamp(struct sock *sk, int flag); 2306 int sock_get_timestamp(struct sock *, struct timeval __user *); 2307 int sock_get_timestampns(struct sock *, struct timespec __user *); 2308 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2309 int type); 2310 2311 bool sk_ns_capable(const struct sock *sk, 2312 struct user_namespace *user_ns, int cap); 2313 bool sk_capable(const struct sock *sk, int cap); 2314 bool sk_net_capable(const struct sock *sk, int cap); 2315 2316 extern __u32 sysctl_wmem_max; 2317 extern __u32 sysctl_rmem_max; 2318 2319 extern int sysctl_tstamp_allow_data; 2320 extern int sysctl_optmem_max; 2321 2322 extern __u32 sysctl_wmem_default; 2323 extern __u32 sysctl_rmem_default; 2324 2325 #endif /* _SOCK_H */ 2326