1/* SPDX-License-Identifier: GPL-2.0-or-later */ 2/* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35#ifndef _SOCK_H 36#define _SOCK_H 37 38#include <linux/hardirq.h> 39#include <linux/kernel.h> 40#include <linux/list.h> 41#include <linux/list_nulls.h> 42#include <linux/timer.h> 43#include <linux/cache.h> 44#include <linux/bitops.h> 45#include <linux/lockdep.h> 46#include <linux/netdevice.h> 47#include <linux/skbuff.h> /* struct sk_buff */ 48#include <linux/mm.h> 49#include <linux/security.h> 50#include <linux/slab.h> 51#include <linux/uaccess.h> 52#include <linux/page_counter.h> 53#include <linux/memcontrol.h> 54#include <linux/static_key.h> 55#include <linux/sched.h> 56#include <linux/wait.h> 57#include <linux/cgroup-defs.h> 58#include <linux/rbtree.h> 59#include <linux/rculist_nulls.h> 60#include <linux/poll.h> 61#include <linux/sockptr.h> 62#include <linux/indirect_call_wrapper.h> 63#include <linux/atomic.h> 64#include <linux/refcount.h> 65#include <linux/llist.h> 66#include <net/dst.h> 67#include <net/checksum.h> 68#include <net/tcp_states.h> 69#include <linux/net_tstamp.h> 70#include <net/l3mdev.h> 71#include <uapi/linux/socket.h> 72 73/* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79/* This is the per-socket lock. The spinlock provides a synchronization 80 * between user contexts and software interrupt processing, whereas the 81 * mini-semaphore synchronizes multiple users amongst themselves. 82 */ 83typedef struct { 84 spinlock_t slock; 85 int owned; 86 wait_queue_head_t wq; 87 /* 88 * We express the mutex-alike socket_lock semantics 89 * to the lock validator by explicitly managing 90 * the slock as a lock variant (in addition to 91 * the slock itself): 92 */ 93#ifdef CONFIG_DEBUG_LOCK_ALLOC 94 struct lockdep_map dep_map; 95#endif 96} socket_lock_t; 97 98struct sock; 99struct proto; 100struct net; 101 102typedef __u32 __bitwise __portpair; 103typedef __u64 __bitwise __addrpair; 104 105/** 106 * struct sock_common - minimal network layer representation of sockets 107 * @skc_daddr: Foreign IPv4 addr 108 * @skc_rcv_saddr: Bound local IPv4 addr 109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 110 * @skc_hash: hash value used with various protocol lookup tables 111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 112 * @skc_dport: placeholder for inet_dport/tw_dport 113 * @skc_num: placeholder for inet_num/tw_num 114 * @skc_portpair: __u32 union of @skc_dport & @skc_num 115 * @skc_family: network address family 116 * @skc_state: Connection state 117 * @skc_reuse: %SO_REUSEADDR setting 118 * @skc_reuseport: %SO_REUSEPORT setting 119 * @skc_ipv6only: socket is IPV6 only 120 * @skc_net_refcnt: socket is using net ref counting 121 * @skc_bound_dev_if: bound device index if != 0 122 * @skc_bind_node: bind hash linkage for various protocol lookup tables 123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 124 * @skc_prot: protocol handlers inside a network family 125 * @skc_net: reference to the network namespace of this socket 126 * @skc_v6_daddr: IPV6 destination address 127 * @skc_v6_rcv_saddr: IPV6 source address 128 * @skc_cookie: socket's cookie value 129 * @skc_node: main hash linkage for various protocol lookup tables 130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 131 * @skc_tx_queue_mapping: tx queue number for this connection 132 * @skc_rx_queue_mapping: rx queue number for this connection 133 * @skc_flags: place holder for sk_flags 134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 136 * @skc_listener: connection request listener socket (aka rsk_listener) 137 * [union with @skc_flags] 138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 139 * [union with @skc_flags] 140 * @skc_incoming_cpu: record/match cpu processing incoming packets 141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 142 * [union with @skc_incoming_cpu] 143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 144 * [union with @skc_incoming_cpu] 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 */ 150struct sock_common { 151 union { 152 __addrpair skc_addrpair; 153 struct { 154 __be32 skc_daddr; 155 __be32 skc_rcv_saddr; 156 }; 157 }; 158 union { 159 unsigned int skc_hash; 160 __u16 skc_u16hashes[2]; 161 }; 162 /* skc_dport && skc_num must be grouped as well */ 163 union { 164 __portpair skc_portpair; 165 struct { 166 __be16 skc_dport; 167 __u16 skc_num; 168 }; 169 }; 170 171 unsigned short skc_family; 172 volatile unsigned char skc_state; 173 unsigned char skc_reuse:4; 174 unsigned char skc_reuseport:1; 175 unsigned char skc_ipv6only:1; 176 unsigned char skc_net_refcnt:1; 177 int skc_bound_dev_if; 178 union { 179 struct hlist_node skc_bind_node; 180 struct hlist_node skc_portaddr_node; 181 }; 182 struct proto *skc_prot; 183 possible_net_t skc_net; 184 185#if IS_ENABLED(CONFIG_IPV6) 186 struct in6_addr skc_v6_daddr; 187 struct in6_addr skc_v6_rcv_saddr; 188#endif 189 190 atomic64_t skc_cookie; 191 192 /* following fields are padding to force 193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 194 * assuming IPV6 is enabled. We use this padding differently 195 * for different kind of 'sockets' 196 */ 197 union { 198 unsigned long skc_flags; 199 struct sock *skc_listener; /* request_sock */ 200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 201 }; 202 /* 203 * fields between dontcopy_begin/dontcopy_end 204 * are not copied in sock_copy() 205 */ 206 /* private: */ 207 int skc_dontcopy_begin[0]; 208 /* public: */ 209 union { 210 struct hlist_node skc_node; 211 struct hlist_nulls_node skc_nulls_node; 212 }; 213 unsigned short skc_tx_queue_mapping; 214#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 215 unsigned short skc_rx_queue_mapping; 216#endif 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 234struct bpf_local_storage; 235struct sk_filter; 236 237/** 238 * struct sock - network layer representation of sockets 239 * @__sk_common: shared layout with inet_timewait_sock 240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 242 * @sk_lock: synchronizer 243 * @sk_kern_sock: True if sock is using kernel lock classes 244 * @sk_rcvbuf: size of receive buffer in bytes 245 * @sk_wq: sock wait queue and async head 246 * @sk_rx_dst: receive input route used by early demux 247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 249 * @sk_dst_cache: destination cache 250 * @sk_dst_pending_confirm: need to confirm neighbour 251 * @sk_policy: flow policy 252 * @sk_receive_queue: incoming packets 253 * @sk_wmem_alloc: transmit queue bytes committed 254 * @sk_tsq_flags: TCP Small Queues flags 255 * @sk_write_queue: Packet sending queue 256 * @sk_omem_alloc: "o" is "option" or "other" 257 * @sk_wmem_queued: persistent queue size 258 * @sk_forward_alloc: space allocated forward 259 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 260 * @sk_napi_id: id of the last napi context to receive data for sk 261 * @sk_ll_usec: usecs to busypoll when there is no data 262 * @sk_allocation: allocation mode 263 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 264 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 265 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 266 * @sk_sndbuf: size of send buffer in bytes 267 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 268 * @sk_no_check_rx: allow zero checksum in RX packets 269 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 270 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 271 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 272 * @sk_gso_max_size: Maximum GSO segment size to build 273 * @sk_gso_max_segs: Maximum number of GSO segments 274 * @sk_pacing_shift: scaling factor for TCP Small Queues 275 * @sk_lingertime: %SO_LINGER l_linger setting 276 * @sk_backlog: always used with the per-socket spinlock held 277 * @sk_callback_lock: used with the callbacks in the end of this struct 278 * @sk_error_queue: rarely used 279 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 280 * IPV6_ADDRFORM for instance) 281 * @sk_err: last error 282 * @sk_err_soft: errors that don't cause failure but are the cause of a 283 * persistent failure not just 'timed out' 284 * @sk_drops: raw/udp drops counter 285 * @sk_ack_backlog: current listen backlog 286 * @sk_max_ack_backlog: listen backlog set in listen() 287 * @sk_uid: user id of owner 288 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 289 * @sk_busy_poll_budget: napi processing budget when busypolling 290 * @sk_priority: %SO_PRIORITY setting 291 * @sk_type: socket type (%SOCK_STREAM, etc) 292 * @sk_protocol: which protocol this socket belongs in this network family 293 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 294 * @sk_peer_pid: &struct pid for this socket's peer 295 * @sk_peer_cred: %SO_PEERCRED setting 296 * @sk_rcvlowat: %SO_RCVLOWAT setting 297 * @sk_rcvtimeo: %SO_RCVTIMEO setting 298 * @sk_sndtimeo: %SO_SNDTIMEO setting 299 * @sk_txhash: computed flow hash for use on transmit 300 * @sk_txrehash: enable TX hash rethink 301 * @sk_filter: socket filtering instructions 302 * @sk_timer: sock cleanup timer 303 * @sk_stamp: time stamp of last packet received 304 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 305 * @sk_tsflags: SO_TIMESTAMPING flags 306 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 307 * Sockets that can be used under memory reclaim should 308 * set this to false. 309 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 310 * for timestamping 311 * @sk_tskey: counter to disambiguate concurrent tstamp requests 312 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 313 * @sk_socket: Identd and reporting IO signals 314 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 315 * @sk_frag: cached page frag 316 * @sk_peek_off: current peek_offset value 317 * @sk_send_head: front of stuff to transmit 318 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 319 * @sk_security: used by security modules 320 * @sk_mark: generic packet mark 321 * @sk_cgrp_data: cgroup data for this cgroup 322 * @sk_memcg: this socket's memory cgroup association 323 * @sk_write_pending: a write to stream socket waits to start 324 * @sk_disconnects: number of disconnect operations performed on this sock 325 * @sk_state_change: callback to indicate change in the state of the sock 326 * @sk_data_ready: callback to indicate there is data to be processed 327 * @sk_write_space: callback to indicate there is bf sending space available 328 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 329 * @sk_backlog_rcv: callback to process the backlog 330 * @sk_validate_xmit_skb: ptr to an optional validate function 331 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 332 * @sk_reuseport_cb: reuseport group container 333 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 334 * @sk_rcu: used during RCU grace period 335 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 336 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 337 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 338 * @sk_txtime_unused: unused txtime flags 339 * @ns_tracker: tracker for netns reference 340 */ 341struct sock { 342 /* 343 * Now struct inet_timewait_sock also uses sock_common, so please just 344 * don't add nothing before this first member (__sk_common) --acme 345 */ 346 struct sock_common __sk_common; 347#define sk_node __sk_common.skc_node 348#define sk_nulls_node __sk_common.skc_nulls_node 349#define sk_refcnt __sk_common.skc_refcnt 350#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 351#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 352#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 353#endif 354 355#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 356#define sk_dontcopy_end __sk_common.skc_dontcopy_end 357#define sk_hash __sk_common.skc_hash 358#define sk_portpair __sk_common.skc_portpair 359#define sk_num __sk_common.skc_num 360#define sk_dport __sk_common.skc_dport 361#define sk_addrpair __sk_common.skc_addrpair 362#define sk_daddr __sk_common.skc_daddr 363#define sk_rcv_saddr __sk_common.skc_rcv_saddr 364#define sk_family __sk_common.skc_family 365#define sk_state __sk_common.skc_state 366#define sk_reuse __sk_common.skc_reuse 367#define sk_reuseport __sk_common.skc_reuseport 368#define sk_ipv6only __sk_common.skc_ipv6only 369#define sk_net_refcnt __sk_common.skc_net_refcnt 370#define sk_bound_dev_if __sk_common.skc_bound_dev_if 371#define sk_bind_node __sk_common.skc_bind_node 372#define sk_prot __sk_common.skc_prot 373#define sk_net __sk_common.skc_net 374#define sk_v6_daddr __sk_common.skc_v6_daddr 375#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 376#define sk_cookie __sk_common.skc_cookie 377#define sk_incoming_cpu __sk_common.skc_incoming_cpu 378#define sk_flags __sk_common.skc_flags 379#define sk_rxhash __sk_common.skc_rxhash 380 381 __cacheline_group_begin(sock_write_rx); 382 383 atomic_t sk_drops; 384 __s32 sk_peek_off; 385 struct sk_buff_head sk_error_queue; 386 struct sk_buff_head sk_receive_queue; 387 /* 388 * The backlog queue is special, it is always used with 389 * the per-socket spinlock held and requires low latency 390 * access. Therefore we special case it's implementation. 391 * Note : rmem_alloc is in this structure to fill a hole 392 * on 64bit arches, not because its logically part of 393 * backlog. 394 */ 395 struct { 396 atomic_t rmem_alloc; 397 int len; 398 struct sk_buff *head; 399 struct sk_buff *tail; 400 } sk_backlog; 401#define sk_rmem_alloc sk_backlog.rmem_alloc 402 403 __cacheline_group_end(sock_write_rx); 404 405 __cacheline_group_begin(sock_read_rx); 406 /* early demux fields */ 407 struct dst_entry __rcu *sk_rx_dst; 408 int sk_rx_dst_ifindex; 409 u32 sk_rx_dst_cookie; 410 411#ifdef CONFIG_NET_RX_BUSY_POLL 412 unsigned int sk_ll_usec; 413 unsigned int sk_napi_id; 414 u16 sk_busy_poll_budget; 415 u8 sk_prefer_busy_poll; 416#endif 417 u8 sk_userlocks; 418 int sk_rcvbuf; 419 420 struct sk_filter __rcu *sk_filter; 421 union { 422 struct socket_wq __rcu *sk_wq; 423 /* private: */ 424 struct socket_wq *sk_wq_raw; 425 /* public: */ 426 }; 427 428 void (*sk_data_ready)(struct sock *sk); 429 long sk_rcvtimeo; 430 int sk_rcvlowat; 431 __cacheline_group_end(sock_read_rx); 432 433 __cacheline_group_begin(sock_read_rxtx); 434 int sk_err; 435 struct socket *sk_socket; 436 struct mem_cgroup *sk_memcg; 437#ifdef CONFIG_XFRM 438 struct xfrm_policy __rcu *sk_policy[2]; 439#endif 440 __cacheline_group_end(sock_read_rxtx); 441 442 __cacheline_group_begin(sock_write_rxtx); 443 socket_lock_t sk_lock; 444 u32 sk_reserved_mem; 445 int sk_forward_alloc; 446 u32 sk_tsflags; 447 __cacheline_group_end(sock_write_rxtx); 448 449 __cacheline_group_begin(sock_write_tx); 450 int sk_write_pending; 451 atomic_t sk_omem_alloc; 452 int sk_sndbuf; 453 454 int sk_wmem_queued; 455 refcount_t sk_wmem_alloc; 456 unsigned long sk_tsq_flags; 457 union { 458 struct sk_buff *sk_send_head; 459 struct rb_root tcp_rtx_queue; 460 }; 461 struct sk_buff_head sk_write_queue; 462 u32 sk_dst_pending_confirm; 463 u32 sk_pacing_status; /* see enum sk_pacing */ 464 struct page_frag sk_frag; 465 struct timer_list sk_timer; 466 467 unsigned long sk_pacing_rate; /* bytes per second */ 468 atomic_t sk_zckey; 469 atomic_t sk_tskey; 470 __cacheline_group_end(sock_write_tx); 471 472 __cacheline_group_begin(sock_read_tx); 473 unsigned long sk_max_pacing_rate; 474 long sk_sndtimeo; 475 u32 sk_priority; 476 u32 sk_mark; 477 struct dst_entry __rcu *sk_dst_cache; 478 netdev_features_t sk_route_caps; 479#ifdef CONFIG_SOCK_VALIDATE_XMIT 480 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 481 struct net_device *dev, 482 struct sk_buff *skb); 483#endif 484 u16 sk_gso_type; 485 u16 sk_gso_max_segs; 486 unsigned int sk_gso_max_size; 487 gfp_t sk_allocation; 488 u32 sk_txhash; 489 u8 sk_pacing_shift; 490 bool sk_use_task_frag; 491 __cacheline_group_end(sock_read_tx); 492 493 /* 494 * Because of non atomicity rules, all 495 * changes are protected by socket lock. 496 */ 497 u8 sk_gso_disabled : 1, 498 sk_kern_sock : 1, 499 sk_no_check_tx : 1, 500 sk_no_check_rx : 1; 501 u8 sk_shutdown; 502 u16 sk_type; 503 u16 sk_protocol; 504 unsigned long sk_lingertime; 505 struct proto *sk_prot_creator; 506 rwlock_t sk_callback_lock; 507 int sk_err_soft; 508 u32 sk_ack_backlog; 509 u32 sk_max_ack_backlog; 510 kuid_t sk_uid; 511 spinlock_t sk_peer_lock; 512 int sk_bind_phc; 513 struct pid *sk_peer_pid; 514 const struct cred *sk_peer_cred; 515 516 ktime_t sk_stamp; 517#if BITS_PER_LONG==32 518 seqlock_t sk_stamp_seq; 519#endif 520 int sk_disconnects; 521 522 u8 sk_txrehash; 523 u8 sk_clockid; 524 u8 sk_txtime_deadline_mode : 1, 525 sk_txtime_report_errors : 1, 526 sk_txtime_unused : 6; 527 528 void *sk_user_data; 529#ifdef CONFIG_SECURITY 530 void *sk_security; 531#endif 532 struct sock_cgroup_data sk_cgrp_data; 533 void (*sk_state_change)(struct sock *sk); 534 void (*sk_write_space)(struct sock *sk); 535 void (*sk_error_report)(struct sock *sk); 536 int (*sk_backlog_rcv)(struct sock *sk, 537 struct sk_buff *skb); 538 void (*sk_destruct)(struct sock *sk); 539 struct sock_reuseport __rcu *sk_reuseport_cb; 540#ifdef CONFIG_BPF_SYSCALL 541 struct bpf_local_storage __rcu *sk_bpf_storage; 542#endif 543 struct rcu_head sk_rcu; 544 netns_tracker ns_tracker; 545}; 546 547enum sk_pacing { 548 SK_PACING_NONE = 0, 549 SK_PACING_NEEDED = 1, 550 SK_PACING_FQ = 2, 551}; 552 553/* flag bits in sk_user_data 554 * 555 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 556 * not be suitable for copying when cloning the socket. For instance, 557 * it can point to a reference counted object. sk_user_data bottom 558 * bit is set if pointer must not be copied. 559 * 560 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 561 * managed/owned by a BPF reuseport array. This bit should be set 562 * when sk_user_data's sk is added to the bpf's reuseport_array. 563 * 564 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 565 * sk_user_data points to psock type. This bit should be set 566 * when sk_user_data is assigned to a psock object. 567 */ 568#define SK_USER_DATA_NOCOPY 1UL 569#define SK_USER_DATA_BPF 2UL 570#define SK_USER_DATA_PSOCK 4UL 571#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 572 SK_USER_DATA_PSOCK) 573 574/** 575 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 576 * @sk: socket 577 */ 578static inline bool sk_user_data_is_nocopy(const struct sock *sk) 579{ 580 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 581} 582 583#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 584 585/** 586 * __locked_read_sk_user_data_with_flags - return the pointer 587 * only if argument flags all has been set in sk_user_data. Otherwise 588 * return NULL 589 * 590 * @sk: socket 591 * @flags: flag bits 592 * 593 * The caller must be holding sk->sk_callback_lock. 594 */ 595static inline void * 596__locked_read_sk_user_data_with_flags(const struct sock *sk, 597 uintptr_t flags) 598{ 599 uintptr_t sk_user_data = 600 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 601 lockdep_is_held(&sk->sk_callback_lock)); 602 603 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 604 605 if ((sk_user_data & flags) == flags) 606 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 607 return NULL; 608} 609 610/** 611 * __rcu_dereference_sk_user_data_with_flags - return the pointer 612 * only if argument flags all has been set in sk_user_data. Otherwise 613 * return NULL 614 * 615 * @sk: socket 616 * @flags: flag bits 617 */ 618static inline void * 619__rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 620 uintptr_t flags) 621{ 622 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 623 624 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 625 626 if ((sk_user_data & flags) == flags) 627 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 628 return NULL; 629} 630 631#define rcu_dereference_sk_user_data(sk) \ 632 __rcu_dereference_sk_user_data_with_flags(sk, 0) 633#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 634({ \ 635 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 636 __tmp2 = (uintptr_t)(flags); \ 637 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 638 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 639 rcu_assign_pointer(__sk_user_data((sk)), \ 640 __tmp1 | __tmp2); \ 641}) 642#define rcu_assign_sk_user_data(sk, ptr) \ 643 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 644 645static inline 646struct net *sock_net(const struct sock *sk) 647{ 648 return read_pnet(&sk->sk_net); 649} 650 651static inline 652void sock_net_set(struct sock *sk, struct net *net) 653{ 654 write_pnet(&sk->sk_net, net); 655} 656 657/* 658 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 659 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 660 * on a socket means that the socket will reuse everybody else's port 661 * without looking at the other's sk_reuse value. 662 */ 663 664#define SK_NO_REUSE 0 665#define SK_CAN_REUSE 1 666#define SK_FORCE_REUSE 2 667 668int sk_set_peek_off(struct sock *sk, int val); 669 670static inline int sk_peek_offset(const struct sock *sk, int flags) 671{ 672 if (unlikely(flags & MSG_PEEK)) { 673 return READ_ONCE(sk->sk_peek_off); 674 } 675 676 return 0; 677} 678 679static inline void sk_peek_offset_bwd(struct sock *sk, int val) 680{ 681 s32 off = READ_ONCE(sk->sk_peek_off); 682 683 if (unlikely(off >= 0)) { 684 off = max_t(s32, off - val, 0); 685 WRITE_ONCE(sk->sk_peek_off, off); 686 } 687} 688 689static inline void sk_peek_offset_fwd(struct sock *sk, int val) 690{ 691 sk_peek_offset_bwd(sk, -val); 692} 693 694/* 695 * Hashed lists helper routines 696 */ 697static inline struct sock *sk_entry(const struct hlist_node *node) 698{ 699 return hlist_entry(node, struct sock, sk_node); 700} 701 702static inline struct sock *__sk_head(const struct hlist_head *head) 703{ 704 return hlist_entry(head->first, struct sock, sk_node); 705} 706 707static inline struct sock *sk_head(const struct hlist_head *head) 708{ 709 return hlist_empty(head) ? NULL : __sk_head(head); 710} 711 712static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 713{ 714 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 715} 716 717static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 718{ 719 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 720} 721 722static inline struct sock *sk_next(const struct sock *sk) 723{ 724 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 725} 726 727static inline struct sock *sk_nulls_next(const struct sock *sk) 728{ 729 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 730 hlist_nulls_entry(sk->sk_nulls_node.next, 731 struct sock, sk_nulls_node) : 732 NULL; 733} 734 735static inline bool sk_unhashed(const struct sock *sk) 736{ 737 return hlist_unhashed(&sk->sk_node); 738} 739 740static inline bool sk_hashed(const struct sock *sk) 741{ 742 return !sk_unhashed(sk); 743} 744 745static inline void sk_node_init(struct hlist_node *node) 746{ 747 node->pprev = NULL; 748} 749 750static inline void __sk_del_node(struct sock *sk) 751{ 752 __hlist_del(&sk->sk_node); 753} 754 755/* NB: equivalent to hlist_del_init_rcu */ 756static inline bool __sk_del_node_init(struct sock *sk) 757{ 758 if (sk_hashed(sk)) { 759 __sk_del_node(sk); 760 sk_node_init(&sk->sk_node); 761 return true; 762 } 763 return false; 764} 765 766/* Grab socket reference count. This operation is valid only 767 when sk is ALREADY grabbed f.e. it is found in hash table 768 or a list and the lookup is made under lock preventing hash table 769 modifications. 770 */ 771 772static __always_inline void sock_hold(struct sock *sk) 773{ 774 refcount_inc(&sk->sk_refcnt); 775} 776 777/* Ungrab socket in the context, which assumes that socket refcnt 778 cannot hit zero, f.e. it is true in context of any socketcall. 779 */ 780static __always_inline void __sock_put(struct sock *sk) 781{ 782 refcount_dec(&sk->sk_refcnt); 783} 784 785static inline bool sk_del_node_init(struct sock *sk) 786{ 787 bool rc = __sk_del_node_init(sk); 788 789 if (rc) { 790 /* paranoid for a while -acme */ 791 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 792 __sock_put(sk); 793 } 794 return rc; 795} 796#define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 797 798static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 799{ 800 if (sk_hashed(sk)) { 801 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 802 return true; 803 } 804 return false; 805} 806 807static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 808{ 809 bool rc = __sk_nulls_del_node_init_rcu(sk); 810 811 if (rc) { 812 /* paranoid for a while -acme */ 813 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 814 __sock_put(sk); 815 } 816 return rc; 817} 818 819static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 820{ 821 hlist_add_head(&sk->sk_node, list); 822} 823 824static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 825{ 826 sock_hold(sk); 827 __sk_add_node(sk, list); 828} 829 830static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 831{ 832 sock_hold(sk); 833 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 834 sk->sk_family == AF_INET6) 835 hlist_add_tail_rcu(&sk->sk_node, list); 836 else 837 hlist_add_head_rcu(&sk->sk_node, list); 838} 839 840static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 841{ 842 sock_hold(sk); 843 hlist_add_tail_rcu(&sk->sk_node, list); 844} 845 846static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 847{ 848 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 849} 850 851static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 852{ 853 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 854} 855 856static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 857{ 858 sock_hold(sk); 859 __sk_nulls_add_node_rcu(sk, list); 860} 861 862static inline void __sk_del_bind_node(struct sock *sk) 863{ 864 __hlist_del(&sk->sk_bind_node); 865} 866 867static inline void sk_add_bind_node(struct sock *sk, 868 struct hlist_head *list) 869{ 870 hlist_add_head(&sk->sk_bind_node, list); 871} 872 873#define sk_for_each(__sk, list) \ 874 hlist_for_each_entry(__sk, list, sk_node) 875#define sk_for_each_rcu(__sk, list) \ 876 hlist_for_each_entry_rcu(__sk, list, sk_node) 877#define sk_nulls_for_each(__sk, node, list) \ 878 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 879#define sk_nulls_for_each_rcu(__sk, node, list) \ 880 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 881#define sk_for_each_from(__sk) \ 882 hlist_for_each_entry_from(__sk, sk_node) 883#define sk_nulls_for_each_from(__sk, node) \ 884 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 885 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 886#define sk_for_each_safe(__sk, tmp, list) \ 887 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 888#define sk_for_each_bound(__sk, list) \ 889 hlist_for_each_entry(__sk, list, sk_bind_node) 890 891/** 892 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 893 * @tpos: the type * to use as a loop cursor. 894 * @pos: the &struct hlist_node to use as a loop cursor. 895 * @head: the head for your list. 896 * @offset: offset of hlist_node within the struct. 897 * 898 */ 899#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 900 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 901 pos != NULL && \ 902 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 903 pos = rcu_dereference(hlist_next_rcu(pos))) 904 905static inline struct user_namespace *sk_user_ns(const struct sock *sk) 906{ 907 /* Careful only use this in a context where these parameters 908 * can not change and must all be valid, such as recvmsg from 909 * userspace. 910 */ 911 return sk->sk_socket->file->f_cred->user_ns; 912} 913 914/* Sock flags */ 915enum sock_flags { 916 SOCK_DEAD, 917 SOCK_DONE, 918 SOCK_URGINLINE, 919 SOCK_KEEPOPEN, 920 SOCK_LINGER, 921 SOCK_DESTROY, 922 SOCK_BROADCAST, 923 SOCK_TIMESTAMP, 924 SOCK_ZAPPED, 925 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 926 SOCK_DBG, /* %SO_DEBUG setting */ 927 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 928 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 929 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 930 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 931 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 932 SOCK_FASYNC, /* fasync() active */ 933 SOCK_RXQ_OVFL, 934 SOCK_ZEROCOPY, /* buffers from userspace */ 935 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 936 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 937 * Will use last 4 bytes of packet sent from 938 * user-space instead. 939 */ 940 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 941 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 942 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 943 SOCK_TXTIME, 944 SOCK_XDP, /* XDP is attached */ 945 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 946 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 947}; 948 949#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 950 951static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 952{ 953 nsk->sk_flags = osk->sk_flags; 954} 955 956static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 957{ 958 __set_bit(flag, &sk->sk_flags); 959} 960 961static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 962{ 963 __clear_bit(flag, &sk->sk_flags); 964} 965 966static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 967 int valbool) 968{ 969 if (valbool) 970 sock_set_flag(sk, bit); 971 else 972 sock_reset_flag(sk, bit); 973} 974 975static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 976{ 977 return test_bit(flag, &sk->sk_flags); 978} 979 980#ifdef CONFIG_NET 981DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 982static inline int sk_memalloc_socks(void) 983{ 984 return static_branch_unlikely(&memalloc_socks_key); 985} 986 987void __receive_sock(struct file *file); 988#else 989 990static inline int sk_memalloc_socks(void) 991{ 992 return 0; 993} 994 995static inline void __receive_sock(struct file *file) 996{ } 997#endif 998 999static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1000{ 1001 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1002} 1003 1004static inline void sk_acceptq_removed(struct sock *sk) 1005{ 1006 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1007} 1008 1009static inline void sk_acceptq_added(struct sock *sk) 1010{ 1011 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1012} 1013 1014/* Note: If you think the test should be: 1015 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1016 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1017 */ 1018static inline bool sk_acceptq_is_full(const struct sock *sk) 1019{ 1020 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1021} 1022 1023/* 1024 * Compute minimal free write space needed to queue new packets. 1025 */ 1026static inline int sk_stream_min_wspace(const struct sock *sk) 1027{ 1028 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1029} 1030 1031static inline int sk_stream_wspace(const struct sock *sk) 1032{ 1033 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1034} 1035 1036static inline void sk_wmem_queued_add(struct sock *sk, int val) 1037{ 1038 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1039} 1040 1041static inline void sk_forward_alloc_add(struct sock *sk, int val) 1042{ 1043 /* Paired with lockless reads of sk->sk_forward_alloc */ 1044 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1045} 1046 1047void sk_stream_write_space(struct sock *sk); 1048 1049/* OOB backlog add */ 1050static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1051{ 1052 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1053 skb_dst_force(skb); 1054 1055 if (!sk->sk_backlog.tail) 1056 WRITE_ONCE(sk->sk_backlog.head, skb); 1057 else 1058 sk->sk_backlog.tail->next = skb; 1059 1060 WRITE_ONCE(sk->sk_backlog.tail, skb); 1061 skb->next = NULL; 1062} 1063 1064/* 1065 * Take into account size of receive queue and backlog queue 1066 * Do not take into account this skb truesize, 1067 * to allow even a single big packet to come. 1068 */ 1069static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1070{ 1071 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1072 1073 return qsize > limit; 1074} 1075 1076/* The per-socket spinlock must be held here. */ 1077static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1078 unsigned int limit) 1079{ 1080 if (sk_rcvqueues_full(sk, limit)) 1081 return -ENOBUFS; 1082 1083 /* 1084 * If the skb was allocated from pfmemalloc reserves, only 1085 * allow SOCK_MEMALLOC sockets to use it as this socket is 1086 * helping free memory 1087 */ 1088 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1089 return -ENOMEM; 1090 1091 __sk_add_backlog(sk, skb); 1092 sk->sk_backlog.len += skb->truesize; 1093 return 0; 1094} 1095 1096int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1097 1098INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1099INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1100 1101static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1102{ 1103 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1104 return __sk_backlog_rcv(sk, skb); 1105 1106 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1107 tcp_v6_do_rcv, 1108 tcp_v4_do_rcv, 1109 sk, skb); 1110} 1111 1112static inline void sk_incoming_cpu_update(struct sock *sk) 1113{ 1114 int cpu = raw_smp_processor_id(); 1115 1116 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1117 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1118} 1119 1120 1121static inline void sock_rps_save_rxhash(struct sock *sk, 1122 const struct sk_buff *skb) 1123{ 1124#ifdef CONFIG_RPS 1125 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1126 * here, and another one in sock_rps_record_flow(). 1127 */ 1128 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1129 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1130#endif 1131} 1132 1133static inline void sock_rps_reset_rxhash(struct sock *sk) 1134{ 1135#ifdef CONFIG_RPS 1136 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1137 WRITE_ONCE(sk->sk_rxhash, 0); 1138#endif 1139} 1140 1141#define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1142 ({ int __rc, __dis = __sk->sk_disconnects; \ 1143 release_sock(__sk); \ 1144 __rc = __condition; \ 1145 if (!__rc) { \ 1146 *(__timeo) = wait_woken(__wait, \ 1147 TASK_INTERRUPTIBLE, \ 1148 *(__timeo)); \ 1149 } \ 1150 sched_annotate_sleep(); \ 1151 lock_sock(__sk); \ 1152 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1153 __rc; \ 1154 }) 1155 1156int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1157int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1158void sk_stream_wait_close(struct sock *sk, long timeo_p); 1159int sk_stream_error(struct sock *sk, int flags, int err); 1160void sk_stream_kill_queues(struct sock *sk); 1161void sk_set_memalloc(struct sock *sk); 1162void sk_clear_memalloc(struct sock *sk); 1163 1164void __sk_flush_backlog(struct sock *sk); 1165 1166static inline bool sk_flush_backlog(struct sock *sk) 1167{ 1168 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1169 __sk_flush_backlog(sk); 1170 return true; 1171 } 1172 return false; 1173} 1174 1175int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1176 1177struct request_sock_ops; 1178struct timewait_sock_ops; 1179struct inet_hashinfo; 1180struct raw_hashinfo; 1181struct smc_hashinfo; 1182struct module; 1183struct sk_psock; 1184 1185/* 1186 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1187 * un-modified. Special care is taken when initializing object to zero. 1188 */ 1189static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1190{ 1191 if (offsetof(struct sock, sk_node.next) != 0) 1192 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1193 memset(&sk->sk_node.pprev, 0, 1194 size - offsetof(struct sock, sk_node.pprev)); 1195} 1196 1197/* Networking protocol blocks we attach to sockets. 1198 * socket layer -> transport layer interface 1199 */ 1200struct proto { 1201 void (*close)(struct sock *sk, 1202 long timeout); 1203 int (*pre_connect)(struct sock *sk, 1204 struct sockaddr *uaddr, 1205 int addr_len); 1206 int (*connect)(struct sock *sk, 1207 struct sockaddr *uaddr, 1208 int addr_len); 1209 int (*disconnect)(struct sock *sk, int flags); 1210 1211 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1212 bool kern); 1213 1214 int (*ioctl)(struct sock *sk, int cmd, 1215 int *karg); 1216 int (*init)(struct sock *sk); 1217 void (*destroy)(struct sock *sk); 1218 void (*shutdown)(struct sock *sk, int how); 1219 int (*setsockopt)(struct sock *sk, int level, 1220 int optname, sockptr_t optval, 1221 unsigned int optlen); 1222 int (*getsockopt)(struct sock *sk, int level, 1223 int optname, char __user *optval, 1224 int __user *option); 1225 void (*keepalive)(struct sock *sk, int valbool); 1226#ifdef CONFIG_COMPAT 1227 int (*compat_ioctl)(struct sock *sk, 1228 unsigned int cmd, unsigned long arg); 1229#endif 1230 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1231 size_t len); 1232 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1233 size_t len, int flags, int *addr_len); 1234 void (*splice_eof)(struct socket *sock); 1235 int (*bind)(struct sock *sk, 1236 struct sockaddr *addr, int addr_len); 1237 int (*bind_add)(struct sock *sk, 1238 struct sockaddr *addr, int addr_len); 1239 1240 int (*backlog_rcv) (struct sock *sk, 1241 struct sk_buff *skb); 1242 bool (*bpf_bypass_getsockopt)(int level, 1243 int optname); 1244 1245 void (*release_cb)(struct sock *sk); 1246 1247 /* Keeping track of sk's, looking them up, and port selection methods. */ 1248 int (*hash)(struct sock *sk); 1249 void (*unhash)(struct sock *sk); 1250 void (*rehash)(struct sock *sk); 1251 int (*get_port)(struct sock *sk, unsigned short snum); 1252 void (*put_port)(struct sock *sk); 1253#ifdef CONFIG_BPF_SYSCALL 1254 int (*psock_update_sk_prot)(struct sock *sk, 1255 struct sk_psock *psock, 1256 bool restore); 1257#endif 1258 1259 /* Keeping track of sockets in use */ 1260#ifdef CONFIG_PROC_FS 1261 unsigned int inuse_idx; 1262#endif 1263 1264#if IS_ENABLED(CONFIG_MPTCP) 1265 int (*forward_alloc_get)(const struct sock *sk); 1266#endif 1267 1268 bool (*stream_memory_free)(const struct sock *sk, int wake); 1269 bool (*sock_is_readable)(struct sock *sk); 1270 /* Memory pressure */ 1271 void (*enter_memory_pressure)(struct sock *sk); 1272 void (*leave_memory_pressure)(struct sock *sk); 1273 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1274 int __percpu *per_cpu_fw_alloc; 1275 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1276 1277 /* 1278 * Pressure flag: try to collapse. 1279 * Technical note: it is used by multiple contexts non atomically. 1280 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1281 * All the __sk_mem_schedule() is of this nature: accounting 1282 * is strict, actions are advisory and have some latency. 1283 */ 1284 unsigned long *memory_pressure; 1285 long *sysctl_mem; 1286 1287 int *sysctl_wmem; 1288 int *sysctl_rmem; 1289 u32 sysctl_wmem_offset; 1290 u32 sysctl_rmem_offset; 1291 1292 int max_header; 1293 bool no_autobind; 1294 1295 struct kmem_cache *slab; 1296 unsigned int obj_size; 1297 unsigned int ipv6_pinfo_offset; 1298 slab_flags_t slab_flags; 1299 unsigned int useroffset; /* Usercopy region offset */ 1300 unsigned int usersize; /* Usercopy region size */ 1301 1302 unsigned int __percpu *orphan_count; 1303 1304 struct request_sock_ops *rsk_prot; 1305 struct timewait_sock_ops *twsk_prot; 1306 1307 union { 1308 struct inet_hashinfo *hashinfo; 1309 struct udp_table *udp_table; 1310 struct raw_hashinfo *raw_hash; 1311 struct smc_hashinfo *smc_hash; 1312 } h; 1313 1314 struct module *owner; 1315 1316 char name[32]; 1317 1318 struct list_head node; 1319 int (*diag_destroy)(struct sock *sk, int err); 1320} __randomize_layout; 1321 1322int proto_register(struct proto *prot, int alloc_slab); 1323void proto_unregister(struct proto *prot); 1324int sock_load_diag_module(int family, int protocol); 1325 1326INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1327 1328static inline int sk_forward_alloc_get(const struct sock *sk) 1329{ 1330#if IS_ENABLED(CONFIG_MPTCP) 1331 if (sk->sk_prot->forward_alloc_get) 1332 return sk->sk_prot->forward_alloc_get(sk); 1333#endif 1334 return READ_ONCE(sk->sk_forward_alloc); 1335} 1336 1337static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1338{ 1339 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1340 return false; 1341 1342 return sk->sk_prot->stream_memory_free ? 1343 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1344 tcp_stream_memory_free, sk, wake) : true; 1345} 1346 1347static inline bool sk_stream_memory_free(const struct sock *sk) 1348{ 1349 return __sk_stream_memory_free(sk, 0); 1350} 1351 1352static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1353{ 1354 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1355 __sk_stream_memory_free(sk, wake); 1356} 1357 1358static inline bool sk_stream_is_writeable(const struct sock *sk) 1359{ 1360 return __sk_stream_is_writeable(sk, 0); 1361} 1362 1363static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1364 struct cgroup *ancestor) 1365{ 1366#ifdef CONFIG_SOCK_CGROUP_DATA 1367 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1368 ancestor); 1369#else 1370 return -ENOTSUPP; 1371#endif 1372} 1373 1374static inline bool sk_has_memory_pressure(const struct sock *sk) 1375{ 1376 return sk->sk_prot->memory_pressure != NULL; 1377} 1378 1379static inline bool sk_under_global_memory_pressure(const struct sock *sk) 1380{ 1381 return sk->sk_prot->memory_pressure && 1382 !!READ_ONCE(*sk->sk_prot->memory_pressure); 1383} 1384 1385static inline bool sk_under_memory_pressure(const struct sock *sk) 1386{ 1387 if (!sk->sk_prot->memory_pressure) 1388 return false; 1389 1390 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1391 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1392 return true; 1393 1394 return !!READ_ONCE(*sk->sk_prot->memory_pressure); 1395} 1396 1397static inline long 1398proto_memory_allocated(const struct proto *prot) 1399{ 1400 return max(0L, atomic_long_read(prot->memory_allocated)); 1401} 1402 1403static inline long 1404sk_memory_allocated(const struct sock *sk) 1405{ 1406 return proto_memory_allocated(sk->sk_prot); 1407} 1408 1409/* 1 MB per cpu, in page units */ 1410#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1411extern int sysctl_mem_pcpu_rsv; 1412 1413static inline void proto_memory_pcpu_drain(struct proto *proto) 1414{ 1415 int val = this_cpu_xchg(*proto->per_cpu_fw_alloc, 0); 1416 1417 if (val) 1418 atomic_long_add(val, proto->memory_allocated); 1419} 1420 1421static inline void 1422sk_memory_allocated_add(const struct sock *sk, int val) 1423{ 1424 struct proto *proto = sk->sk_prot; 1425 1426 val = this_cpu_add_return(*proto->per_cpu_fw_alloc, val); 1427 1428 if (unlikely(val >= READ_ONCE(sysctl_mem_pcpu_rsv))) 1429 proto_memory_pcpu_drain(proto); 1430} 1431 1432static inline void 1433sk_memory_allocated_sub(const struct sock *sk, int val) 1434{ 1435 struct proto *proto = sk->sk_prot; 1436 1437 val = this_cpu_sub_return(*proto->per_cpu_fw_alloc, val); 1438 1439 if (unlikely(val <= -READ_ONCE(sysctl_mem_pcpu_rsv))) 1440 proto_memory_pcpu_drain(proto); 1441} 1442 1443#define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1444 1445static inline void sk_sockets_allocated_dec(struct sock *sk) 1446{ 1447 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1448 SK_ALLOC_PERCPU_COUNTER_BATCH); 1449} 1450 1451static inline void sk_sockets_allocated_inc(struct sock *sk) 1452{ 1453 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1454 SK_ALLOC_PERCPU_COUNTER_BATCH); 1455} 1456 1457static inline u64 1458sk_sockets_allocated_read_positive(struct sock *sk) 1459{ 1460 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1461} 1462 1463static inline int 1464proto_sockets_allocated_sum_positive(struct proto *prot) 1465{ 1466 return percpu_counter_sum_positive(prot->sockets_allocated); 1467} 1468 1469static inline bool 1470proto_memory_pressure(struct proto *prot) 1471{ 1472 if (!prot->memory_pressure) 1473 return false; 1474 return !!READ_ONCE(*prot->memory_pressure); 1475} 1476 1477 1478#ifdef CONFIG_PROC_FS 1479#define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1480struct prot_inuse { 1481 int all; 1482 int val[PROTO_INUSE_NR]; 1483}; 1484 1485static inline void sock_prot_inuse_add(const struct net *net, 1486 const struct proto *prot, int val) 1487{ 1488 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1489} 1490 1491static inline void sock_inuse_add(const struct net *net, int val) 1492{ 1493 this_cpu_add(net->core.prot_inuse->all, val); 1494} 1495 1496int sock_prot_inuse_get(struct net *net, struct proto *proto); 1497int sock_inuse_get(struct net *net); 1498#else 1499static inline void sock_prot_inuse_add(const struct net *net, 1500 const struct proto *prot, int val) 1501{ 1502} 1503 1504static inline void sock_inuse_add(const struct net *net, int val) 1505{ 1506} 1507#endif 1508 1509 1510/* With per-bucket locks this operation is not-atomic, so that 1511 * this version is not worse. 1512 */ 1513static inline int __sk_prot_rehash(struct sock *sk) 1514{ 1515 sk->sk_prot->unhash(sk); 1516 return sk->sk_prot->hash(sk); 1517} 1518 1519/* About 10 seconds */ 1520#define SOCK_DESTROY_TIME (10*HZ) 1521 1522/* Sockets 0-1023 can't be bound to unless you are superuser */ 1523#define PROT_SOCK 1024 1524 1525#define SHUTDOWN_MASK 3 1526#define RCV_SHUTDOWN 1 1527#define SEND_SHUTDOWN 2 1528 1529#define SOCK_BINDADDR_LOCK 4 1530#define SOCK_BINDPORT_LOCK 8 1531 1532struct socket_alloc { 1533 struct socket socket; 1534 struct inode vfs_inode; 1535}; 1536 1537static inline struct socket *SOCKET_I(struct inode *inode) 1538{ 1539 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1540} 1541 1542static inline struct inode *SOCK_INODE(struct socket *socket) 1543{ 1544 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1545} 1546 1547/* 1548 * Functions for memory accounting 1549 */ 1550int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1551int __sk_mem_schedule(struct sock *sk, int size, int kind); 1552void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1553void __sk_mem_reclaim(struct sock *sk, int amount); 1554 1555#define SK_MEM_SEND 0 1556#define SK_MEM_RECV 1 1557 1558/* sysctl_mem values are in pages */ 1559static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1560{ 1561 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1562} 1563 1564static inline int sk_mem_pages(int amt) 1565{ 1566 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1567} 1568 1569static inline bool sk_has_account(struct sock *sk) 1570{ 1571 /* return true if protocol supports memory accounting */ 1572 return !!sk->sk_prot->memory_allocated; 1573} 1574 1575static inline bool sk_wmem_schedule(struct sock *sk, int size) 1576{ 1577 int delta; 1578 1579 if (!sk_has_account(sk)) 1580 return true; 1581 delta = size - sk->sk_forward_alloc; 1582 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1583} 1584 1585static inline bool 1586sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1587{ 1588 int delta; 1589 1590 if (!sk_has_account(sk)) 1591 return true; 1592 delta = size - sk->sk_forward_alloc; 1593 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1594 skb_pfmemalloc(skb); 1595} 1596 1597static inline int sk_unused_reserved_mem(const struct sock *sk) 1598{ 1599 int unused_mem; 1600 1601 if (likely(!sk->sk_reserved_mem)) 1602 return 0; 1603 1604 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1605 atomic_read(&sk->sk_rmem_alloc); 1606 1607 return unused_mem > 0 ? unused_mem : 0; 1608} 1609 1610static inline void sk_mem_reclaim(struct sock *sk) 1611{ 1612 int reclaimable; 1613 1614 if (!sk_has_account(sk)) 1615 return; 1616 1617 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1618 1619 if (reclaimable >= (int)PAGE_SIZE) 1620 __sk_mem_reclaim(sk, reclaimable); 1621} 1622 1623static inline void sk_mem_reclaim_final(struct sock *sk) 1624{ 1625 sk->sk_reserved_mem = 0; 1626 sk_mem_reclaim(sk); 1627} 1628 1629static inline void sk_mem_charge(struct sock *sk, int size) 1630{ 1631 if (!sk_has_account(sk)) 1632 return; 1633 sk_forward_alloc_add(sk, -size); 1634} 1635 1636static inline void sk_mem_uncharge(struct sock *sk, int size) 1637{ 1638 if (!sk_has_account(sk)) 1639 return; 1640 sk_forward_alloc_add(sk, size); 1641 sk_mem_reclaim(sk); 1642} 1643 1644/* 1645 * Macro so as to not evaluate some arguments when 1646 * lockdep is not enabled. 1647 * 1648 * Mark both the sk_lock and the sk_lock.slock as a 1649 * per-address-family lock class. 1650 */ 1651#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1652do { \ 1653 sk->sk_lock.owned = 0; \ 1654 init_waitqueue_head(&sk->sk_lock.wq); \ 1655 spin_lock_init(&(sk)->sk_lock.slock); \ 1656 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1657 sizeof((sk)->sk_lock)); \ 1658 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1659 (skey), (sname)); \ 1660 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1661} while (0) 1662 1663static inline bool lockdep_sock_is_held(const struct sock *sk) 1664{ 1665 return lockdep_is_held(&sk->sk_lock) || 1666 lockdep_is_held(&sk->sk_lock.slock); 1667} 1668 1669void lock_sock_nested(struct sock *sk, int subclass); 1670 1671static inline void lock_sock(struct sock *sk) 1672{ 1673 lock_sock_nested(sk, 0); 1674} 1675 1676void __lock_sock(struct sock *sk); 1677void __release_sock(struct sock *sk); 1678void release_sock(struct sock *sk); 1679 1680/* BH context may only use the following locking interface. */ 1681#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1682#define bh_lock_sock_nested(__sk) \ 1683 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1684 SINGLE_DEPTH_NESTING) 1685#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1686 1687bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1688 1689/** 1690 * lock_sock_fast - fast version of lock_sock 1691 * @sk: socket 1692 * 1693 * This version should be used for very small section, where process wont block 1694 * return false if fast path is taken: 1695 * 1696 * sk_lock.slock locked, owned = 0, BH disabled 1697 * 1698 * return true if slow path is taken: 1699 * 1700 * sk_lock.slock unlocked, owned = 1, BH enabled 1701 */ 1702static inline bool lock_sock_fast(struct sock *sk) 1703{ 1704 /* The sk_lock has mutex_lock() semantics here. */ 1705 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1706 1707 return __lock_sock_fast(sk); 1708} 1709 1710/* fast socket lock variant for caller already holding a [different] socket lock */ 1711static inline bool lock_sock_fast_nested(struct sock *sk) 1712{ 1713 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1714 1715 return __lock_sock_fast(sk); 1716} 1717 1718/** 1719 * unlock_sock_fast - complement of lock_sock_fast 1720 * @sk: socket 1721 * @slow: slow mode 1722 * 1723 * fast unlock socket for user context. 1724 * If slow mode is on, we call regular release_sock() 1725 */ 1726static inline void unlock_sock_fast(struct sock *sk, bool slow) 1727 __releases(&sk->sk_lock.slock) 1728{ 1729 if (slow) { 1730 release_sock(sk); 1731 __release(&sk->sk_lock.slock); 1732 } else { 1733 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1734 spin_unlock_bh(&sk->sk_lock.slock); 1735 } 1736} 1737 1738void sockopt_lock_sock(struct sock *sk); 1739void sockopt_release_sock(struct sock *sk); 1740bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1741bool sockopt_capable(int cap); 1742 1743/* Used by processes to "lock" a socket state, so that 1744 * interrupts and bottom half handlers won't change it 1745 * from under us. It essentially blocks any incoming 1746 * packets, so that we won't get any new data or any 1747 * packets that change the state of the socket. 1748 * 1749 * While locked, BH processing will add new packets to 1750 * the backlog queue. This queue is processed by the 1751 * owner of the socket lock right before it is released. 1752 * 1753 * Since ~2.3.5 it is also exclusive sleep lock serializing 1754 * accesses from user process context. 1755 */ 1756 1757static inline void sock_owned_by_me(const struct sock *sk) 1758{ 1759#ifdef CONFIG_LOCKDEP 1760 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1761#endif 1762} 1763 1764static inline void sock_not_owned_by_me(const struct sock *sk) 1765{ 1766#ifdef CONFIG_LOCKDEP 1767 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1768#endif 1769} 1770 1771static inline bool sock_owned_by_user(const struct sock *sk) 1772{ 1773 sock_owned_by_me(sk); 1774 return sk->sk_lock.owned; 1775} 1776 1777static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1778{ 1779 return sk->sk_lock.owned; 1780} 1781 1782static inline void sock_release_ownership(struct sock *sk) 1783{ 1784 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1785 sk->sk_lock.owned = 0; 1786 1787 /* The sk_lock has mutex_unlock() semantics: */ 1788 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1789} 1790 1791/* no reclassification while locks are held */ 1792static inline bool sock_allow_reclassification(const struct sock *csk) 1793{ 1794 struct sock *sk = (struct sock *)csk; 1795 1796 return !sock_owned_by_user_nocheck(sk) && 1797 !spin_is_locked(&sk->sk_lock.slock); 1798} 1799 1800struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1801 struct proto *prot, int kern); 1802void sk_free(struct sock *sk); 1803void sk_destruct(struct sock *sk); 1804struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1805void sk_free_unlock_clone(struct sock *sk); 1806 1807struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1808 gfp_t priority); 1809void __sock_wfree(struct sk_buff *skb); 1810void sock_wfree(struct sk_buff *skb); 1811struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1812 gfp_t priority); 1813void skb_orphan_partial(struct sk_buff *skb); 1814void sock_rfree(struct sk_buff *skb); 1815void sock_efree(struct sk_buff *skb); 1816#ifdef CONFIG_INET 1817void sock_edemux(struct sk_buff *skb); 1818void sock_pfree(struct sk_buff *skb); 1819#else 1820#define sock_edemux sock_efree 1821#endif 1822 1823int sk_setsockopt(struct sock *sk, int level, int optname, 1824 sockptr_t optval, unsigned int optlen); 1825int sock_setsockopt(struct socket *sock, int level, int op, 1826 sockptr_t optval, unsigned int optlen); 1827int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1828 int optname, sockptr_t optval, int optlen); 1829int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1830 int optname, sockptr_t optval, sockptr_t optlen); 1831 1832int sk_getsockopt(struct sock *sk, int level, int optname, 1833 sockptr_t optval, sockptr_t optlen); 1834int sock_gettstamp(struct socket *sock, void __user *userstamp, 1835 bool timeval, bool time32); 1836struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1837 unsigned long data_len, int noblock, 1838 int *errcode, int max_page_order); 1839 1840static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1841 unsigned long size, 1842 int noblock, int *errcode) 1843{ 1844 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1845} 1846 1847void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1848void sock_kfree_s(struct sock *sk, void *mem, int size); 1849void sock_kzfree_s(struct sock *sk, void *mem, int size); 1850void sk_send_sigurg(struct sock *sk); 1851 1852static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1853{ 1854 if (sk->sk_socket) 1855 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1856 WRITE_ONCE(sk->sk_prot, proto); 1857} 1858 1859struct sockcm_cookie { 1860 u64 transmit_time; 1861 u32 mark; 1862 u32 tsflags; 1863}; 1864 1865static inline void sockcm_init(struct sockcm_cookie *sockc, 1866 const struct sock *sk) 1867{ 1868 *sockc = (struct sockcm_cookie) { 1869 .tsflags = READ_ONCE(sk->sk_tsflags) 1870 }; 1871} 1872 1873int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1874 struct sockcm_cookie *sockc); 1875int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1876 struct sockcm_cookie *sockc); 1877 1878/* 1879 * Functions to fill in entries in struct proto_ops when a protocol 1880 * does not implement a particular function. 1881 */ 1882int sock_no_bind(struct socket *, struct sockaddr *, int); 1883int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1884int sock_no_socketpair(struct socket *, struct socket *); 1885int sock_no_accept(struct socket *, struct socket *, int, bool); 1886int sock_no_getname(struct socket *, struct sockaddr *, int); 1887int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1888int sock_no_listen(struct socket *, int); 1889int sock_no_shutdown(struct socket *, int); 1890int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1891int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1892int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1893int sock_no_mmap(struct file *file, struct socket *sock, 1894 struct vm_area_struct *vma); 1895 1896/* 1897 * Functions to fill in entries in struct proto_ops when a protocol 1898 * uses the inet style. 1899 */ 1900int sock_common_getsockopt(struct socket *sock, int level, int optname, 1901 char __user *optval, int __user *optlen); 1902int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1903 int flags); 1904int sock_common_setsockopt(struct socket *sock, int level, int optname, 1905 sockptr_t optval, unsigned int optlen); 1906 1907void sk_common_release(struct sock *sk); 1908 1909/* 1910 * Default socket callbacks and setup code 1911 */ 1912 1913/* Initialise core socket variables using an explicit uid. */ 1914void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1915 1916/* Initialise core socket variables. 1917 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1918 */ 1919void sock_init_data(struct socket *sock, struct sock *sk); 1920 1921/* 1922 * Socket reference counting postulates. 1923 * 1924 * * Each user of socket SHOULD hold a reference count. 1925 * * Each access point to socket (an hash table bucket, reference from a list, 1926 * running timer, skb in flight MUST hold a reference count. 1927 * * When reference count hits 0, it means it will never increase back. 1928 * * When reference count hits 0, it means that no references from 1929 * outside exist to this socket and current process on current CPU 1930 * is last user and may/should destroy this socket. 1931 * * sk_free is called from any context: process, BH, IRQ. When 1932 * it is called, socket has no references from outside -> sk_free 1933 * may release descendant resources allocated by the socket, but 1934 * to the time when it is called, socket is NOT referenced by any 1935 * hash tables, lists etc. 1936 * * Packets, delivered from outside (from network or from another process) 1937 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1938 * when they sit in queue. Otherwise, packets will leak to hole, when 1939 * socket is looked up by one cpu and unhasing is made by another CPU. 1940 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1941 * (leak to backlog). Packet socket does all the processing inside 1942 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1943 * use separate SMP lock, so that they are prone too. 1944 */ 1945 1946/* Ungrab socket and destroy it, if it was the last reference. */ 1947static inline void sock_put(struct sock *sk) 1948{ 1949 if (refcount_dec_and_test(&sk->sk_refcnt)) 1950 sk_free(sk); 1951} 1952/* Generic version of sock_put(), dealing with all sockets 1953 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1954 */ 1955void sock_gen_put(struct sock *sk); 1956 1957int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1958 unsigned int trim_cap, bool refcounted); 1959static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1960 const int nested) 1961{ 1962 return __sk_receive_skb(sk, skb, nested, 1, true); 1963} 1964 1965static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1966{ 1967 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1968 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1969 return; 1970 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1971 * other WRITE_ONCE() because socket lock might be not held. 1972 */ 1973 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 1974} 1975 1976#define NO_QUEUE_MAPPING USHRT_MAX 1977 1978static inline void sk_tx_queue_clear(struct sock *sk) 1979{ 1980 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1981 * other WRITE_ONCE() because socket lock might be not held. 1982 */ 1983 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 1984} 1985 1986static inline int sk_tx_queue_get(const struct sock *sk) 1987{ 1988 if (sk) { 1989 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 1990 * and sk_tx_queue_set(). 1991 */ 1992 int val = READ_ONCE(sk->sk_tx_queue_mapping); 1993 1994 if (val != NO_QUEUE_MAPPING) 1995 return val; 1996 } 1997 return -1; 1998} 1999 2000static inline void __sk_rx_queue_set(struct sock *sk, 2001 const struct sk_buff *skb, 2002 bool force_set) 2003{ 2004#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2005 if (skb_rx_queue_recorded(skb)) { 2006 u16 rx_queue = skb_get_rx_queue(skb); 2007 2008 if (force_set || 2009 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2010 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2011 } 2012#endif 2013} 2014 2015static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2016{ 2017 __sk_rx_queue_set(sk, skb, true); 2018} 2019 2020static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2021{ 2022 __sk_rx_queue_set(sk, skb, false); 2023} 2024 2025static inline void sk_rx_queue_clear(struct sock *sk) 2026{ 2027#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2028 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2029#endif 2030} 2031 2032static inline int sk_rx_queue_get(const struct sock *sk) 2033{ 2034#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2035 if (sk) { 2036 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2037 2038 if (res != NO_QUEUE_MAPPING) 2039 return res; 2040 } 2041#endif 2042 2043 return -1; 2044} 2045 2046static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2047{ 2048 sk->sk_socket = sock; 2049} 2050 2051static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2052{ 2053 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2054 return &rcu_dereference_raw(sk->sk_wq)->wait; 2055} 2056/* Detach socket from process context. 2057 * Announce socket dead, detach it from wait queue and inode. 2058 * Note that parent inode held reference count on this struct sock, 2059 * we do not release it in this function, because protocol 2060 * probably wants some additional cleanups or even continuing 2061 * to work with this socket (TCP). 2062 */ 2063static inline void sock_orphan(struct sock *sk) 2064{ 2065 write_lock_bh(&sk->sk_callback_lock); 2066 sock_set_flag(sk, SOCK_DEAD); 2067 sk_set_socket(sk, NULL); 2068 sk->sk_wq = NULL; 2069 write_unlock_bh(&sk->sk_callback_lock); 2070} 2071 2072static inline void sock_graft(struct sock *sk, struct socket *parent) 2073{ 2074 WARN_ON(parent->sk); 2075 write_lock_bh(&sk->sk_callback_lock); 2076 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2077 parent->sk = sk; 2078 sk_set_socket(sk, parent); 2079 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2080 security_sock_graft(sk, parent); 2081 write_unlock_bh(&sk->sk_callback_lock); 2082} 2083 2084kuid_t sock_i_uid(struct sock *sk); 2085unsigned long __sock_i_ino(struct sock *sk); 2086unsigned long sock_i_ino(struct sock *sk); 2087 2088static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2089{ 2090 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2091} 2092 2093static inline u32 net_tx_rndhash(void) 2094{ 2095 u32 v = get_random_u32(); 2096 2097 return v ?: 1; 2098} 2099 2100static inline void sk_set_txhash(struct sock *sk) 2101{ 2102 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2103 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2104} 2105 2106static inline bool sk_rethink_txhash(struct sock *sk) 2107{ 2108 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2109 sk_set_txhash(sk); 2110 return true; 2111 } 2112 return false; 2113} 2114 2115static inline struct dst_entry * 2116__sk_dst_get(const struct sock *sk) 2117{ 2118 return rcu_dereference_check(sk->sk_dst_cache, 2119 lockdep_sock_is_held(sk)); 2120} 2121 2122static inline struct dst_entry * 2123sk_dst_get(const struct sock *sk) 2124{ 2125 struct dst_entry *dst; 2126 2127 rcu_read_lock(); 2128 dst = rcu_dereference(sk->sk_dst_cache); 2129 if (dst && !rcuref_get(&dst->__rcuref)) 2130 dst = NULL; 2131 rcu_read_unlock(); 2132 return dst; 2133} 2134 2135static inline void __dst_negative_advice(struct sock *sk) 2136{ 2137 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2138 2139 if (dst && dst->ops->negative_advice) { 2140 ndst = dst->ops->negative_advice(dst); 2141 2142 if (ndst != dst) { 2143 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2144 sk_tx_queue_clear(sk); 2145 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2146 } 2147 } 2148} 2149 2150static inline void dst_negative_advice(struct sock *sk) 2151{ 2152 sk_rethink_txhash(sk); 2153 __dst_negative_advice(sk); 2154} 2155 2156static inline void 2157__sk_dst_set(struct sock *sk, struct dst_entry *dst) 2158{ 2159 struct dst_entry *old_dst; 2160 2161 sk_tx_queue_clear(sk); 2162 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2163 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2164 lockdep_sock_is_held(sk)); 2165 rcu_assign_pointer(sk->sk_dst_cache, dst); 2166 dst_release(old_dst); 2167} 2168 2169static inline void 2170sk_dst_set(struct sock *sk, struct dst_entry *dst) 2171{ 2172 struct dst_entry *old_dst; 2173 2174 sk_tx_queue_clear(sk); 2175 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2176 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2177 dst_release(old_dst); 2178} 2179 2180static inline void 2181__sk_dst_reset(struct sock *sk) 2182{ 2183 __sk_dst_set(sk, NULL); 2184} 2185 2186static inline void 2187sk_dst_reset(struct sock *sk) 2188{ 2189 sk_dst_set(sk, NULL); 2190} 2191 2192struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2193 2194struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2195 2196static inline void sk_dst_confirm(struct sock *sk) 2197{ 2198 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2199 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2200} 2201 2202static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2203{ 2204 if (skb_get_dst_pending_confirm(skb)) { 2205 struct sock *sk = skb->sk; 2206 2207 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2208 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2209 neigh_confirm(n); 2210 } 2211} 2212 2213bool sk_mc_loop(const struct sock *sk); 2214 2215static inline bool sk_can_gso(const struct sock *sk) 2216{ 2217 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2218} 2219 2220void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2221 2222static inline void sk_gso_disable(struct sock *sk) 2223{ 2224 sk->sk_gso_disabled = 1; 2225 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2226} 2227 2228static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2229 struct iov_iter *from, char *to, 2230 int copy, int offset) 2231{ 2232 if (skb->ip_summed == CHECKSUM_NONE) { 2233 __wsum csum = 0; 2234 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2235 return -EFAULT; 2236 skb->csum = csum_block_add(skb->csum, csum, offset); 2237 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2238 if (!copy_from_iter_full_nocache(to, copy, from)) 2239 return -EFAULT; 2240 } else if (!copy_from_iter_full(to, copy, from)) 2241 return -EFAULT; 2242 2243 return 0; 2244} 2245 2246static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2247 struct iov_iter *from, int copy) 2248{ 2249 int err, offset = skb->len; 2250 2251 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2252 copy, offset); 2253 if (err) 2254 __skb_trim(skb, offset); 2255 2256 return err; 2257} 2258 2259static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2260 struct sk_buff *skb, 2261 struct page *page, 2262 int off, int copy) 2263{ 2264 int err; 2265 2266 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2267 copy, skb->len); 2268 if (err) 2269 return err; 2270 2271 skb_len_add(skb, copy); 2272 sk_wmem_queued_add(sk, copy); 2273 sk_mem_charge(sk, copy); 2274 return 0; 2275} 2276 2277/** 2278 * sk_wmem_alloc_get - returns write allocations 2279 * @sk: socket 2280 * 2281 * Return: sk_wmem_alloc minus initial offset of one 2282 */ 2283static inline int sk_wmem_alloc_get(const struct sock *sk) 2284{ 2285 return refcount_read(&sk->sk_wmem_alloc) - 1; 2286} 2287 2288/** 2289 * sk_rmem_alloc_get - returns read allocations 2290 * @sk: socket 2291 * 2292 * Return: sk_rmem_alloc 2293 */ 2294static inline int sk_rmem_alloc_get(const struct sock *sk) 2295{ 2296 return atomic_read(&sk->sk_rmem_alloc); 2297} 2298 2299/** 2300 * sk_has_allocations - check if allocations are outstanding 2301 * @sk: socket 2302 * 2303 * Return: true if socket has write or read allocations 2304 */ 2305static inline bool sk_has_allocations(const struct sock *sk) 2306{ 2307 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2308} 2309 2310/** 2311 * skwq_has_sleeper - check if there are any waiting processes 2312 * @wq: struct socket_wq 2313 * 2314 * Return: true if socket_wq has waiting processes 2315 * 2316 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2317 * barrier call. They were added due to the race found within the tcp code. 2318 * 2319 * Consider following tcp code paths:: 2320 * 2321 * CPU1 CPU2 2322 * sys_select receive packet 2323 * ... ... 2324 * __add_wait_queue update tp->rcv_nxt 2325 * ... ... 2326 * tp->rcv_nxt check sock_def_readable 2327 * ... { 2328 * schedule rcu_read_lock(); 2329 * wq = rcu_dereference(sk->sk_wq); 2330 * if (wq && waitqueue_active(&wq->wait)) 2331 * wake_up_interruptible(&wq->wait) 2332 * ... 2333 * } 2334 * 2335 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2336 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2337 * could then endup calling schedule and sleep forever if there are no more 2338 * data on the socket. 2339 * 2340 */ 2341static inline bool skwq_has_sleeper(struct socket_wq *wq) 2342{ 2343 return wq && wq_has_sleeper(&wq->wait); 2344} 2345 2346/** 2347 * sock_poll_wait - place memory barrier behind the poll_wait call. 2348 * @filp: file 2349 * @sock: socket to wait on 2350 * @p: poll_table 2351 * 2352 * See the comments in the wq_has_sleeper function. 2353 */ 2354static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2355 poll_table *p) 2356{ 2357 if (!poll_does_not_wait(p)) { 2358 poll_wait(filp, &sock->wq.wait, p); 2359 /* We need to be sure we are in sync with the 2360 * socket flags modification. 2361 * 2362 * This memory barrier is paired in the wq_has_sleeper. 2363 */ 2364 smp_mb(); 2365 } 2366} 2367 2368static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2369{ 2370 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2371 u32 txhash = READ_ONCE(sk->sk_txhash); 2372 2373 if (txhash) { 2374 skb->l4_hash = 1; 2375 skb->hash = txhash; 2376 } 2377} 2378 2379void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2380 2381/* 2382 * Queue a received datagram if it will fit. Stream and sequenced 2383 * protocols can't normally use this as they need to fit buffers in 2384 * and play with them. 2385 * 2386 * Inlined as it's very short and called for pretty much every 2387 * packet ever received. 2388 */ 2389static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2390{ 2391 skb_orphan(skb); 2392 skb->sk = sk; 2393 skb->destructor = sock_rfree; 2394 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2395 sk_mem_charge(sk, skb->truesize); 2396} 2397 2398static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2399{ 2400 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2401 skb_orphan(skb); 2402 skb->destructor = sock_efree; 2403 skb->sk = sk; 2404 return true; 2405 } 2406 return false; 2407} 2408 2409static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2410{ 2411 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2412 if (skb) { 2413 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2414 skb_set_owner_r(skb, sk); 2415 return skb; 2416 } 2417 __kfree_skb(skb); 2418 } 2419 return NULL; 2420} 2421 2422static inline void skb_prepare_for_gro(struct sk_buff *skb) 2423{ 2424 if (skb->destructor != sock_wfree) { 2425 skb_orphan(skb); 2426 return; 2427 } 2428 skb->slow_gro = 1; 2429} 2430 2431void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2432 unsigned long expires); 2433 2434void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2435 2436void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2437 2438int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2439 struct sk_buff *skb, unsigned int flags, 2440 void (*destructor)(struct sock *sk, 2441 struct sk_buff *skb)); 2442int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2443 2444int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2445 enum skb_drop_reason *reason); 2446 2447static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2448{ 2449 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2450} 2451 2452int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2453struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2454 2455/* 2456 * Recover an error report and clear atomically 2457 */ 2458 2459static inline int sock_error(struct sock *sk) 2460{ 2461 int err; 2462 2463 /* Avoid an atomic operation for the common case. 2464 * This is racy since another cpu/thread can change sk_err under us. 2465 */ 2466 if (likely(data_race(!sk->sk_err))) 2467 return 0; 2468 2469 err = xchg(&sk->sk_err, 0); 2470 return -err; 2471} 2472 2473void sk_error_report(struct sock *sk); 2474 2475static inline unsigned long sock_wspace(struct sock *sk) 2476{ 2477 int amt = 0; 2478 2479 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2480 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2481 if (amt < 0) 2482 amt = 0; 2483 } 2484 return amt; 2485} 2486 2487/* Note: 2488 * We use sk->sk_wq_raw, from contexts knowing this 2489 * pointer is not NULL and cannot disappear/change. 2490 */ 2491static inline void sk_set_bit(int nr, struct sock *sk) 2492{ 2493 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2494 !sock_flag(sk, SOCK_FASYNC)) 2495 return; 2496 2497 set_bit(nr, &sk->sk_wq_raw->flags); 2498} 2499 2500static inline void sk_clear_bit(int nr, struct sock *sk) 2501{ 2502 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2503 !sock_flag(sk, SOCK_FASYNC)) 2504 return; 2505 2506 clear_bit(nr, &sk->sk_wq_raw->flags); 2507} 2508 2509static inline void sk_wake_async(const struct sock *sk, int how, int band) 2510{ 2511 if (sock_flag(sk, SOCK_FASYNC)) { 2512 rcu_read_lock(); 2513 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2514 rcu_read_unlock(); 2515 } 2516} 2517 2518/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2519 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2520 * Note: for send buffers, TCP works better if we can build two skbs at 2521 * minimum. 2522 */ 2523#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2524 2525#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2526#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2527 2528static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2529{ 2530 u32 val; 2531 2532 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2533 return; 2534 2535 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2536 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2537 2538 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2539} 2540 2541/** 2542 * sk_page_frag - return an appropriate page_frag 2543 * @sk: socket 2544 * 2545 * Use the per task page_frag instead of the per socket one for 2546 * optimization when we know that we're in process context and own 2547 * everything that's associated with %current. 2548 * 2549 * Both direct reclaim and page faults can nest inside other 2550 * socket operations and end up recursing into sk_page_frag() 2551 * while it's already in use: explicitly avoid task page_frag 2552 * when users disable sk_use_task_frag. 2553 * 2554 * Return: a per task page_frag if context allows that, 2555 * otherwise a per socket one. 2556 */ 2557static inline struct page_frag *sk_page_frag(struct sock *sk) 2558{ 2559 if (sk->sk_use_task_frag) 2560 return ¤t->task_frag; 2561 2562 return &sk->sk_frag; 2563} 2564 2565bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2566 2567/* 2568 * Default write policy as shown to user space via poll/select/SIGIO 2569 */ 2570static inline bool sock_writeable(const struct sock *sk) 2571{ 2572 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2573} 2574 2575static inline gfp_t gfp_any(void) 2576{ 2577 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2578} 2579 2580static inline gfp_t gfp_memcg_charge(void) 2581{ 2582 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2583} 2584 2585static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2586{ 2587 return noblock ? 0 : sk->sk_rcvtimeo; 2588} 2589 2590static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2591{ 2592 return noblock ? 0 : sk->sk_sndtimeo; 2593} 2594 2595static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2596{ 2597 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2598 2599 return v ?: 1; 2600} 2601 2602/* Alas, with timeout socket operations are not restartable. 2603 * Compare this to poll(). 2604 */ 2605static inline int sock_intr_errno(long timeo) 2606{ 2607 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2608} 2609 2610struct sock_skb_cb { 2611 u32 dropcount; 2612}; 2613 2614/* Store sock_skb_cb at the end of skb->cb[] so protocol families 2615 * using skb->cb[] would keep using it directly and utilize its 2616 * alignement guarantee. 2617 */ 2618#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2619 sizeof(struct sock_skb_cb))) 2620 2621#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2622 SOCK_SKB_CB_OFFSET)) 2623 2624#define sock_skb_cb_check_size(size) \ 2625 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2626 2627static inline void 2628sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2629{ 2630 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2631 atomic_read(&sk->sk_drops) : 0; 2632} 2633 2634static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2635{ 2636 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2637 2638 atomic_add(segs, &sk->sk_drops); 2639} 2640 2641static inline ktime_t sock_read_timestamp(struct sock *sk) 2642{ 2643#if BITS_PER_LONG==32 2644 unsigned int seq; 2645 ktime_t kt; 2646 2647 do { 2648 seq = read_seqbegin(&sk->sk_stamp_seq); 2649 kt = sk->sk_stamp; 2650 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2651 2652 return kt; 2653#else 2654 return READ_ONCE(sk->sk_stamp); 2655#endif 2656} 2657 2658static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2659{ 2660#if BITS_PER_LONG==32 2661 write_seqlock(&sk->sk_stamp_seq); 2662 sk->sk_stamp = kt; 2663 write_sequnlock(&sk->sk_stamp_seq); 2664#else 2665 WRITE_ONCE(sk->sk_stamp, kt); 2666#endif 2667} 2668 2669void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2670 struct sk_buff *skb); 2671void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2672 struct sk_buff *skb); 2673 2674static inline void 2675sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2676{ 2677 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2678 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2679 ktime_t kt = skb->tstamp; 2680 /* 2681 * generate control messages if 2682 * - receive time stamping in software requested 2683 * - software time stamp available and wanted 2684 * - hardware time stamps available and wanted 2685 */ 2686 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2687 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2688 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2689 (hwtstamps->hwtstamp && 2690 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2691 __sock_recv_timestamp(msg, sk, skb); 2692 else 2693 sock_write_timestamp(sk, kt); 2694 2695 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2696 __sock_recv_wifi_status(msg, sk, skb); 2697} 2698 2699void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2700 struct sk_buff *skb); 2701 2702#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2703static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2704 struct sk_buff *skb) 2705{ 2706#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2707 (1UL << SOCK_RCVTSTAMP) | \ 2708 (1UL << SOCK_RCVMARK)) 2709#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2710 SOF_TIMESTAMPING_RAW_HARDWARE) 2711 2712 if (sk->sk_flags & FLAGS_RECV_CMSGS || 2713 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY) 2714 __sock_recv_cmsgs(msg, sk, skb); 2715 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2716 sock_write_timestamp(sk, skb->tstamp); 2717 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2718 sock_write_timestamp(sk, 0); 2719} 2720 2721void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2722 2723/** 2724 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2725 * @sk: socket sending this packet 2726 * @tsflags: timestamping flags to use 2727 * @tx_flags: completed with instructions for time stamping 2728 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2729 * 2730 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2731 */ 2732static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2733 __u8 *tx_flags, __u32 *tskey) 2734{ 2735 if (unlikely(tsflags)) { 2736 __sock_tx_timestamp(tsflags, tx_flags); 2737 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2738 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2739 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2740 } 2741 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2742 *tx_flags |= SKBTX_WIFI_STATUS; 2743} 2744 2745static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2746 __u8 *tx_flags) 2747{ 2748 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2749} 2750 2751static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2752{ 2753 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2754 &skb_shinfo(skb)->tskey); 2755} 2756 2757static inline bool sk_is_inet(const struct sock *sk) 2758{ 2759 int family = READ_ONCE(sk->sk_family); 2760 2761 return family == AF_INET || family == AF_INET6; 2762} 2763 2764static inline bool sk_is_tcp(const struct sock *sk) 2765{ 2766 return sk_is_inet(sk) && 2767 sk->sk_type == SOCK_STREAM && 2768 sk->sk_protocol == IPPROTO_TCP; 2769} 2770 2771static inline bool sk_is_udp(const struct sock *sk) 2772{ 2773 return sk_is_inet(sk) && 2774 sk->sk_type == SOCK_DGRAM && 2775 sk->sk_protocol == IPPROTO_UDP; 2776} 2777 2778static inline bool sk_is_stream_unix(const struct sock *sk) 2779{ 2780 return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM; 2781} 2782 2783/** 2784 * sk_eat_skb - Release a skb if it is no longer needed 2785 * @sk: socket to eat this skb from 2786 * @skb: socket buffer to eat 2787 * 2788 * This routine must be called with interrupts disabled or with the socket 2789 * locked so that the sk_buff queue operation is ok. 2790*/ 2791static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2792{ 2793 __skb_unlink(skb, &sk->sk_receive_queue); 2794 __kfree_skb(skb); 2795} 2796 2797static inline bool 2798skb_sk_is_prefetched(struct sk_buff *skb) 2799{ 2800#ifdef CONFIG_INET 2801 return skb->destructor == sock_pfree; 2802#else 2803 return false; 2804#endif /* CONFIG_INET */ 2805} 2806 2807/* This helper checks if a socket is a full socket, 2808 * ie _not_ a timewait or request socket. 2809 */ 2810static inline bool sk_fullsock(const struct sock *sk) 2811{ 2812 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2813} 2814 2815static inline bool 2816sk_is_refcounted(struct sock *sk) 2817{ 2818 /* Only full sockets have sk->sk_flags. */ 2819 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2820} 2821 2822/* Checks if this SKB belongs to an HW offloaded socket 2823 * and whether any SW fallbacks are required based on dev. 2824 * Check decrypted mark in case skb_orphan() cleared socket. 2825 */ 2826static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2827 struct net_device *dev) 2828{ 2829#ifdef CONFIG_SOCK_VALIDATE_XMIT 2830 struct sock *sk = skb->sk; 2831 2832 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2833 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2834#ifdef CONFIG_TLS_DEVICE 2835 } else if (unlikely(skb->decrypted)) { 2836 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2837 kfree_skb(skb); 2838 skb = NULL; 2839#endif 2840 } 2841#endif 2842 2843 return skb; 2844} 2845 2846/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2847 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2848 */ 2849static inline bool sk_listener(const struct sock *sk) 2850{ 2851 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2852} 2853 2854void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2855int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2856 int type); 2857 2858bool sk_ns_capable(const struct sock *sk, 2859 struct user_namespace *user_ns, int cap); 2860bool sk_capable(const struct sock *sk, int cap); 2861bool sk_net_capable(const struct sock *sk, int cap); 2862 2863void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2864 2865/* Take into consideration the size of the struct sk_buff overhead in the 2866 * determination of these values, since that is non-constant across 2867 * platforms. This makes socket queueing behavior and performance 2868 * not depend upon such differences. 2869 */ 2870#define _SK_MEM_PACKETS 256 2871#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2872#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2873#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2874 2875extern __u32 sysctl_wmem_max; 2876extern __u32 sysctl_rmem_max; 2877 2878extern int sysctl_tstamp_allow_data; 2879 2880extern __u32 sysctl_wmem_default; 2881extern __u32 sysctl_rmem_default; 2882 2883#define SKB_FRAG_PAGE_ORDER get_order(32768) 2884DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2885 2886static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2887{ 2888 /* Does this proto have per netns sysctl_wmem ? */ 2889 if (proto->sysctl_wmem_offset) 2890 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2891 2892 return READ_ONCE(*proto->sysctl_wmem); 2893} 2894 2895static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2896{ 2897 /* Does this proto have per netns sysctl_rmem ? */ 2898 if (proto->sysctl_rmem_offset) 2899 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2900 2901 return READ_ONCE(*proto->sysctl_rmem); 2902} 2903 2904/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2905 * Some wifi drivers need to tweak it to get more chunks. 2906 * They can use this helper from their ndo_start_xmit() 2907 */ 2908static inline void sk_pacing_shift_update(struct sock *sk, int val) 2909{ 2910 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2911 return; 2912 WRITE_ONCE(sk->sk_pacing_shift, val); 2913} 2914 2915/* if a socket is bound to a device, check that the given device 2916 * index is either the same or that the socket is bound to an L3 2917 * master device and the given device index is also enslaved to 2918 * that L3 master 2919 */ 2920static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2921{ 2922 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2923 int mdif; 2924 2925 if (!bound_dev_if || bound_dev_if == dif) 2926 return true; 2927 2928 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2929 if (mdif && mdif == bound_dev_if) 2930 return true; 2931 2932 return false; 2933} 2934 2935void sock_def_readable(struct sock *sk); 2936 2937int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2938void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2939int sock_set_timestamping(struct sock *sk, int optname, 2940 struct so_timestamping timestamping); 2941 2942void sock_enable_timestamps(struct sock *sk); 2943void sock_no_linger(struct sock *sk); 2944void sock_set_keepalive(struct sock *sk); 2945void sock_set_priority(struct sock *sk, u32 priority); 2946void sock_set_rcvbuf(struct sock *sk, int val); 2947void sock_set_mark(struct sock *sk, u32 val); 2948void sock_set_reuseaddr(struct sock *sk); 2949void sock_set_reuseport(struct sock *sk); 2950void sock_set_sndtimeo(struct sock *sk, s64 secs); 2951 2952int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2953 2954int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2955int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2956 sockptr_t optval, int optlen, bool old_timeval); 2957 2958int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 2959 void __user *arg, void *karg, size_t size); 2960int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 2961static inline bool sk_is_readable(struct sock *sk) 2962{ 2963 if (sk->sk_prot->sock_is_readable) 2964 return sk->sk_prot->sock_is_readable(sk); 2965 return false; 2966} 2967#endif /* _SOCK_H */ 2968