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 #include <linux/rbtree.h> 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_pacing_shift: scaling factor for TCP Small Queues 271 * @sk_lingertime: %SO_LINGER l_linger setting 272 * @sk_backlog: always used with the per-socket spinlock held 273 * @sk_callback_lock: used with the callbacks in the end of this struct 274 * @sk_error_queue: rarely used 275 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 276 * IPV6_ADDRFORM for instance) 277 * @sk_err: last error 278 * @sk_err_soft: errors that don't cause failure but are the cause of a 279 * persistent failure not just 'timed out' 280 * @sk_drops: raw/udp drops counter 281 * @sk_ack_backlog: current listen backlog 282 * @sk_max_ack_backlog: listen backlog set in listen() 283 * @sk_uid: user id of owner 284 * @sk_priority: %SO_PRIORITY setting 285 * @sk_type: socket type (%SOCK_STREAM, etc) 286 * @sk_protocol: which protocol this socket belongs in this network family 287 * @sk_peer_pid: &struct pid for this socket's peer 288 * @sk_peer_cred: %SO_PEERCRED setting 289 * @sk_rcvlowat: %SO_RCVLOWAT setting 290 * @sk_rcvtimeo: %SO_RCVTIMEO setting 291 * @sk_sndtimeo: %SO_SNDTIMEO setting 292 * @sk_txhash: computed flow hash for use on transmit 293 * @sk_filter: socket filtering instructions 294 * @sk_timer: sock cleanup timer 295 * @sk_stamp: time stamp of last packet received 296 * @sk_tsflags: SO_TIMESTAMPING socket options 297 * @sk_tskey: counter to disambiguate concurrent tstamp requests 298 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 299 * @sk_socket: Identd and reporting IO signals 300 * @sk_user_data: RPC layer private data 301 * @sk_frag: cached page frag 302 * @sk_peek_off: current peek_offset value 303 * @sk_send_head: front of stuff to transmit 304 * @sk_security: used by security modules 305 * @sk_mark: generic packet mark 306 * @sk_cgrp_data: cgroup data for this cgroup 307 * @sk_memcg: this socket's memory cgroup association 308 * @sk_write_pending: a write to stream socket waits to start 309 * @sk_state_change: callback to indicate change in the state of the sock 310 * @sk_data_ready: callback to indicate there is data to be processed 311 * @sk_write_space: callback to indicate there is bf sending space available 312 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 313 * @sk_backlog_rcv: callback to process the backlog 314 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 315 * @sk_reuseport_cb: reuseport group container 316 * @sk_rcu: used during RCU grace period 317 */ 318 struct sock { 319 /* 320 * Now struct inet_timewait_sock also uses sock_common, so please just 321 * don't add nothing before this first member (__sk_common) --acme 322 */ 323 struct sock_common __sk_common; 324 #define sk_node __sk_common.skc_node 325 #define sk_nulls_node __sk_common.skc_nulls_node 326 #define sk_refcnt __sk_common.skc_refcnt 327 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 328 329 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 330 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 331 #define sk_hash __sk_common.skc_hash 332 #define sk_portpair __sk_common.skc_portpair 333 #define sk_num __sk_common.skc_num 334 #define sk_dport __sk_common.skc_dport 335 #define sk_addrpair __sk_common.skc_addrpair 336 #define sk_daddr __sk_common.skc_daddr 337 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 338 #define sk_family __sk_common.skc_family 339 #define sk_state __sk_common.skc_state 340 #define sk_reuse __sk_common.skc_reuse 341 #define sk_reuseport __sk_common.skc_reuseport 342 #define sk_ipv6only __sk_common.skc_ipv6only 343 #define sk_net_refcnt __sk_common.skc_net_refcnt 344 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 345 #define sk_bind_node __sk_common.skc_bind_node 346 #define sk_prot __sk_common.skc_prot 347 #define sk_net __sk_common.skc_net 348 #define sk_v6_daddr __sk_common.skc_v6_daddr 349 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 350 #define sk_cookie __sk_common.skc_cookie 351 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 352 #define sk_flags __sk_common.skc_flags 353 #define sk_rxhash __sk_common.skc_rxhash 354 355 socket_lock_t sk_lock; 356 atomic_t sk_drops; 357 int sk_rcvlowat; 358 struct sk_buff_head sk_error_queue; 359 struct sk_buff_head sk_receive_queue; 360 /* 361 * The backlog queue is special, it is always used with 362 * the per-socket spinlock held and requires low latency 363 * access. Therefore we special case it's implementation. 364 * Note : rmem_alloc is in this structure to fill a hole 365 * on 64bit arches, not because its logically part of 366 * backlog. 367 */ 368 struct { 369 atomic_t rmem_alloc; 370 int len; 371 struct sk_buff *head; 372 struct sk_buff *tail; 373 } sk_backlog; 374 #define sk_rmem_alloc sk_backlog.rmem_alloc 375 376 int sk_forward_alloc; 377 #ifdef CONFIG_NET_RX_BUSY_POLL 378 unsigned int sk_ll_usec; 379 /* ===== mostly read cache line ===== */ 380 unsigned int sk_napi_id; 381 #endif 382 int sk_rcvbuf; 383 384 struct sk_filter __rcu *sk_filter; 385 union { 386 struct socket_wq __rcu *sk_wq; 387 struct socket_wq *sk_wq_raw; 388 }; 389 #ifdef CONFIG_XFRM 390 struct xfrm_policy __rcu *sk_policy[2]; 391 #endif 392 struct dst_entry *sk_rx_dst; 393 struct dst_entry __rcu *sk_dst_cache; 394 atomic_t sk_omem_alloc; 395 int sk_sndbuf; 396 397 /* ===== cache line for TX ===== */ 398 int sk_wmem_queued; 399 refcount_t sk_wmem_alloc; 400 unsigned long sk_tsq_flags; 401 union { 402 struct sk_buff *sk_send_head; 403 struct rb_root tcp_rtx_queue; 404 }; 405 struct sk_buff_head sk_write_queue; 406 __s32 sk_peek_off; 407 int sk_write_pending; 408 __u32 sk_dst_pending_confirm; 409 u32 sk_pacing_status; /* see enum sk_pacing */ 410 long sk_sndtimeo; 411 struct timer_list sk_timer; 412 __u32 sk_priority; 413 __u32 sk_mark; 414 u32 sk_pacing_rate; /* bytes per second */ 415 u32 sk_max_pacing_rate; 416 struct page_frag sk_frag; 417 netdev_features_t sk_route_caps; 418 netdev_features_t sk_route_nocaps; 419 int sk_gso_type; 420 unsigned int sk_gso_max_size; 421 gfp_t sk_allocation; 422 __u32 sk_txhash; 423 424 /* 425 * Because of non atomicity rules, all 426 * changes are protected by socket lock. 427 */ 428 unsigned int __sk_flags_offset[0]; 429 #ifdef __BIG_ENDIAN_BITFIELD 430 #define SK_FL_PROTO_SHIFT 16 431 #define SK_FL_PROTO_MASK 0x00ff0000 432 433 #define SK_FL_TYPE_SHIFT 0 434 #define SK_FL_TYPE_MASK 0x0000ffff 435 #else 436 #define SK_FL_PROTO_SHIFT 8 437 #define SK_FL_PROTO_MASK 0x0000ff00 438 439 #define SK_FL_TYPE_SHIFT 16 440 #define SK_FL_TYPE_MASK 0xffff0000 441 #endif 442 443 unsigned int sk_padding : 1, 444 sk_kern_sock : 1, 445 sk_no_check_tx : 1, 446 sk_no_check_rx : 1, 447 sk_userlocks : 4, 448 sk_protocol : 8, 449 sk_type : 16; 450 #define SK_PROTOCOL_MAX U8_MAX 451 u16 sk_gso_max_segs; 452 u8 sk_pacing_shift; 453 unsigned long sk_lingertime; 454 struct proto *sk_prot_creator; 455 rwlock_t sk_callback_lock; 456 int sk_err, 457 sk_err_soft; 458 u32 sk_ack_backlog; 459 u32 sk_max_ack_backlog; 460 kuid_t sk_uid; 461 struct pid *sk_peer_pid; 462 const struct cred *sk_peer_cred; 463 long sk_rcvtimeo; 464 ktime_t sk_stamp; 465 u16 sk_tsflags; 466 u8 sk_shutdown; 467 u32 sk_tskey; 468 atomic_t sk_zckey; 469 struct socket *sk_socket; 470 void *sk_user_data; 471 #ifdef CONFIG_SECURITY 472 void *sk_security; 473 #endif 474 struct sock_cgroup_data sk_cgrp_data; 475 struct mem_cgroup *sk_memcg; 476 void (*sk_state_change)(struct sock *sk); 477 void (*sk_data_ready)(struct sock *sk); 478 void (*sk_write_space)(struct sock *sk); 479 void (*sk_error_report)(struct sock *sk); 480 int (*sk_backlog_rcv)(struct sock *sk, 481 struct sk_buff *skb); 482 void (*sk_destruct)(struct sock *sk); 483 struct sock_reuseport __rcu *sk_reuseport_cb; 484 struct rcu_head sk_rcu; 485 }; 486 487 enum sk_pacing { 488 SK_PACING_NONE = 0, 489 SK_PACING_NEEDED = 1, 490 SK_PACING_FQ = 2, 491 }; 492 493 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 494 495 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 496 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 497 498 /* 499 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 500 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 501 * on a socket means that the socket will reuse everybody else's port 502 * without looking at the other's sk_reuse value. 503 */ 504 505 #define SK_NO_REUSE 0 506 #define SK_CAN_REUSE 1 507 #define SK_FORCE_REUSE 2 508 509 int sk_set_peek_off(struct sock *sk, int val); 510 511 static inline int sk_peek_offset(struct sock *sk, int flags) 512 { 513 if (unlikely(flags & MSG_PEEK)) { 514 return READ_ONCE(sk->sk_peek_off); 515 } 516 517 return 0; 518 } 519 520 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 521 { 522 s32 off = READ_ONCE(sk->sk_peek_off); 523 524 if (unlikely(off >= 0)) { 525 off = max_t(s32, off - val, 0); 526 WRITE_ONCE(sk->sk_peek_off, off); 527 } 528 } 529 530 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 531 { 532 sk_peek_offset_bwd(sk, -val); 533 } 534 535 /* 536 * Hashed lists helper routines 537 */ 538 static inline struct sock *sk_entry(const struct hlist_node *node) 539 { 540 return hlist_entry(node, struct sock, sk_node); 541 } 542 543 static inline struct sock *__sk_head(const struct hlist_head *head) 544 { 545 return hlist_entry(head->first, struct sock, sk_node); 546 } 547 548 static inline struct sock *sk_head(const struct hlist_head *head) 549 { 550 return hlist_empty(head) ? NULL : __sk_head(head); 551 } 552 553 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 554 { 555 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 556 } 557 558 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 559 { 560 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 561 } 562 563 static inline struct sock *sk_next(const struct sock *sk) 564 { 565 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 566 } 567 568 static inline struct sock *sk_nulls_next(const struct sock *sk) 569 { 570 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 571 hlist_nulls_entry(sk->sk_nulls_node.next, 572 struct sock, sk_nulls_node) : 573 NULL; 574 } 575 576 static inline bool sk_unhashed(const struct sock *sk) 577 { 578 return hlist_unhashed(&sk->sk_node); 579 } 580 581 static inline bool sk_hashed(const struct sock *sk) 582 { 583 return !sk_unhashed(sk); 584 } 585 586 static inline void sk_node_init(struct hlist_node *node) 587 { 588 node->pprev = NULL; 589 } 590 591 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 592 { 593 node->pprev = NULL; 594 } 595 596 static inline void __sk_del_node(struct sock *sk) 597 { 598 __hlist_del(&sk->sk_node); 599 } 600 601 /* NB: equivalent to hlist_del_init_rcu */ 602 static inline bool __sk_del_node_init(struct sock *sk) 603 { 604 if (sk_hashed(sk)) { 605 __sk_del_node(sk); 606 sk_node_init(&sk->sk_node); 607 return true; 608 } 609 return false; 610 } 611 612 /* Grab socket reference count. This operation is valid only 613 when sk is ALREADY grabbed f.e. it is found in hash table 614 or a list and the lookup is made under lock preventing hash table 615 modifications. 616 */ 617 618 static __always_inline void sock_hold(struct sock *sk) 619 { 620 refcount_inc(&sk->sk_refcnt); 621 } 622 623 /* Ungrab socket in the context, which assumes that socket refcnt 624 cannot hit zero, f.e. it is true in context of any socketcall. 625 */ 626 static __always_inline void __sock_put(struct sock *sk) 627 { 628 refcount_dec(&sk->sk_refcnt); 629 } 630 631 static inline bool sk_del_node_init(struct sock *sk) 632 { 633 bool rc = __sk_del_node_init(sk); 634 635 if (rc) { 636 /* paranoid for a while -acme */ 637 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 638 __sock_put(sk); 639 } 640 return rc; 641 } 642 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 643 644 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 645 { 646 if (sk_hashed(sk)) { 647 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 648 return true; 649 } 650 return false; 651 } 652 653 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 654 { 655 bool rc = __sk_nulls_del_node_init_rcu(sk); 656 657 if (rc) { 658 /* paranoid for a while -acme */ 659 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 660 __sock_put(sk); 661 } 662 return rc; 663 } 664 665 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 666 { 667 hlist_add_head(&sk->sk_node, list); 668 } 669 670 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 671 { 672 sock_hold(sk); 673 __sk_add_node(sk, list); 674 } 675 676 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 677 { 678 sock_hold(sk); 679 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 680 sk->sk_family == AF_INET6) 681 hlist_add_tail_rcu(&sk->sk_node, list); 682 else 683 hlist_add_head_rcu(&sk->sk_node, list); 684 } 685 686 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 687 { 688 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 689 sk->sk_family == AF_INET6) 690 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 691 else 692 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 693 } 694 695 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 696 { 697 sock_hold(sk); 698 __sk_nulls_add_node_rcu(sk, list); 699 } 700 701 static inline void __sk_del_bind_node(struct sock *sk) 702 { 703 __hlist_del(&sk->sk_bind_node); 704 } 705 706 static inline void sk_add_bind_node(struct sock *sk, 707 struct hlist_head *list) 708 { 709 hlist_add_head(&sk->sk_bind_node, list); 710 } 711 712 #define sk_for_each(__sk, list) \ 713 hlist_for_each_entry(__sk, list, sk_node) 714 #define sk_for_each_rcu(__sk, list) \ 715 hlist_for_each_entry_rcu(__sk, list, sk_node) 716 #define sk_nulls_for_each(__sk, node, list) \ 717 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 718 #define sk_nulls_for_each_rcu(__sk, node, list) \ 719 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 720 #define sk_for_each_from(__sk) \ 721 hlist_for_each_entry_from(__sk, sk_node) 722 #define sk_nulls_for_each_from(__sk, node) \ 723 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 724 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 725 #define sk_for_each_safe(__sk, tmp, list) \ 726 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 727 #define sk_for_each_bound(__sk, list) \ 728 hlist_for_each_entry(__sk, list, sk_bind_node) 729 730 /** 731 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 732 * @tpos: the type * to use as a loop cursor. 733 * @pos: the &struct hlist_node to use as a loop cursor. 734 * @head: the head for your list. 735 * @offset: offset of hlist_node within the struct. 736 * 737 */ 738 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 739 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 740 pos != NULL && \ 741 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 742 pos = rcu_dereference(hlist_next_rcu(pos))) 743 744 static inline struct user_namespace *sk_user_ns(struct sock *sk) 745 { 746 /* Careful only use this in a context where these parameters 747 * can not change and must all be valid, such as recvmsg from 748 * userspace. 749 */ 750 return sk->sk_socket->file->f_cred->user_ns; 751 } 752 753 /* Sock flags */ 754 enum sock_flags { 755 SOCK_DEAD, 756 SOCK_DONE, 757 SOCK_URGINLINE, 758 SOCK_KEEPOPEN, 759 SOCK_LINGER, 760 SOCK_DESTROY, 761 SOCK_BROADCAST, 762 SOCK_TIMESTAMP, 763 SOCK_ZAPPED, 764 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 765 SOCK_DBG, /* %SO_DEBUG setting */ 766 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 767 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 768 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 769 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 770 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 771 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 772 SOCK_FASYNC, /* fasync() active */ 773 SOCK_RXQ_OVFL, 774 SOCK_ZEROCOPY, /* buffers from userspace */ 775 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 776 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 777 * Will use last 4 bytes of packet sent from 778 * user-space instead. 779 */ 780 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 781 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 782 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 783 }; 784 785 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 786 787 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 788 { 789 nsk->sk_flags = osk->sk_flags; 790 } 791 792 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 793 { 794 __set_bit(flag, &sk->sk_flags); 795 } 796 797 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 798 { 799 __clear_bit(flag, &sk->sk_flags); 800 } 801 802 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 803 { 804 return test_bit(flag, &sk->sk_flags); 805 } 806 807 #ifdef CONFIG_NET 808 extern struct static_key memalloc_socks; 809 static inline int sk_memalloc_socks(void) 810 { 811 return static_key_false(&memalloc_socks); 812 } 813 #else 814 815 static inline int sk_memalloc_socks(void) 816 { 817 return 0; 818 } 819 820 #endif 821 822 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 823 { 824 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 825 } 826 827 static inline void sk_acceptq_removed(struct sock *sk) 828 { 829 sk->sk_ack_backlog--; 830 } 831 832 static inline void sk_acceptq_added(struct sock *sk) 833 { 834 sk->sk_ack_backlog++; 835 } 836 837 static inline bool sk_acceptq_is_full(const struct sock *sk) 838 { 839 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 840 } 841 842 /* 843 * Compute minimal free write space needed to queue new packets. 844 */ 845 static inline int sk_stream_min_wspace(const struct sock *sk) 846 { 847 return sk->sk_wmem_queued >> 1; 848 } 849 850 static inline int sk_stream_wspace(const struct sock *sk) 851 { 852 return sk->sk_sndbuf - sk->sk_wmem_queued; 853 } 854 855 void sk_stream_write_space(struct sock *sk); 856 857 /* OOB backlog add */ 858 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 859 { 860 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 861 skb_dst_force(skb); 862 863 if (!sk->sk_backlog.tail) 864 sk->sk_backlog.head = skb; 865 else 866 sk->sk_backlog.tail->next = skb; 867 868 sk->sk_backlog.tail = skb; 869 skb->next = NULL; 870 } 871 872 /* 873 * Take into account size of receive queue and backlog queue 874 * Do not take into account this skb truesize, 875 * to allow even a single big packet to come. 876 */ 877 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 878 { 879 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 880 881 return qsize > limit; 882 } 883 884 /* The per-socket spinlock must be held here. */ 885 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 886 unsigned int limit) 887 { 888 if (sk_rcvqueues_full(sk, limit)) 889 return -ENOBUFS; 890 891 /* 892 * If the skb was allocated from pfmemalloc reserves, only 893 * allow SOCK_MEMALLOC sockets to use it as this socket is 894 * helping free memory 895 */ 896 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 897 return -ENOMEM; 898 899 __sk_add_backlog(sk, skb); 900 sk->sk_backlog.len += skb->truesize; 901 return 0; 902 } 903 904 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 905 906 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 907 { 908 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 909 return __sk_backlog_rcv(sk, skb); 910 911 return sk->sk_backlog_rcv(sk, skb); 912 } 913 914 static inline void sk_incoming_cpu_update(struct sock *sk) 915 { 916 int cpu = raw_smp_processor_id(); 917 918 if (unlikely(sk->sk_incoming_cpu != cpu)) 919 sk->sk_incoming_cpu = cpu; 920 } 921 922 static inline void sock_rps_record_flow_hash(__u32 hash) 923 { 924 #ifdef CONFIG_RPS 925 struct rps_sock_flow_table *sock_flow_table; 926 927 rcu_read_lock(); 928 sock_flow_table = rcu_dereference(rps_sock_flow_table); 929 rps_record_sock_flow(sock_flow_table, hash); 930 rcu_read_unlock(); 931 #endif 932 } 933 934 static inline void sock_rps_record_flow(const struct sock *sk) 935 { 936 #ifdef CONFIG_RPS 937 if (static_key_false(&rfs_needed)) { 938 /* Reading sk->sk_rxhash might incur an expensive cache line 939 * miss. 940 * 941 * TCP_ESTABLISHED does cover almost all states where RFS 942 * might be useful, and is cheaper [1] than testing : 943 * IPv4: inet_sk(sk)->inet_daddr 944 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 945 * OR an additional socket flag 946 * [1] : sk_state and sk_prot are in the same cache line. 947 */ 948 if (sk->sk_state == TCP_ESTABLISHED) 949 sock_rps_record_flow_hash(sk->sk_rxhash); 950 } 951 #endif 952 } 953 954 static inline void sock_rps_save_rxhash(struct sock *sk, 955 const struct sk_buff *skb) 956 { 957 #ifdef CONFIG_RPS 958 if (unlikely(sk->sk_rxhash != skb->hash)) 959 sk->sk_rxhash = skb->hash; 960 #endif 961 } 962 963 static inline void sock_rps_reset_rxhash(struct sock *sk) 964 { 965 #ifdef CONFIG_RPS 966 sk->sk_rxhash = 0; 967 #endif 968 } 969 970 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 971 ({ int __rc; \ 972 release_sock(__sk); \ 973 __rc = __condition; \ 974 if (!__rc) { \ 975 *(__timeo) = wait_woken(__wait, \ 976 TASK_INTERRUPTIBLE, \ 977 *(__timeo)); \ 978 } \ 979 sched_annotate_sleep(); \ 980 lock_sock(__sk); \ 981 __rc = __condition; \ 982 __rc; \ 983 }) 984 985 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 986 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 987 void sk_stream_wait_close(struct sock *sk, long timeo_p); 988 int sk_stream_error(struct sock *sk, int flags, int err); 989 void sk_stream_kill_queues(struct sock *sk); 990 void sk_set_memalloc(struct sock *sk); 991 void sk_clear_memalloc(struct sock *sk); 992 993 void __sk_flush_backlog(struct sock *sk); 994 995 static inline bool sk_flush_backlog(struct sock *sk) 996 { 997 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 998 __sk_flush_backlog(sk); 999 return true; 1000 } 1001 return false; 1002 } 1003 1004 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1005 1006 struct request_sock_ops; 1007 struct timewait_sock_ops; 1008 struct inet_hashinfo; 1009 struct raw_hashinfo; 1010 struct smc_hashinfo; 1011 struct module; 1012 1013 /* 1014 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1015 * un-modified. Special care is taken when initializing object to zero. 1016 */ 1017 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1018 { 1019 if (offsetof(struct sock, sk_node.next) != 0) 1020 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1021 memset(&sk->sk_node.pprev, 0, 1022 size - offsetof(struct sock, sk_node.pprev)); 1023 } 1024 1025 /* Networking protocol blocks we attach to sockets. 1026 * socket layer -> transport layer interface 1027 */ 1028 struct proto { 1029 void (*close)(struct sock *sk, 1030 long timeout); 1031 int (*connect)(struct sock *sk, 1032 struct sockaddr *uaddr, 1033 int addr_len); 1034 int (*disconnect)(struct sock *sk, int flags); 1035 1036 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1037 bool kern); 1038 1039 int (*ioctl)(struct sock *sk, int cmd, 1040 unsigned long arg); 1041 int (*init)(struct sock *sk); 1042 void (*destroy)(struct sock *sk); 1043 void (*shutdown)(struct sock *sk, int how); 1044 int (*setsockopt)(struct sock *sk, int level, 1045 int optname, char __user *optval, 1046 unsigned int optlen); 1047 int (*getsockopt)(struct sock *sk, int level, 1048 int optname, char __user *optval, 1049 int __user *option); 1050 void (*keepalive)(struct sock *sk, int valbool); 1051 #ifdef CONFIG_COMPAT 1052 int (*compat_setsockopt)(struct sock *sk, 1053 int level, 1054 int optname, char __user *optval, 1055 unsigned int optlen); 1056 int (*compat_getsockopt)(struct sock *sk, 1057 int level, 1058 int optname, char __user *optval, 1059 int __user *option); 1060 int (*compat_ioctl)(struct sock *sk, 1061 unsigned int cmd, unsigned long arg); 1062 #endif 1063 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1064 size_t len); 1065 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1066 size_t len, int noblock, int flags, 1067 int *addr_len); 1068 int (*sendpage)(struct sock *sk, struct page *page, 1069 int offset, size_t size, int flags); 1070 int (*bind)(struct sock *sk, 1071 struct sockaddr *uaddr, int addr_len); 1072 1073 int (*backlog_rcv) (struct sock *sk, 1074 struct sk_buff *skb); 1075 1076 void (*release_cb)(struct sock *sk); 1077 1078 /* Keeping track of sk's, looking them up, and port selection methods. */ 1079 int (*hash)(struct sock *sk); 1080 void (*unhash)(struct sock *sk); 1081 void (*rehash)(struct sock *sk); 1082 int (*get_port)(struct sock *sk, unsigned short snum); 1083 1084 /* Keeping track of sockets in use */ 1085 #ifdef CONFIG_PROC_FS 1086 unsigned int inuse_idx; 1087 #endif 1088 1089 bool (*stream_memory_free)(const struct sock *sk); 1090 /* Memory pressure */ 1091 void (*enter_memory_pressure)(struct sock *sk); 1092 void (*leave_memory_pressure)(struct sock *sk); 1093 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1094 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1095 /* 1096 * Pressure flag: try to collapse. 1097 * Technical note: it is used by multiple contexts non atomically. 1098 * All the __sk_mem_schedule() is of this nature: accounting 1099 * is strict, actions are advisory and have some latency. 1100 */ 1101 unsigned long *memory_pressure; 1102 long *sysctl_mem; 1103 1104 int *sysctl_wmem; 1105 int *sysctl_rmem; 1106 u32 sysctl_wmem_offset; 1107 u32 sysctl_rmem_offset; 1108 1109 int max_header; 1110 bool no_autobind; 1111 1112 struct kmem_cache *slab; 1113 unsigned int obj_size; 1114 slab_flags_t slab_flags; 1115 1116 struct percpu_counter *orphan_count; 1117 1118 struct request_sock_ops *rsk_prot; 1119 struct timewait_sock_ops *twsk_prot; 1120 1121 union { 1122 struct inet_hashinfo *hashinfo; 1123 struct udp_table *udp_table; 1124 struct raw_hashinfo *raw_hash; 1125 struct smc_hashinfo *smc_hash; 1126 } h; 1127 1128 struct module *owner; 1129 1130 char name[32]; 1131 1132 struct list_head node; 1133 #ifdef SOCK_REFCNT_DEBUG 1134 atomic_t socks; 1135 #endif 1136 int (*diag_destroy)(struct sock *sk, int err); 1137 } __randomize_layout; 1138 1139 int proto_register(struct proto *prot, int alloc_slab); 1140 void proto_unregister(struct proto *prot); 1141 1142 #ifdef SOCK_REFCNT_DEBUG 1143 static inline void sk_refcnt_debug_inc(struct sock *sk) 1144 { 1145 atomic_inc(&sk->sk_prot->socks); 1146 } 1147 1148 static inline void sk_refcnt_debug_dec(struct sock *sk) 1149 { 1150 atomic_dec(&sk->sk_prot->socks); 1151 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1152 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1153 } 1154 1155 static inline void sk_refcnt_debug_release(const struct sock *sk) 1156 { 1157 if (refcount_read(&sk->sk_refcnt) != 1) 1158 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1159 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1160 } 1161 #else /* SOCK_REFCNT_DEBUG */ 1162 #define sk_refcnt_debug_inc(sk) do { } while (0) 1163 #define sk_refcnt_debug_dec(sk) do { } while (0) 1164 #define sk_refcnt_debug_release(sk) do { } while (0) 1165 #endif /* SOCK_REFCNT_DEBUG */ 1166 1167 static inline bool sk_stream_memory_free(const struct sock *sk) 1168 { 1169 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1170 return false; 1171 1172 return sk->sk_prot->stream_memory_free ? 1173 sk->sk_prot->stream_memory_free(sk) : true; 1174 } 1175 1176 static inline bool sk_stream_is_writeable(const struct sock *sk) 1177 { 1178 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1179 sk_stream_memory_free(sk); 1180 } 1181 1182 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1183 struct cgroup *ancestor) 1184 { 1185 #ifdef CONFIG_SOCK_CGROUP_DATA 1186 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1187 ancestor); 1188 #else 1189 return -ENOTSUPP; 1190 #endif 1191 } 1192 1193 static inline bool sk_has_memory_pressure(const struct sock *sk) 1194 { 1195 return sk->sk_prot->memory_pressure != NULL; 1196 } 1197 1198 static inline bool sk_under_memory_pressure(const struct sock *sk) 1199 { 1200 if (!sk->sk_prot->memory_pressure) 1201 return false; 1202 1203 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1204 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1205 return true; 1206 1207 return !!*sk->sk_prot->memory_pressure; 1208 } 1209 1210 static inline long 1211 sk_memory_allocated(const struct sock *sk) 1212 { 1213 return atomic_long_read(sk->sk_prot->memory_allocated); 1214 } 1215 1216 static inline long 1217 sk_memory_allocated_add(struct sock *sk, int amt) 1218 { 1219 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1220 } 1221 1222 static inline void 1223 sk_memory_allocated_sub(struct sock *sk, int amt) 1224 { 1225 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1226 } 1227 1228 static inline void sk_sockets_allocated_dec(struct sock *sk) 1229 { 1230 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1231 } 1232 1233 static inline void sk_sockets_allocated_inc(struct sock *sk) 1234 { 1235 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1236 } 1237 1238 static inline int 1239 sk_sockets_allocated_read_positive(struct sock *sk) 1240 { 1241 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1242 } 1243 1244 static inline int 1245 proto_sockets_allocated_sum_positive(struct proto *prot) 1246 { 1247 return percpu_counter_sum_positive(prot->sockets_allocated); 1248 } 1249 1250 static inline long 1251 proto_memory_allocated(struct proto *prot) 1252 { 1253 return atomic_long_read(prot->memory_allocated); 1254 } 1255 1256 static inline bool 1257 proto_memory_pressure(struct proto *prot) 1258 { 1259 if (!prot->memory_pressure) 1260 return false; 1261 return !!*prot->memory_pressure; 1262 } 1263 1264 1265 #ifdef CONFIG_PROC_FS 1266 /* Called with local bh disabled */ 1267 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1268 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1269 #else 1270 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1271 int inc) 1272 { 1273 } 1274 #endif 1275 1276 1277 /* With per-bucket locks this operation is not-atomic, so that 1278 * this version is not worse. 1279 */ 1280 static inline int __sk_prot_rehash(struct sock *sk) 1281 { 1282 sk->sk_prot->unhash(sk); 1283 return sk->sk_prot->hash(sk); 1284 } 1285 1286 /* About 10 seconds */ 1287 #define SOCK_DESTROY_TIME (10*HZ) 1288 1289 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1290 #define PROT_SOCK 1024 1291 1292 #define SHUTDOWN_MASK 3 1293 #define RCV_SHUTDOWN 1 1294 #define SEND_SHUTDOWN 2 1295 1296 #define SOCK_SNDBUF_LOCK 1 1297 #define SOCK_RCVBUF_LOCK 2 1298 #define SOCK_BINDADDR_LOCK 4 1299 #define SOCK_BINDPORT_LOCK 8 1300 1301 struct socket_alloc { 1302 struct socket socket; 1303 struct inode vfs_inode; 1304 }; 1305 1306 static inline struct socket *SOCKET_I(struct inode *inode) 1307 { 1308 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1309 } 1310 1311 static inline struct inode *SOCK_INODE(struct socket *socket) 1312 { 1313 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1314 } 1315 1316 /* 1317 * Functions for memory accounting 1318 */ 1319 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1320 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1321 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1322 void __sk_mem_reclaim(struct sock *sk, int amount); 1323 1324 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1325 * do not necessarily have 16x time more memory than 4KB ones. 1326 */ 1327 #define SK_MEM_QUANTUM 4096 1328 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1329 #define SK_MEM_SEND 0 1330 #define SK_MEM_RECV 1 1331 1332 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1333 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1334 { 1335 long val = sk->sk_prot->sysctl_mem[index]; 1336 1337 #if PAGE_SIZE > SK_MEM_QUANTUM 1338 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1339 #elif PAGE_SIZE < SK_MEM_QUANTUM 1340 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1341 #endif 1342 return val; 1343 } 1344 1345 static inline int sk_mem_pages(int amt) 1346 { 1347 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1348 } 1349 1350 static inline bool sk_has_account(struct sock *sk) 1351 { 1352 /* return true if protocol supports memory accounting */ 1353 return !!sk->sk_prot->memory_allocated; 1354 } 1355 1356 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1357 { 1358 if (!sk_has_account(sk)) 1359 return true; 1360 return size <= sk->sk_forward_alloc || 1361 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1362 } 1363 1364 static inline bool 1365 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1366 { 1367 if (!sk_has_account(sk)) 1368 return true; 1369 return size<= sk->sk_forward_alloc || 1370 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1371 skb_pfmemalloc(skb); 1372 } 1373 1374 static inline void sk_mem_reclaim(struct sock *sk) 1375 { 1376 if (!sk_has_account(sk)) 1377 return; 1378 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1379 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1380 } 1381 1382 static inline void sk_mem_reclaim_partial(struct sock *sk) 1383 { 1384 if (!sk_has_account(sk)) 1385 return; 1386 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1387 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1388 } 1389 1390 static inline void sk_mem_charge(struct sock *sk, int size) 1391 { 1392 if (!sk_has_account(sk)) 1393 return; 1394 sk->sk_forward_alloc -= size; 1395 } 1396 1397 static inline void sk_mem_uncharge(struct sock *sk, int size) 1398 { 1399 if (!sk_has_account(sk)) 1400 return; 1401 sk->sk_forward_alloc += size; 1402 1403 /* Avoid a possible overflow. 1404 * TCP send queues can make this happen, if sk_mem_reclaim() 1405 * is not called and more than 2 GBytes are released at once. 1406 * 1407 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1408 * no need to hold that much forward allocation anyway. 1409 */ 1410 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1411 __sk_mem_reclaim(sk, 1 << 20); 1412 } 1413 1414 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1415 { 1416 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1417 sk->sk_wmem_queued -= skb->truesize; 1418 sk_mem_uncharge(sk, skb->truesize); 1419 __kfree_skb(skb); 1420 } 1421 1422 static inline void sock_release_ownership(struct sock *sk) 1423 { 1424 if (sk->sk_lock.owned) { 1425 sk->sk_lock.owned = 0; 1426 1427 /* The sk_lock has mutex_unlock() semantics: */ 1428 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1429 } 1430 } 1431 1432 /* 1433 * Macro so as to not evaluate some arguments when 1434 * lockdep is not enabled. 1435 * 1436 * Mark both the sk_lock and the sk_lock.slock as a 1437 * per-address-family lock class. 1438 */ 1439 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1440 do { \ 1441 sk->sk_lock.owned = 0; \ 1442 init_waitqueue_head(&sk->sk_lock.wq); \ 1443 spin_lock_init(&(sk)->sk_lock.slock); \ 1444 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1445 sizeof((sk)->sk_lock)); \ 1446 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1447 (skey), (sname)); \ 1448 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1449 } while (0) 1450 1451 #ifdef CONFIG_LOCKDEP 1452 static inline bool lockdep_sock_is_held(const struct sock *csk) 1453 { 1454 struct sock *sk = (struct sock *)csk; 1455 1456 return lockdep_is_held(&sk->sk_lock) || 1457 lockdep_is_held(&sk->sk_lock.slock); 1458 } 1459 #endif 1460 1461 void lock_sock_nested(struct sock *sk, int subclass); 1462 1463 static inline void lock_sock(struct sock *sk) 1464 { 1465 lock_sock_nested(sk, 0); 1466 } 1467 1468 void release_sock(struct sock *sk); 1469 1470 /* BH context may only use the following locking interface. */ 1471 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1472 #define bh_lock_sock_nested(__sk) \ 1473 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1474 SINGLE_DEPTH_NESTING) 1475 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1476 1477 bool lock_sock_fast(struct sock *sk); 1478 /** 1479 * unlock_sock_fast - complement of lock_sock_fast 1480 * @sk: socket 1481 * @slow: slow mode 1482 * 1483 * fast unlock socket for user context. 1484 * If slow mode is on, we call regular release_sock() 1485 */ 1486 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1487 { 1488 if (slow) 1489 release_sock(sk); 1490 else 1491 spin_unlock_bh(&sk->sk_lock.slock); 1492 } 1493 1494 /* Used by processes to "lock" a socket state, so that 1495 * interrupts and bottom half handlers won't change it 1496 * from under us. It essentially blocks any incoming 1497 * packets, so that we won't get any new data or any 1498 * packets that change the state of the socket. 1499 * 1500 * While locked, BH processing will add new packets to 1501 * the backlog queue. This queue is processed by the 1502 * owner of the socket lock right before it is released. 1503 * 1504 * Since ~2.3.5 it is also exclusive sleep lock serializing 1505 * accesses from user process context. 1506 */ 1507 1508 static inline void sock_owned_by_me(const struct sock *sk) 1509 { 1510 #ifdef CONFIG_LOCKDEP 1511 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1512 #endif 1513 } 1514 1515 static inline bool sock_owned_by_user(const struct sock *sk) 1516 { 1517 sock_owned_by_me(sk); 1518 return sk->sk_lock.owned; 1519 } 1520 1521 /* no reclassification while locks are held */ 1522 static inline bool sock_allow_reclassification(const struct sock *csk) 1523 { 1524 struct sock *sk = (struct sock *)csk; 1525 1526 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1527 } 1528 1529 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1530 struct proto *prot, int kern); 1531 void sk_free(struct sock *sk); 1532 void sk_destruct(struct sock *sk); 1533 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1534 void sk_free_unlock_clone(struct sock *sk); 1535 1536 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1537 gfp_t priority); 1538 void __sock_wfree(struct sk_buff *skb); 1539 void sock_wfree(struct sk_buff *skb); 1540 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1541 gfp_t priority); 1542 void skb_orphan_partial(struct sk_buff *skb); 1543 void sock_rfree(struct sk_buff *skb); 1544 void sock_efree(struct sk_buff *skb); 1545 #ifdef CONFIG_INET 1546 void sock_edemux(struct sk_buff *skb); 1547 #else 1548 #define sock_edemux sock_efree 1549 #endif 1550 1551 int sock_setsockopt(struct socket *sock, int level, int op, 1552 char __user *optval, unsigned int optlen); 1553 1554 int sock_getsockopt(struct socket *sock, int level, int op, 1555 char __user *optval, int __user *optlen); 1556 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1557 int noblock, int *errcode); 1558 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1559 unsigned long data_len, int noblock, 1560 int *errcode, int max_page_order); 1561 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1562 void sock_kfree_s(struct sock *sk, void *mem, int size); 1563 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1564 void sk_send_sigurg(struct sock *sk); 1565 1566 struct sockcm_cookie { 1567 u32 mark; 1568 u16 tsflags; 1569 }; 1570 1571 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1572 struct sockcm_cookie *sockc); 1573 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1574 struct sockcm_cookie *sockc); 1575 1576 /* 1577 * Functions to fill in entries in struct proto_ops when a protocol 1578 * does not implement a particular function. 1579 */ 1580 int sock_no_bind(struct socket *, struct sockaddr *, int); 1581 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1582 int sock_no_socketpair(struct socket *, struct socket *); 1583 int sock_no_accept(struct socket *, struct socket *, int, bool); 1584 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1585 unsigned int sock_no_poll(struct file *, struct socket *, 1586 struct poll_table_struct *); 1587 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1588 int sock_no_listen(struct socket *, int); 1589 int sock_no_shutdown(struct socket *, int); 1590 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1591 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1592 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1593 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1594 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1595 int sock_no_mmap(struct file *file, struct socket *sock, 1596 struct vm_area_struct *vma); 1597 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1598 size_t size, int flags); 1599 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1600 int offset, size_t size, int flags); 1601 1602 /* 1603 * Functions to fill in entries in struct proto_ops when a protocol 1604 * uses the inet style. 1605 */ 1606 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1607 char __user *optval, int __user *optlen); 1608 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1609 int flags); 1610 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1611 char __user *optval, unsigned int optlen); 1612 int compat_sock_common_getsockopt(struct socket *sock, int level, 1613 int optname, char __user *optval, int __user *optlen); 1614 int compat_sock_common_setsockopt(struct socket *sock, int level, 1615 int optname, char __user *optval, unsigned int optlen); 1616 1617 void sk_common_release(struct sock *sk); 1618 1619 /* 1620 * Default socket callbacks and setup code 1621 */ 1622 1623 /* Initialise core socket variables */ 1624 void sock_init_data(struct socket *sock, struct sock *sk); 1625 1626 /* 1627 * Socket reference counting postulates. 1628 * 1629 * * Each user of socket SHOULD hold a reference count. 1630 * * Each access point to socket (an hash table bucket, reference from a list, 1631 * running timer, skb in flight MUST hold a reference count. 1632 * * When reference count hits 0, it means it will never increase back. 1633 * * When reference count hits 0, it means that no references from 1634 * outside exist to this socket and current process on current CPU 1635 * is last user and may/should destroy this socket. 1636 * * sk_free is called from any context: process, BH, IRQ. When 1637 * it is called, socket has no references from outside -> sk_free 1638 * may release descendant resources allocated by the socket, but 1639 * to the time when it is called, socket is NOT referenced by any 1640 * hash tables, lists etc. 1641 * * Packets, delivered from outside (from network or from another process) 1642 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1643 * when they sit in queue. Otherwise, packets will leak to hole, when 1644 * socket is looked up by one cpu and unhasing is made by another CPU. 1645 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1646 * (leak to backlog). Packet socket does all the processing inside 1647 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1648 * use separate SMP lock, so that they are prone too. 1649 */ 1650 1651 /* Ungrab socket and destroy it, if it was the last reference. */ 1652 static inline void sock_put(struct sock *sk) 1653 { 1654 if (refcount_dec_and_test(&sk->sk_refcnt)) 1655 sk_free(sk); 1656 } 1657 /* Generic version of sock_put(), dealing with all sockets 1658 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1659 */ 1660 void sock_gen_put(struct sock *sk); 1661 1662 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1663 unsigned int trim_cap, bool refcounted); 1664 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1665 const int nested) 1666 { 1667 return __sk_receive_skb(sk, skb, nested, 1, true); 1668 } 1669 1670 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1671 { 1672 sk->sk_tx_queue_mapping = tx_queue; 1673 } 1674 1675 static inline void sk_tx_queue_clear(struct sock *sk) 1676 { 1677 sk->sk_tx_queue_mapping = -1; 1678 } 1679 1680 static inline int sk_tx_queue_get(const struct sock *sk) 1681 { 1682 return sk ? sk->sk_tx_queue_mapping : -1; 1683 } 1684 1685 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1686 { 1687 sk_tx_queue_clear(sk); 1688 sk->sk_socket = sock; 1689 } 1690 1691 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1692 { 1693 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1694 return &rcu_dereference_raw(sk->sk_wq)->wait; 1695 } 1696 /* Detach socket from process context. 1697 * Announce socket dead, detach it from wait queue and inode. 1698 * Note that parent inode held reference count on this struct sock, 1699 * we do not release it in this function, because protocol 1700 * probably wants some additional cleanups or even continuing 1701 * to work with this socket (TCP). 1702 */ 1703 static inline void sock_orphan(struct sock *sk) 1704 { 1705 write_lock_bh(&sk->sk_callback_lock); 1706 sock_set_flag(sk, SOCK_DEAD); 1707 sk_set_socket(sk, NULL); 1708 sk->sk_wq = NULL; 1709 write_unlock_bh(&sk->sk_callback_lock); 1710 } 1711 1712 static inline void sock_graft(struct sock *sk, struct socket *parent) 1713 { 1714 WARN_ON(parent->sk); 1715 write_lock_bh(&sk->sk_callback_lock); 1716 sk->sk_wq = parent->wq; 1717 parent->sk = sk; 1718 sk_set_socket(sk, parent); 1719 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1720 security_sock_graft(sk, parent); 1721 write_unlock_bh(&sk->sk_callback_lock); 1722 } 1723 1724 kuid_t sock_i_uid(struct sock *sk); 1725 unsigned long sock_i_ino(struct sock *sk); 1726 1727 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1728 { 1729 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1730 } 1731 1732 static inline u32 net_tx_rndhash(void) 1733 { 1734 u32 v = prandom_u32(); 1735 1736 return v ?: 1; 1737 } 1738 1739 static inline void sk_set_txhash(struct sock *sk) 1740 { 1741 sk->sk_txhash = net_tx_rndhash(); 1742 } 1743 1744 static inline void sk_rethink_txhash(struct sock *sk) 1745 { 1746 if (sk->sk_txhash) 1747 sk_set_txhash(sk); 1748 } 1749 1750 static inline struct dst_entry * 1751 __sk_dst_get(struct sock *sk) 1752 { 1753 return rcu_dereference_check(sk->sk_dst_cache, 1754 lockdep_sock_is_held(sk)); 1755 } 1756 1757 static inline struct dst_entry * 1758 sk_dst_get(struct sock *sk) 1759 { 1760 struct dst_entry *dst; 1761 1762 rcu_read_lock(); 1763 dst = rcu_dereference(sk->sk_dst_cache); 1764 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1765 dst = NULL; 1766 rcu_read_unlock(); 1767 return dst; 1768 } 1769 1770 static inline void dst_negative_advice(struct sock *sk) 1771 { 1772 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1773 1774 sk_rethink_txhash(sk); 1775 1776 if (dst && dst->ops->negative_advice) { 1777 ndst = dst->ops->negative_advice(dst); 1778 1779 if (ndst != dst) { 1780 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1781 sk_tx_queue_clear(sk); 1782 sk->sk_dst_pending_confirm = 0; 1783 } 1784 } 1785 } 1786 1787 static inline void 1788 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1789 { 1790 struct dst_entry *old_dst; 1791 1792 sk_tx_queue_clear(sk); 1793 sk->sk_dst_pending_confirm = 0; 1794 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1795 lockdep_sock_is_held(sk)); 1796 rcu_assign_pointer(sk->sk_dst_cache, dst); 1797 dst_release(old_dst); 1798 } 1799 1800 static inline void 1801 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1802 { 1803 struct dst_entry *old_dst; 1804 1805 sk_tx_queue_clear(sk); 1806 sk->sk_dst_pending_confirm = 0; 1807 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1808 dst_release(old_dst); 1809 } 1810 1811 static inline void 1812 __sk_dst_reset(struct sock *sk) 1813 { 1814 __sk_dst_set(sk, NULL); 1815 } 1816 1817 static inline void 1818 sk_dst_reset(struct sock *sk) 1819 { 1820 sk_dst_set(sk, NULL); 1821 } 1822 1823 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1824 1825 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1826 1827 static inline void sk_dst_confirm(struct sock *sk) 1828 { 1829 if (!sk->sk_dst_pending_confirm) 1830 sk->sk_dst_pending_confirm = 1; 1831 } 1832 1833 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1834 { 1835 if (skb_get_dst_pending_confirm(skb)) { 1836 struct sock *sk = skb->sk; 1837 unsigned long now = jiffies; 1838 1839 /* avoid dirtying neighbour */ 1840 if (n->confirmed != now) 1841 n->confirmed = now; 1842 if (sk && sk->sk_dst_pending_confirm) 1843 sk->sk_dst_pending_confirm = 0; 1844 } 1845 } 1846 1847 bool sk_mc_loop(struct sock *sk); 1848 1849 static inline bool sk_can_gso(const struct sock *sk) 1850 { 1851 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1852 } 1853 1854 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1855 1856 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1857 { 1858 sk->sk_route_nocaps |= flags; 1859 sk->sk_route_caps &= ~flags; 1860 } 1861 1862 static inline bool sk_check_csum_caps(struct sock *sk) 1863 { 1864 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1865 (sk->sk_family == PF_INET && 1866 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1867 (sk->sk_family == PF_INET6 && 1868 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1869 } 1870 1871 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1872 struct iov_iter *from, char *to, 1873 int copy, int offset) 1874 { 1875 if (skb->ip_summed == CHECKSUM_NONE) { 1876 __wsum csum = 0; 1877 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1878 return -EFAULT; 1879 skb->csum = csum_block_add(skb->csum, csum, offset); 1880 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1881 if (!copy_from_iter_full_nocache(to, copy, from)) 1882 return -EFAULT; 1883 } else if (!copy_from_iter_full(to, copy, from)) 1884 return -EFAULT; 1885 1886 return 0; 1887 } 1888 1889 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1890 struct iov_iter *from, int copy) 1891 { 1892 int err, offset = skb->len; 1893 1894 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1895 copy, offset); 1896 if (err) 1897 __skb_trim(skb, offset); 1898 1899 return err; 1900 } 1901 1902 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1903 struct sk_buff *skb, 1904 struct page *page, 1905 int off, int copy) 1906 { 1907 int err; 1908 1909 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1910 copy, skb->len); 1911 if (err) 1912 return err; 1913 1914 skb->len += copy; 1915 skb->data_len += copy; 1916 skb->truesize += copy; 1917 sk->sk_wmem_queued += copy; 1918 sk_mem_charge(sk, copy); 1919 return 0; 1920 } 1921 1922 /** 1923 * sk_wmem_alloc_get - returns write allocations 1924 * @sk: socket 1925 * 1926 * Returns sk_wmem_alloc minus initial offset of one 1927 */ 1928 static inline int sk_wmem_alloc_get(const struct sock *sk) 1929 { 1930 return refcount_read(&sk->sk_wmem_alloc) - 1; 1931 } 1932 1933 /** 1934 * sk_rmem_alloc_get - returns read allocations 1935 * @sk: socket 1936 * 1937 * Returns sk_rmem_alloc 1938 */ 1939 static inline int sk_rmem_alloc_get(const struct sock *sk) 1940 { 1941 return atomic_read(&sk->sk_rmem_alloc); 1942 } 1943 1944 /** 1945 * sk_has_allocations - check if allocations are outstanding 1946 * @sk: socket 1947 * 1948 * Returns true if socket has write or read allocations 1949 */ 1950 static inline bool sk_has_allocations(const struct sock *sk) 1951 { 1952 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1953 } 1954 1955 /** 1956 * skwq_has_sleeper - check if there are any waiting processes 1957 * @wq: struct socket_wq 1958 * 1959 * Returns true if socket_wq has waiting processes 1960 * 1961 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1962 * barrier call. They were added due to the race found within the tcp code. 1963 * 1964 * Consider following tcp code paths:: 1965 * 1966 * CPU1 CPU2 1967 * sys_select receive packet 1968 * ... ... 1969 * __add_wait_queue update tp->rcv_nxt 1970 * ... ... 1971 * tp->rcv_nxt check sock_def_readable 1972 * ... { 1973 * schedule rcu_read_lock(); 1974 * wq = rcu_dereference(sk->sk_wq); 1975 * if (wq && waitqueue_active(&wq->wait)) 1976 * wake_up_interruptible(&wq->wait) 1977 * ... 1978 * } 1979 * 1980 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1981 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1982 * could then endup calling schedule and sleep forever if there are no more 1983 * data on the socket. 1984 * 1985 */ 1986 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1987 { 1988 return wq && wq_has_sleeper(&wq->wait); 1989 } 1990 1991 /** 1992 * sock_poll_wait - place memory barrier behind the poll_wait call. 1993 * @filp: file 1994 * @wait_address: socket wait queue 1995 * @p: poll_table 1996 * 1997 * See the comments in the wq_has_sleeper function. 1998 */ 1999 static inline void sock_poll_wait(struct file *filp, 2000 wait_queue_head_t *wait_address, poll_table *p) 2001 { 2002 if (!poll_does_not_wait(p) && wait_address) { 2003 poll_wait(filp, wait_address, p); 2004 /* We need to be sure we are in sync with the 2005 * socket flags modification. 2006 * 2007 * This memory barrier is paired in the wq_has_sleeper. 2008 */ 2009 smp_mb(); 2010 } 2011 } 2012 2013 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2014 { 2015 if (sk->sk_txhash) { 2016 skb->l4_hash = 1; 2017 skb->hash = sk->sk_txhash; 2018 } 2019 } 2020 2021 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2022 2023 /* 2024 * Queue a received datagram if it will fit. Stream and sequenced 2025 * protocols can't normally use this as they need to fit buffers in 2026 * and play with them. 2027 * 2028 * Inlined as it's very short and called for pretty much every 2029 * packet ever received. 2030 */ 2031 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2032 { 2033 skb_orphan(skb); 2034 skb->sk = sk; 2035 skb->destructor = sock_rfree; 2036 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2037 sk_mem_charge(sk, skb->truesize); 2038 } 2039 2040 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2041 unsigned long expires); 2042 2043 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2044 2045 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2046 struct sk_buff *skb, unsigned int flags, 2047 void (*destructor)(struct sock *sk, 2048 struct sk_buff *skb)); 2049 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2050 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2051 2052 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2053 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2054 2055 /* 2056 * Recover an error report and clear atomically 2057 */ 2058 2059 static inline int sock_error(struct sock *sk) 2060 { 2061 int err; 2062 if (likely(!sk->sk_err)) 2063 return 0; 2064 err = xchg(&sk->sk_err, 0); 2065 return -err; 2066 } 2067 2068 static inline unsigned long sock_wspace(struct sock *sk) 2069 { 2070 int amt = 0; 2071 2072 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2073 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2074 if (amt < 0) 2075 amt = 0; 2076 } 2077 return amt; 2078 } 2079 2080 /* Note: 2081 * We use sk->sk_wq_raw, from contexts knowing this 2082 * pointer is not NULL and cannot disappear/change. 2083 */ 2084 static inline void sk_set_bit(int nr, struct sock *sk) 2085 { 2086 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2087 !sock_flag(sk, SOCK_FASYNC)) 2088 return; 2089 2090 set_bit(nr, &sk->sk_wq_raw->flags); 2091 } 2092 2093 static inline void sk_clear_bit(int nr, struct sock *sk) 2094 { 2095 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2096 !sock_flag(sk, SOCK_FASYNC)) 2097 return; 2098 2099 clear_bit(nr, &sk->sk_wq_raw->flags); 2100 } 2101 2102 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2103 { 2104 if (sock_flag(sk, SOCK_FASYNC)) { 2105 rcu_read_lock(); 2106 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2107 rcu_read_unlock(); 2108 } 2109 } 2110 2111 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2112 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2113 * Note: for send buffers, TCP works better if we can build two skbs at 2114 * minimum. 2115 */ 2116 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2117 2118 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2119 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2120 2121 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2122 { 2123 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2124 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2125 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2126 } 2127 } 2128 2129 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2130 bool force_schedule); 2131 2132 /** 2133 * sk_page_frag - return an appropriate page_frag 2134 * @sk: socket 2135 * 2136 * If socket allocation mode allows current thread to sleep, it means its 2137 * safe to use the per task page_frag instead of the per socket one. 2138 */ 2139 static inline struct page_frag *sk_page_frag(struct sock *sk) 2140 { 2141 if (gfpflags_allow_blocking(sk->sk_allocation)) 2142 return ¤t->task_frag; 2143 2144 return &sk->sk_frag; 2145 } 2146 2147 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2148 2149 /* 2150 * Default write policy as shown to user space via poll/select/SIGIO 2151 */ 2152 static inline bool sock_writeable(const struct sock *sk) 2153 { 2154 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2155 } 2156 2157 static inline gfp_t gfp_any(void) 2158 { 2159 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2160 } 2161 2162 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2163 { 2164 return noblock ? 0 : sk->sk_rcvtimeo; 2165 } 2166 2167 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2168 { 2169 return noblock ? 0 : sk->sk_sndtimeo; 2170 } 2171 2172 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2173 { 2174 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2175 } 2176 2177 /* Alas, with timeout socket operations are not restartable. 2178 * Compare this to poll(). 2179 */ 2180 static inline int sock_intr_errno(long timeo) 2181 { 2182 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2183 } 2184 2185 struct sock_skb_cb { 2186 u32 dropcount; 2187 }; 2188 2189 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2190 * using skb->cb[] would keep using it directly and utilize its 2191 * alignement guarantee. 2192 */ 2193 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2194 sizeof(struct sock_skb_cb))) 2195 2196 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2197 SOCK_SKB_CB_OFFSET)) 2198 2199 #define sock_skb_cb_check_size(size) \ 2200 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2201 2202 static inline void 2203 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2204 { 2205 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2206 atomic_read(&sk->sk_drops) : 0; 2207 } 2208 2209 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2210 { 2211 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2212 2213 atomic_add(segs, &sk->sk_drops); 2214 } 2215 2216 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2217 struct sk_buff *skb); 2218 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2219 struct sk_buff *skb); 2220 2221 static inline void 2222 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2223 { 2224 ktime_t kt = skb->tstamp; 2225 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2226 2227 /* 2228 * generate control messages if 2229 * - receive time stamping in software requested 2230 * - software time stamp available and wanted 2231 * - hardware time stamps available and wanted 2232 */ 2233 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2234 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2235 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2236 (hwtstamps->hwtstamp && 2237 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2238 __sock_recv_timestamp(msg, sk, skb); 2239 else 2240 sk->sk_stamp = kt; 2241 2242 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2243 __sock_recv_wifi_status(msg, sk, skb); 2244 } 2245 2246 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2247 struct sk_buff *skb); 2248 2249 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2250 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2251 struct sk_buff *skb) 2252 { 2253 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2254 (1UL << SOCK_RCVTSTAMP)) 2255 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2256 SOF_TIMESTAMPING_RAW_HARDWARE) 2257 2258 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2259 __sock_recv_ts_and_drops(msg, sk, skb); 2260 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2261 sk->sk_stamp = skb->tstamp; 2262 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2263 sk->sk_stamp = 0; 2264 } 2265 2266 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2267 2268 /** 2269 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2270 * @sk: socket sending this packet 2271 * @tsflags: timestamping flags to use 2272 * @tx_flags: completed with instructions for time stamping 2273 * 2274 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2275 */ 2276 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2277 __u8 *tx_flags) 2278 { 2279 if (unlikely(tsflags)) 2280 __sock_tx_timestamp(tsflags, tx_flags); 2281 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2282 *tx_flags |= SKBTX_WIFI_STATUS; 2283 } 2284 2285 /** 2286 * sk_eat_skb - Release a skb if it is no longer needed 2287 * @sk: socket to eat this skb from 2288 * @skb: socket buffer to eat 2289 * 2290 * This routine must be called with interrupts disabled or with the socket 2291 * locked so that the sk_buff queue operation is ok. 2292 */ 2293 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2294 { 2295 __skb_unlink(skb, &sk->sk_receive_queue); 2296 __kfree_skb(skb); 2297 } 2298 2299 static inline 2300 struct net *sock_net(const struct sock *sk) 2301 { 2302 return read_pnet(&sk->sk_net); 2303 } 2304 2305 static inline 2306 void sock_net_set(struct sock *sk, struct net *net) 2307 { 2308 write_pnet(&sk->sk_net, net); 2309 } 2310 2311 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2312 { 2313 if (skb->sk) { 2314 struct sock *sk = skb->sk; 2315 2316 skb->destructor = NULL; 2317 skb->sk = NULL; 2318 return sk; 2319 } 2320 return NULL; 2321 } 2322 2323 /* This helper checks if a socket is a full socket, 2324 * ie _not_ a timewait or request socket. 2325 */ 2326 static inline bool sk_fullsock(const struct sock *sk) 2327 { 2328 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2329 } 2330 2331 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2332 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2333 */ 2334 static inline bool sk_listener(const struct sock *sk) 2335 { 2336 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2337 } 2338 2339 /** 2340 * sk_state_load - read sk->sk_state for lockless contexts 2341 * @sk: socket pointer 2342 * 2343 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2344 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2345 */ 2346 static inline int sk_state_load(const struct sock *sk) 2347 { 2348 return smp_load_acquire(&sk->sk_state); 2349 } 2350 2351 /** 2352 * sk_state_store - update sk->sk_state 2353 * @sk: socket pointer 2354 * @newstate: new state 2355 * 2356 * Paired with sk_state_load(). Should be used in contexts where 2357 * state change might impact lockless readers. 2358 */ 2359 static inline void sk_state_store(struct sock *sk, int newstate) 2360 { 2361 smp_store_release(&sk->sk_state, newstate); 2362 } 2363 2364 void sock_enable_timestamp(struct sock *sk, int flag); 2365 int sock_get_timestamp(struct sock *, struct timeval __user *); 2366 int sock_get_timestampns(struct sock *, struct timespec __user *); 2367 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2368 int type); 2369 2370 bool sk_ns_capable(const struct sock *sk, 2371 struct user_namespace *user_ns, int cap); 2372 bool sk_capable(const struct sock *sk, int cap); 2373 bool sk_net_capable(const struct sock *sk, int cap); 2374 2375 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2376 2377 /* Take into consideration the size of the struct sk_buff overhead in the 2378 * determination of these values, since that is non-constant across 2379 * platforms. This makes socket queueing behavior and performance 2380 * not depend upon such differences. 2381 */ 2382 #define _SK_MEM_PACKETS 256 2383 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2384 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2385 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2386 2387 extern __u32 sysctl_wmem_max; 2388 extern __u32 sysctl_rmem_max; 2389 2390 extern int sysctl_tstamp_allow_data; 2391 extern int sysctl_optmem_max; 2392 2393 extern __u32 sysctl_wmem_default; 2394 extern __u32 sysctl_rmem_default; 2395 2396 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2397 { 2398 /* Does this proto have per netns sysctl_wmem ? */ 2399 if (proto->sysctl_wmem_offset) 2400 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2401 2402 return *proto->sysctl_wmem; 2403 } 2404 2405 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2406 { 2407 /* Does this proto have per netns sysctl_rmem ? */ 2408 if (proto->sysctl_rmem_offset) 2409 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2410 2411 return *proto->sysctl_rmem; 2412 } 2413 2414 #endif /* _SOCK_H */ 2415