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/filter.h> 60 #include <linux/rculist_nulls.h> 61 #include <linux/poll.h> 62 #include <linux/sockptr.h> 63 #include <linux/indirect_call_wrapper.h> 64 #include <linux/atomic.h> 65 #include <linux/refcount.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 72 /* 73 * This structure really needs to be cleaned up. 74 * Most of it is for TCP, and not used by any of 75 * the other protocols. 76 */ 77 78 /* Define this to get the SOCK_DBG debugging facility. */ 79 #define SOCK_DEBUGGING 80 #ifdef SOCK_DEBUGGING 81 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 82 printk(KERN_DEBUG msg); } while (0) 83 #else 84 /* Validate arguments and do nothing */ 85 static inline __printf(2, 3) 86 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 87 { 88 } 89 #endif 90 91 /* This is the per-socket lock. The spinlock provides a synchronization 92 * between user contexts and software interrupt processing, whereas the 93 * mini-semaphore synchronizes multiple users amongst themselves. 94 */ 95 typedef struct { 96 spinlock_t slock; 97 int owned; 98 wait_queue_head_t wq; 99 /* 100 * We express the mutex-alike socket_lock semantics 101 * to the lock validator by explicitly managing 102 * the slock as a lock variant (in addition to 103 * the slock itself): 104 */ 105 #ifdef CONFIG_DEBUG_LOCK_ALLOC 106 struct lockdep_map dep_map; 107 #endif 108 } socket_lock_t; 109 110 struct sock; 111 struct proto; 112 struct net; 113 114 typedef __u32 __bitwise __portpair; 115 typedef __u64 __bitwise __addrpair; 116 117 /** 118 * struct sock_common - minimal network layer representation of sockets 119 * @skc_daddr: Foreign IPv4 addr 120 * @skc_rcv_saddr: Bound local IPv4 addr 121 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 122 * @skc_hash: hash value used with various protocol lookup tables 123 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 124 * @skc_dport: placeholder for inet_dport/tw_dport 125 * @skc_num: placeholder for inet_num/tw_num 126 * @skc_portpair: __u32 union of @skc_dport & @skc_num 127 * @skc_family: network address family 128 * @skc_state: Connection state 129 * @skc_reuse: %SO_REUSEADDR setting 130 * @skc_reuseport: %SO_REUSEPORT setting 131 * @skc_ipv6only: socket is IPV6 only 132 * @skc_net_refcnt: socket is using net ref counting 133 * @skc_bound_dev_if: bound device index if != 0 134 * @skc_bind_node: bind hash linkage for various protocol lookup tables 135 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 136 * @skc_prot: protocol handlers inside a network family 137 * @skc_net: reference to the network namespace of this socket 138 * @skc_v6_daddr: IPV6 destination address 139 * @skc_v6_rcv_saddr: IPV6 source address 140 * @skc_cookie: socket's cookie value 141 * @skc_node: main hash linkage for various protocol lookup tables 142 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 143 * @skc_tx_queue_mapping: tx queue number for this connection 144 * @skc_rx_queue_mapping: rx queue number for this connection 145 * @skc_flags: place holder for sk_flags 146 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 147 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 148 * @skc_listener: connection request listener socket (aka rsk_listener) 149 * [union with @skc_flags] 150 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 151 * [union with @skc_flags] 152 * @skc_incoming_cpu: record/match cpu processing incoming packets 153 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 154 * [union with @skc_incoming_cpu] 155 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 156 * [union with @skc_incoming_cpu] 157 * @skc_refcnt: reference count 158 * 159 * This is the minimal network layer representation of sockets, the header 160 * for struct sock and struct inet_timewait_sock. 161 */ 162 struct sock_common { 163 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 164 * address on 64bit arches : cf INET_MATCH() 165 */ 166 union { 167 __addrpair skc_addrpair; 168 struct { 169 __be32 skc_daddr; 170 __be32 skc_rcv_saddr; 171 }; 172 }; 173 union { 174 unsigned int skc_hash; 175 __u16 skc_u16hashes[2]; 176 }; 177 /* skc_dport && skc_num must be grouped as well */ 178 union { 179 __portpair skc_portpair; 180 struct { 181 __be16 skc_dport; 182 __u16 skc_num; 183 }; 184 }; 185 186 unsigned short skc_family; 187 volatile unsigned char skc_state; 188 unsigned char skc_reuse:4; 189 unsigned char skc_reuseport:1; 190 unsigned char skc_ipv6only:1; 191 unsigned char skc_net_refcnt:1; 192 int skc_bound_dev_if; 193 union { 194 struct hlist_node skc_bind_node; 195 struct hlist_node skc_portaddr_node; 196 }; 197 struct proto *skc_prot; 198 possible_net_t skc_net; 199 200 #if IS_ENABLED(CONFIG_IPV6) 201 struct in6_addr skc_v6_daddr; 202 struct in6_addr skc_v6_rcv_saddr; 203 #endif 204 205 atomic64_t skc_cookie; 206 207 /* following fields are padding to force 208 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 209 * assuming IPV6 is enabled. We use this padding differently 210 * for different kind of 'sockets' 211 */ 212 union { 213 unsigned long skc_flags; 214 struct sock *skc_listener; /* request_sock */ 215 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 216 }; 217 /* 218 * fields between dontcopy_begin/dontcopy_end 219 * are not copied in sock_copy() 220 */ 221 /* private: */ 222 int skc_dontcopy_begin[0]; 223 /* public: */ 224 union { 225 struct hlist_node skc_node; 226 struct hlist_nulls_node skc_nulls_node; 227 }; 228 unsigned short skc_tx_queue_mapping; 229 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 230 unsigned short skc_rx_queue_mapping; 231 #endif 232 union { 233 int skc_incoming_cpu; 234 u32 skc_rcv_wnd; 235 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 236 }; 237 238 refcount_t skc_refcnt; 239 /* private: */ 240 int skc_dontcopy_end[0]; 241 union { 242 u32 skc_rxhash; 243 u32 skc_window_clamp; 244 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 245 }; 246 /* public: */ 247 }; 248 249 struct bpf_local_storage; 250 251 /** 252 * struct sock - network layer representation of sockets 253 * @__sk_common: shared layout with inet_timewait_sock 254 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 255 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 256 * @sk_lock: synchronizer 257 * @sk_kern_sock: True if sock is using kernel lock classes 258 * @sk_rcvbuf: size of receive buffer in bytes 259 * @sk_wq: sock wait queue and async head 260 * @sk_rx_dst: receive input route used by early demux 261 * @sk_dst_cache: destination cache 262 * @sk_dst_pending_confirm: need to confirm neighbour 263 * @sk_policy: flow policy 264 * @sk_rx_skb_cache: cache copy of recently accessed RX skb 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_napi_id: id of the last napi context to receive data for sk 273 * @sk_ll_usec: usecs to busypoll when there is no data 274 * @sk_allocation: allocation mode 275 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 276 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 277 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 278 * @sk_sndbuf: size of send buffer in bytes 279 * @__sk_flags_offset: empty field used to determine location of bitfield 280 * @sk_padding: unused element for alignment 281 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 282 * @sk_no_check_rx: allow zero checksum in RX packets 283 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 284 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 285 * @sk_route_forced_caps: static, forced route capabilities 286 * (set in tcp_init_sock()) 287 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 288 * @sk_gso_max_size: Maximum GSO segment size to build 289 * @sk_gso_max_segs: Maximum number of GSO segments 290 * @sk_pacing_shift: scaling factor for TCP Small Queues 291 * @sk_lingertime: %SO_LINGER l_linger setting 292 * @sk_backlog: always used with the per-socket spinlock held 293 * @sk_callback_lock: used with the callbacks in the end of this struct 294 * @sk_error_queue: rarely used 295 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 296 * IPV6_ADDRFORM for instance) 297 * @sk_err: last error 298 * @sk_err_soft: errors that don't cause failure but are the cause of a 299 * persistent failure not just 'timed out' 300 * @sk_drops: raw/udp drops counter 301 * @sk_ack_backlog: current listen backlog 302 * @sk_max_ack_backlog: listen backlog set in listen() 303 * @sk_uid: user id of owner 304 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 305 * @sk_busy_poll_budget: napi processing budget when busypolling 306 * @sk_priority: %SO_PRIORITY setting 307 * @sk_type: socket type (%SOCK_STREAM, etc) 308 * @sk_protocol: which protocol this socket belongs in this network family 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_filter: socket filtering instructions 316 * @sk_timer: sock cleanup timer 317 * @sk_stamp: time stamp of last packet received 318 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 319 * @sk_tsflags: SO_TIMESTAMPING socket options 320 * @sk_tskey: counter to disambiguate concurrent tstamp requests 321 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 322 * @sk_socket: Identd and reporting IO signals 323 * @sk_user_data: RPC layer private data 324 * @sk_frag: cached page frag 325 * @sk_peek_off: current peek_offset value 326 * @sk_send_head: front of stuff to transmit 327 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 328 * @sk_tx_skb_cache: cache copy of recently accessed TX skb 329 * @sk_security: used by security modules 330 * @sk_mark: generic packet mark 331 * @sk_cgrp_data: cgroup data for this cgroup 332 * @sk_memcg: this socket's memory cgroup association 333 * @sk_write_pending: a write to stream socket waits to start 334 * @sk_state_change: callback to indicate change in the state of the sock 335 * @sk_data_ready: callback to indicate there is data to be processed 336 * @sk_write_space: callback to indicate there is bf sending space available 337 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 338 * @sk_backlog_rcv: callback to process the backlog 339 * @sk_validate_xmit_skb: ptr to an optional validate function 340 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 341 * @sk_reuseport_cb: reuseport group container 342 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 343 * @sk_rcu: used during RCU grace period 344 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 345 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 346 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 347 * @sk_txtime_unused: unused txtime flags 348 */ 349 struct sock { 350 /* 351 * Now struct inet_timewait_sock also uses sock_common, so please just 352 * don't add nothing before this first member (__sk_common) --acme 353 */ 354 struct sock_common __sk_common; 355 #define sk_node __sk_common.skc_node 356 #define sk_nulls_node __sk_common.skc_nulls_node 357 #define sk_refcnt __sk_common.skc_refcnt 358 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 359 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 360 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 361 #endif 362 363 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 364 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 365 #define sk_hash __sk_common.skc_hash 366 #define sk_portpair __sk_common.skc_portpair 367 #define sk_num __sk_common.skc_num 368 #define sk_dport __sk_common.skc_dport 369 #define sk_addrpair __sk_common.skc_addrpair 370 #define sk_daddr __sk_common.skc_daddr 371 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 372 #define sk_family __sk_common.skc_family 373 #define sk_state __sk_common.skc_state 374 #define sk_reuse __sk_common.skc_reuse 375 #define sk_reuseport __sk_common.skc_reuseport 376 #define sk_ipv6only __sk_common.skc_ipv6only 377 #define sk_net_refcnt __sk_common.skc_net_refcnt 378 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 379 #define sk_bind_node __sk_common.skc_bind_node 380 #define sk_prot __sk_common.skc_prot 381 #define sk_net __sk_common.skc_net 382 #define sk_v6_daddr __sk_common.skc_v6_daddr 383 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 384 #define sk_cookie __sk_common.skc_cookie 385 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 386 #define sk_flags __sk_common.skc_flags 387 #define sk_rxhash __sk_common.skc_rxhash 388 389 socket_lock_t sk_lock; 390 atomic_t sk_drops; 391 int sk_rcvlowat; 392 struct sk_buff_head sk_error_queue; 393 struct sk_buff *sk_rx_skb_cache; 394 struct sk_buff_head sk_receive_queue; 395 /* 396 * The backlog queue is special, it is always used with 397 * the per-socket spinlock held and requires low latency 398 * access. Therefore we special case it's implementation. 399 * Note : rmem_alloc is in this structure to fill a hole 400 * on 64bit arches, not because its logically part of 401 * backlog. 402 */ 403 struct { 404 atomic_t rmem_alloc; 405 int len; 406 struct sk_buff *head; 407 struct sk_buff *tail; 408 } sk_backlog; 409 #define sk_rmem_alloc sk_backlog.rmem_alloc 410 411 int sk_forward_alloc; 412 #ifdef CONFIG_NET_RX_BUSY_POLL 413 unsigned int sk_ll_usec; 414 /* ===== mostly read cache line ===== */ 415 unsigned int sk_napi_id; 416 #endif 417 int sk_rcvbuf; 418 419 struct sk_filter __rcu *sk_filter; 420 union { 421 struct socket_wq __rcu *sk_wq; 422 /* private: */ 423 struct socket_wq *sk_wq_raw; 424 /* public: */ 425 }; 426 #ifdef CONFIG_XFRM 427 struct xfrm_policy __rcu *sk_policy[2]; 428 #endif 429 struct dst_entry *sk_rx_dst; 430 struct dst_entry __rcu *sk_dst_cache; 431 atomic_t sk_omem_alloc; 432 int sk_sndbuf; 433 434 /* ===== cache line for TX ===== */ 435 int sk_wmem_queued; 436 refcount_t sk_wmem_alloc; 437 unsigned long sk_tsq_flags; 438 union { 439 struct sk_buff *sk_send_head; 440 struct rb_root tcp_rtx_queue; 441 }; 442 struct sk_buff *sk_tx_skb_cache; 443 struct sk_buff_head sk_write_queue; 444 __s32 sk_peek_off; 445 int sk_write_pending; 446 __u32 sk_dst_pending_confirm; 447 u32 sk_pacing_status; /* see enum sk_pacing */ 448 long sk_sndtimeo; 449 struct timer_list sk_timer; 450 __u32 sk_priority; 451 __u32 sk_mark; 452 unsigned long sk_pacing_rate; /* bytes per second */ 453 unsigned long sk_max_pacing_rate; 454 struct page_frag sk_frag; 455 netdev_features_t sk_route_caps; 456 netdev_features_t sk_route_nocaps; 457 netdev_features_t sk_route_forced_caps; 458 int sk_gso_type; 459 unsigned int sk_gso_max_size; 460 gfp_t sk_allocation; 461 __u32 sk_txhash; 462 463 /* 464 * Because of non atomicity rules, all 465 * changes are protected by socket lock. 466 */ 467 u8 sk_padding : 1, 468 sk_kern_sock : 1, 469 sk_no_check_tx : 1, 470 sk_no_check_rx : 1, 471 sk_userlocks : 4; 472 u8 sk_pacing_shift; 473 u16 sk_type; 474 u16 sk_protocol; 475 u16 sk_gso_max_segs; 476 unsigned long sk_lingertime; 477 struct proto *sk_prot_creator; 478 rwlock_t sk_callback_lock; 479 int sk_err, 480 sk_err_soft; 481 u32 sk_ack_backlog; 482 u32 sk_max_ack_backlog; 483 kuid_t sk_uid; 484 #ifdef CONFIG_NET_RX_BUSY_POLL 485 u8 sk_prefer_busy_poll; 486 u16 sk_busy_poll_budget; 487 #endif 488 struct pid *sk_peer_pid; 489 const struct cred *sk_peer_cred; 490 long sk_rcvtimeo; 491 ktime_t sk_stamp; 492 #if BITS_PER_LONG==32 493 seqlock_t sk_stamp_seq; 494 #endif 495 u16 sk_tsflags; 496 u8 sk_shutdown; 497 u32 sk_tskey; 498 atomic_t sk_zckey; 499 500 u8 sk_clockid; 501 u8 sk_txtime_deadline_mode : 1, 502 sk_txtime_report_errors : 1, 503 sk_txtime_unused : 6; 504 505 struct socket *sk_socket; 506 void *sk_user_data; 507 #ifdef CONFIG_SECURITY 508 void *sk_security; 509 #endif 510 struct sock_cgroup_data sk_cgrp_data; 511 struct mem_cgroup *sk_memcg; 512 void (*sk_state_change)(struct sock *sk); 513 void (*sk_data_ready)(struct sock *sk); 514 void (*sk_write_space)(struct sock *sk); 515 void (*sk_error_report)(struct sock *sk); 516 int (*sk_backlog_rcv)(struct sock *sk, 517 struct sk_buff *skb); 518 #ifdef CONFIG_SOCK_VALIDATE_XMIT 519 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 520 struct net_device *dev, 521 struct sk_buff *skb); 522 #endif 523 void (*sk_destruct)(struct sock *sk); 524 struct sock_reuseport __rcu *sk_reuseport_cb; 525 #ifdef CONFIG_BPF_SYSCALL 526 struct bpf_local_storage __rcu *sk_bpf_storage; 527 #endif 528 struct rcu_head sk_rcu; 529 }; 530 531 enum sk_pacing { 532 SK_PACING_NONE = 0, 533 SK_PACING_NEEDED = 1, 534 SK_PACING_FQ = 2, 535 }; 536 537 /* Pointer stored in sk_user_data might not be suitable for copying 538 * when cloning the socket. For instance, it can point to a reference 539 * counted object. sk_user_data bottom bit is set if pointer must not 540 * be copied. 541 */ 542 #define SK_USER_DATA_NOCOPY 1UL 543 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 544 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 545 546 /** 547 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 548 * @sk: socket 549 */ 550 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 551 { 552 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 553 } 554 555 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 556 557 #define rcu_dereference_sk_user_data(sk) \ 558 ({ \ 559 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 560 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 561 }) 562 #define rcu_assign_sk_user_data(sk, ptr) \ 563 ({ \ 564 uintptr_t __tmp = (uintptr_t)(ptr); \ 565 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 566 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 567 }) 568 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 569 ({ \ 570 uintptr_t __tmp = (uintptr_t)(ptr); \ 571 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 572 rcu_assign_pointer(__sk_user_data((sk)), \ 573 __tmp | SK_USER_DATA_NOCOPY); \ 574 }) 575 576 /* 577 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 578 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 579 * on a socket means that the socket will reuse everybody else's port 580 * without looking at the other's sk_reuse value. 581 */ 582 583 #define SK_NO_REUSE 0 584 #define SK_CAN_REUSE 1 585 #define SK_FORCE_REUSE 2 586 587 int sk_set_peek_off(struct sock *sk, int val); 588 589 static inline int sk_peek_offset(struct sock *sk, int flags) 590 { 591 if (unlikely(flags & MSG_PEEK)) { 592 return READ_ONCE(sk->sk_peek_off); 593 } 594 595 return 0; 596 } 597 598 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 599 { 600 s32 off = READ_ONCE(sk->sk_peek_off); 601 602 if (unlikely(off >= 0)) { 603 off = max_t(s32, off - val, 0); 604 WRITE_ONCE(sk->sk_peek_off, off); 605 } 606 } 607 608 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 609 { 610 sk_peek_offset_bwd(sk, -val); 611 } 612 613 /* 614 * Hashed lists helper routines 615 */ 616 static inline struct sock *sk_entry(const struct hlist_node *node) 617 { 618 return hlist_entry(node, struct sock, sk_node); 619 } 620 621 static inline struct sock *__sk_head(const struct hlist_head *head) 622 { 623 return hlist_entry(head->first, struct sock, sk_node); 624 } 625 626 static inline struct sock *sk_head(const struct hlist_head *head) 627 { 628 return hlist_empty(head) ? NULL : __sk_head(head); 629 } 630 631 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 632 { 633 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 634 } 635 636 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 637 { 638 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 639 } 640 641 static inline struct sock *sk_next(const struct sock *sk) 642 { 643 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 644 } 645 646 static inline struct sock *sk_nulls_next(const struct sock *sk) 647 { 648 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 649 hlist_nulls_entry(sk->sk_nulls_node.next, 650 struct sock, sk_nulls_node) : 651 NULL; 652 } 653 654 static inline bool sk_unhashed(const struct sock *sk) 655 { 656 return hlist_unhashed(&sk->sk_node); 657 } 658 659 static inline bool sk_hashed(const struct sock *sk) 660 { 661 return !sk_unhashed(sk); 662 } 663 664 static inline void sk_node_init(struct hlist_node *node) 665 { 666 node->pprev = NULL; 667 } 668 669 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 670 { 671 node->pprev = NULL; 672 } 673 674 static inline void __sk_del_node(struct sock *sk) 675 { 676 __hlist_del(&sk->sk_node); 677 } 678 679 /* NB: equivalent to hlist_del_init_rcu */ 680 static inline bool __sk_del_node_init(struct sock *sk) 681 { 682 if (sk_hashed(sk)) { 683 __sk_del_node(sk); 684 sk_node_init(&sk->sk_node); 685 return true; 686 } 687 return false; 688 } 689 690 /* Grab socket reference count. This operation is valid only 691 when sk is ALREADY grabbed f.e. it is found in hash table 692 or a list and the lookup is made under lock preventing hash table 693 modifications. 694 */ 695 696 static __always_inline void sock_hold(struct sock *sk) 697 { 698 refcount_inc(&sk->sk_refcnt); 699 } 700 701 /* Ungrab socket in the context, which assumes that socket refcnt 702 cannot hit zero, f.e. it is true in context of any socketcall. 703 */ 704 static __always_inline void __sock_put(struct sock *sk) 705 { 706 refcount_dec(&sk->sk_refcnt); 707 } 708 709 static inline bool sk_del_node_init(struct sock *sk) 710 { 711 bool rc = __sk_del_node_init(sk); 712 713 if (rc) { 714 /* paranoid for a while -acme */ 715 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 716 __sock_put(sk); 717 } 718 return rc; 719 } 720 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 721 722 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 723 { 724 if (sk_hashed(sk)) { 725 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 726 return true; 727 } 728 return false; 729 } 730 731 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 732 { 733 bool rc = __sk_nulls_del_node_init_rcu(sk); 734 735 if (rc) { 736 /* paranoid for a while -acme */ 737 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 738 __sock_put(sk); 739 } 740 return rc; 741 } 742 743 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 744 { 745 hlist_add_head(&sk->sk_node, list); 746 } 747 748 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 749 { 750 sock_hold(sk); 751 __sk_add_node(sk, list); 752 } 753 754 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 755 { 756 sock_hold(sk); 757 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 758 sk->sk_family == AF_INET6) 759 hlist_add_tail_rcu(&sk->sk_node, list); 760 else 761 hlist_add_head_rcu(&sk->sk_node, list); 762 } 763 764 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 765 { 766 sock_hold(sk); 767 hlist_add_tail_rcu(&sk->sk_node, list); 768 } 769 770 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 771 { 772 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 773 } 774 775 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 776 { 777 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 778 } 779 780 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 781 { 782 sock_hold(sk); 783 __sk_nulls_add_node_rcu(sk, list); 784 } 785 786 static inline void __sk_del_bind_node(struct sock *sk) 787 { 788 __hlist_del(&sk->sk_bind_node); 789 } 790 791 static inline void sk_add_bind_node(struct sock *sk, 792 struct hlist_head *list) 793 { 794 hlist_add_head(&sk->sk_bind_node, list); 795 } 796 797 #define sk_for_each(__sk, list) \ 798 hlist_for_each_entry(__sk, list, sk_node) 799 #define sk_for_each_rcu(__sk, list) \ 800 hlist_for_each_entry_rcu(__sk, list, sk_node) 801 #define sk_nulls_for_each(__sk, node, list) \ 802 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 803 #define sk_nulls_for_each_rcu(__sk, node, list) \ 804 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 805 #define sk_for_each_from(__sk) \ 806 hlist_for_each_entry_from(__sk, sk_node) 807 #define sk_nulls_for_each_from(__sk, node) \ 808 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 809 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 810 #define sk_for_each_safe(__sk, tmp, list) \ 811 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 812 #define sk_for_each_bound(__sk, list) \ 813 hlist_for_each_entry(__sk, list, sk_bind_node) 814 815 /** 816 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 817 * @tpos: the type * to use as a loop cursor. 818 * @pos: the &struct hlist_node to use as a loop cursor. 819 * @head: the head for your list. 820 * @offset: offset of hlist_node within the struct. 821 * 822 */ 823 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 824 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 825 pos != NULL && \ 826 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 827 pos = rcu_dereference(hlist_next_rcu(pos))) 828 829 static inline struct user_namespace *sk_user_ns(struct sock *sk) 830 { 831 /* Careful only use this in a context where these parameters 832 * can not change and must all be valid, such as recvmsg from 833 * userspace. 834 */ 835 return sk->sk_socket->file->f_cred->user_ns; 836 } 837 838 /* Sock flags */ 839 enum sock_flags { 840 SOCK_DEAD, 841 SOCK_DONE, 842 SOCK_URGINLINE, 843 SOCK_KEEPOPEN, 844 SOCK_LINGER, 845 SOCK_DESTROY, 846 SOCK_BROADCAST, 847 SOCK_TIMESTAMP, 848 SOCK_ZAPPED, 849 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 850 SOCK_DBG, /* %SO_DEBUG setting */ 851 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 852 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 853 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 854 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 855 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 856 SOCK_FASYNC, /* fasync() active */ 857 SOCK_RXQ_OVFL, 858 SOCK_ZEROCOPY, /* buffers from userspace */ 859 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 860 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 861 * Will use last 4 bytes of packet sent from 862 * user-space instead. 863 */ 864 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 865 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 866 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 867 SOCK_TXTIME, 868 SOCK_XDP, /* XDP is attached */ 869 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 870 }; 871 872 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 873 874 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 875 { 876 nsk->sk_flags = osk->sk_flags; 877 } 878 879 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 880 { 881 __set_bit(flag, &sk->sk_flags); 882 } 883 884 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 885 { 886 __clear_bit(flag, &sk->sk_flags); 887 } 888 889 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 890 int valbool) 891 { 892 if (valbool) 893 sock_set_flag(sk, bit); 894 else 895 sock_reset_flag(sk, bit); 896 } 897 898 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 899 { 900 return test_bit(flag, &sk->sk_flags); 901 } 902 903 #ifdef CONFIG_NET 904 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 905 static inline int sk_memalloc_socks(void) 906 { 907 return static_branch_unlikely(&memalloc_socks_key); 908 } 909 910 void __receive_sock(struct file *file); 911 #else 912 913 static inline int sk_memalloc_socks(void) 914 { 915 return 0; 916 } 917 918 static inline void __receive_sock(struct file *file) 919 { } 920 #endif 921 922 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 923 { 924 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 925 } 926 927 static inline void sk_acceptq_removed(struct sock *sk) 928 { 929 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 930 } 931 932 static inline void sk_acceptq_added(struct sock *sk) 933 { 934 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 935 } 936 937 static inline bool sk_acceptq_is_full(const struct sock *sk) 938 { 939 return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 940 } 941 942 /* 943 * Compute minimal free write space needed to queue new packets. 944 */ 945 static inline int sk_stream_min_wspace(const struct sock *sk) 946 { 947 return READ_ONCE(sk->sk_wmem_queued) >> 1; 948 } 949 950 static inline int sk_stream_wspace(const struct sock *sk) 951 { 952 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 953 } 954 955 static inline void sk_wmem_queued_add(struct sock *sk, int val) 956 { 957 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 958 } 959 960 void sk_stream_write_space(struct sock *sk); 961 962 /* OOB backlog add */ 963 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 964 { 965 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 966 skb_dst_force(skb); 967 968 if (!sk->sk_backlog.tail) 969 WRITE_ONCE(sk->sk_backlog.head, skb); 970 else 971 sk->sk_backlog.tail->next = skb; 972 973 WRITE_ONCE(sk->sk_backlog.tail, skb); 974 skb->next = NULL; 975 } 976 977 /* 978 * Take into account size of receive queue and backlog queue 979 * Do not take into account this skb truesize, 980 * to allow even a single big packet to come. 981 */ 982 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 983 { 984 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 985 986 return qsize > limit; 987 } 988 989 /* The per-socket spinlock must be held here. */ 990 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 991 unsigned int limit) 992 { 993 if (sk_rcvqueues_full(sk, limit)) 994 return -ENOBUFS; 995 996 /* 997 * If the skb was allocated from pfmemalloc reserves, only 998 * allow SOCK_MEMALLOC sockets to use it as this socket is 999 * helping free memory 1000 */ 1001 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1002 return -ENOMEM; 1003 1004 __sk_add_backlog(sk, skb); 1005 sk->sk_backlog.len += skb->truesize; 1006 return 0; 1007 } 1008 1009 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1010 1011 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1012 { 1013 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1014 return __sk_backlog_rcv(sk, skb); 1015 1016 return sk->sk_backlog_rcv(sk, skb); 1017 } 1018 1019 static inline void sk_incoming_cpu_update(struct sock *sk) 1020 { 1021 int cpu = raw_smp_processor_id(); 1022 1023 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1024 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1025 } 1026 1027 static inline void sock_rps_record_flow_hash(__u32 hash) 1028 { 1029 #ifdef CONFIG_RPS 1030 struct rps_sock_flow_table *sock_flow_table; 1031 1032 rcu_read_lock(); 1033 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1034 rps_record_sock_flow(sock_flow_table, hash); 1035 rcu_read_unlock(); 1036 #endif 1037 } 1038 1039 static inline void sock_rps_record_flow(const struct sock *sk) 1040 { 1041 #ifdef CONFIG_RPS 1042 if (static_branch_unlikely(&rfs_needed)) { 1043 /* Reading sk->sk_rxhash might incur an expensive cache line 1044 * miss. 1045 * 1046 * TCP_ESTABLISHED does cover almost all states where RFS 1047 * might be useful, and is cheaper [1] than testing : 1048 * IPv4: inet_sk(sk)->inet_daddr 1049 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1050 * OR an additional socket flag 1051 * [1] : sk_state and sk_prot are in the same cache line. 1052 */ 1053 if (sk->sk_state == TCP_ESTABLISHED) 1054 sock_rps_record_flow_hash(sk->sk_rxhash); 1055 } 1056 #endif 1057 } 1058 1059 static inline void sock_rps_save_rxhash(struct sock *sk, 1060 const struct sk_buff *skb) 1061 { 1062 #ifdef CONFIG_RPS 1063 if (unlikely(sk->sk_rxhash != skb->hash)) 1064 sk->sk_rxhash = skb->hash; 1065 #endif 1066 } 1067 1068 static inline void sock_rps_reset_rxhash(struct sock *sk) 1069 { 1070 #ifdef CONFIG_RPS 1071 sk->sk_rxhash = 0; 1072 #endif 1073 } 1074 1075 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1076 ({ int __rc; \ 1077 release_sock(__sk); \ 1078 __rc = __condition; \ 1079 if (!__rc) { \ 1080 *(__timeo) = wait_woken(__wait, \ 1081 TASK_INTERRUPTIBLE, \ 1082 *(__timeo)); \ 1083 } \ 1084 sched_annotate_sleep(); \ 1085 lock_sock(__sk); \ 1086 __rc = __condition; \ 1087 __rc; \ 1088 }) 1089 1090 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1091 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1092 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1093 int sk_stream_error(struct sock *sk, int flags, int err); 1094 void sk_stream_kill_queues(struct sock *sk); 1095 void sk_set_memalloc(struct sock *sk); 1096 void sk_clear_memalloc(struct sock *sk); 1097 1098 void __sk_flush_backlog(struct sock *sk); 1099 1100 static inline bool sk_flush_backlog(struct sock *sk) 1101 { 1102 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1103 __sk_flush_backlog(sk); 1104 return true; 1105 } 1106 return false; 1107 } 1108 1109 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1110 1111 struct request_sock_ops; 1112 struct timewait_sock_ops; 1113 struct inet_hashinfo; 1114 struct raw_hashinfo; 1115 struct smc_hashinfo; 1116 struct module; 1117 1118 /* 1119 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1120 * un-modified. Special care is taken when initializing object to zero. 1121 */ 1122 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1123 { 1124 if (offsetof(struct sock, sk_node.next) != 0) 1125 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1126 memset(&sk->sk_node.pprev, 0, 1127 size - offsetof(struct sock, sk_node.pprev)); 1128 } 1129 1130 /* Networking protocol blocks we attach to sockets. 1131 * socket layer -> transport layer interface 1132 */ 1133 struct proto { 1134 void (*close)(struct sock *sk, 1135 long timeout); 1136 int (*pre_connect)(struct sock *sk, 1137 struct sockaddr *uaddr, 1138 int addr_len); 1139 int (*connect)(struct sock *sk, 1140 struct sockaddr *uaddr, 1141 int addr_len); 1142 int (*disconnect)(struct sock *sk, int flags); 1143 1144 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1145 bool kern); 1146 1147 int (*ioctl)(struct sock *sk, int cmd, 1148 unsigned long arg); 1149 int (*init)(struct sock *sk); 1150 void (*destroy)(struct sock *sk); 1151 void (*shutdown)(struct sock *sk, int how); 1152 int (*setsockopt)(struct sock *sk, int level, 1153 int optname, sockptr_t optval, 1154 unsigned int optlen); 1155 int (*getsockopt)(struct sock *sk, int level, 1156 int optname, char __user *optval, 1157 int __user *option); 1158 void (*keepalive)(struct sock *sk, int valbool); 1159 #ifdef CONFIG_COMPAT 1160 int (*compat_ioctl)(struct sock *sk, 1161 unsigned int cmd, unsigned long arg); 1162 #endif 1163 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1164 size_t len); 1165 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1166 size_t len, int noblock, int flags, 1167 int *addr_len); 1168 int (*sendpage)(struct sock *sk, struct page *page, 1169 int offset, size_t size, int flags); 1170 int (*bind)(struct sock *sk, 1171 struct sockaddr *addr, int addr_len); 1172 int (*bind_add)(struct sock *sk, 1173 struct sockaddr *addr, int addr_len); 1174 1175 int (*backlog_rcv) (struct sock *sk, 1176 struct sk_buff *skb); 1177 bool (*bpf_bypass_getsockopt)(int level, 1178 int optname); 1179 1180 void (*release_cb)(struct sock *sk); 1181 1182 /* Keeping track of sk's, looking them up, and port selection methods. */ 1183 int (*hash)(struct sock *sk); 1184 void (*unhash)(struct sock *sk); 1185 void (*rehash)(struct sock *sk); 1186 int (*get_port)(struct sock *sk, unsigned short snum); 1187 #ifdef CONFIG_BPF_SYSCALL 1188 int (*psock_update_sk_prot)(struct sock *sk, bool restore); 1189 #endif 1190 1191 /* Keeping track of sockets in use */ 1192 #ifdef CONFIG_PROC_FS 1193 unsigned int inuse_idx; 1194 #endif 1195 1196 bool (*stream_memory_free)(const struct sock *sk, int wake); 1197 bool (*stream_memory_read)(const struct sock *sk); 1198 /* Memory pressure */ 1199 void (*enter_memory_pressure)(struct sock *sk); 1200 void (*leave_memory_pressure)(struct sock *sk); 1201 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1202 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1203 /* 1204 * Pressure flag: try to collapse. 1205 * Technical note: it is used by multiple contexts non atomically. 1206 * All the __sk_mem_schedule() is of this nature: accounting 1207 * is strict, actions are advisory and have some latency. 1208 */ 1209 unsigned long *memory_pressure; 1210 long *sysctl_mem; 1211 1212 int *sysctl_wmem; 1213 int *sysctl_rmem; 1214 u32 sysctl_wmem_offset; 1215 u32 sysctl_rmem_offset; 1216 1217 int max_header; 1218 bool no_autobind; 1219 1220 struct kmem_cache *slab; 1221 unsigned int obj_size; 1222 slab_flags_t slab_flags; 1223 unsigned int useroffset; /* Usercopy region offset */ 1224 unsigned int usersize; /* Usercopy region size */ 1225 1226 struct percpu_counter *orphan_count; 1227 1228 struct request_sock_ops *rsk_prot; 1229 struct timewait_sock_ops *twsk_prot; 1230 1231 union { 1232 struct inet_hashinfo *hashinfo; 1233 struct udp_table *udp_table; 1234 struct raw_hashinfo *raw_hash; 1235 struct smc_hashinfo *smc_hash; 1236 } h; 1237 1238 struct module *owner; 1239 1240 char name[32]; 1241 1242 struct list_head node; 1243 #ifdef SOCK_REFCNT_DEBUG 1244 atomic_t socks; 1245 #endif 1246 int (*diag_destroy)(struct sock *sk, int err); 1247 } __randomize_layout; 1248 1249 int proto_register(struct proto *prot, int alloc_slab); 1250 void proto_unregister(struct proto *prot); 1251 int sock_load_diag_module(int family, int protocol); 1252 1253 #ifdef SOCK_REFCNT_DEBUG 1254 static inline void sk_refcnt_debug_inc(struct sock *sk) 1255 { 1256 atomic_inc(&sk->sk_prot->socks); 1257 } 1258 1259 static inline void sk_refcnt_debug_dec(struct sock *sk) 1260 { 1261 atomic_dec(&sk->sk_prot->socks); 1262 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1263 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1264 } 1265 1266 static inline void sk_refcnt_debug_release(const struct sock *sk) 1267 { 1268 if (refcount_read(&sk->sk_refcnt) != 1) 1269 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1270 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1271 } 1272 #else /* SOCK_REFCNT_DEBUG */ 1273 #define sk_refcnt_debug_inc(sk) do { } while (0) 1274 #define sk_refcnt_debug_dec(sk) do { } while (0) 1275 #define sk_refcnt_debug_release(sk) do { } while (0) 1276 #endif /* SOCK_REFCNT_DEBUG */ 1277 1278 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1279 1280 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1281 { 1282 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1283 return false; 1284 1285 #ifdef CONFIG_INET 1286 return sk->sk_prot->stream_memory_free ? 1287 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free, 1288 tcp_stream_memory_free, 1289 sk, wake) : true; 1290 #else 1291 return sk->sk_prot->stream_memory_free ? 1292 sk->sk_prot->stream_memory_free(sk, wake) : true; 1293 #endif 1294 } 1295 1296 static inline bool sk_stream_memory_free(const struct sock *sk) 1297 { 1298 return __sk_stream_memory_free(sk, 0); 1299 } 1300 1301 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1302 { 1303 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1304 __sk_stream_memory_free(sk, wake); 1305 } 1306 1307 static inline bool sk_stream_is_writeable(const struct sock *sk) 1308 { 1309 return __sk_stream_is_writeable(sk, 0); 1310 } 1311 1312 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1313 struct cgroup *ancestor) 1314 { 1315 #ifdef CONFIG_SOCK_CGROUP_DATA 1316 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1317 ancestor); 1318 #else 1319 return -ENOTSUPP; 1320 #endif 1321 } 1322 1323 static inline bool sk_has_memory_pressure(const struct sock *sk) 1324 { 1325 return sk->sk_prot->memory_pressure != NULL; 1326 } 1327 1328 static inline bool sk_under_memory_pressure(const struct sock *sk) 1329 { 1330 if (!sk->sk_prot->memory_pressure) 1331 return false; 1332 1333 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1334 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1335 return true; 1336 1337 return !!*sk->sk_prot->memory_pressure; 1338 } 1339 1340 static inline long 1341 sk_memory_allocated(const struct sock *sk) 1342 { 1343 return atomic_long_read(sk->sk_prot->memory_allocated); 1344 } 1345 1346 static inline long 1347 sk_memory_allocated_add(struct sock *sk, int amt) 1348 { 1349 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1350 } 1351 1352 static inline void 1353 sk_memory_allocated_sub(struct sock *sk, int amt) 1354 { 1355 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1356 } 1357 1358 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1359 1360 static inline void sk_sockets_allocated_dec(struct sock *sk) 1361 { 1362 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1363 SK_ALLOC_PERCPU_COUNTER_BATCH); 1364 } 1365 1366 static inline void sk_sockets_allocated_inc(struct sock *sk) 1367 { 1368 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1369 SK_ALLOC_PERCPU_COUNTER_BATCH); 1370 } 1371 1372 static inline u64 1373 sk_sockets_allocated_read_positive(struct sock *sk) 1374 { 1375 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1376 } 1377 1378 static inline int 1379 proto_sockets_allocated_sum_positive(struct proto *prot) 1380 { 1381 return percpu_counter_sum_positive(prot->sockets_allocated); 1382 } 1383 1384 static inline long 1385 proto_memory_allocated(struct proto *prot) 1386 { 1387 return atomic_long_read(prot->memory_allocated); 1388 } 1389 1390 static inline bool 1391 proto_memory_pressure(struct proto *prot) 1392 { 1393 if (!prot->memory_pressure) 1394 return false; 1395 return !!*prot->memory_pressure; 1396 } 1397 1398 1399 #ifdef CONFIG_PROC_FS 1400 /* Called with local bh disabled */ 1401 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1402 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1403 int sock_inuse_get(struct net *net); 1404 #else 1405 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1406 int inc) 1407 { 1408 } 1409 #endif 1410 1411 1412 /* With per-bucket locks this operation is not-atomic, so that 1413 * this version is not worse. 1414 */ 1415 static inline int __sk_prot_rehash(struct sock *sk) 1416 { 1417 sk->sk_prot->unhash(sk); 1418 return sk->sk_prot->hash(sk); 1419 } 1420 1421 /* About 10 seconds */ 1422 #define SOCK_DESTROY_TIME (10*HZ) 1423 1424 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1425 #define PROT_SOCK 1024 1426 1427 #define SHUTDOWN_MASK 3 1428 #define RCV_SHUTDOWN 1 1429 #define SEND_SHUTDOWN 2 1430 1431 #define SOCK_SNDBUF_LOCK 1 1432 #define SOCK_RCVBUF_LOCK 2 1433 #define SOCK_BINDADDR_LOCK 4 1434 #define SOCK_BINDPORT_LOCK 8 1435 1436 struct socket_alloc { 1437 struct socket socket; 1438 struct inode vfs_inode; 1439 }; 1440 1441 static inline struct socket *SOCKET_I(struct inode *inode) 1442 { 1443 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1444 } 1445 1446 static inline struct inode *SOCK_INODE(struct socket *socket) 1447 { 1448 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1449 } 1450 1451 /* 1452 * Functions for memory accounting 1453 */ 1454 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1455 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1456 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1457 void __sk_mem_reclaim(struct sock *sk, int amount); 1458 1459 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1460 * do not necessarily have 16x time more memory than 4KB ones. 1461 */ 1462 #define SK_MEM_QUANTUM 4096 1463 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1464 #define SK_MEM_SEND 0 1465 #define SK_MEM_RECV 1 1466 1467 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1468 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1469 { 1470 long val = sk->sk_prot->sysctl_mem[index]; 1471 1472 #if PAGE_SIZE > SK_MEM_QUANTUM 1473 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1474 #elif PAGE_SIZE < SK_MEM_QUANTUM 1475 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1476 #endif 1477 return val; 1478 } 1479 1480 static inline int sk_mem_pages(int amt) 1481 { 1482 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1483 } 1484 1485 static inline bool sk_has_account(struct sock *sk) 1486 { 1487 /* return true if protocol supports memory accounting */ 1488 return !!sk->sk_prot->memory_allocated; 1489 } 1490 1491 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1492 { 1493 if (!sk_has_account(sk)) 1494 return true; 1495 return size <= sk->sk_forward_alloc || 1496 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1497 } 1498 1499 static inline bool 1500 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1501 { 1502 if (!sk_has_account(sk)) 1503 return true; 1504 return size <= sk->sk_forward_alloc || 1505 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1506 skb_pfmemalloc(skb); 1507 } 1508 1509 static inline void sk_mem_reclaim(struct sock *sk) 1510 { 1511 if (!sk_has_account(sk)) 1512 return; 1513 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1514 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1515 } 1516 1517 static inline void sk_mem_reclaim_partial(struct sock *sk) 1518 { 1519 if (!sk_has_account(sk)) 1520 return; 1521 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1522 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1523 } 1524 1525 static inline void sk_mem_charge(struct sock *sk, int size) 1526 { 1527 if (!sk_has_account(sk)) 1528 return; 1529 sk->sk_forward_alloc -= size; 1530 } 1531 1532 static inline void sk_mem_uncharge(struct sock *sk, int size) 1533 { 1534 if (!sk_has_account(sk)) 1535 return; 1536 sk->sk_forward_alloc += size; 1537 1538 /* Avoid a possible overflow. 1539 * TCP send queues can make this happen, if sk_mem_reclaim() 1540 * is not called and more than 2 GBytes are released at once. 1541 * 1542 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1543 * no need to hold that much forward allocation anyway. 1544 */ 1545 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1546 __sk_mem_reclaim(sk, 1 << 20); 1547 } 1548 1549 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key); 1550 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1551 { 1552 sk_wmem_queued_add(sk, -skb->truesize); 1553 sk_mem_uncharge(sk, skb->truesize); 1554 if (static_branch_unlikely(&tcp_tx_skb_cache_key) && 1555 !sk->sk_tx_skb_cache && !skb_cloned(skb)) { 1556 skb_ext_reset(skb); 1557 skb_zcopy_clear(skb, true); 1558 sk->sk_tx_skb_cache = skb; 1559 return; 1560 } 1561 __kfree_skb(skb); 1562 } 1563 1564 static inline void sock_release_ownership(struct sock *sk) 1565 { 1566 if (sk->sk_lock.owned) { 1567 sk->sk_lock.owned = 0; 1568 1569 /* The sk_lock has mutex_unlock() semantics: */ 1570 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1571 } 1572 } 1573 1574 /* 1575 * Macro so as to not evaluate some arguments when 1576 * lockdep is not enabled. 1577 * 1578 * Mark both the sk_lock and the sk_lock.slock as a 1579 * per-address-family lock class. 1580 */ 1581 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1582 do { \ 1583 sk->sk_lock.owned = 0; \ 1584 init_waitqueue_head(&sk->sk_lock.wq); \ 1585 spin_lock_init(&(sk)->sk_lock.slock); \ 1586 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1587 sizeof((sk)->sk_lock)); \ 1588 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1589 (skey), (sname)); \ 1590 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1591 } while (0) 1592 1593 static inline bool lockdep_sock_is_held(const struct sock *sk) 1594 { 1595 return lockdep_is_held(&sk->sk_lock) || 1596 lockdep_is_held(&sk->sk_lock.slock); 1597 } 1598 1599 void lock_sock_nested(struct sock *sk, int subclass); 1600 1601 static inline void lock_sock(struct sock *sk) 1602 { 1603 lock_sock_nested(sk, 0); 1604 } 1605 1606 void __lock_sock(struct sock *sk); 1607 void __release_sock(struct sock *sk); 1608 void release_sock(struct sock *sk); 1609 1610 /* BH context may only use the following locking interface. */ 1611 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1612 #define bh_lock_sock_nested(__sk) \ 1613 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1614 SINGLE_DEPTH_NESTING) 1615 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1616 1617 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1618 1619 /** 1620 * unlock_sock_fast - complement of lock_sock_fast 1621 * @sk: socket 1622 * @slow: slow mode 1623 * 1624 * fast unlock socket for user context. 1625 * If slow mode is on, we call regular release_sock() 1626 */ 1627 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1628 __releases(&sk->sk_lock.slock) 1629 { 1630 if (slow) { 1631 release_sock(sk); 1632 __release(&sk->sk_lock.slock); 1633 } else { 1634 spin_unlock_bh(&sk->sk_lock.slock); 1635 } 1636 } 1637 1638 /* Used by processes to "lock" a socket state, so that 1639 * interrupts and bottom half handlers won't change it 1640 * from under us. It essentially blocks any incoming 1641 * packets, so that we won't get any new data or any 1642 * packets that change the state of the socket. 1643 * 1644 * While locked, BH processing will add new packets to 1645 * the backlog queue. This queue is processed by the 1646 * owner of the socket lock right before it is released. 1647 * 1648 * Since ~2.3.5 it is also exclusive sleep lock serializing 1649 * accesses from user process context. 1650 */ 1651 1652 static inline void sock_owned_by_me(const struct sock *sk) 1653 { 1654 #ifdef CONFIG_LOCKDEP 1655 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1656 #endif 1657 } 1658 1659 static inline bool sock_owned_by_user(const struct sock *sk) 1660 { 1661 sock_owned_by_me(sk); 1662 return sk->sk_lock.owned; 1663 } 1664 1665 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1666 { 1667 return sk->sk_lock.owned; 1668 } 1669 1670 /* no reclassification while locks are held */ 1671 static inline bool sock_allow_reclassification(const struct sock *csk) 1672 { 1673 struct sock *sk = (struct sock *)csk; 1674 1675 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1676 } 1677 1678 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1679 struct proto *prot, int kern); 1680 void sk_free(struct sock *sk); 1681 void sk_destruct(struct sock *sk); 1682 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1683 void sk_free_unlock_clone(struct sock *sk); 1684 1685 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1686 gfp_t priority); 1687 void __sock_wfree(struct sk_buff *skb); 1688 void sock_wfree(struct sk_buff *skb); 1689 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1690 gfp_t priority); 1691 void skb_orphan_partial(struct sk_buff *skb); 1692 void sock_rfree(struct sk_buff *skb); 1693 void sock_efree(struct sk_buff *skb); 1694 #ifdef CONFIG_INET 1695 void sock_edemux(struct sk_buff *skb); 1696 void sock_pfree(struct sk_buff *skb); 1697 #else 1698 #define sock_edemux sock_efree 1699 #endif 1700 1701 int sock_setsockopt(struct socket *sock, int level, int op, 1702 sockptr_t optval, unsigned int optlen); 1703 1704 int sock_getsockopt(struct socket *sock, int level, int op, 1705 char __user *optval, int __user *optlen); 1706 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1707 bool timeval, bool time32); 1708 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1709 int noblock, int *errcode); 1710 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1711 unsigned long data_len, int noblock, 1712 int *errcode, int max_page_order); 1713 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1714 void sock_kfree_s(struct sock *sk, void *mem, int size); 1715 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1716 void sk_send_sigurg(struct sock *sk); 1717 1718 struct sockcm_cookie { 1719 u64 transmit_time; 1720 u32 mark; 1721 u16 tsflags; 1722 }; 1723 1724 static inline void sockcm_init(struct sockcm_cookie *sockc, 1725 const struct sock *sk) 1726 { 1727 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1728 } 1729 1730 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1731 struct sockcm_cookie *sockc); 1732 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1733 struct sockcm_cookie *sockc); 1734 1735 /* 1736 * Functions to fill in entries in struct proto_ops when a protocol 1737 * does not implement a particular function. 1738 */ 1739 int sock_no_bind(struct socket *, struct sockaddr *, int); 1740 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1741 int sock_no_socketpair(struct socket *, struct socket *); 1742 int sock_no_accept(struct socket *, struct socket *, int, bool); 1743 int sock_no_getname(struct socket *, struct sockaddr *, int); 1744 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1745 int sock_no_listen(struct socket *, int); 1746 int sock_no_shutdown(struct socket *, int); 1747 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1748 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1749 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1750 int sock_no_mmap(struct file *file, struct socket *sock, 1751 struct vm_area_struct *vma); 1752 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1753 size_t size, int flags); 1754 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1755 int offset, size_t size, int flags); 1756 1757 /* 1758 * Functions to fill in entries in struct proto_ops when a protocol 1759 * uses the inet style. 1760 */ 1761 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1762 char __user *optval, int __user *optlen); 1763 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1764 int flags); 1765 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1766 sockptr_t optval, unsigned int optlen); 1767 1768 void sk_common_release(struct sock *sk); 1769 1770 /* 1771 * Default socket callbacks and setup code 1772 */ 1773 1774 /* Initialise core socket variables */ 1775 void sock_init_data(struct socket *sock, struct sock *sk); 1776 1777 /* 1778 * Socket reference counting postulates. 1779 * 1780 * * Each user of socket SHOULD hold a reference count. 1781 * * Each access point to socket (an hash table bucket, reference from a list, 1782 * running timer, skb in flight MUST hold a reference count. 1783 * * When reference count hits 0, it means it will never increase back. 1784 * * When reference count hits 0, it means that no references from 1785 * outside exist to this socket and current process on current CPU 1786 * is last user and may/should destroy this socket. 1787 * * sk_free is called from any context: process, BH, IRQ. When 1788 * it is called, socket has no references from outside -> sk_free 1789 * may release descendant resources allocated by the socket, but 1790 * to the time when it is called, socket is NOT referenced by any 1791 * hash tables, lists etc. 1792 * * Packets, delivered from outside (from network or from another process) 1793 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1794 * when they sit in queue. Otherwise, packets will leak to hole, when 1795 * socket is looked up by one cpu and unhasing is made by another CPU. 1796 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1797 * (leak to backlog). Packet socket does all the processing inside 1798 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1799 * use separate SMP lock, so that they are prone too. 1800 */ 1801 1802 /* Ungrab socket and destroy it, if it was the last reference. */ 1803 static inline void sock_put(struct sock *sk) 1804 { 1805 if (refcount_dec_and_test(&sk->sk_refcnt)) 1806 sk_free(sk); 1807 } 1808 /* Generic version of sock_put(), dealing with all sockets 1809 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1810 */ 1811 void sock_gen_put(struct sock *sk); 1812 1813 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1814 unsigned int trim_cap, bool refcounted); 1815 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1816 const int nested) 1817 { 1818 return __sk_receive_skb(sk, skb, nested, 1, true); 1819 } 1820 1821 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1822 { 1823 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1824 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1825 return; 1826 sk->sk_tx_queue_mapping = tx_queue; 1827 } 1828 1829 #define NO_QUEUE_MAPPING USHRT_MAX 1830 1831 static inline void sk_tx_queue_clear(struct sock *sk) 1832 { 1833 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1834 } 1835 1836 static inline int sk_tx_queue_get(const struct sock *sk) 1837 { 1838 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1839 return sk->sk_tx_queue_mapping; 1840 1841 return -1; 1842 } 1843 1844 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1845 { 1846 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1847 if (skb_rx_queue_recorded(skb)) { 1848 u16 rx_queue = skb_get_rx_queue(skb); 1849 1850 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING)) 1851 return; 1852 1853 sk->sk_rx_queue_mapping = rx_queue; 1854 } 1855 #endif 1856 } 1857 1858 static inline void sk_rx_queue_clear(struct sock *sk) 1859 { 1860 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1861 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING; 1862 #endif 1863 } 1864 1865 static inline int sk_rx_queue_get(const struct sock *sk) 1866 { 1867 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1868 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING) 1869 return sk->sk_rx_queue_mapping; 1870 #endif 1871 1872 return -1; 1873 } 1874 1875 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1876 { 1877 sk->sk_socket = sock; 1878 } 1879 1880 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1881 { 1882 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1883 return &rcu_dereference_raw(sk->sk_wq)->wait; 1884 } 1885 /* Detach socket from process context. 1886 * Announce socket dead, detach it from wait queue and inode. 1887 * Note that parent inode held reference count on this struct sock, 1888 * we do not release it in this function, because protocol 1889 * probably wants some additional cleanups or even continuing 1890 * to work with this socket (TCP). 1891 */ 1892 static inline void sock_orphan(struct sock *sk) 1893 { 1894 write_lock_bh(&sk->sk_callback_lock); 1895 sock_set_flag(sk, SOCK_DEAD); 1896 sk_set_socket(sk, NULL); 1897 sk->sk_wq = NULL; 1898 write_unlock_bh(&sk->sk_callback_lock); 1899 } 1900 1901 static inline void sock_graft(struct sock *sk, struct socket *parent) 1902 { 1903 WARN_ON(parent->sk); 1904 write_lock_bh(&sk->sk_callback_lock); 1905 rcu_assign_pointer(sk->sk_wq, &parent->wq); 1906 parent->sk = sk; 1907 sk_set_socket(sk, parent); 1908 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1909 security_sock_graft(sk, parent); 1910 write_unlock_bh(&sk->sk_callback_lock); 1911 } 1912 1913 kuid_t sock_i_uid(struct sock *sk); 1914 unsigned long sock_i_ino(struct sock *sk); 1915 1916 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1917 { 1918 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1919 } 1920 1921 static inline u32 net_tx_rndhash(void) 1922 { 1923 u32 v = prandom_u32(); 1924 1925 return v ?: 1; 1926 } 1927 1928 static inline void sk_set_txhash(struct sock *sk) 1929 { 1930 sk->sk_txhash = net_tx_rndhash(); 1931 } 1932 1933 static inline bool sk_rethink_txhash(struct sock *sk) 1934 { 1935 if (sk->sk_txhash) { 1936 sk_set_txhash(sk); 1937 return true; 1938 } 1939 return false; 1940 } 1941 1942 static inline struct dst_entry * 1943 __sk_dst_get(struct sock *sk) 1944 { 1945 return rcu_dereference_check(sk->sk_dst_cache, 1946 lockdep_sock_is_held(sk)); 1947 } 1948 1949 static inline struct dst_entry * 1950 sk_dst_get(struct sock *sk) 1951 { 1952 struct dst_entry *dst; 1953 1954 rcu_read_lock(); 1955 dst = rcu_dereference(sk->sk_dst_cache); 1956 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1957 dst = NULL; 1958 rcu_read_unlock(); 1959 return dst; 1960 } 1961 1962 static inline void __dst_negative_advice(struct sock *sk) 1963 { 1964 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1965 1966 if (dst && dst->ops->negative_advice) { 1967 ndst = dst->ops->negative_advice(dst); 1968 1969 if (ndst != dst) { 1970 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1971 sk_tx_queue_clear(sk); 1972 sk->sk_dst_pending_confirm = 0; 1973 } 1974 } 1975 } 1976 1977 static inline void dst_negative_advice(struct sock *sk) 1978 { 1979 sk_rethink_txhash(sk); 1980 __dst_negative_advice(sk); 1981 } 1982 1983 static inline void 1984 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1985 { 1986 struct dst_entry *old_dst; 1987 1988 sk_tx_queue_clear(sk); 1989 sk->sk_dst_pending_confirm = 0; 1990 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1991 lockdep_sock_is_held(sk)); 1992 rcu_assign_pointer(sk->sk_dst_cache, dst); 1993 dst_release(old_dst); 1994 } 1995 1996 static inline void 1997 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1998 { 1999 struct dst_entry *old_dst; 2000 2001 sk_tx_queue_clear(sk); 2002 sk->sk_dst_pending_confirm = 0; 2003 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2004 dst_release(old_dst); 2005 } 2006 2007 static inline void 2008 __sk_dst_reset(struct sock *sk) 2009 { 2010 __sk_dst_set(sk, NULL); 2011 } 2012 2013 static inline void 2014 sk_dst_reset(struct sock *sk) 2015 { 2016 sk_dst_set(sk, NULL); 2017 } 2018 2019 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2020 2021 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2022 2023 static inline void sk_dst_confirm(struct sock *sk) 2024 { 2025 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2026 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2027 } 2028 2029 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2030 { 2031 if (skb_get_dst_pending_confirm(skb)) { 2032 struct sock *sk = skb->sk; 2033 unsigned long now = jiffies; 2034 2035 /* avoid dirtying neighbour */ 2036 if (READ_ONCE(n->confirmed) != now) 2037 WRITE_ONCE(n->confirmed, now); 2038 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2039 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2040 } 2041 } 2042 2043 bool sk_mc_loop(struct sock *sk); 2044 2045 static inline bool sk_can_gso(const struct sock *sk) 2046 { 2047 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2048 } 2049 2050 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2051 2052 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 2053 { 2054 sk->sk_route_nocaps |= flags; 2055 sk->sk_route_caps &= ~flags; 2056 } 2057 2058 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2059 struct iov_iter *from, char *to, 2060 int copy, int offset) 2061 { 2062 if (skb->ip_summed == CHECKSUM_NONE) { 2063 __wsum csum = 0; 2064 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2065 return -EFAULT; 2066 skb->csum = csum_block_add(skb->csum, csum, offset); 2067 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2068 if (!copy_from_iter_full_nocache(to, copy, from)) 2069 return -EFAULT; 2070 } else if (!copy_from_iter_full(to, copy, from)) 2071 return -EFAULT; 2072 2073 return 0; 2074 } 2075 2076 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2077 struct iov_iter *from, int copy) 2078 { 2079 int err, offset = skb->len; 2080 2081 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2082 copy, offset); 2083 if (err) 2084 __skb_trim(skb, offset); 2085 2086 return err; 2087 } 2088 2089 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2090 struct sk_buff *skb, 2091 struct page *page, 2092 int off, int copy) 2093 { 2094 int err; 2095 2096 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2097 copy, skb->len); 2098 if (err) 2099 return err; 2100 2101 skb->len += copy; 2102 skb->data_len += copy; 2103 skb->truesize += copy; 2104 sk_wmem_queued_add(sk, copy); 2105 sk_mem_charge(sk, copy); 2106 return 0; 2107 } 2108 2109 /** 2110 * sk_wmem_alloc_get - returns write allocations 2111 * @sk: socket 2112 * 2113 * Return: sk_wmem_alloc minus initial offset of one 2114 */ 2115 static inline int sk_wmem_alloc_get(const struct sock *sk) 2116 { 2117 return refcount_read(&sk->sk_wmem_alloc) - 1; 2118 } 2119 2120 /** 2121 * sk_rmem_alloc_get - returns read allocations 2122 * @sk: socket 2123 * 2124 * Return: sk_rmem_alloc 2125 */ 2126 static inline int sk_rmem_alloc_get(const struct sock *sk) 2127 { 2128 return atomic_read(&sk->sk_rmem_alloc); 2129 } 2130 2131 /** 2132 * sk_has_allocations - check if allocations are outstanding 2133 * @sk: socket 2134 * 2135 * Return: true if socket has write or read allocations 2136 */ 2137 static inline bool sk_has_allocations(const struct sock *sk) 2138 { 2139 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2140 } 2141 2142 /** 2143 * skwq_has_sleeper - check if there are any waiting processes 2144 * @wq: struct socket_wq 2145 * 2146 * Return: true if socket_wq has waiting processes 2147 * 2148 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2149 * barrier call. They were added due to the race found within the tcp code. 2150 * 2151 * Consider following tcp code paths:: 2152 * 2153 * CPU1 CPU2 2154 * sys_select receive packet 2155 * ... ... 2156 * __add_wait_queue update tp->rcv_nxt 2157 * ... ... 2158 * tp->rcv_nxt check sock_def_readable 2159 * ... { 2160 * schedule rcu_read_lock(); 2161 * wq = rcu_dereference(sk->sk_wq); 2162 * if (wq && waitqueue_active(&wq->wait)) 2163 * wake_up_interruptible(&wq->wait) 2164 * ... 2165 * } 2166 * 2167 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2168 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2169 * could then endup calling schedule and sleep forever if there are no more 2170 * data on the socket. 2171 * 2172 */ 2173 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2174 { 2175 return wq && wq_has_sleeper(&wq->wait); 2176 } 2177 2178 /** 2179 * sock_poll_wait - place memory barrier behind the poll_wait call. 2180 * @filp: file 2181 * @sock: socket to wait on 2182 * @p: poll_table 2183 * 2184 * See the comments in the wq_has_sleeper function. 2185 */ 2186 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2187 poll_table *p) 2188 { 2189 if (!poll_does_not_wait(p)) { 2190 poll_wait(filp, &sock->wq.wait, p); 2191 /* We need to be sure we are in sync with the 2192 * socket flags modification. 2193 * 2194 * This memory barrier is paired in the wq_has_sleeper. 2195 */ 2196 smp_mb(); 2197 } 2198 } 2199 2200 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2201 { 2202 if (sk->sk_txhash) { 2203 skb->l4_hash = 1; 2204 skb->hash = sk->sk_txhash; 2205 } 2206 } 2207 2208 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2209 2210 /* 2211 * Queue a received datagram if it will fit. Stream and sequenced 2212 * protocols can't normally use this as they need to fit buffers in 2213 * and play with them. 2214 * 2215 * Inlined as it's very short and called for pretty much every 2216 * packet ever received. 2217 */ 2218 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2219 { 2220 skb_orphan(skb); 2221 skb->sk = sk; 2222 skb->destructor = sock_rfree; 2223 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2224 sk_mem_charge(sk, skb->truesize); 2225 } 2226 2227 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2228 unsigned long expires); 2229 2230 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2231 2232 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2233 2234 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2235 struct sk_buff *skb, unsigned int flags, 2236 void (*destructor)(struct sock *sk, 2237 struct sk_buff *skb)); 2238 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2239 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2240 2241 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2242 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2243 2244 /* 2245 * Recover an error report and clear atomically 2246 */ 2247 2248 static inline int sock_error(struct sock *sk) 2249 { 2250 int err; 2251 if (likely(!sk->sk_err)) 2252 return 0; 2253 err = xchg(&sk->sk_err, 0); 2254 return -err; 2255 } 2256 2257 static inline unsigned long sock_wspace(struct sock *sk) 2258 { 2259 int amt = 0; 2260 2261 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2262 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2263 if (amt < 0) 2264 amt = 0; 2265 } 2266 return amt; 2267 } 2268 2269 /* Note: 2270 * We use sk->sk_wq_raw, from contexts knowing this 2271 * pointer is not NULL and cannot disappear/change. 2272 */ 2273 static inline void sk_set_bit(int nr, struct sock *sk) 2274 { 2275 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2276 !sock_flag(sk, SOCK_FASYNC)) 2277 return; 2278 2279 set_bit(nr, &sk->sk_wq_raw->flags); 2280 } 2281 2282 static inline void sk_clear_bit(int nr, struct sock *sk) 2283 { 2284 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2285 !sock_flag(sk, SOCK_FASYNC)) 2286 return; 2287 2288 clear_bit(nr, &sk->sk_wq_raw->flags); 2289 } 2290 2291 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2292 { 2293 if (sock_flag(sk, SOCK_FASYNC)) { 2294 rcu_read_lock(); 2295 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2296 rcu_read_unlock(); 2297 } 2298 } 2299 2300 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2301 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2302 * Note: for send buffers, TCP works better if we can build two skbs at 2303 * minimum. 2304 */ 2305 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2306 2307 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2308 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2309 2310 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2311 { 2312 u32 val; 2313 2314 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2315 return; 2316 2317 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2318 2319 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2320 } 2321 2322 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2323 bool force_schedule); 2324 2325 /** 2326 * sk_page_frag - return an appropriate page_frag 2327 * @sk: socket 2328 * 2329 * Use the per task page_frag instead of the per socket one for 2330 * optimization when we know that we're in the normal context and owns 2331 * everything that's associated with %current. 2332 * 2333 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest 2334 * inside other socket operations and end up recursing into sk_page_frag() 2335 * while it's already in use. 2336 * 2337 * Return: a per task page_frag if context allows that, 2338 * otherwise a per socket one. 2339 */ 2340 static inline struct page_frag *sk_page_frag(struct sock *sk) 2341 { 2342 if (gfpflags_normal_context(sk->sk_allocation)) 2343 return ¤t->task_frag; 2344 2345 return &sk->sk_frag; 2346 } 2347 2348 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2349 2350 /* 2351 * Default write policy as shown to user space via poll/select/SIGIO 2352 */ 2353 static inline bool sock_writeable(const struct sock *sk) 2354 { 2355 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2356 } 2357 2358 static inline gfp_t gfp_any(void) 2359 { 2360 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2361 } 2362 2363 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2364 { 2365 return noblock ? 0 : sk->sk_rcvtimeo; 2366 } 2367 2368 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2369 { 2370 return noblock ? 0 : sk->sk_sndtimeo; 2371 } 2372 2373 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2374 { 2375 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2376 2377 return v ?: 1; 2378 } 2379 2380 /* Alas, with timeout socket operations are not restartable. 2381 * Compare this to poll(). 2382 */ 2383 static inline int sock_intr_errno(long timeo) 2384 { 2385 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2386 } 2387 2388 struct sock_skb_cb { 2389 u32 dropcount; 2390 }; 2391 2392 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2393 * using skb->cb[] would keep using it directly and utilize its 2394 * alignement guarantee. 2395 */ 2396 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2397 sizeof(struct sock_skb_cb))) 2398 2399 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2400 SOCK_SKB_CB_OFFSET)) 2401 2402 #define sock_skb_cb_check_size(size) \ 2403 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2404 2405 static inline void 2406 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2407 { 2408 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2409 atomic_read(&sk->sk_drops) : 0; 2410 } 2411 2412 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2413 { 2414 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2415 2416 atomic_add(segs, &sk->sk_drops); 2417 } 2418 2419 static inline ktime_t sock_read_timestamp(struct sock *sk) 2420 { 2421 #if BITS_PER_LONG==32 2422 unsigned int seq; 2423 ktime_t kt; 2424 2425 do { 2426 seq = read_seqbegin(&sk->sk_stamp_seq); 2427 kt = sk->sk_stamp; 2428 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2429 2430 return kt; 2431 #else 2432 return READ_ONCE(sk->sk_stamp); 2433 #endif 2434 } 2435 2436 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2437 { 2438 #if BITS_PER_LONG==32 2439 write_seqlock(&sk->sk_stamp_seq); 2440 sk->sk_stamp = kt; 2441 write_sequnlock(&sk->sk_stamp_seq); 2442 #else 2443 WRITE_ONCE(sk->sk_stamp, kt); 2444 #endif 2445 } 2446 2447 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2448 struct sk_buff *skb); 2449 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2450 struct sk_buff *skb); 2451 2452 static inline void 2453 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2454 { 2455 ktime_t kt = skb->tstamp; 2456 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2457 2458 /* 2459 * generate control messages if 2460 * - receive time stamping in software requested 2461 * - software time stamp available and wanted 2462 * - hardware time stamps available and wanted 2463 */ 2464 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2465 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2466 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2467 (hwtstamps->hwtstamp && 2468 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2469 __sock_recv_timestamp(msg, sk, skb); 2470 else 2471 sock_write_timestamp(sk, kt); 2472 2473 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2474 __sock_recv_wifi_status(msg, sk, skb); 2475 } 2476 2477 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2478 struct sk_buff *skb); 2479 2480 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2481 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2482 struct sk_buff *skb) 2483 { 2484 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2485 (1UL << SOCK_RCVTSTAMP)) 2486 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2487 SOF_TIMESTAMPING_RAW_HARDWARE) 2488 2489 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2490 __sock_recv_ts_and_drops(msg, sk, skb); 2491 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2492 sock_write_timestamp(sk, skb->tstamp); 2493 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2494 sock_write_timestamp(sk, 0); 2495 } 2496 2497 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2498 2499 /** 2500 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2501 * @sk: socket sending this packet 2502 * @tsflags: timestamping flags to use 2503 * @tx_flags: completed with instructions for time stamping 2504 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2505 * 2506 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2507 */ 2508 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2509 __u8 *tx_flags, __u32 *tskey) 2510 { 2511 if (unlikely(tsflags)) { 2512 __sock_tx_timestamp(tsflags, tx_flags); 2513 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2514 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2515 *tskey = sk->sk_tskey++; 2516 } 2517 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2518 *tx_flags |= SKBTX_WIFI_STATUS; 2519 } 2520 2521 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2522 __u8 *tx_flags) 2523 { 2524 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2525 } 2526 2527 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2528 { 2529 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2530 &skb_shinfo(skb)->tskey); 2531 } 2532 2533 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key); 2534 /** 2535 * sk_eat_skb - Release a skb if it is no longer needed 2536 * @sk: socket to eat this skb from 2537 * @skb: socket buffer to eat 2538 * 2539 * This routine must be called with interrupts disabled or with the socket 2540 * locked so that the sk_buff queue operation is ok. 2541 */ 2542 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2543 { 2544 __skb_unlink(skb, &sk->sk_receive_queue); 2545 if (static_branch_unlikely(&tcp_rx_skb_cache_key) && 2546 !sk->sk_rx_skb_cache) { 2547 sk->sk_rx_skb_cache = skb; 2548 skb_orphan(skb); 2549 return; 2550 } 2551 __kfree_skb(skb); 2552 } 2553 2554 static inline 2555 struct net *sock_net(const struct sock *sk) 2556 { 2557 return read_pnet(&sk->sk_net); 2558 } 2559 2560 static inline 2561 void sock_net_set(struct sock *sk, struct net *net) 2562 { 2563 write_pnet(&sk->sk_net, net); 2564 } 2565 2566 static inline bool 2567 skb_sk_is_prefetched(struct sk_buff *skb) 2568 { 2569 #ifdef CONFIG_INET 2570 return skb->destructor == sock_pfree; 2571 #else 2572 return false; 2573 #endif /* CONFIG_INET */ 2574 } 2575 2576 /* This helper checks if a socket is a full socket, 2577 * ie _not_ a timewait or request socket. 2578 */ 2579 static inline bool sk_fullsock(const struct sock *sk) 2580 { 2581 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2582 } 2583 2584 static inline bool 2585 sk_is_refcounted(struct sock *sk) 2586 { 2587 /* Only full sockets have sk->sk_flags. */ 2588 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2589 } 2590 2591 /** 2592 * skb_steal_sock - steal a socket from an sk_buff 2593 * @skb: sk_buff to steal the socket from 2594 * @refcounted: is set to true if the socket is reference-counted 2595 */ 2596 static inline struct sock * 2597 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2598 { 2599 if (skb->sk) { 2600 struct sock *sk = skb->sk; 2601 2602 *refcounted = true; 2603 if (skb_sk_is_prefetched(skb)) 2604 *refcounted = sk_is_refcounted(sk); 2605 skb->destructor = NULL; 2606 skb->sk = NULL; 2607 return sk; 2608 } 2609 *refcounted = false; 2610 return NULL; 2611 } 2612 2613 /* Checks if this SKB belongs to an HW offloaded socket 2614 * and whether any SW fallbacks are required based on dev. 2615 * Check decrypted mark in case skb_orphan() cleared socket. 2616 */ 2617 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2618 struct net_device *dev) 2619 { 2620 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2621 struct sock *sk = skb->sk; 2622 2623 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2624 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2625 #ifdef CONFIG_TLS_DEVICE 2626 } else if (unlikely(skb->decrypted)) { 2627 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2628 kfree_skb(skb); 2629 skb = NULL; 2630 #endif 2631 } 2632 #endif 2633 2634 return skb; 2635 } 2636 2637 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2638 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2639 */ 2640 static inline bool sk_listener(const struct sock *sk) 2641 { 2642 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2643 } 2644 2645 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2646 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2647 int type); 2648 2649 bool sk_ns_capable(const struct sock *sk, 2650 struct user_namespace *user_ns, int cap); 2651 bool sk_capable(const struct sock *sk, int cap); 2652 bool sk_net_capable(const struct sock *sk, int cap); 2653 2654 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2655 2656 /* Take into consideration the size of the struct sk_buff overhead in the 2657 * determination of these values, since that is non-constant across 2658 * platforms. This makes socket queueing behavior and performance 2659 * not depend upon such differences. 2660 */ 2661 #define _SK_MEM_PACKETS 256 2662 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2663 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2664 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2665 2666 extern __u32 sysctl_wmem_max; 2667 extern __u32 sysctl_rmem_max; 2668 2669 extern int sysctl_tstamp_allow_data; 2670 extern int sysctl_optmem_max; 2671 2672 extern __u32 sysctl_wmem_default; 2673 extern __u32 sysctl_rmem_default; 2674 2675 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2676 2677 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2678 { 2679 /* Does this proto have per netns sysctl_wmem ? */ 2680 if (proto->sysctl_wmem_offset) 2681 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2682 2683 return *proto->sysctl_wmem; 2684 } 2685 2686 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2687 { 2688 /* Does this proto have per netns sysctl_rmem ? */ 2689 if (proto->sysctl_rmem_offset) 2690 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2691 2692 return *proto->sysctl_rmem; 2693 } 2694 2695 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2696 * Some wifi drivers need to tweak it to get more chunks. 2697 * They can use this helper from their ndo_start_xmit() 2698 */ 2699 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2700 { 2701 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2702 return; 2703 WRITE_ONCE(sk->sk_pacing_shift, val); 2704 } 2705 2706 /* if a socket is bound to a device, check that the given device 2707 * index is either the same or that the socket is bound to an L3 2708 * master device and the given device index is also enslaved to 2709 * that L3 master 2710 */ 2711 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2712 { 2713 int mdif; 2714 2715 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2716 return true; 2717 2718 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2719 if (mdif && mdif == sk->sk_bound_dev_if) 2720 return true; 2721 2722 return false; 2723 } 2724 2725 void sock_def_readable(struct sock *sk); 2726 2727 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2728 void sock_enable_timestamps(struct sock *sk); 2729 void sock_no_linger(struct sock *sk); 2730 void sock_set_keepalive(struct sock *sk); 2731 void sock_set_priority(struct sock *sk, u32 priority); 2732 void sock_set_rcvbuf(struct sock *sk, int val); 2733 void sock_set_mark(struct sock *sk, u32 val); 2734 void sock_set_reuseaddr(struct sock *sk); 2735 void sock_set_reuseport(struct sock *sk); 2736 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2737 2738 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2739 2740 #endif /* _SOCK_H */ 2741