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 #include <net/l3mdev.h> 76 77 /* 78 * This structure really needs to be cleaned up. 79 * Most of it is for TCP, and not used by any of 80 * the other protocols. 81 */ 82 83 /* Define this to get the SOCK_DBG debugging facility. */ 84 #define SOCK_DEBUGGING 85 #ifdef SOCK_DEBUGGING 86 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 87 printk(KERN_DEBUG msg); } while (0) 88 #else 89 /* Validate arguments and do nothing */ 90 static inline __printf(2, 3) 91 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 92 { 93 } 94 #endif 95 96 /* This is the per-socket lock. The spinlock provides a synchronization 97 * between user contexts and software interrupt processing, whereas the 98 * mini-semaphore synchronizes multiple users amongst themselves. 99 */ 100 typedef struct { 101 spinlock_t slock; 102 int owned; 103 wait_queue_head_t wq; 104 /* 105 * We express the mutex-alike socket_lock semantics 106 * to the lock validator by explicitly managing 107 * the slock as a lock variant (in addition to 108 * the slock itself): 109 */ 110 #ifdef CONFIG_DEBUG_LOCK_ALLOC 111 struct lockdep_map dep_map; 112 #endif 113 } socket_lock_t; 114 115 struct sock; 116 struct proto; 117 struct net; 118 119 typedef __u32 __bitwise __portpair; 120 typedef __u64 __bitwise __addrpair; 121 122 /** 123 * struct sock_common - minimal network layer representation of sockets 124 * @skc_daddr: Foreign IPv4 addr 125 * @skc_rcv_saddr: Bound local IPv4 addr 126 * @skc_hash: hash value used with various protocol lookup tables 127 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 128 * @skc_dport: placeholder for inet_dport/tw_dport 129 * @skc_num: placeholder for inet_num/tw_num 130 * @skc_family: network address family 131 * @skc_state: Connection state 132 * @skc_reuse: %SO_REUSEADDR setting 133 * @skc_reuseport: %SO_REUSEPORT setting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_node: main hash linkage for various protocol lookup tables 140 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 141 * @skc_tx_queue_mapping: tx queue number for this connection 142 * @skc_flags: place holder for sk_flags 143 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 144 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 145 * @skc_incoming_cpu: record/match cpu processing incoming packets 146 * @skc_refcnt: reference count 147 * 148 * This is the minimal network layer representation of sockets, the header 149 * for struct sock and struct inet_timewait_sock. 150 */ 151 struct sock_common { 152 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 153 * address on 64bit arches : cf INET_MATCH() 154 */ 155 union { 156 __addrpair skc_addrpair; 157 struct { 158 __be32 skc_daddr; 159 __be32 skc_rcv_saddr; 160 }; 161 }; 162 union { 163 unsigned int skc_hash; 164 __u16 skc_u16hashes[2]; 165 }; 166 /* skc_dport && skc_num must be grouped as well */ 167 union { 168 __portpair skc_portpair; 169 struct { 170 __be16 skc_dport; 171 __u16 skc_num; 172 }; 173 }; 174 175 unsigned short skc_family; 176 volatile unsigned char skc_state; 177 unsigned char skc_reuse:4; 178 unsigned char skc_reuseport:1; 179 unsigned char skc_ipv6only:1; 180 unsigned char skc_net_refcnt:1; 181 int skc_bound_dev_if; 182 union { 183 struct hlist_node skc_bind_node; 184 struct hlist_node skc_portaddr_node; 185 }; 186 struct proto *skc_prot; 187 possible_net_t skc_net; 188 189 #if IS_ENABLED(CONFIG_IPV6) 190 struct in6_addr skc_v6_daddr; 191 struct in6_addr skc_v6_rcv_saddr; 192 #endif 193 194 atomic64_t skc_cookie; 195 196 /* following fields are padding to force 197 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 198 * assuming IPV6 is enabled. We use this padding differently 199 * for different kind of 'sockets' 200 */ 201 union { 202 unsigned long skc_flags; 203 struct sock *skc_listener; /* request_sock */ 204 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 205 }; 206 /* 207 * fields between dontcopy_begin/dontcopy_end 208 * are not copied in sock_copy() 209 */ 210 /* private: */ 211 int skc_dontcopy_begin[0]; 212 /* public: */ 213 union { 214 struct hlist_node skc_node; 215 struct hlist_nulls_node skc_nulls_node; 216 }; 217 int skc_tx_queue_mapping; 218 union { 219 int skc_incoming_cpu; 220 u32 skc_rcv_wnd; 221 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 222 }; 223 224 refcount_t skc_refcnt; 225 /* private: */ 226 int skc_dontcopy_end[0]; 227 union { 228 u32 skc_rxhash; 229 u32 skc_window_clamp; 230 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 231 }; 232 /* public: */ 233 }; 234 235 /** 236 * struct sock - network layer representation of sockets 237 * @__sk_common: shared layout with inet_timewait_sock 238 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 239 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 240 * @sk_lock: synchronizer 241 * @sk_kern_sock: True if sock is using kernel lock classes 242 * @sk_rcvbuf: size of receive buffer in bytes 243 * @sk_wq: sock wait queue and async head 244 * @sk_rx_dst: receive input route used by early demux 245 * @sk_dst_cache: destination cache 246 * @sk_dst_pending_confirm: need to confirm neighbour 247 * @sk_policy: flow policy 248 * @sk_receive_queue: incoming packets 249 * @sk_wmem_alloc: transmit queue bytes committed 250 * @sk_tsq_flags: TCP Small Queues flags 251 * @sk_write_queue: Packet sending queue 252 * @sk_omem_alloc: "o" is "option" or "other" 253 * @sk_wmem_queued: persistent queue size 254 * @sk_forward_alloc: space allocated forward 255 * @sk_napi_id: id of the last napi context to receive data for sk 256 * @sk_ll_usec: usecs to busypoll when there is no data 257 * @sk_allocation: allocation mode 258 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 259 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 260 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 261 * @sk_sndbuf: size of send buffer in bytes 262 * @__sk_flags_offset: empty field used to determine location of bitfield 263 * @sk_padding: unused element for alignment 264 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 265 * @sk_no_check_rx: allow zero checksum in RX packets 266 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 267 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 268 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 269 * @sk_gso_max_size: Maximum GSO segment size to build 270 * @sk_gso_max_segs: Maximum number of GSO segments 271 * @sk_pacing_shift: scaling factor for TCP Small Queues 272 * @sk_lingertime: %SO_LINGER l_linger setting 273 * @sk_backlog: always used with the per-socket spinlock held 274 * @sk_callback_lock: used with the callbacks in the end of this struct 275 * @sk_error_queue: rarely used 276 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 277 * IPV6_ADDRFORM for instance) 278 * @sk_err: last error 279 * @sk_err_soft: errors that don't cause failure but are the cause of a 280 * persistent failure not just 'timed out' 281 * @sk_drops: raw/udp drops counter 282 * @sk_ack_backlog: current listen backlog 283 * @sk_max_ack_backlog: listen backlog set in listen() 284 * @sk_uid: user id of owner 285 * @sk_priority: %SO_PRIORITY setting 286 * @sk_type: socket type (%SOCK_STREAM, etc) 287 * @sk_protocol: which protocol this socket belongs in this network family 288 * @sk_peer_pid: &struct pid for this socket's peer 289 * @sk_peer_cred: %SO_PEERCRED setting 290 * @sk_rcvlowat: %SO_RCVLOWAT setting 291 * @sk_rcvtimeo: %SO_RCVTIMEO setting 292 * @sk_sndtimeo: %SO_SNDTIMEO setting 293 * @sk_txhash: computed flow hash for use on transmit 294 * @sk_filter: socket filtering instructions 295 * @sk_timer: sock cleanup timer 296 * @sk_stamp: time stamp of last packet received 297 * @sk_tsflags: SO_TIMESTAMPING socket options 298 * @sk_tskey: counter to disambiguate concurrent tstamp requests 299 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 300 * @sk_socket: Identd and reporting IO signals 301 * @sk_user_data: RPC layer private data 302 * @sk_frag: cached page frag 303 * @sk_peek_off: current peek_offset value 304 * @sk_send_head: front of stuff to transmit 305 * @sk_security: used by security modules 306 * @sk_mark: generic packet mark 307 * @sk_cgrp_data: cgroup data for this cgroup 308 * @sk_memcg: this socket's memory cgroup association 309 * @sk_write_pending: a write to stream socket waits to start 310 * @sk_state_change: callback to indicate change in the state of the sock 311 * @sk_data_ready: callback to indicate there is data to be processed 312 * @sk_write_space: callback to indicate there is bf sending space available 313 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 314 * @sk_backlog_rcv: callback to process the backlog 315 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 316 * @sk_reuseport_cb: reuseport group container 317 * @sk_rcu: used during RCU grace period 318 */ 319 struct sock { 320 /* 321 * Now struct inet_timewait_sock also uses sock_common, so please just 322 * don't add nothing before this first member (__sk_common) --acme 323 */ 324 struct sock_common __sk_common; 325 #define sk_node __sk_common.skc_node 326 #define sk_nulls_node __sk_common.skc_nulls_node 327 #define sk_refcnt __sk_common.skc_refcnt 328 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 329 330 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 331 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 332 #define sk_hash __sk_common.skc_hash 333 #define sk_portpair __sk_common.skc_portpair 334 #define sk_num __sk_common.skc_num 335 #define sk_dport __sk_common.skc_dport 336 #define sk_addrpair __sk_common.skc_addrpair 337 #define sk_daddr __sk_common.skc_daddr 338 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 339 #define sk_family __sk_common.skc_family 340 #define sk_state __sk_common.skc_state 341 #define sk_reuse __sk_common.skc_reuse 342 #define sk_reuseport __sk_common.skc_reuseport 343 #define sk_ipv6only __sk_common.skc_ipv6only 344 #define sk_net_refcnt __sk_common.skc_net_refcnt 345 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 346 #define sk_bind_node __sk_common.skc_bind_node 347 #define sk_prot __sk_common.skc_prot 348 #define sk_net __sk_common.skc_net 349 #define sk_v6_daddr __sk_common.skc_v6_daddr 350 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 351 #define sk_cookie __sk_common.skc_cookie 352 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 353 #define sk_flags __sk_common.skc_flags 354 #define sk_rxhash __sk_common.skc_rxhash 355 356 socket_lock_t sk_lock; 357 atomic_t sk_drops; 358 int sk_rcvlowat; 359 struct sk_buff_head sk_error_queue; 360 struct sk_buff_head sk_receive_queue; 361 /* 362 * The backlog queue is special, it is always used with 363 * the per-socket spinlock held and requires low latency 364 * access. Therefore we special case it's implementation. 365 * Note : rmem_alloc is in this structure to fill a hole 366 * on 64bit arches, not because its logically part of 367 * backlog. 368 */ 369 struct { 370 atomic_t rmem_alloc; 371 int len; 372 struct sk_buff *head; 373 struct sk_buff *tail; 374 } sk_backlog; 375 #define sk_rmem_alloc sk_backlog.rmem_alloc 376 377 int sk_forward_alloc; 378 #ifdef CONFIG_NET_RX_BUSY_POLL 379 unsigned int sk_ll_usec; 380 /* ===== mostly read cache line ===== */ 381 unsigned int sk_napi_id; 382 #endif 383 int sk_rcvbuf; 384 385 struct sk_filter __rcu *sk_filter; 386 union { 387 struct socket_wq __rcu *sk_wq; 388 struct socket_wq *sk_wq_raw; 389 }; 390 #ifdef CONFIG_XFRM 391 struct xfrm_policy __rcu *sk_policy[2]; 392 #endif 393 struct dst_entry *sk_rx_dst; 394 struct dst_entry __rcu *sk_dst_cache; 395 atomic_t sk_omem_alloc; 396 int sk_sndbuf; 397 398 /* ===== cache line for TX ===== */ 399 int sk_wmem_queued; 400 refcount_t sk_wmem_alloc; 401 unsigned long sk_tsq_flags; 402 union { 403 struct sk_buff *sk_send_head; 404 struct rb_root tcp_rtx_queue; 405 }; 406 struct sk_buff_head sk_write_queue; 407 __s32 sk_peek_off; 408 int sk_write_pending; 409 __u32 sk_dst_pending_confirm; 410 u32 sk_pacing_status; /* see enum sk_pacing */ 411 long sk_sndtimeo; 412 struct timer_list sk_timer; 413 __u32 sk_priority; 414 __u32 sk_mark; 415 u32 sk_pacing_rate; /* bytes per second */ 416 u32 sk_max_pacing_rate; 417 struct page_frag sk_frag; 418 netdev_features_t sk_route_caps; 419 netdev_features_t sk_route_nocaps; 420 netdev_features_t sk_route_forced_caps; 421 int sk_gso_type; 422 unsigned int sk_gso_max_size; 423 gfp_t sk_allocation; 424 __u32 sk_txhash; 425 426 /* 427 * Because of non atomicity rules, all 428 * changes are protected by socket lock. 429 */ 430 unsigned int __sk_flags_offset[0]; 431 #ifdef __BIG_ENDIAN_BITFIELD 432 #define SK_FL_PROTO_SHIFT 16 433 #define SK_FL_PROTO_MASK 0x00ff0000 434 435 #define SK_FL_TYPE_SHIFT 0 436 #define SK_FL_TYPE_MASK 0x0000ffff 437 #else 438 #define SK_FL_PROTO_SHIFT 8 439 #define SK_FL_PROTO_MASK 0x0000ff00 440 441 #define SK_FL_TYPE_SHIFT 16 442 #define SK_FL_TYPE_MASK 0xffff0000 443 #endif 444 445 unsigned int sk_padding : 1, 446 sk_kern_sock : 1, 447 sk_no_check_tx : 1, 448 sk_no_check_rx : 1, 449 sk_userlocks : 4, 450 sk_protocol : 8, 451 sk_type : 16; 452 #define SK_PROTOCOL_MAX U8_MAX 453 u16 sk_gso_max_segs; 454 u8 sk_pacing_shift; 455 unsigned long sk_lingertime; 456 struct proto *sk_prot_creator; 457 rwlock_t sk_callback_lock; 458 int sk_err, 459 sk_err_soft; 460 u32 sk_ack_backlog; 461 u32 sk_max_ack_backlog; 462 kuid_t sk_uid; 463 struct pid *sk_peer_pid; 464 const struct cred *sk_peer_cred; 465 long sk_rcvtimeo; 466 ktime_t sk_stamp; 467 u16 sk_tsflags; 468 u8 sk_shutdown; 469 u32 sk_tskey; 470 atomic_t sk_zckey; 471 struct socket *sk_socket; 472 void *sk_user_data; 473 #ifdef CONFIG_SECURITY 474 void *sk_security; 475 #endif 476 struct sock_cgroup_data sk_cgrp_data; 477 struct mem_cgroup *sk_memcg; 478 void (*sk_state_change)(struct sock *sk); 479 void (*sk_data_ready)(struct sock *sk); 480 void (*sk_write_space)(struct sock *sk); 481 void (*sk_error_report)(struct sock *sk); 482 int (*sk_backlog_rcv)(struct sock *sk, 483 struct sk_buff *skb); 484 void (*sk_destruct)(struct sock *sk); 485 struct sock_reuseport __rcu *sk_reuseport_cb; 486 struct rcu_head sk_rcu; 487 }; 488 489 enum sk_pacing { 490 SK_PACING_NONE = 0, 491 SK_PACING_NEEDED = 1, 492 SK_PACING_FQ = 2, 493 }; 494 495 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 496 497 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 498 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 499 500 /* 501 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 502 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 503 * on a socket means that the socket will reuse everybody else's port 504 * without looking at the other's sk_reuse value. 505 */ 506 507 #define SK_NO_REUSE 0 508 #define SK_CAN_REUSE 1 509 #define SK_FORCE_REUSE 2 510 511 int sk_set_peek_off(struct sock *sk, int val); 512 513 static inline int sk_peek_offset(struct sock *sk, int flags) 514 { 515 if (unlikely(flags & MSG_PEEK)) { 516 return READ_ONCE(sk->sk_peek_off); 517 } 518 519 return 0; 520 } 521 522 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 523 { 524 s32 off = READ_ONCE(sk->sk_peek_off); 525 526 if (unlikely(off >= 0)) { 527 off = max_t(s32, off - val, 0); 528 WRITE_ONCE(sk->sk_peek_off, off); 529 } 530 } 531 532 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 533 { 534 sk_peek_offset_bwd(sk, -val); 535 } 536 537 /* 538 * Hashed lists helper routines 539 */ 540 static inline struct sock *sk_entry(const struct hlist_node *node) 541 { 542 return hlist_entry(node, struct sock, sk_node); 543 } 544 545 static inline struct sock *__sk_head(const struct hlist_head *head) 546 { 547 return hlist_entry(head->first, struct sock, sk_node); 548 } 549 550 static inline struct sock *sk_head(const struct hlist_head *head) 551 { 552 return hlist_empty(head) ? NULL : __sk_head(head); 553 } 554 555 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 556 { 557 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 558 } 559 560 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 561 { 562 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 563 } 564 565 static inline struct sock *sk_next(const struct sock *sk) 566 { 567 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 568 } 569 570 static inline struct sock *sk_nulls_next(const struct sock *sk) 571 { 572 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 573 hlist_nulls_entry(sk->sk_nulls_node.next, 574 struct sock, sk_nulls_node) : 575 NULL; 576 } 577 578 static inline bool sk_unhashed(const struct sock *sk) 579 { 580 return hlist_unhashed(&sk->sk_node); 581 } 582 583 static inline bool sk_hashed(const struct sock *sk) 584 { 585 return !sk_unhashed(sk); 586 } 587 588 static inline void sk_node_init(struct hlist_node *node) 589 { 590 node->pprev = NULL; 591 } 592 593 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 594 { 595 node->pprev = NULL; 596 } 597 598 static inline void __sk_del_node(struct sock *sk) 599 { 600 __hlist_del(&sk->sk_node); 601 } 602 603 /* NB: equivalent to hlist_del_init_rcu */ 604 static inline bool __sk_del_node_init(struct sock *sk) 605 { 606 if (sk_hashed(sk)) { 607 __sk_del_node(sk); 608 sk_node_init(&sk->sk_node); 609 return true; 610 } 611 return false; 612 } 613 614 /* Grab socket reference count. This operation is valid only 615 when sk is ALREADY grabbed f.e. it is found in hash table 616 or a list and the lookup is made under lock preventing hash table 617 modifications. 618 */ 619 620 static __always_inline void sock_hold(struct sock *sk) 621 { 622 refcount_inc(&sk->sk_refcnt); 623 } 624 625 /* Ungrab socket in the context, which assumes that socket refcnt 626 cannot hit zero, f.e. it is true in context of any socketcall. 627 */ 628 static __always_inline void __sock_put(struct sock *sk) 629 { 630 refcount_dec(&sk->sk_refcnt); 631 } 632 633 static inline bool sk_del_node_init(struct sock *sk) 634 { 635 bool rc = __sk_del_node_init(sk); 636 637 if (rc) { 638 /* paranoid for a while -acme */ 639 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 640 __sock_put(sk); 641 } 642 return rc; 643 } 644 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 645 646 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 647 { 648 if (sk_hashed(sk)) { 649 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 650 return true; 651 } 652 return false; 653 } 654 655 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 656 { 657 bool rc = __sk_nulls_del_node_init_rcu(sk); 658 659 if (rc) { 660 /* paranoid for a while -acme */ 661 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 662 __sock_put(sk); 663 } 664 return rc; 665 } 666 667 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 668 { 669 hlist_add_head(&sk->sk_node, list); 670 } 671 672 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 673 { 674 sock_hold(sk); 675 __sk_add_node(sk, list); 676 } 677 678 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 679 { 680 sock_hold(sk); 681 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 682 sk->sk_family == AF_INET6) 683 hlist_add_tail_rcu(&sk->sk_node, list); 684 else 685 hlist_add_head_rcu(&sk->sk_node, list); 686 } 687 688 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 689 { 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(hlist_first_rcu(head)); \ 738 pos != NULL && \ 739 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 740 pos = rcu_dereference(hlist_next_rcu(pos))) 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 1102 int *sysctl_wmem; 1103 int *sysctl_rmem; 1104 u32 sysctl_wmem_offset; 1105 u32 sysctl_rmem_offset; 1106 1107 int max_header; 1108 bool no_autobind; 1109 1110 struct kmem_cache *slab; 1111 unsigned int obj_size; 1112 slab_flags_t slab_flags; 1113 size_t useroffset; /* Usercopy region offset */ 1114 size_t usersize; /* Usercopy region size */ 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 int sock_inuse_get(struct net *net); 1270 #else 1271 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1272 int inc) 1273 { 1274 } 1275 #endif 1276 1277 1278 /* With per-bucket locks this operation is not-atomic, so that 1279 * this version is not worse. 1280 */ 1281 static inline int __sk_prot_rehash(struct sock *sk) 1282 { 1283 sk->sk_prot->unhash(sk); 1284 return sk->sk_prot->hash(sk); 1285 } 1286 1287 /* About 10 seconds */ 1288 #define SOCK_DESTROY_TIME (10*HZ) 1289 1290 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1291 #define PROT_SOCK 1024 1292 1293 #define SHUTDOWN_MASK 3 1294 #define RCV_SHUTDOWN 1 1295 #define SEND_SHUTDOWN 2 1296 1297 #define SOCK_SNDBUF_LOCK 1 1298 #define SOCK_RCVBUF_LOCK 2 1299 #define SOCK_BINDADDR_LOCK 4 1300 #define SOCK_BINDPORT_LOCK 8 1301 1302 struct socket_alloc { 1303 struct socket socket; 1304 struct inode vfs_inode; 1305 }; 1306 1307 static inline struct socket *SOCKET_I(struct inode *inode) 1308 { 1309 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1310 } 1311 1312 static inline struct inode *SOCK_INODE(struct socket *socket) 1313 { 1314 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1315 } 1316 1317 /* 1318 * Functions for memory accounting 1319 */ 1320 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1321 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1322 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1323 void __sk_mem_reclaim(struct sock *sk, int amount); 1324 1325 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1326 * do not necessarily have 16x time more memory than 4KB ones. 1327 */ 1328 #define SK_MEM_QUANTUM 4096 1329 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1330 #define SK_MEM_SEND 0 1331 #define SK_MEM_RECV 1 1332 1333 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1334 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1335 { 1336 long val = sk->sk_prot->sysctl_mem[index]; 1337 1338 #if PAGE_SIZE > SK_MEM_QUANTUM 1339 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1340 #elif PAGE_SIZE < SK_MEM_QUANTUM 1341 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1342 #endif 1343 return val; 1344 } 1345 1346 static inline int sk_mem_pages(int amt) 1347 { 1348 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1349 } 1350 1351 static inline bool sk_has_account(struct sock *sk) 1352 { 1353 /* return true if protocol supports memory accounting */ 1354 return !!sk->sk_prot->memory_allocated; 1355 } 1356 1357 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1358 { 1359 if (!sk_has_account(sk)) 1360 return true; 1361 return size <= sk->sk_forward_alloc || 1362 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1363 } 1364 1365 static inline bool 1366 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1367 { 1368 if (!sk_has_account(sk)) 1369 return true; 1370 return size<= sk->sk_forward_alloc || 1371 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1372 skb_pfmemalloc(skb); 1373 } 1374 1375 static inline void sk_mem_reclaim(struct sock *sk) 1376 { 1377 if (!sk_has_account(sk)) 1378 return; 1379 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1380 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1381 } 1382 1383 static inline void sk_mem_reclaim_partial(struct sock *sk) 1384 { 1385 if (!sk_has_account(sk)) 1386 return; 1387 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1388 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1389 } 1390 1391 static inline void sk_mem_charge(struct sock *sk, int size) 1392 { 1393 if (!sk_has_account(sk)) 1394 return; 1395 sk->sk_forward_alloc -= size; 1396 } 1397 1398 static inline void sk_mem_uncharge(struct sock *sk, int size) 1399 { 1400 if (!sk_has_account(sk)) 1401 return; 1402 sk->sk_forward_alloc += size; 1403 1404 /* Avoid a possible overflow. 1405 * TCP send queues can make this happen, if sk_mem_reclaim() 1406 * is not called and more than 2 GBytes are released at once. 1407 * 1408 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1409 * no need to hold that much forward allocation anyway. 1410 */ 1411 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1412 __sk_mem_reclaim(sk, 1 << 20); 1413 } 1414 1415 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1416 { 1417 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1418 sk->sk_wmem_queued -= skb->truesize; 1419 sk_mem_uncharge(sk, skb->truesize); 1420 __kfree_skb(skb); 1421 } 1422 1423 static inline void sock_release_ownership(struct sock *sk) 1424 { 1425 if (sk->sk_lock.owned) { 1426 sk->sk_lock.owned = 0; 1427 1428 /* The sk_lock has mutex_unlock() semantics: */ 1429 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1430 } 1431 } 1432 1433 /* 1434 * Macro so as to not evaluate some arguments when 1435 * lockdep is not enabled. 1436 * 1437 * Mark both the sk_lock and the sk_lock.slock as a 1438 * per-address-family lock class. 1439 */ 1440 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1441 do { \ 1442 sk->sk_lock.owned = 0; \ 1443 init_waitqueue_head(&sk->sk_lock.wq); \ 1444 spin_lock_init(&(sk)->sk_lock.slock); \ 1445 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1446 sizeof((sk)->sk_lock)); \ 1447 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1448 (skey), (sname)); \ 1449 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1450 } while (0) 1451 1452 #ifdef CONFIG_LOCKDEP 1453 static inline bool lockdep_sock_is_held(const struct sock *sk) 1454 { 1455 return lockdep_is_held(&sk->sk_lock) || 1456 lockdep_is_held(&sk->sk_lock.slock); 1457 } 1458 #endif 1459 1460 void lock_sock_nested(struct sock *sk, int subclass); 1461 1462 static inline void lock_sock(struct sock *sk) 1463 { 1464 lock_sock_nested(sk, 0); 1465 } 1466 1467 void release_sock(struct sock *sk); 1468 1469 /* BH context may only use the following locking interface. */ 1470 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1471 #define bh_lock_sock_nested(__sk) \ 1472 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1473 SINGLE_DEPTH_NESTING) 1474 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1475 1476 bool lock_sock_fast(struct sock *sk); 1477 /** 1478 * unlock_sock_fast - complement of lock_sock_fast 1479 * @sk: socket 1480 * @slow: slow mode 1481 * 1482 * fast unlock socket for user context. 1483 * If slow mode is on, we call regular release_sock() 1484 */ 1485 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1486 { 1487 if (slow) 1488 release_sock(sk); 1489 else 1490 spin_unlock_bh(&sk->sk_lock.slock); 1491 } 1492 1493 /* Used by processes to "lock" a socket state, so that 1494 * interrupts and bottom half handlers won't change it 1495 * from under us. It essentially blocks any incoming 1496 * packets, so that we won't get any new data or any 1497 * packets that change the state of the socket. 1498 * 1499 * While locked, BH processing will add new packets to 1500 * the backlog queue. This queue is processed by the 1501 * owner of the socket lock right before it is released. 1502 * 1503 * Since ~2.3.5 it is also exclusive sleep lock serializing 1504 * accesses from user process context. 1505 */ 1506 1507 static inline void sock_owned_by_me(const struct sock *sk) 1508 { 1509 #ifdef CONFIG_LOCKDEP 1510 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1511 #endif 1512 } 1513 1514 static inline bool sock_owned_by_user(const struct sock *sk) 1515 { 1516 sock_owned_by_me(sk); 1517 return sk->sk_lock.owned; 1518 } 1519 1520 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1521 { 1522 return sk->sk_lock.owned; 1523 } 1524 1525 /* no reclassification while locks are held */ 1526 static inline bool sock_allow_reclassification(const struct sock *csk) 1527 { 1528 struct sock *sk = (struct sock *)csk; 1529 1530 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1531 } 1532 1533 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1534 struct proto *prot, int kern); 1535 void sk_free(struct sock *sk); 1536 void sk_destruct(struct sock *sk); 1537 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1538 void sk_free_unlock_clone(struct sock *sk); 1539 1540 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1541 gfp_t priority); 1542 void __sock_wfree(struct sk_buff *skb); 1543 void sock_wfree(struct sk_buff *skb); 1544 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1545 gfp_t priority); 1546 void skb_orphan_partial(struct sk_buff *skb); 1547 void sock_rfree(struct sk_buff *skb); 1548 void sock_efree(struct sk_buff *skb); 1549 #ifdef CONFIG_INET 1550 void sock_edemux(struct sk_buff *skb); 1551 #else 1552 #define sock_edemux sock_efree 1553 #endif 1554 1555 int sock_setsockopt(struct socket *sock, int level, int op, 1556 char __user *optval, unsigned int optlen); 1557 1558 int sock_getsockopt(struct socket *sock, int level, int op, 1559 char __user *optval, int __user *optlen); 1560 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1561 int noblock, int *errcode); 1562 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1563 unsigned long data_len, int noblock, 1564 int *errcode, int max_page_order); 1565 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1566 void sock_kfree_s(struct sock *sk, void *mem, int size); 1567 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1568 void sk_send_sigurg(struct sock *sk); 1569 1570 struct sockcm_cookie { 1571 u32 mark; 1572 u16 tsflags; 1573 }; 1574 1575 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1576 struct sockcm_cookie *sockc); 1577 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1578 struct sockcm_cookie *sockc); 1579 1580 /* 1581 * Functions to fill in entries in struct proto_ops when a protocol 1582 * does not implement a particular function. 1583 */ 1584 int sock_no_bind(struct socket *, struct sockaddr *, int); 1585 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1586 int sock_no_socketpair(struct socket *, struct socket *); 1587 int sock_no_accept(struct socket *, struct socket *, int, bool); 1588 int sock_no_getname(struct socket *, struct sockaddr *, int); 1589 __poll_t sock_no_poll(struct file *, struct socket *, 1590 struct poll_table_struct *); 1591 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1592 int sock_no_listen(struct socket *, int); 1593 int sock_no_shutdown(struct socket *, int); 1594 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1595 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1596 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1597 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1598 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1599 int sock_no_mmap(struct file *file, struct socket *sock, 1600 struct vm_area_struct *vma); 1601 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1602 size_t size, int flags); 1603 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1604 int offset, size_t size, int flags); 1605 1606 /* 1607 * Functions to fill in entries in struct proto_ops when a protocol 1608 * uses the inet style. 1609 */ 1610 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1611 char __user *optval, int __user *optlen); 1612 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1613 int flags); 1614 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1615 char __user *optval, unsigned int optlen); 1616 int compat_sock_common_getsockopt(struct socket *sock, int level, 1617 int optname, char __user *optval, int __user *optlen); 1618 int compat_sock_common_setsockopt(struct socket *sock, int level, 1619 int optname, char __user *optval, unsigned int optlen); 1620 1621 void sk_common_release(struct sock *sk); 1622 1623 /* 1624 * Default socket callbacks and setup code 1625 */ 1626 1627 /* Initialise core socket variables */ 1628 void sock_init_data(struct socket *sock, struct sock *sk); 1629 1630 /* 1631 * Socket reference counting postulates. 1632 * 1633 * * Each user of socket SHOULD hold a reference count. 1634 * * Each access point to socket (an hash table bucket, reference from a list, 1635 * running timer, skb in flight MUST hold a reference count. 1636 * * When reference count hits 0, it means it will never increase back. 1637 * * When reference count hits 0, it means that no references from 1638 * outside exist to this socket and current process on current CPU 1639 * is last user and may/should destroy this socket. 1640 * * sk_free is called from any context: process, BH, IRQ. When 1641 * it is called, socket has no references from outside -> sk_free 1642 * may release descendant resources allocated by the socket, but 1643 * to the time when it is called, socket is NOT referenced by any 1644 * hash tables, lists etc. 1645 * * Packets, delivered from outside (from network or from another process) 1646 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1647 * when they sit in queue. Otherwise, packets will leak to hole, when 1648 * socket is looked up by one cpu and unhasing is made by another CPU. 1649 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1650 * (leak to backlog). Packet socket does all the processing inside 1651 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1652 * use separate SMP lock, so that they are prone too. 1653 */ 1654 1655 /* Ungrab socket and destroy it, if it was the last reference. */ 1656 static inline void sock_put(struct sock *sk) 1657 { 1658 if (refcount_dec_and_test(&sk->sk_refcnt)) 1659 sk_free(sk); 1660 } 1661 /* Generic version of sock_put(), dealing with all sockets 1662 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1663 */ 1664 void sock_gen_put(struct sock *sk); 1665 1666 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1667 unsigned int trim_cap, bool refcounted); 1668 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1669 const int nested) 1670 { 1671 return __sk_receive_skb(sk, skb, nested, 1, true); 1672 } 1673 1674 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1675 { 1676 sk->sk_tx_queue_mapping = tx_queue; 1677 } 1678 1679 static inline void sk_tx_queue_clear(struct sock *sk) 1680 { 1681 sk->sk_tx_queue_mapping = -1; 1682 } 1683 1684 static inline int sk_tx_queue_get(const struct sock *sk) 1685 { 1686 return sk ? sk->sk_tx_queue_mapping : -1; 1687 } 1688 1689 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1690 { 1691 sk_tx_queue_clear(sk); 1692 sk->sk_socket = sock; 1693 } 1694 1695 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1696 { 1697 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1698 return &rcu_dereference_raw(sk->sk_wq)->wait; 1699 } 1700 /* Detach socket from process context. 1701 * Announce socket dead, detach it from wait queue and inode. 1702 * Note that parent inode held reference count on this struct sock, 1703 * we do not release it in this function, because protocol 1704 * probably wants some additional cleanups or even continuing 1705 * to work with this socket (TCP). 1706 */ 1707 static inline void sock_orphan(struct sock *sk) 1708 { 1709 write_lock_bh(&sk->sk_callback_lock); 1710 sock_set_flag(sk, SOCK_DEAD); 1711 sk_set_socket(sk, NULL); 1712 sk->sk_wq = NULL; 1713 write_unlock_bh(&sk->sk_callback_lock); 1714 } 1715 1716 static inline void sock_graft(struct sock *sk, struct socket *parent) 1717 { 1718 WARN_ON(parent->sk); 1719 write_lock_bh(&sk->sk_callback_lock); 1720 sk->sk_wq = parent->wq; 1721 parent->sk = sk; 1722 sk_set_socket(sk, parent); 1723 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1724 security_sock_graft(sk, parent); 1725 write_unlock_bh(&sk->sk_callback_lock); 1726 } 1727 1728 kuid_t sock_i_uid(struct sock *sk); 1729 unsigned long sock_i_ino(struct sock *sk); 1730 1731 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1732 { 1733 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1734 } 1735 1736 static inline u32 net_tx_rndhash(void) 1737 { 1738 u32 v = prandom_u32(); 1739 1740 return v ?: 1; 1741 } 1742 1743 static inline void sk_set_txhash(struct sock *sk) 1744 { 1745 sk->sk_txhash = net_tx_rndhash(); 1746 } 1747 1748 static inline void sk_rethink_txhash(struct sock *sk) 1749 { 1750 if (sk->sk_txhash) 1751 sk_set_txhash(sk); 1752 } 1753 1754 static inline struct dst_entry * 1755 __sk_dst_get(struct sock *sk) 1756 { 1757 return rcu_dereference_check(sk->sk_dst_cache, 1758 lockdep_sock_is_held(sk)); 1759 } 1760 1761 static inline struct dst_entry * 1762 sk_dst_get(struct sock *sk) 1763 { 1764 struct dst_entry *dst; 1765 1766 rcu_read_lock(); 1767 dst = rcu_dereference(sk->sk_dst_cache); 1768 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1769 dst = NULL; 1770 rcu_read_unlock(); 1771 return dst; 1772 } 1773 1774 static inline void dst_negative_advice(struct sock *sk) 1775 { 1776 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1777 1778 sk_rethink_txhash(sk); 1779 1780 if (dst && dst->ops->negative_advice) { 1781 ndst = dst->ops->negative_advice(dst); 1782 1783 if (ndst != dst) { 1784 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1785 sk_tx_queue_clear(sk); 1786 sk->sk_dst_pending_confirm = 0; 1787 } 1788 } 1789 } 1790 1791 static inline void 1792 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1793 { 1794 struct dst_entry *old_dst; 1795 1796 sk_tx_queue_clear(sk); 1797 sk->sk_dst_pending_confirm = 0; 1798 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1799 lockdep_sock_is_held(sk)); 1800 rcu_assign_pointer(sk->sk_dst_cache, dst); 1801 dst_release(old_dst); 1802 } 1803 1804 static inline void 1805 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1806 { 1807 struct dst_entry *old_dst; 1808 1809 sk_tx_queue_clear(sk); 1810 sk->sk_dst_pending_confirm = 0; 1811 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1812 dst_release(old_dst); 1813 } 1814 1815 static inline void 1816 __sk_dst_reset(struct sock *sk) 1817 { 1818 __sk_dst_set(sk, NULL); 1819 } 1820 1821 static inline void 1822 sk_dst_reset(struct sock *sk) 1823 { 1824 sk_dst_set(sk, NULL); 1825 } 1826 1827 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1828 1829 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1830 1831 static inline void sk_dst_confirm(struct sock *sk) 1832 { 1833 if (!sk->sk_dst_pending_confirm) 1834 sk->sk_dst_pending_confirm = 1; 1835 } 1836 1837 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1838 { 1839 if (skb_get_dst_pending_confirm(skb)) { 1840 struct sock *sk = skb->sk; 1841 unsigned long now = jiffies; 1842 1843 /* avoid dirtying neighbour */ 1844 if (n->confirmed != now) 1845 n->confirmed = now; 1846 if (sk && sk->sk_dst_pending_confirm) 1847 sk->sk_dst_pending_confirm = 0; 1848 } 1849 } 1850 1851 bool sk_mc_loop(struct sock *sk); 1852 1853 static inline bool sk_can_gso(const struct sock *sk) 1854 { 1855 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1856 } 1857 1858 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1859 1860 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1861 { 1862 sk->sk_route_nocaps |= flags; 1863 sk->sk_route_caps &= ~flags; 1864 } 1865 1866 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1867 struct iov_iter *from, char *to, 1868 int copy, int offset) 1869 { 1870 if (skb->ip_summed == CHECKSUM_NONE) { 1871 __wsum csum = 0; 1872 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1873 return -EFAULT; 1874 skb->csum = csum_block_add(skb->csum, csum, offset); 1875 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1876 if (!copy_from_iter_full_nocache(to, copy, from)) 1877 return -EFAULT; 1878 } else if (!copy_from_iter_full(to, copy, from)) 1879 return -EFAULT; 1880 1881 return 0; 1882 } 1883 1884 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1885 struct iov_iter *from, int copy) 1886 { 1887 int err, offset = skb->len; 1888 1889 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1890 copy, offset); 1891 if (err) 1892 __skb_trim(skb, offset); 1893 1894 return err; 1895 } 1896 1897 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1898 struct sk_buff *skb, 1899 struct page *page, 1900 int off, int copy) 1901 { 1902 int err; 1903 1904 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1905 copy, skb->len); 1906 if (err) 1907 return err; 1908 1909 skb->len += copy; 1910 skb->data_len += copy; 1911 skb->truesize += copy; 1912 sk->sk_wmem_queued += copy; 1913 sk_mem_charge(sk, copy); 1914 return 0; 1915 } 1916 1917 /** 1918 * sk_wmem_alloc_get - returns write allocations 1919 * @sk: socket 1920 * 1921 * Returns sk_wmem_alloc minus initial offset of one 1922 */ 1923 static inline int sk_wmem_alloc_get(const struct sock *sk) 1924 { 1925 return refcount_read(&sk->sk_wmem_alloc) - 1; 1926 } 1927 1928 /** 1929 * sk_rmem_alloc_get - returns read allocations 1930 * @sk: socket 1931 * 1932 * Returns sk_rmem_alloc 1933 */ 1934 static inline int sk_rmem_alloc_get(const struct sock *sk) 1935 { 1936 return atomic_read(&sk->sk_rmem_alloc); 1937 } 1938 1939 /** 1940 * sk_has_allocations - check if allocations are outstanding 1941 * @sk: socket 1942 * 1943 * Returns true if socket has write or read allocations 1944 */ 1945 static inline bool sk_has_allocations(const struct sock *sk) 1946 { 1947 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1948 } 1949 1950 /** 1951 * skwq_has_sleeper - check if there are any waiting processes 1952 * @wq: struct socket_wq 1953 * 1954 * Returns true if socket_wq has waiting processes 1955 * 1956 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1957 * barrier call. They were added due to the race found within the tcp code. 1958 * 1959 * Consider following tcp code paths:: 1960 * 1961 * CPU1 CPU2 1962 * sys_select receive packet 1963 * ... ... 1964 * __add_wait_queue update tp->rcv_nxt 1965 * ... ... 1966 * tp->rcv_nxt check sock_def_readable 1967 * ... { 1968 * schedule rcu_read_lock(); 1969 * wq = rcu_dereference(sk->sk_wq); 1970 * if (wq && waitqueue_active(&wq->wait)) 1971 * wake_up_interruptible(&wq->wait) 1972 * ... 1973 * } 1974 * 1975 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1976 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1977 * could then endup calling schedule and sleep forever if there are no more 1978 * data on the socket. 1979 * 1980 */ 1981 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1982 { 1983 return wq && wq_has_sleeper(&wq->wait); 1984 } 1985 1986 /** 1987 * sock_poll_wait - place memory barrier behind the poll_wait call. 1988 * @filp: file 1989 * @wait_address: socket wait queue 1990 * @p: poll_table 1991 * 1992 * See the comments in the wq_has_sleeper function. 1993 */ 1994 static inline void sock_poll_wait(struct file *filp, 1995 wait_queue_head_t *wait_address, poll_table *p) 1996 { 1997 if (!poll_does_not_wait(p) && wait_address) { 1998 poll_wait(filp, wait_address, p); 1999 /* We need to be sure we are in sync with the 2000 * socket flags modification. 2001 * 2002 * This memory barrier is paired in the wq_has_sleeper. 2003 */ 2004 smp_mb(); 2005 } 2006 } 2007 2008 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2009 { 2010 if (sk->sk_txhash) { 2011 skb->l4_hash = 1; 2012 skb->hash = sk->sk_txhash; 2013 } 2014 } 2015 2016 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2017 2018 /* 2019 * Queue a received datagram if it will fit. Stream and sequenced 2020 * protocols can't normally use this as they need to fit buffers in 2021 * and play with them. 2022 * 2023 * Inlined as it's very short and called for pretty much every 2024 * packet ever received. 2025 */ 2026 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2027 { 2028 skb_orphan(skb); 2029 skb->sk = sk; 2030 skb->destructor = sock_rfree; 2031 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2032 sk_mem_charge(sk, skb->truesize); 2033 } 2034 2035 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2036 unsigned long expires); 2037 2038 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2039 2040 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2041 struct sk_buff *skb, unsigned int flags, 2042 void (*destructor)(struct sock *sk, 2043 struct sk_buff *skb)); 2044 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2045 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2046 2047 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2048 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2049 2050 /* 2051 * Recover an error report and clear atomically 2052 */ 2053 2054 static inline int sock_error(struct sock *sk) 2055 { 2056 int err; 2057 if (likely(!sk->sk_err)) 2058 return 0; 2059 err = xchg(&sk->sk_err, 0); 2060 return -err; 2061 } 2062 2063 static inline unsigned long sock_wspace(struct sock *sk) 2064 { 2065 int amt = 0; 2066 2067 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2068 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2069 if (amt < 0) 2070 amt = 0; 2071 } 2072 return amt; 2073 } 2074 2075 /* Note: 2076 * We use sk->sk_wq_raw, from contexts knowing this 2077 * pointer is not NULL and cannot disappear/change. 2078 */ 2079 static inline void sk_set_bit(int nr, struct sock *sk) 2080 { 2081 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2082 !sock_flag(sk, SOCK_FASYNC)) 2083 return; 2084 2085 set_bit(nr, &sk->sk_wq_raw->flags); 2086 } 2087 2088 static inline void sk_clear_bit(int nr, struct sock *sk) 2089 { 2090 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2091 !sock_flag(sk, SOCK_FASYNC)) 2092 return; 2093 2094 clear_bit(nr, &sk->sk_wq_raw->flags); 2095 } 2096 2097 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2098 { 2099 if (sock_flag(sk, SOCK_FASYNC)) { 2100 rcu_read_lock(); 2101 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2102 rcu_read_unlock(); 2103 } 2104 } 2105 2106 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2107 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2108 * Note: for send buffers, TCP works better if we can build two skbs at 2109 * minimum. 2110 */ 2111 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2112 2113 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2114 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2115 2116 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2117 { 2118 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2119 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2120 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2121 } 2122 } 2123 2124 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2125 bool force_schedule); 2126 2127 /** 2128 * sk_page_frag - return an appropriate page_frag 2129 * @sk: socket 2130 * 2131 * If socket allocation mode allows current thread to sleep, it means its 2132 * safe to use the per task page_frag instead of the per socket one. 2133 */ 2134 static inline struct page_frag *sk_page_frag(struct sock *sk) 2135 { 2136 if (gfpflags_allow_blocking(sk->sk_allocation)) 2137 return ¤t->task_frag; 2138 2139 return &sk->sk_frag; 2140 } 2141 2142 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2143 2144 /* 2145 * Default write policy as shown to user space via poll/select/SIGIO 2146 */ 2147 static inline bool sock_writeable(const struct sock *sk) 2148 { 2149 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2150 } 2151 2152 static inline gfp_t gfp_any(void) 2153 { 2154 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2155 } 2156 2157 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2158 { 2159 return noblock ? 0 : sk->sk_rcvtimeo; 2160 } 2161 2162 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2163 { 2164 return noblock ? 0 : sk->sk_sndtimeo; 2165 } 2166 2167 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2168 { 2169 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2170 } 2171 2172 /* Alas, with timeout socket operations are not restartable. 2173 * Compare this to poll(). 2174 */ 2175 static inline int sock_intr_errno(long timeo) 2176 { 2177 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2178 } 2179 2180 struct sock_skb_cb { 2181 u32 dropcount; 2182 }; 2183 2184 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2185 * using skb->cb[] would keep using it directly and utilize its 2186 * alignement guarantee. 2187 */ 2188 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2189 sizeof(struct sock_skb_cb))) 2190 2191 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2192 SOCK_SKB_CB_OFFSET)) 2193 2194 #define sock_skb_cb_check_size(size) \ 2195 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2196 2197 static inline void 2198 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2199 { 2200 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2201 atomic_read(&sk->sk_drops) : 0; 2202 } 2203 2204 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2205 { 2206 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2207 2208 atomic_add(segs, &sk->sk_drops); 2209 } 2210 2211 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2212 struct sk_buff *skb); 2213 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2214 struct sk_buff *skb); 2215 2216 static inline void 2217 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2218 { 2219 ktime_t kt = skb->tstamp; 2220 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2221 2222 /* 2223 * generate control messages if 2224 * - receive time stamping in software requested 2225 * - software time stamp available and wanted 2226 * - hardware time stamps available and wanted 2227 */ 2228 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2229 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2230 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2231 (hwtstamps->hwtstamp && 2232 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2233 __sock_recv_timestamp(msg, sk, skb); 2234 else 2235 sk->sk_stamp = kt; 2236 2237 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2238 __sock_recv_wifi_status(msg, sk, skb); 2239 } 2240 2241 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2242 struct sk_buff *skb); 2243 2244 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2245 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2246 struct sk_buff *skb) 2247 { 2248 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2249 (1UL << SOCK_RCVTSTAMP)) 2250 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2251 SOF_TIMESTAMPING_RAW_HARDWARE) 2252 2253 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2254 __sock_recv_ts_and_drops(msg, sk, skb); 2255 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2256 sk->sk_stamp = skb->tstamp; 2257 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2258 sk->sk_stamp = 0; 2259 } 2260 2261 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2262 2263 /** 2264 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2265 * @sk: socket sending this packet 2266 * @tsflags: timestamping flags to use 2267 * @tx_flags: completed with instructions for time stamping 2268 * 2269 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2270 */ 2271 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags, 2272 __u8 *tx_flags) 2273 { 2274 if (unlikely(tsflags)) 2275 __sock_tx_timestamp(tsflags, tx_flags); 2276 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2277 *tx_flags |= SKBTX_WIFI_STATUS; 2278 } 2279 2280 /** 2281 * sk_eat_skb - Release a skb if it is no longer needed 2282 * @sk: socket to eat this skb from 2283 * @skb: socket buffer to eat 2284 * 2285 * This routine must be called with interrupts disabled or with the socket 2286 * locked so that the sk_buff queue operation is ok. 2287 */ 2288 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2289 { 2290 __skb_unlink(skb, &sk->sk_receive_queue); 2291 __kfree_skb(skb); 2292 } 2293 2294 static inline 2295 struct net *sock_net(const struct sock *sk) 2296 { 2297 return read_pnet(&sk->sk_net); 2298 } 2299 2300 static inline 2301 void sock_net_set(struct sock *sk, struct net *net) 2302 { 2303 write_pnet(&sk->sk_net, net); 2304 } 2305 2306 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2307 { 2308 if (skb->sk) { 2309 struct sock *sk = skb->sk; 2310 2311 skb->destructor = NULL; 2312 skb->sk = NULL; 2313 return sk; 2314 } 2315 return NULL; 2316 } 2317 2318 /* This helper checks if a socket is a full socket, 2319 * ie _not_ a timewait or request socket. 2320 */ 2321 static inline bool sk_fullsock(const struct sock *sk) 2322 { 2323 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2324 } 2325 2326 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2327 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2328 */ 2329 static inline bool sk_listener(const struct sock *sk) 2330 { 2331 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2332 } 2333 2334 void sock_enable_timestamp(struct sock *sk, int flag); 2335 int sock_get_timestamp(struct sock *, struct timeval __user *); 2336 int sock_get_timestampns(struct sock *, struct timespec __user *); 2337 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2338 int type); 2339 2340 bool sk_ns_capable(const struct sock *sk, 2341 struct user_namespace *user_ns, int cap); 2342 bool sk_capable(const struct sock *sk, int cap); 2343 bool sk_net_capable(const struct sock *sk, int cap); 2344 2345 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2346 2347 /* Take into consideration the size of the struct sk_buff overhead in the 2348 * determination of these values, since that is non-constant across 2349 * platforms. This makes socket queueing behavior and performance 2350 * not depend upon such differences. 2351 */ 2352 #define _SK_MEM_PACKETS 256 2353 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2354 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2355 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2356 2357 extern __u32 sysctl_wmem_max; 2358 extern __u32 sysctl_rmem_max; 2359 2360 extern int sysctl_tstamp_allow_data; 2361 extern int sysctl_optmem_max; 2362 2363 extern __u32 sysctl_wmem_default; 2364 extern __u32 sysctl_rmem_default; 2365 2366 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2367 { 2368 /* Does this proto have per netns sysctl_wmem ? */ 2369 if (proto->sysctl_wmem_offset) 2370 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2371 2372 return *proto->sysctl_wmem; 2373 } 2374 2375 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2376 { 2377 /* Does this proto have per netns sysctl_rmem ? */ 2378 if (proto->sysctl_rmem_offset) 2379 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2380 2381 return *proto->sysctl_rmem; 2382 } 2383 2384 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2385 * Some wifi drivers need to tweak it to get more chunks. 2386 * They can use this helper from their ndo_start_xmit() 2387 */ 2388 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2389 { 2390 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val) 2391 return; 2392 sk->sk_pacing_shift = val; 2393 } 2394 2395 /* if a socket is bound to a device, check that the given device 2396 * index is either the same or that the socket is bound to an L3 2397 * master device and the given device index is also enslaved to 2398 * that L3 master 2399 */ 2400 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2401 { 2402 int mdif; 2403 2404 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2405 return true; 2406 2407 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2408 if (mdif && mdif == sk->sk_bound_dev_if) 2409 return true; 2410 2411 return false; 2412 } 2413 2414 #endif /* _SOCK_H */ 2415