1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the AF_INET socket handler. 7 * 8 * Version: @(#)sock.h 1.0.4 05/13/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Corey Minyard <wf-rch!minyard@relay.EU.net> 13 * Florian La Roche <flla@stud.uni-sb.de> 14 * 15 * Fixes: 16 * Alan Cox : Volatiles in skbuff pointers. See 17 * skbuff comments. May be overdone, 18 * better to prove they can be removed 19 * than the reverse. 20 * Alan Cox : Added a zapped field for tcp to note 21 * a socket is reset and must stay shut up 22 * Alan Cox : New fields for options 23 * Pauline Middelink : identd support 24 * Alan Cox : Eliminate low level recv/recvfrom 25 * David S. Miller : New socket lookup architecture. 26 * Steve Whitehouse: Default routines for sock_ops 27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 28 * protinfo be just a void pointer, as the 29 * protocol specific parts were moved to 30 * respective headers and ipv4/v6, etc now 31 * use private slabcaches for its socks 32 * Pedro Hortas : New flags field for socket options 33 * 34 * 35 * This program is free software; you can redistribute it and/or 36 * modify it under the terms of the GNU General Public License 37 * as published by the Free Software Foundation; either version 38 * 2 of the License, or (at your option) any later version. 39 */ 40 #ifndef _SOCK_H 41 #define _SOCK_H 42 43 #include <linux/hardirq.h> 44 #include <linux/kernel.h> 45 #include <linux/list.h> 46 #include <linux/list_nulls.h> 47 #include <linux/timer.h> 48 #include <linux/cache.h> 49 #include <linux/bitops.h> 50 #include <linux/lockdep.h> 51 #include <linux/netdevice.h> 52 #include <linux/skbuff.h> /* struct sk_buff */ 53 #include <linux/mm.h> 54 #include <linux/security.h> 55 #include <linux/slab.h> 56 #include <linux/uaccess.h> 57 #include <linux/page_counter.h> 58 #include <linux/memcontrol.h> 59 #include <linux/static_key.h> 60 #include <linux/sched.h> 61 #include <linux/wait.h> 62 #include <linux/cgroup-defs.h> 63 64 #include <linux/filter.h> 65 #include <linux/rculist_nulls.h> 66 #include <linux/poll.h> 67 68 #include <linux/atomic.h> 69 #include <linux/refcount.h> 70 #include <net/dst.h> 71 #include <net/checksum.h> 72 #include <net/tcp_states.h> 73 #include <linux/net_tstamp.h> 74 #include <net/smc.h> 75 76 /* 77 * This structure really needs to be cleaned up. 78 * Most of it is for TCP, and not used by any of 79 * the other protocols. 80 */ 81 82 /* Define this to get the SOCK_DBG debugging facility. */ 83 #define SOCK_DEBUGGING 84 #ifdef SOCK_DEBUGGING 85 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 86 printk(KERN_DEBUG msg); } while (0) 87 #else 88 /* Validate arguments and do nothing */ 89 static inline __printf(2, 3) 90 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 91 { 92 } 93 #endif 94 95 /* This is the per-socket lock. The spinlock provides a synchronization 96 * between user contexts and software interrupt processing, whereas the 97 * mini-semaphore synchronizes multiple users amongst themselves. 98 */ 99 typedef struct { 100 spinlock_t slock; 101 int owned; 102 wait_queue_head_t wq; 103 /* 104 * We express the mutex-alike socket_lock semantics 105 * to the lock validator by explicitly managing 106 * the slock as a lock variant (in addition to 107 * the slock itself): 108 */ 109 #ifdef CONFIG_DEBUG_LOCK_ALLOC 110 struct lockdep_map dep_map; 111 #endif 112 } socket_lock_t; 113 114 struct sock; 115 struct proto; 116 struct net; 117 118 typedef __u32 __bitwise __portpair; 119 typedef __u64 __bitwise __addrpair; 120 121 /** 122 * struct sock_common - minimal network layer representation of sockets 123 * @skc_daddr: Foreign IPv4 addr 124 * @skc_rcv_saddr: Bound local IPv4 addr 125 * @skc_hash: hash value used with various protocol lookup tables 126 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 127 * @skc_dport: placeholder for inet_dport/tw_dport 128 * @skc_num: placeholder for inet_num/tw_num 129 * @skc_family: network address family 130 * @skc_state: Connection state 131 * @skc_reuse: %SO_REUSEADDR setting 132 * @skc_reuseport: %SO_REUSEPORT setting 133 * @skc_bound_dev_if: bound device index if != 0 134 * @skc_bind_node: bind hash linkage for various protocol lookup tables 135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 136 * @skc_prot: protocol handlers inside a network family 137 * @skc_net: reference to the network namespace of this socket 138 * @skc_node: main hash linkage for various protocol lookup tables 139 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 140 * @skc_tx_queue_mapping: tx queue number for this connection 141 * @skc_flags: place holder for sk_flags 142 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 143 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 144 * @skc_incoming_cpu: record/match cpu processing incoming packets 145 * @skc_refcnt: reference count 146 * 147 * This is the minimal network layer representation of sockets, the header 148 * for struct sock and struct inet_timewait_sock. 149 */ 150 struct sock_common { 151 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 152 * address on 64bit arches : cf INET_MATCH() 153 */ 154 union { 155 __addrpair skc_addrpair; 156 struct { 157 __be32 skc_daddr; 158 __be32 skc_rcv_saddr; 159 }; 160 }; 161 union { 162 unsigned int skc_hash; 163 __u16 skc_u16hashes[2]; 164 }; 165 /* skc_dport && skc_num must be grouped as well */ 166 union { 167 __portpair skc_portpair; 168 struct { 169 __be16 skc_dport; 170 __u16 skc_num; 171 }; 172 }; 173 174 unsigned short skc_family; 175 volatile unsigned char skc_state; 176 unsigned char skc_reuse:4; 177 unsigned char skc_reuseport:1; 178 unsigned char skc_ipv6only:1; 179 unsigned char skc_net_refcnt:1; 180 int skc_bound_dev_if; 181 union { 182 struct hlist_node skc_bind_node; 183 struct hlist_node skc_portaddr_node; 184 }; 185 struct proto *skc_prot; 186 possible_net_t skc_net; 187 188 #if IS_ENABLED(CONFIG_IPV6) 189 struct in6_addr skc_v6_daddr; 190 struct in6_addr skc_v6_rcv_saddr; 191 #endif 192 193 atomic64_t skc_cookie; 194 195 /* following fields are padding to force 196 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 197 * assuming IPV6 is enabled. We use this padding differently 198 * for different kind of 'sockets' 199 */ 200 union { 201 unsigned long skc_flags; 202 struct sock *skc_listener; /* request_sock */ 203 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 204 }; 205 /* 206 * fields between dontcopy_begin/dontcopy_end 207 * are not copied in sock_copy() 208 */ 209 /* private: */ 210 int skc_dontcopy_begin[0]; 211 /* public: */ 212 union { 213 struct hlist_node skc_node; 214 struct hlist_nulls_node skc_nulls_node; 215 }; 216 int skc_tx_queue_mapping; 217 union { 218 int skc_incoming_cpu; 219 u32 skc_rcv_wnd; 220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 221 }; 222 223 refcount_t skc_refcnt; 224 /* private: */ 225 int skc_dontcopy_end[0]; 226 union { 227 u32 skc_rxhash; 228 u32 skc_window_clamp; 229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 230 }; 231 /* public: */ 232 }; 233 234 /** 235 * struct sock - network layer representation of sockets 236 * @__sk_common: shared layout with inet_timewait_sock 237 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 238 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 239 * @sk_lock: synchronizer 240 * @sk_kern_sock: True if sock is using kernel lock classes 241 * @sk_rcvbuf: size of receive buffer in bytes 242 * @sk_wq: sock wait queue and async head 243 * @sk_rx_dst: receive input route used by early demux 244 * @sk_dst_cache: destination cache 245 * @sk_dst_pending_confirm: need to confirm neighbour 246 * @sk_policy: flow policy 247 * @sk_receive_queue: incoming packets 248 * @sk_wmem_alloc: transmit queue bytes committed 249 * @sk_tsq_flags: TCP Small Queues flags 250 * @sk_write_queue: Packet sending queue 251 * @sk_omem_alloc: "o" is "option" or "other" 252 * @sk_wmem_queued: persistent queue size 253 * @sk_forward_alloc: space allocated forward 254 * @sk_napi_id: id of the last napi context to receive data for sk 255 * @sk_ll_usec: usecs to busypoll when there is no data 256 * @sk_allocation: allocation mode 257 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 258 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 259 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 260 * @sk_sndbuf: size of send buffer in bytes 261 * @__sk_flags_offset: empty field used to determine location of bitfield 262 * @sk_padding: unused element for alignment 263 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 264 * @sk_no_check_rx: allow zero checksum in RX packets 265 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 266 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 267 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 268 * @sk_gso_max_size: Maximum GSO segment size to build 269 * @sk_gso_max_segs: Maximum number of GSO segments 270 * @sk_lingertime: %SO_LINGER l_linger setting 271 * @sk_backlog: always used with the per-socket spinlock held 272 * @sk_callback_lock: used with the callbacks in the end of this struct 273 * @sk_error_queue: rarely used 274 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 275 * IPV6_ADDRFORM for instance) 276 * @sk_err: last error 277 * @sk_err_soft: errors that don't cause failure but are the cause of a 278 * persistent failure not just 'timed out' 279 * @sk_drops: raw/udp drops counter 280 * @sk_ack_backlog: current listen backlog 281 * @sk_max_ack_backlog: listen backlog set in listen() 282 * @sk_uid: user id of owner 283 * @sk_priority: %SO_PRIORITY setting 284 * @sk_type: socket type (%SOCK_STREAM, etc) 285 * @sk_protocol: which protocol this socket belongs in this network family 286 * @sk_peer_pid: &struct pid for this socket's peer 287 * @sk_peer_cred: %SO_PEERCRED setting 288 * @sk_rcvlowat: %SO_RCVLOWAT setting 289 * @sk_rcvtimeo: %SO_RCVTIMEO setting 290 * @sk_sndtimeo: %SO_SNDTIMEO setting 291 * @sk_txhash: computed flow hash for use on transmit 292 * @sk_filter: socket filtering instructions 293 * @sk_timer: sock cleanup timer 294 * @sk_stamp: time stamp of last packet received 295 * @sk_tsflags: SO_TIMESTAMPING socket options 296 * @sk_tskey: counter to disambiguate concurrent tstamp requests 297 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 298 * @sk_socket: Identd and reporting IO signals 299 * @sk_user_data: RPC layer private data 300 * @sk_frag: cached page frag 301 * @sk_peek_off: current peek_offset value 302 * @sk_send_head: front of stuff to transmit 303 * @sk_security: used by security modules 304 * @sk_mark: generic packet mark 305 * @sk_cgrp_data: cgroup data for this cgroup 306 * @sk_memcg: this socket's memory cgroup association 307 * @sk_write_pending: a write to stream socket waits to start 308 * @sk_state_change: callback to indicate change in the state of the sock 309 * @sk_data_ready: callback to indicate there is data to be processed 310 * @sk_write_space: callback to indicate there is bf sending space available 311 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 312 * @sk_backlog_rcv: callback to process the backlog 313 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 314 * @sk_reuseport_cb: reuseport group container 315 * @sk_rcu: used during RCU grace period 316 */ 317 struct sock { 318 /* 319 * Now struct inet_timewait_sock also uses sock_common, so please just 320 * don't add nothing before this first member (__sk_common) --acme 321 */ 322 struct sock_common __sk_common; 323 #define sk_node __sk_common.skc_node 324 #define sk_nulls_node __sk_common.skc_nulls_node 325 #define sk_refcnt __sk_common.skc_refcnt 326 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 327 328 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 329 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 330 #define sk_hash __sk_common.skc_hash 331 #define sk_portpair __sk_common.skc_portpair 332 #define sk_num __sk_common.skc_num 333 #define sk_dport __sk_common.skc_dport 334 #define sk_addrpair __sk_common.skc_addrpair 335 #define sk_daddr __sk_common.skc_daddr 336 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 337 #define sk_family __sk_common.skc_family 338 #define sk_state __sk_common.skc_state 339 #define sk_reuse __sk_common.skc_reuse 340 #define sk_reuseport __sk_common.skc_reuseport 341 #define sk_ipv6only __sk_common.skc_ipv6only 342 #define sk_net_refcnt __sk_common.skc_net_refcnt 343 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 344 #define sk_bind_node __sk_common.skc_bind_node 345 #define sk_prot __sk_common.skc_prot 346 #define sk_net __sk_common.skc_net 347 #define sk_v6_daddr __sk_common.skc_v6_daddr 348 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 349 #define sk_cookie __sk_common.skc_cookie 350 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 351 #define sk_flags __sk_common.skc_flags 352 #define sk_rxhash __sk_common.skc_rxhash 353 354 socket_lock_t sk_lock; 355 atomic_t sk_drops; 356 int sk_rcvlowat; 357 struct sk_buff_head sk_error_queue; 358 struct sk_buff_head sk_receive_queue; 359 /* 360 * The backlog queue is special, it is always used with 361 * the per-socket spinlock held and requires low latency 362 * access. Therefore we special case it's implementation. 363 * Note : rmem_alloc is in this structure to fill a hole 364 * on 64bit arches, not because its logically part of 365 * backlog. 366 */ 367 struct { 368 atomic_t rmem_alloc; 369 int len; 370 struct sk_buff *head; 371 struct sk_buff *tail; 372 } sk_backlog; 373 #define sk_rmem_alloc sk_backlog.rmem_alloc 374 375 int sk_forward_alloc; 376 #ifdef CONFIG_NET_RX_BUSY_POLL 377 unsigned int sk_ll_usec; 378 /* ===== mostly read cache line ===== */ 379 unsigned int sk_napi_id; 380 #endif 381 int sk_rcvbuf; 382 383 struct sk_filter __rcu *sk_filter; 384 union { 385 struct socket_wq __rcu *sk_wq; 386 struct socket_wq *sk_wq_raw; 387 }; 388 #ifdef CONFIG_XFRM 389 struct xfrm_policy __rcu *sk_policy[2]; 390 #endif 391 struct dst_entry *sk_rx_dst; 392 struct dst_entry __rcu *sk_dst_cache; 393 atomic_t sk_omem_alloc; 394 int sk_sndbuf; 395 396 /* ===== cache line for TX ===== */ 397 int sk_wmem_queued; 398 refcount_t sk_wmem_alloc; 399 unsigned long sk_tsq_flags; 400 struct sk_buff *sk_send_head; 401 struct sk_buff_head sk_write_queue; 402 __s32 sk_peek_off; 403 int sk_write_pending; 404 __u32 sk_dst_pending_confirm; 405 u32 sk_pacing_status; /* see enum sk_pacing */ 406 long sk_sndtimeo; 407 struct timer_list sk_timer; 408 __u32 sk_priority; 409 __u32 sk_mark; 410 u32 sk_pacing_rate; /* bytes per second */ 411 u32 sk_max_pacing_rate; 412 struct page_frag sk_frag; 413 netdev_features_t sk_route_caps; 414 netdev_features_t sk_route_nocaps; 415 int sk_gso_type; 416 unsigned int sk_gso_max_size; 417 gfp_t sk_allocation; 418 __u32 sk_txhash; 419 420 /* 421 * Because of non atomicity rules, all 422 * changes are protected by socket lock. 423 */ 424 unsigned int __sk_flags_offset[0]; 425 #ifdef __BIG_ENDIAN_BITFIELD 426 #define SK_FL_PROTO_SHIFT 16 427 #define SK_FL_PROTO_MASK 0x00ff0000 428 429 #define SK_FL_TYPE_SHIFT 0 430 #define SK_FL_TYPE_MASK 0x0000ffff 431 #else 432 #define SK_FL_PROTO_SHIFT 8 433 #define SK_FL_PROTO_MASK 0x0000ff00 434 435 #define SK_FL_TYPE_SHIFT 16 436 #define SK_FL_TYPE_MASK 0xffff0000 437 #endif 438 439 kmemcheck_bitfield_begin(flags); 440 unsigned int sk_padding : 1, 441 sk_kern_sock : 1, 442 sk_no_check_tx : 1, 443 sk_no_check_rx : 1, 444 sk_userlocks : 4, 445 sk_protocol : 8, 446 sk_type : 16; 447 #define SK_PROTOCOL_MAX U8_MAX 448 kmemcheck_bitfield_end(flags); 449 450 u16 sk_gso_max_segs; 451 unsigned long sk_lingertime; 452 struct proto *sk_prot_creator; 453 rwlock_t sk_callback_lock; 454 int sk_err, 455 sk_err_soft; 456 u32 sk_ack_backlog; 457 u32 sk_max_ack_backlog; 458 kuid_t sk_uid; 459 struct pid *sk_peer_pid; 460 const struct cred *sk_peer_cred; 461 long sk_rcvtimeo; 462 ktime_t sk_stamp; 463 u16 sk_tsflags; 464 u8 sk_shutdown; 465 u32 sk_tskey; 466 atomic_t sk_zckey; 467 struct socket *sk_socket; 468 void *sk_user_data; 469 #ifdef CONFIG_SECURITY 470 void *sk_security; 471 #endif 472 struct sock_cgroup_data sk_cgrp_data; 473 struct mem_cgroup *sk_memcg; 474 void (*sk_state_change)(struct sock *sk); 475 void (*sk_data_ready)(struct sock *sk); 476 void (*sk_write_space)(struct sock *sk); 477 void (*sk_error_report)(struct sock *sk); 478 int (*sk_backlog_rcv)(struct sock *sk, 479 struct sk_buff *skb); 480 void (*sk_destruct)(struct sock *sk); 481 struct sock_reuseport __rcu *sk_reuseport_cb; 482 struct rcu_head sk_rcu; 483 }; 484 485 enum sk_pacing { 486 SK_PACING_NONE = 0, 487 SK_PACING_NEEDED = 1, 488 SK_PACING_FQ = 2, 489 }; 490 491 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 492 493 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 494 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 495 496 /* 497 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 498 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 499 * on a socket means that the socket will reuse everybody else's port 500 * without looking at the other's sk_reuse value. 501 */ 502 503 #define SK_NO_REUSE 0 504 #define SK_CAN_REUSE 1 505 #define SK_FORCE_REUSE 2 506 507 int sk_set_peek_off(struct sock *sk, int val); 508 509 static inline int sk_peek_offset(struct sock *sk, int flags) 510 { 511 if (unlikely(flags & MSG_PEEK)) { 512 return READ_ONCE(sk->sk_peek_off); 513 } 514 515 return 0; 516 } 517 518 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 519 { 520 s32 off = READ_ONCE(sk->sk_peek_off); 521 522 if (unlikely(off >= 0)) { 523 off = max_t(s32, off - val, 0); 524 WRITE_ONCE(sk->sk_peek_off, off); 525 } 526 } 527 528 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 529 { 530 sk_peek_offset_bwd(sk, -val); 531 } 532 533 /* 534 * Hashed lists helper routines 535 */ 536 static inline struct sock *sk_entry(const struct hlist_node *node) 537 { 538 return hlist_entry(node, struct sock, sk_node); 539 } 540 541 static inline struct sock *__sk_head(const struct hlist_head *head) 542 { 543 return hlist_entry(head->first, struct sock, sk_node); 544 } 545 546 static inline struct sock *sk_head(const struct hlist_head *head) 547 { 548 return hlist_empty(head) ? NULL : __sk_head(head); 549 } 550 551 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 552 { 553 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 554 } 555 556 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 557 { 558 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 559 } 560 561 static inline struct sock *sk_next(const struct sock *sk) 562 { 563 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 564 } 565 566 static inline struct sock *sk_nulls_next(const struct sock *sk) 567 { 568 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 569 hlist_nulls_entry(sk->sk_nulls_node.next, 570 struct sock, sk_nulls_node) : 571 NULL; 572 } 573 574 static inline bool sk_unhashed(const struct sock *sk) 575 { 576 return hlist_unhashed(&sk->sk_node); 577 } 578 579 static inline bool sk_hashed(const struct sock *sk) 580 { 581 return !sk_unhashed(sk); 582 } 583 584 static inline void sk_node_init(struct hlist_node *node) 585 { 586 node->pprev = NULL; 587 } 588 589 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 590 { 591 node->pprev = NULL; 592 } 593 594 static inline void __sk_del_node(struct sock *sk) 595 { 596 __hlist_del(&sk->sk_node); 597 } 598 599 /* NB: equivalent to hlist_del_init_rcu */ 600 static inline bool __sk_del_node_init(struct sock *sk) 601 { 602 if (sk_hashed(sk)) { 603 __sk_del_node(sk); 604 sk_node_init(&sk->sk_node); 605 return true; 606 } 607 return false; 608 } 609 610 /* Grab socket reference count. This operation is valid only 611 when sk is ALREADY grabbed f.e. it is found in hash table 612 or a list and the lookup is made under lock preventing hash table 613 modifications. 614 */ 615 616 static __always_inline void sock_hold(struct sock *sk) 617 { 618 refcount_inc(&sk->sk_refcnt); 619 } 620 621 /* Ungrab socket in the context, which assumes that socket refcnt 622 cannot hit zero, f.e. it is true in context of any socketcall. 623 */ 624 static __always_inline void __sock_put(struct sock *sk) 625 { 626 refcount_dec(&sk->sk_refcnt); 627 } 628 629 static inline bool sk_del_node_init(struct sock *sk) 630 { 631 bool rc = __sk_del_node_init(sk); 632 633 if (rc) { 634 /* paranoid for a while -acme */ 635 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 636 __sock_put(sk); 637 } 638 return rc; 639 } 640 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 641 642 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 643 { 644 if (sk_hashed(sk)) { 645 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 646 return true; 647 } 648 return false; 649 } 650 651 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 652 { 653 bool rc = __sk_nulls_del_node_init_rcu(sk); 654 655 if (rc) { 656 /* paranoid for a while -acme */ 657 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 658 __sock_put(sk); 659 } 660 return rc; 661 } 662 663 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 664 { 665 hlist_add_head(&sk->sk_node, list); 666 } 667 668 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 669 { 670 sock_hold(sk); 671 __sk_add_node(sk, list); 672 } 673 674 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 675 { 676 sock_hold(sk); 677 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 678 sk->sk_family == AF_INET6) 679 hlist_add_tail_rcu(&sk->sk_node, list); 680 else 681 hlist_add_head_rcu(&sk->sk_node, list); 682 } 683 684 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 685 { 686 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 687 sk->sk_family == AF_INET6) 688 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 689 else 690 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 691 } 692 693 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 694 { 695 sock_hold(sk); 696 __sk_nulls_add_node_rcu(sk, list); 697 } 698 699 static inline void __sk_del_bind_node(struct sock *sk) 700 { 701 __hlist_del(&sk->sk_bind_node); 702 } 703 704 static inline void sk_add_bind_node(struct sock *sk, 705 struct hlist_head *list) 706 { 707 hlist_add_head(&sk->sk_bind_node, list); 708 } 709 710 #define sk_for_each(__sk, list) \ 711 hlist_for_each_entry(__sk, list, sk_node) 712 #define sk_for_each_rcu(__sk, list) \ 713 hlist_for_each_entry_rcu(__sk, list, sk_node) 714 #define sk_nulls_for_each(__sk, node, list) \ 715 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 716 #define sk_nulls_for_each_rcu(__sk, node, list) \ 717 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 718 #define sk_for_each_from(__sk) \ 719 hlist_for_each_entry_from(__sk, sk_node) 720 #define sk_nulls_for_each_from(__sk, node) \ 721 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 722 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 723 #define sk_for_each_safe(__sk, tmp, list) \ 724 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 725 #define sk_for_each_bound(__sk, list) \ 726 hlist_for_each_entry(__sk, list, sk_bind_node) 727 728 /** 729 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 730 * @tpos: the type * to use as a loop cursor. 731 * @pos: the &struct hlist_node to use as a loop cursor. 732 * @head: the head for your list. 733 * @offset: offset of hlist_node within the struct. 734 * 735 */ 736 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 737 for (pos = rcu_dereference((head)->first); \ 738 pos != NULL && \ 739 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 740 pos = rcu_dereference(pos->next)) 741 742 static inline struct user_namespace *sk_user_ns(struct sock *sk) 743 { 744 /* Careful only use this in a context where these parameters 745 * can not change and must all be valid, such as recvmsg from 746 * userspace. 747 */ 748 return sk->sk_socket->file->f_cred->user_ns; 749 } 750 751 /* Sock flags */ 752 enum sock_flags { 753 SOCK_DEAD, 754 SOCK_DONE, 755 SOCK_URGINLINE, 756 SOCK_KEEPOPEN, 757 SOCK_LINGER, 758 SOCK_DESTROY, 759 SOCK_BROADCAST, 760 SOCK_TIMESTAMP, 761 SOCK_ZAPPED, 762 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 763 SOCK_DBG, /* %SO_DEBUG setting */ 764 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 765 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 766 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 767 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 768 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 769 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 770 SOCK_FASYNC, /* fasync() active */ 771 SOCK_RXQ_OVFL, 772 SOCK_ZEROCOPY, /* buffers from userspace */ 773 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 774 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 775 * Will use last 4 bytes of packet sent from 776 * user-space instead. 777 */ 778 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 779 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 780 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 781 }; 782 783 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 784 785 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 786 { 787 nsk->sk_flags = osk->sk_flags; 788 } 789 790 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 791 { 792 __set_bit(flag, &sk->sk_flags); 793 } 794 795 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 796 { 797 __clear_bit(flag, &sk->sk_flags); 798 } 799 800 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 801 { 802 return test_bit(flag, &sk->sk_flags); 803 } 804 805 #ifdef CONFIG_NET 806 extern struct static_key memalloc_socks; 807 static inline int sk_memalloc_socks(void) 808 { 809 return static_key_false(&memalloc_socks); 810 } 811 #else 812 813 static inline int sk_memalloc_socks(void) 814 { 815 return 0; 816 } 817 818 #endif 819 820 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 821 { 822 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 823 } 824 825 static inline void sk_acceptq_removed(struct sock *sk) 826 { 827 sk->sk_ack_backlog--; 828 } 829 830 static inline void sk_acceptq_added(struct sock *sk) 831 { 832 sk->sk_ack_backlog++; 833 } 834 835 static inline bool sk_acceptq_is_full(const struct sock *sk) 836 { 837 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 838 } 839 840 /* 841 * Compute minimal free write space needed to queue new packets. 842 */ 843 static inline int sk_stream_min_wspace(const struct sock *sk) 844 { 845 return sk->sk_wmem_queued >> 1; 846 } 847 848 static inline int sk_stream_wspace(const struct sock *sk) 849 { 850 return sk->sk_sndbuf - sk->sk_wmem_queued; 851 } 852 853 void sk_stream_write_space(struct sock *sk); 854 855 /* OOB backlog add */ 856 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 857 { 858 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 859 skb_dst_force(skb); 860 861 if (!sk->sk_backlog.tail) 862 sk->sk_backlog.head = skb; 863 else 864 sk->sk_backlog.tail->next = skb; 865 866 sk->sk_backlog.tail = skb; 867 skb->next = NULL; 868 } 869 870 /* 871 * Take into account size of receive queue and backlog queue 872 * Do not take into account this skb truesize, 873 * to allow even a single big packet to come. 874 */ 875 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 876 { 877 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 878 879 return qsize > limit; 880 } 881 882 /* The per-socket spinlock must be held here. */ 883 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 884 unsigned int limit) 885 { 886 if (sk_rcvqueues_full(sk, limit)) 887 return -ENOBUFS; 888 889 /* 890 * If the skb was allocated from pfmemalloc reserves, only 891 * allow SOCK_MEMALLOC sockets to use it as this socket is 892 * helping free memory 893 */ 894 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 895 return -ENOMEM; 896 897 __sk_add_backlog(sk, skb); 898 sk->sk_backlog.len += skb->truesize; 899 return 0; 900 } 901 902 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 903 904 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 905 { 906 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 907 return __sk_backlog_rcv(sk, skb); 908 909 return sk->sk_backlog_rcv(sk, skb); 910 } 911 912 static inline void sk_incoming_cpu_update(struct sock *sk) 913 { 914 int cpu = raw_smp_processor_id(); 915 916 if (unlikely(sk->sk_incoming_cpu != cpu)) 917 sk->sk_incoming_cpu = cpu; 918 } 919 920 static inline void sock_rps_record_flow_hash(__u32 hash) 921 { 922 #ifdef CONFIG_RPS 923 struct rps_sock_flow_table *sock_flow_table; 924 925 rcu_read_lock(); 926 sock_flow_table = rcu_dereference(rps_sock_flow_table); 927 rps_record_sock_flow(sock_flow_table, hash); 928 rcu_read_unlock(); 929 #endif 930 } 931 932 static inline void sock_rps_record_flow(const struct sock *sk) 933 { 934 #ifdef CONFIG_RPS 935 if (static_key_false(&rfs_needed)) { 936 /* Reading sk->sk_rxhash might incur an expensive cache line 937 * miss. 938 * 939 * TCP_ESTABLISHED does cover almost all states where RFS 940 * might be useful, and is cheaper [1] than testing : 941 * IPv4: inet_sk(sk)->inet_daddr 942 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 943 * OR an additional socket flag 944 * [1] : sk_state and sk_prot are in the same cache line. 945 */ 946 if (sk->sk_state == TCP_ESTABLISHED) 947 sock_rps_record_flow_hash(sk->sk_rxhash); 948 } 949 #endif 950 } 951 952 static inline void sock_rps_save_rxhash(struct sock *sk, 953 const struct sk_buff *skb) 954 { 955 #ifdef CONFIG_RPS 956 if (unlikely(sk->sk_rxhash != skb->hash)) 957 sk->sk_rxhash = skb->hash; 958 #endif 959 } 960 961 static inline void sock_rps_reset_rxhash(struct sock *sk) 962 { 963 #ifdef CONFIG_RPS 964 sk->sk_rxhash = 0; 965 #endif 966 } 967 968 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 969 ({ int __rc; \ 970 release_sock(__sk); \ 971 __rc = __condition; \ 972 if (!__rc) { \ 973 *(__timeo) = wait_woken(__wait, \ 974 TASK_INTERRUPTIBLE, \ 975 *(__timeo)); \ 976 } \ 977 sched_annotate_sleep(); \ 978 lock_sock(__sk); \ 979 __rc = __condition; \ 980 __rc; \ 981 }) 982 983 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 984 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 985 void sk_stream_wait_close(struct sock *sk, long timeo_p); 986 int sk_stream_error(struct sock *sk, int flags, int err); 987 void sk_stream_kill_queues(struct sock *sk); 988 void sk_set_memalloc(struct sock *sk); 989 void sk_clear_memalloc(struct sock *sk); 990 991 void __sk_flush_backlog(struct sock *sk); 992 993 static inline bool sk_flush_backlog(struct sock *sk) 994 { 995 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 996 __sk_flush_backlog(sk); 997 return true; 998 } 999 return false; 1000 } 1001 1002 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1003 1004 struct request_sock_ops; 1005 struct timewait_sock_ops; 1006 struct inet_hashinfo; 1007 struct raw_hashinfo; 1008 struct smc_hashinfo; 1009 struct module; 1010 1011 /* 1012 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1013 * un-modified. Special care is taken when initializing object to zero. 1014 */ 1015 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1016 { 1017 if (offsetof(struct sock, sk_node.next) != 0) 1018 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1019 memset(&sk->sk_node.pprev, 0, 1020 size - offsetof(struct sock, sk_node.pprev)); 1021 } 1022 1023 /* Networking protocol blocks we attach to sockets. 1024 * socket layer -> transport layer interface 1025 */ 1026 struct proto { 1027 void (*close)(struct sock *sk, 1028 long timeout); 1029 int (*connect)(struct sock *sk, 1030 struct sockaddr *uaddr, 1031 int addr_len); 1032 int (*disconnect)(struct sock *sk, int flags); 1033 1034 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1035 bool kern); 1036 1037 int (*ioctl)(struct sock *sk, int cmd, 1038 unsigned long arg); 1039 int (*init)(struct sock *sk); 1040 void (*destroy)(struct sock *sk); 1041 void (*shutdown)(struct sock *sk, int how); 1042 int (*setsockopt)(struct sock *sk, int level, 1043 int optname, char __user *optval, 1044 unsigned int optlen); 1045 int (*getsockopt)(struct sock *sk, int level, 1046 int optname, char __user *optval, 1047 int __user *option); 1048 void (*keepalive)(struct sock *sk, int valbool); 1049 #ifdef CONFIG_COMPAT 1050 int (*compat_setsockopt)(struct sock *sk, 1051 int level, 1052 int optname, char __user *optval, 1053 unsigned int optlen); 1054 int (*compat_getsockopt)(struct sock *sk, 1055 int level, 1056 int optname, char __user *optval, 1057 int __user *option); 1058 int (*compat_ioctl)(struct sock *sk, 1059 unsigned int cmd, unsigned long arg); 1060 #endif 1061 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1062 size_t len); 1063 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1064 size_t len, int noblock, int flags, 1065 int *addr_len); 1066 int (*sendpage)(struct sock *sk, struct page *page, 1067 int offset, size_t size, int flags); 1068 int (*bind)(struct sock *sk, 1069 struct sockaddr *uaddr, int addr_len); 1070 1071 int (*backlog_rcv) (struct sock *sk, 1072 struct sk_buff *skb); 1073 1074 void (*release_cb)(struct sock *sk); 1075 1076 /* Keeping track of sk's, looking them up, and port selection methods. */ 1077 int (*hash)(struct sock *sk); 1078 void (*unhash)(struct sock *sk); 1079 void (*rehash)(struct sock *sk); 1080 int (*get_port)(struct sock *sk, unsigned short snum); 1081 1082 /* Keeping track of sockets in use */ 1083 #ifdef CONFIG_PROC_FS 1084 unsigned int inuse_idx; 1085 #endif 1086 1087 bool (*stream_memory_free)(const struct sock *sk); 1088 /* Memory pressure */ 1089 void (*enter_memory_pressure)(struct sock *sk); 1090 void (*leave_memory_pressure)(struct sock *sk); 1091 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1092 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1093 /* 1094 * Pressure flag: try to collapse. 1095 * Technical note: it is used by multiple contexts non atomically. 1096 * All the __sk_mem_schedule() is of this nature: accounting 1097 * is strict, actions are advisory and have some latency. 1098 */ 1099 unsigned long *memory_pressure; 1100 long *sysctl_mem; 1101 int *sysctl_wmem; 1102 int *sysctl_rmem; 1103 int max_header; 1104 bool no_autobind; 1105 1106 struct kmem_cache *slab; 1107 unsigned int obj_size; 1108 int slab_flags; 1109 1110 struct percpu_counter *orphan_count; 1111 1112 struct request_sock_ops *rsk_prot; 1113 struct timewait_sock_ops *twsk_prot; 1114 1115 union { 1116 struct inet_hashinfo *hashinfo; 1117 struct udp_table *udp_table; 1118 struct raw_hashinfo *raw_hash; 1119 struct smc_hashinfo *smc_hash; 1120 } h; 1121 1122 struct module *owner; 1123 1124 char name[32]; 1125 1126 struct list_head node; 1127 #ifdef SOCK_REFCNT_DEBUG 1128 atomic_t socks; 1129 #endif 1130 int (*diag_destroy)(struct sock *sk, int err); 1131 } __randomize_layout; 1132 1133 int proto_register(struct proto *prot, int alloc_slab); 1134 void proto_unregister(struct proto *prot); 1135 1136 #ifdef SOCK_REFCNT_DEBUG 1137 static inline void sk_refcnt_debug_inc(struct sock *sk) 1138 { 1139 atomic_inc(&sk->sk_prot->socks); 1140 } 1141 1142 static inline void sk_refcnt_debug_dec(struct sock *sk) 1143 { 1144 atomic_dec(&sk->sk_prot->socks); 1145 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1146 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1147 } 1148 1149 static inline void sk_refcnt_debug_release(const struct sock *sk) 1150 { 1151 if (refcount_read(&sk->sk_refcnt) != 1) 1152 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1153 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1154 } 1155 #else /* SOCK_REFCNT_DEBUG */ 1156 #define sk_refcnt_debug_inc(sk) do { } while (0) 1157 #define sk_refcnt_debug_dec(sk) do { } while (0) 1158 #define sk_refcnt_debug_release(sk) do { } while (0) 1159 #endif /* SOCK_REFCNT_DEBUG */ 1160 1161 static inline bool sk_stream_memory_free(const struct sock *sk) 1162 { 1163 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1164 return false; 1165 1166 return sk->sk_prot->stream_memory_free ? 1167 sk->sk_prot->stream_memory_free(sk) : true; 1168 } 1169 1170 static inline bool sk_stream_is_writeable(const struct sock *sk) 1171 { 1172 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1173 sk_stream_memory_free(sk); 1174 } 1175 1176 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1177 struct cgroup *ancestor) 1178 { 1179 #ifdef CONFIG_SOCK_CGROUP_DATA 1180 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1181 ancestor); 1182 #else 1183 return -ENOTSUPP; 1184 #endif 1185 } 1186 1187 static inline bool sk_has_memory_pressure(const struct sock *sk) 1188 { 1189 return sk->sk_prot->memory_pressure != NULL; 1190 } 1191 1192 static inline bool sk_under_memory_pressure(const struct sock *sk) 1193 { 1194 if (!sk->sk_prot->memory_pressure) 1195 return false; 1196 1197 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1198 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1199 return true; 1200 1201 return !!*sk->sk_prot->memory_pressure; 1202 } 1203 1204 static inline long 1205 sk_memory_allocated(const struct sock *sk) 1206 { 1207 return atomic_long_read(sk->sk_prot->memory_allocated); 1208 } 1209 1210 static inline long 1211 sk_memory_allocated_add(struct sock *sk, int amt) 1212 { 1213 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1214 } 1215 1216 static inline void 1217 sk_memory_allocated_sub(struct sock *sk, int amt) 1218 { 1219 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1220 } 1221 1222 static inline void sk_sockets_allocated_dec(struct sock *sk) 1223 { 1224 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1225 } 1226 1227 static inline void sk_sockets_allocated_inc(struct sock *sk) 1228 { 1229 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1230 } 1231 1232 static inline int 1233 sk_sockets_allocated_read_positive(struct sock *sk) 1234 { 1235 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1236 } 1237 1238 static inline int 1239 proto_sockets_allocated_sum_positive(struct proto *prot) 1240 { 1241 return percpu_counter_sum_positive(prot->sockets_allocated); 1242 } 1243 1244 static inline long 1245 proto_memory_allocated(struct proto *prot) 1246 { 1247 return atomic_long_read(prot->memory_allocated); 1248 } 1249 1250 static inline bool 1251 proto_memory_pressure(struct proto *prot) 1252 { 1253 if (!prot->memory_pressure) 1254 return false; 1255 return !!*prot->memory_pressure; 1256 } 1257 1258 1259 #ifdef CONFIG_PROC_FS 1260 /* Called with local bh disabled */ 1261 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1262 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1263 #else 1264 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1265 int inc) 1266 { 1267 } 1268 #endif 1269 1270 1271 /* With per-bucket locks this operation is not-atomic, so that 1272 * this version is not worse. 1273 */ 1274 static inline int __sk_prot_rehash(struct sock *sk) 1275 { 1276 sk->sk_prot->unhash(sk); 1277 return sk->sk_prot->hash(sk); 1278 } 1279 1280 /* About 10 seconds */ 1281 #define SOCK_DESTROY_TIME (10*HZ) 1282 1283 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1284 #define PROT_SOCK 1024 1285 1286 #define SHUTDOWN_MASK 3 1287 #define RCV_SHUTDOWN 1 1288 #define SEND_SHUTDOWN 2 1289 1290 #define SOCK_SNDBUF_LOCK 1 1291 #define SOCK_RCVBUF_LOCK 2 1292 #define SOCK_BINDADDR_LOCK 4 1293 #define SOCK_BINDPORT_LOCK 8 1294 1295 struct socket_alloc { 1296 struct socket socket; 1297 struct inode vfs_inode; 1298 }; 1299 1300 static inline struct socket *SOCKET_I(struct inode *inode) 1301 { 1302 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1303 } 1304 1305 static inline struct inode *SOCK_INODE(struct socket *socket) 1306 { 1307 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1308 } 1309 1310 /* 1311 * Functions for memory accounting 1312 */ 1313 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1314 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1315 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1316 void __sk_mem_reclaim(struct sock *sk, int amount); 1317 1318 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1319 * do not necessarily have 16x time more memory than 4KB ones. 1320 */ 1321 #define SK_MEM_QUANTUM 4096 1322 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1323 #define SK_MEM_SEND 0 1324 #define SK_MEM_RECV 1 1325 1326 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1327 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1328 { 1329 long val = sk->sk_prot->sysctl_mem[index]; 1330 1331 #if PAGE_SIZE > SK_MEM_QUANTUM 1332 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1333 #elif PAGE_SIZE < SK_MEM_QUANTUM 1334 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1335 #endif 1336 return val; 1337 } 1338 1339 static inline int sk_mem_pages(int amt) 1340 { 1341 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1342 } 1343 1344 static inline bool sk_has_account(struct sock *sk) 1345 { 1346 /* return true if protocol supports memory accounting */ 1347 return !!sk->sk_prot->memory_allocated; 1348 } 1349 1350 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1351 { 1352 if (!sk_has_account(sk)) 1353 return true; 1354 return size <= sk->sk_forward_alloc || 1355 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1356 } 1357 1358 static inline bool 1359 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1360 { 1361 if (!sk_has_account(sk)) 1362 return true; 1363 return size<= sk->sk_forward_alloc || 1364 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1365 skb_pfmemalloc(skb); 1366 } 1367 1368 static inline void sk_mem_reclaim(struct sock *sk) 1369 { 1370 if (!sk_has_account(sk)) 1371 return; 1372 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1373 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1374 } 1375 1376 static inline void sk_mem_reclaim_partial(struct sock *sk) 1377 { 1378 if (!sk_has_account(sk)) 1379 return; 1380 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1381 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1382 } 1383 1384 static inline void sk_mem_charge(struct sock *sk, int size) 1385 { 1386 if (!sk_has_account(sk)) 1387 return; 1388 sk->sk_forward_alloc -= size; 1389 } 1390 1391 static inline void sk_mem_uncharge(struct sock *sk, int size) 1392 { 1393 if (!sk_has_account(sk)) 1394 return; 1395 sk->sk_forward_alloc += size; 1396 1397 /* Avoid a possible overflow. 1398 * TCP send queues can make this happen, if sk_mem_reclaim() 1399 * is not called and more than 2 GBytes are released at once. 1400 * 1401 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1402 * no need to hold that much forward allocation anyway. 1403 */ 1404 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1405 __sk_mem_reclaim(sk, 1 << 20); 1406 } 1407 1408 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1409 { 1410 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1411 sk->sk_wmem_queued -= skb->truesize; 1412 sk_mem_uncharge(sk, skb->truesize); 1413 __kfree_skb(skb); 1414 } 1415 1416 static inline void sock_release_ownership(struct sock *sk) 1417 { 1418 if (sk->sk_lock.owned) { 1419 sk->sk_lock.owned = 0; 1420 1421 /* The sk_lock has mutex_unlock() semantics: */ 1422 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1423 } 1424 } 1425 1426 /* 1427 * Macro so as to not evaluate some arguments when 1428 * lockdep is not enabled. 1429 * 1430 * Mark both the sk_lock and the sk_lock.slock as a 1431 * per-address-family lock class. 1432 */ 1433 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1434 do { \ 1435 sk->sk_lock.owned = 0; \ 1436 init_waitqueue_head(&sk->sk_lock.wq); \ 1437 spin_lock_init(&(sk)->sk_lock.slock); \ 1438 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1439 sizeof((sk)->sk_lock)); \ 1440 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1441 (skey), (sname)); \ 1442 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1443 } while (0) 1444 1445 #ifdef CONFIG_LOCKDEP 1446 static inline bool lockdep_sock_is_held(const struct sock *csk) 1447 { 1448 struct sock *sk = (struct sock *)csk; 1449 1450 return lockdep_is_held(&sk->sk_lock) || 1451 lockdep_is_held(&sk->sk_lock.slock); 1452 } 1453 #endif 1454 1455 void lock_sock_nested(struct sock *sk, int subclass); 1456 1457 static inline void lock_sock(struct sock *sk) 1458 { 1459 lock_sock_nested(sk, 0); 1460 } 1461 1462 void release_sock(struct sock *sk); 1463 1464 /* BH context may only use the following locking interface. */ 1465 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1466 #define bh_lock_sock_nested(__sk) \ 1467 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1468 SINGLE_DEPTH_NESTING) 1469 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1470 1471 bool lock_sock_fast(struct sock *sk); 1472 /** 1473 * unlock_sock_fast - complement of lock_sock_fast 1474 * @sk: socket 1475 * @slow: slow mode 1476 * 1477 * fast unlock socket for user context. 1478 * If slow mode is on, we call regular release_sock() 1479 */ 1480 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1481 { 1482 if (slow) 1483 release_sock(sk); 1484 else 1485 spin_unlock_bh(&sk->sk_lock.slock); 1486 } 1487 1488 /* Used by processes to "lock" a socket state, so that 1489 * interrupts and bottom half handlers won't change it 1490 * from under us. It essentially blocks any incoming 1491 * packets, so that we won't get any new data or any 1492 * packets that change the state of the socket. 1493 * 1494 * While locked, BH processing will add new packets to 1495 * the backlog queue. This queue is processed by the 1496 * owner of the socket lock right before it is released. 1497 * 1498 * Since ~2.3.5 it is also exclusive sleep lock serializing 1499 * accesses from user process context. 1500 */ 1501 1502 static inline void sock_owned_by_me(const struct sock *sk) 1503 { 1504 #ifdef CONFIG_LOCKDEP 1505 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1506 #endif 1507 } 1508 1509 static inline bool sock_owned_by_user(const struct sock *sk) 1510 { 1511 sock_owned_by_me(sk); 1512 return sk->sk_lock.owned; 1513 } 1514 1515 /* no reclassification while locks are held */ 1516 static inline bool sock_allow_reclassification(const struct sock *csk) 1517 { 1518 struct sock *sk = (struct sock *)csk; 1519 1520 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1521 } 1522 1523 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1524 struct proto *prot, int kern); 1525 void sk_free(struct sock *sk); 1526 void sk_destruct(struct sock *sk); 1527 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1528 void sk_free_unlock_clone(struct sock *sk); 1529 1530 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1531 gfp_t priority); 1532 void __sock_wfree(struct sk_buff *skb); 1533 void sock_wfree(struct sk_buff *skb); 1534 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1535 gfp_t priority); 1536 void skb_orphan_partial(struct sk_buff *skb); 1537 void sock_rfree(struct sk_buff *skb); 1538 void sock_efree(struct sk_buff *skb); 1539 #ifdef CONFIG_INET 1540 void sock_edemux(struct sk_buff *skb); 1541 #else 1542 #define sock_edemux sock_efree 1543 #endif 1544 1545 int sock_setsockopt(struct socket *sock, int level, int op, 1546 char __user *optval, unsigned int optlen); 1547 1548 int sock_getsockopt(struct socket *sock, int level, int op, 1549 char __user *optval, int __user *optlen); 1550 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1551 int noblock, int *errcode); 1552 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1553 unsigned long data_len, int noblock, 1554 int *errcode, int max_page_order); 1555 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1556 void sock_kfree_s(struct sock *sk, void *mem, int size); 1557 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1558 void sk_send_sigurg(struct sock *sk); 1559 1560 struct sockcm_cookie { 1561 u32 mark; 1562 u16 tsflags; 1563 }; 1564 1565 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1566 struct sockcm_cookie *sockc); 1567 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1568 struct sockcm_cookie *sockc); 1569 1570 /* 1571 * Functions to fill in entries in struct proto_ops when a protocol 1572 * does not implement a particular function. 1573 */ 1574 int sock_no_bind(struct socket *, struct sockaddr *, int); 1575 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1576 int sock_no_socketpair(struct socket *, struct socket *); 1577 int sock_no_accept(struct socket *, struct socket *, int, bool); 1578 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1579 unsigned int sock_no_poll(struct file *, struct socket *, 1580 struct poll_table_struct *); 1581 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1582 int sock_no_listen(struct socket *, int); 1583 int sock_no_shutdown(struct socket *, int); 1584 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1585 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1586 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1587 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1588 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1589 int sock_no_mmap(struct file *file, struct socket *sock, 1590 struct vm_area_struct *vma); 1591 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1592 size_t size, int flags); 1593 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1594 int offset, size_t size, int flags); 1595 1596 /* 1597 * Functions to fill in entries in struct proto_ops when a protocol 1598 * uses the inet style. 1599 */ 1600 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1601 char __user *optval, int __user *optlen); 1602 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1603 int flags); 1604 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1605 char __user *optval, unsigned int optlen); 1606 int compat_sock_common_getsockopt(struct socket *sock, int level, 1607 int optname, char __user *optval, int __user *optlen); 1608 int compat_sock_common_setsockopt(struct socket *sock, int level, 1609 int optname, char __user *optval, unsigned int optlen); 1610 1611 void sk_common_release(struct sock *sk); 1612 1613 /* 1614 * Default socket callbacks and setup code 1615 */ 1616 1617 /* Initialise core socket variables */ 1618 void sock_init_data(struct socket *sock, struct sock *sk); 1619 1620 /* 1621 * Socket reference counting postulates. 1622 * 1623 * * Each user of socket SHOULD hold a reference count. 1624 * * Each access point to socket (an hash table bucket, reference from a list, 1625 * running timer, skb in flight MUST hold a reference count. 1626 * * When reference count hits 0, it means it will never increase back. 1627 * * When reference count hits 0, it means that no references from 1628 * outside exist to this socket and current process on current CPU 1629 * is last user and may/should destroy this socket. 1630 * * sk_free is called from any context: process, BH, IRQ. When 1631 * it is called, socket has no references from outside -> sk_free 1632 * may release descendant resources allocated by the socket, but 1633 * to the time when it is called, socket is NOT referenced by any 1634 * hash tables, lists etc. 1635 * * Packets, delivered from outside (from network or from another process) 1636 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1637 * when they sit in queue. Otherwise, packets will leak to hole, when 1638 * socket is looked up by one cpu and unhasing is made by another CPU. 1639 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1640 * (leak to backlog). Packet socket does all the processing inside 1641 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1642 * use separate SMP lock, so that they are prone too. 1643 */ 1644 1645 /* Ungrab socket and destroy it, if it was the last reference. */ 1646 static inline void sock_put(struct sock *sk) 1647 { 1648 if (refcount_dec_and_test(&sk->sk_refcnt)) 1649 sk_free(sk); 1650 } 1651 /* Generic version of sock_put(), dealing with all sockets 1652 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1653 */ 1654 void sock_gen_put(struct sock *sk); 1655 1656 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1657 unsigned int trim_cap, bool refcounted); 1658 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1659 const int nested) 1660 { 1661 return __sk_receive_skb(sk, skb, nested, 1, true); 1662 } 1663 1664 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1665 { 1666 sk->sk_tx_queue_mapping = tx_queue; 1667 } 1668 1669 static inline void sk_tx_queue_clear(struct sock *sk) 1670 { 1671 sk->sk_tx_queue_mapping = -1; 1672 } 1673 1674 static inline int sk_tx_queue_get(const struct sock *sk) 1675 { 1676 return sk ? sk->sk_tx_queue_mapping : -1; 1677 } 1678 1679 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1680 { 1681 sk_tx_queue_clear(sk); 1682 sk->sk_socket = sock; 1683 } 1684 1685 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1686 { 1687 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1688 return &rcu_dereference_raw(sk->sk_wq)->wait; 1689 } 1690 /* Detach socket from process context. 1691 * Announce socket dead, detach it from wait queue and inode. 1692 * Note that parent inode held reference count on this struct sock, 1693 * we do not release it in this function, because protocol 1694 * probably wants some additional cleanups or even continuing 1695 * to work with this socket (TCP). 1696 */ 1697 static inline void sock_orphan(struct sock *sk) 1698 { 1699 write_lock_bh(&sk->sk_callback_lock); 1700 sock_set_flag(sk, SOCK_DEAD); 1701 sk_set_socket(sk, NULL); 1702 sk->sk_wq = NULL; 1703 write_unlock_bh(&sk->sk_callback_lock); 1704 } 1705 1706 static inline void sock_graft(struct sock *sk, struct socket *parent) 1707 { 1708 WARN_ON(parent->sk); 1709 write_lock_bh(&sk->sk_callback_lock); 1710 sk->sk_wq = parent->wq; 1711 parent->sk = sk; 1712 sk_set_socket(sk, parent); 1713 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1714 security_sock_graft(sk, parent); 1715 write_unlock_bh(&sk->sk_callback_lock); 1716 } 1717 1718 kuid_t sock_i_uid(struct sock *sk); 1719 unsigned long sock_i_ino(struct sock *sk); 1720 1721 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1722 { 1723 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1724 } 1725 1726 static inline u32 net_tx_rndhash(void) 1727 { 1728 u32 v = prandom_u32(); 1729 1730 return v ?: 1; 1731 } 1732 1733 static inline void sk_set_txhash(struct sock *sk) 1734 { 1735 sk->sk_txhash = net_tx_rndhash(); 1736 } 1737 1738 static inline void sk_rethink_txhash(struct sock *sk) 1739 { 1740 if (sk->sk_txhash) 1741 sk_set_txhash(sk); 1742 } 1743 1744 static inline struct dst_entry * 1745 __sk_dst_get(struct sock *sk) 1746 { 1747 return rcu_dereference_check(sk->sk_dst_cache, 1748 lockdep_sock_is_held(sk)); 1749 } 1750 1751 static inline struct dst_entry * 1752 sk_dst_get(struct sock *sk) 1753 { 1754 struct dst_entry *dst; 1755 1756 rcu_read_lock(); 1757 dst = rcu_dereference(sk->sk_dst_cache); 1758 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1759 dst = NULL; 1760 rcu_read_unlock(); 1761 return dst; 1762 } 1763 1764 static inline void dst_negative_advice(struct sock *sk) 1765 { 1766 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1767 1768 sk_rethink_txhash(sk); 1769 1770 if (dst && dst->ops->negative_advice) { 1771 ndst = dst->ops->negative_advice(dst); 1772 1773 if (ndst != dst) { 1774 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1775 sk_tx_queue_clear(sk); 1776 sk->sk_dst_pending_confirm = 0; 1777 } 1778 } 1779 } 1780 1781 static inline void 1782 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1783 { 1784 struct dst_entry *old_dst; 1785 1786 sk_tx_queue_clear(sk); 1787 sk->sk_dst_pending_confirm = 0; 1788 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1789 lockdep_sock_is_held(sk)); 1790 rcu_assign_pointer(sk->sk_dst_cache, dst); 1791 dst_release(old_dst); 1792 } 1793 1794 static inline void 1795 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1796 { 1797 struct dst_entry *old_dst; 1798 1799 sk_tx_queue_clear(sk); 1800 sk->sk_dst_pending_confirm = 0; 1801 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1802 dst_release(old_dst); 1803 } 1804 1805 static inline void 1806 __sk_dst_reset(struct sock *sk) 1807 { 1808 __sk_dst_set(sk, NULL); 1809 } 1810 1811 static inline void 1812 sk_dst_reset(struct sock *sk) 1813 { 1814 sk_dst_set(sk, NULL); 1815 } 1816 1817 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1818 1819 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1820 1821 static inline void sk_dst_confirm(struct sock *sk) 1822 { 1823 if (!sk->sk_dst_pending_confirm) 1824 sk->sk_dst_pending_confirm = 1; 1825 } 1826 1827 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1828 { 1829 if (skb_get_dst_pending_confirm(skb)) { 1830 struct sock *sk = skb->sk; 1831 unsigned long now = jiffies; 1832 1833 /* avoid dirtying neighbour */ 1834 if (n->confirmed != now) 1835 n->confirmed = now; 1836 if (sk && sk->sk_dst_pending_confirm) 1837 sk->sk_dst_pending_confirm = 0; 1838 } 1839 } 1840 1841 bool sk_mc_loop(struct sock *sk); 1842 1843 static inline bool sk_can_gso(const struct sock *sk) 1844 { 1845 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1846 } 1847 1848 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1849 1850 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1851 { 1852 sk->sk_route_nocaps |= flags; 1853 sk->sk_route_caps &= ~flags; 1854 } 1855 1856 static inline bool sk_check_csum_caps(struct sock *sk) 1857 { 1858 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1859 (sk->sk_family == PF_INET && 1860 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1861 (sk->sk_family == PF_INET6 && 1862 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1863 } 1864 1865 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1866 struct iov_iter *from, char *to, 1867 int copy, int offset) 1868 { 1869 if (skb->ip_summed == CHECKSUM_NONE) { 1870 __wsum csum = 0; 1871 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1872 return -EFAULT; 1873 skb->csum = csum_block_add(skb->csum, csum, offset); 1874 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1875 if (!copy_from_iter_full_nocache(to, copy, from)) 1876 return -EFAULT; 1877 } else if (!copy_from_iter_full(to, copy, from)) 1878 return -EFAULT; 1879 1880 return 0; 1881 } 1882 1883 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1884 struct iov_iter *from, int copy) 1885 { 1886 int err, offset = skb->len; 1887 1888 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1889 copy, offset); 1890 if (err) 1891 __skb_trim(skb, offset); 1892 1893 return err; 1894 } 1895 1896 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1897 struct sk_buff *skb, 1898 struct page *page, 1899 int off, int copy) 1900 { 1901 int err; 1902 1903 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1904 copy, skb->len); 1905 if (err) 1906 return err; 1907 1908 skb->len += copy; 1909 skb->data_len += copy; 1910 skb->truesize += copy; 1911 sk->sk_wmem_queued += copy; 1912 sk_mem_charge(sk, copy); 1913 return 0; 1914 } 1915 1916 /** 1917 * sk_wmem_alloc_get - returns write allocations 1918 * @sk: socket 1919 * 1920 * Returns sk_wmem_alloc minus initial offset of one 1921 */ 1922 static inline int sk_wmem_alloc_get(const struct sock *sk) 1923 { 1924 return refcount_read(&sk->sk_wmem_alloc) - 1; 1925 } 1926 1927 /** 1928 * sk_rmem_alloc_get - returns read allocations 1929 * @sk: socket 1930 * 1931 * Returns sk_rmem_alloc 1932 */ 1933 static inline int sk_rmem_alloc_get(const struct sock *sk) 1934 { 1935 return atomic_read(&sk->sk_rmem_alloc); 1936 } 1937 1938 /** 1939 * sk_has_allocations - check if allocations are outstanding 1940 * @sk: socket 1941 * 1942 * Returns true if socket has write or read allocations 1943 */ 1944 static inline bool sk_has_allocations(const struct sock *sk) 1945 { 1946 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1947 } 1948 1949 /** 1950 * skwq_has_sleeper - check if there are any waiting processes 1951 * @wq: struct socket_wq 1952 * 1953 * Returns true if socket_wq has waiting processes 1954 * 1955 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1956 * barrier call. They were added due to the race found within the tcp code. 1957 * 1958 * Consider following tcp code paths:: 1959 * 1960 * CPU1 CPU2 1961 * sys_select receive packet 1962 * ... ... 1963 * __add_wait_queue update tp->rcv_nxt 1964 * ... ... 1965 * tp->rcv_nxt check sock_def_readable 1966 * ... { 1967 * schedule rcu_read_lock(); 1968 * wq = rcu_dereference(sk->sk_wq); 1969 * if (wq && waitqueue_active(&wq->wait)) 1970 * wake_up_interruptible(&wq->wait) 1971 * ... 1972 * } 1973 * 1974 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1975 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1976 * could then endup calling schedule and sleep forever if there are no more 1977 * data on the socket. 1978 * 1979 */ 1980 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1981 { 1982 return wq && wq_has_sleeper(&wq->wait); 1983 } 1984 1985 /** 1986 * sock_poll_wait - place memory barrier behind the poll_wait call. 1987 * @filp: file 1988 * @wait_address: socket wait queue 1989 * @p: poll_table 1990 * 1991 * See the comments in the wq_has_sleeper function. 1992 */ 1993 static inline void sock_poll_wait(struct file *filp, 1994 wait_queue_head_t *wait_address, poll_table *p) 1995 { 1996 if (!poll_does_not_wait(p) && wait_address) { 1997 poll_wait(filp, wait_address, p); 1998 /* We need to be sure we are in sync with the 1999 * socket flags modification. 2000 * 2001 * This memory barrier is paired in the wq_has_sleeper. 2002 */ 2003 smp_mb(); 2004 } 2005 } 2006 2007 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2008 { 2009 if (sk->sk_txhash) { 2010 skb->l4_hash = 1; 2011 skb->hash = sk->sk_txhash; 2012 } 2013 } 2014 2015 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2016 2017 /* 2018 * Queue a received datagram if it will fit. Stream and sequenced 2019 * protocols can't normally use this as they need to fit buffers in 2020 * and play with them. 2021 * 2022 * Inlined as it's very short and called for pretty much every 2023 * packet ever received. 2024 */ 2025 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2026 { 2027 skb_orphan(skb); 2028 skb->sk = sk; 2029 skb->destructor = sock_rfree; 2030 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2031 sk_mem_charge(sk, skb->truesize); 2032 } 2033 2034 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2035 unsigned long expires); 2036 2037 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2038 2039 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2040 struct sk_buff *skb, unsigned int flags, 2041 void (*destructor)(struct sock *sk, 2042 struct sk_buff *skb)); 2043 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2044 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2045 2046 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2047 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2048 2049 /* 2050 * Recover an error report and clear atomically 2051 */ 2052 2053 static inline int sock_error(struct sock *sk) 2054 { 2055 int err; 2056 if (likely(!sk->sk_err)) 2057 return 0; 2058 err = xchg(&sk->sk_err, 0); 2059 return -err; 2060 } 2061 2062 static inline unsigned long sock_wspace(struct sock *sk) 2063 { 2064 int amt = 0; 2065 2066 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2067 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2068 if (amt < 0) 2069 amt = 0; 2070 } 2071 return amt; 2072 } 2073 2074 /* Note: 2075 * We use sk->sk_wq_raw, from contexts knowing this 2076 * pointer is not NULL and cannot disappear/change. 2077 */ 2078 static inline void sk_set_bit(int nr, struct sock *sk) 2079 { 2080 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2081 !sock_flag(sk, SOCK_FASYNC)) 2082 return; 2083 2084 set_bit(nr, &sk->sk_wq_raw->flags); 2085 } 2086 2087 static inline void sk_clear_bit(int nr, struct sock *sk) 2088 { 2089 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2090 !sock_flag(sk, SOCK_FASYNC)) 2091 return; 2092 2093 clear_bit(nr, &sk->sk_wq_raw->flags); 2094 } 2095 2096 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2097 { 2098 if (sock_flag(sk, SOCK_FASYNC)) { 2099 rcu_read_lock(); 2100 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2101 rcu_read_unlock(); 2102 } 2103 } 2104 2105 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2106 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2107 * Note: for send buffers, TCP works better if we can build two skbs at 2108 * minimum. 2109 */ 2110 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2111 2112 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2113 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2114 2115 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2116 { 2117 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2118 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2119 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2120 } 2121 } 2122 2123 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2124 bool force_schedule); 2125 2126 /** 2127 * sk_page_frag - return an appropriate page_frag 2128 * @sk: socket 2129 * 2130 * If socket allocation mode allows current thread to sleep, it means its 2131 * safe to use the per task page_frag instead of the per socket one. 2132 */ 2133 static inline struct page_frag *sk_page_frag(struct sock *sk) 2134 { 2135 if (gfpflags_allow_blocking(sk->sk_allocation)) 2136 return ¤t->task_frag; 2137 2138 return &sk->sk_frag; 2139 } 2140 2141 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2142 2143 /* 2144 * Default write policy as shown to user space via poll/select/SIGIO 2145 */ 2146 static inline bool sock_writeable(const struct sock *sk) 2147 { 2148 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2149 } 2150 2151 static inline gfp_t gfp_any(void) 2152 { 2153 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2154 } 2155 2156 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2157 { 2158 return noblock ? 0 : sk->sk_rcvtimeo; 2159 } 2160 2161 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2162 { 2163 return noblock ? 0 : sk->sk_sndtimeo; 2164 } 2165 2166 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2167 { 2168 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2169 } 2170 2171 /* Alas, with timeout socket operations are not restartable. 2172 * Compare this to poll(). 2173 */ 2174 static inline int sock_intr_errno(long timeo) 2175 { 2176 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2177 } 2178 2179 struct sock_skb_cb { 2180 u32 dropcount; 2181 }; 2182 2183 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2184 * using skb->cb[] would keep using it directly and utilize its 2185 * alignement guarantee. 2186 */ 2187 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2188 sizeof(struct sock_skb_cb))) 2189 2190 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2191 SOCK_SKB_CB_OFFSET)) 2192 2193 #define sock_skb_cb_check_size(size) \ 2194 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2195 2196 static inline void 2197 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2198 { 2199 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2200 atomic_read(&sk->sk_drops) : 0; 2201 } 2202 2203 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2204 { 2205 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2206 2207 atomic_add(segs, &sk->sk_drops); 2208 } 2209 2210 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2211 struct sk_buff *skb); 2212 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2213 struct sk_buff *skb); 2214 2215 static inline void 2216 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2217 { 2218 ktime_t kt = skb->tstamp; 2219 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2220 2221 /* 2222 * generate control messages if 2223 * - receive time stamping in software requested 2224 * - software time stamp available and wanted 2225 * - hardware time stamps available and wanted 2226 */ 2227 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2228 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2229 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2230 (hwtstamps->hwtstamp && 2231 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2232 __sock_recv_timestamp(msg, sk, skb); 2233 else 2234 sk->sk_stamp = kt; 2235 2236 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2237 __sock_recv_wifi_status(msg, sk, skb); 2238 } 2239 2240 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2241 struct sk_buff *skb); 2242 2243 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2244 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2245 struct sk_buff *skb) 2246 { 2247 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2248 (1UL << SOCK_RCVTSTAMP)) 2249 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2250 SOF_TIMESTAMPING_RAW_HARDWARE) 2251 2252 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2253 __sock_recv_ts_and_drops(msg, sk, skb); 2254 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2255 sk->sk_stamp = skb->tstamp; 2256 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2257 sk->sk_stamp = 0; 2258 } 2259 2260 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2261 2262 /** 2263 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2264 * @sk: socket sending this packet 2265 * @tsflags: timestamping flags to use 2266 * @tx_flags: completed with instructions for time stamping 2267 * 2268 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2269 */ 2270 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2271 __u8 *tx_flags) 2272 { 2273 if (unlikely(tsflags)) 2274 __sock_tx_timestamp(tsflags, tx_flags); 2275 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2276 *tx_flags |= SKBTX_WIFI_STATUS; 2277 } 2278 2279 /** 2280 * sk_eat_skb - Release a skb if it is no longer needed 2281 * @sk: socket to eat this skb from 2282 * @skb: socket buffer to eat 2283 * 2284 * This routine must be called with interrupts disabled or with the socket 2285 * locked so that the sk_buff queue operation is ok. 2286 */ 2287 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2288 { 2289 __skb_unlink(skb, &sk->sk_receive_queue); 2290 __kfree_skb(skb); 2291 } 2292 2293 static inline 2294 struct net *sock_net(const struct sock *sk) 2295 { 2296 return read_pnet(&sk->sk_net); 2297 } 2298 2299 static inline 2300 void sock_net_set(struct sock *sk, struct net *net) 2301 { 2302 write_pnet(&sk->sk_net, net); 2303 } 2304 2305 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2306 { 2307 if (skb->sk) { 2308 struct sock *sk = skb->sk; 2309 2310 skb->destructor = NULL; 2311 skb->sk = NULL; 2312 return sk; 2313 } 2314 return NULL; 2315 } 2316 2317 /* This helper checks if a socket is a full socket, 2318 * ie _not_ a timewait or request socket. 2319 */ 2320 static inline bool sk_fullsock(const struct sock *sk) 2321 { 2322 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2323 } 2324 2325 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2326 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2327 */ 2328 static inline bool sk_listener(const struct sock *sk) 2329 { 2330 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2331 } 2332 2333 /** 2334 * sk_state_load - read sk->sk_state for lockless contexts 2335 * @sk: socket pointer 2336 * 2337 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2338 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2339 */ 2340 static inline int sk_state_load(const struct sock *sk) 2341 { 2342 return smp_load_acquire(&sk->sk_state); 2343 } 2344 2345 /** 2346 * sk_state_store - update sk->sk_state 2347 * @sk: socket pointer 2348 * @newstate: new state 2349 * 2350 * Paired with sk_state_load(). Should be used in contexts where 2351 * state change might impact lockless readers. 2352 */ 2353 static inline void sk_state_store(struct sock *sk, int newstate) 2354 { 2355 smp_store_release(&sk->sk_state, newstate); 2356 } 2357 2358 void sock_enable_timestamp(struct sock *sk, int flag); 2359 int sock_get_timestamp(struct sock *, struct timeval __user *); 2360 int sock_get_timestampns(struct sock *, struct timespec __user *); 2361 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2362 int type); 2363 2364 bool sk_ns_capable(const struct sock *sk, 2365 struct user_namespace *user_ns, int cap); 2366 bool sk_capable(const struct sock *sk, int cap); 2367 bool sk_net_capable(const struct sock *sk, int cap); 2368 2369 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2370 2371 /* Take into consideration the size of the struct sk_buff overhead in the 2372 * determination of these values, since that is non-constant across 2373 * platforms. This makes socket queueing behavior and performance 2374 * not depend upon such differences. 2375 */ 2376 #define _SK_MEM_PACKETS 256 2377 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2378 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2379 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2380 2381 extern __u32 sysctl_wmem_max; 2382 extern __u32 sysctl_rmem_max; 2383 2384 extern int sysctl_tstamp_allow_data; 2385 extern int sysctl_optmem_max; 2386 2387 extern __u32 sysctl_wmem_default; 2388 extern __u32 sysctl_rmem_default; 2389 2390 #endif /* _SOCK_H */ 2391