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 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <net/dst.h> 66 #include <net/checksum.h> 67 #include <net/tcp_states.h> 68 #include <linux/net_tstamp.h> 69 #include <net/l3mdev.h> 70 71 /* 72 * This structure really needs to be cleaned up. 73 * Most of it is for TCP, and not used by any of 74 * the other protocols. 75 */ 76 77 /* Define this to get the SOCK_DBG debugging facility. */ 78 #define SOCK_DEBUGGING 79 #ifdef SOCK_DEBUGGING 80 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 81 printk(KERN_DEBUG msg); } while (0) 82 #else 83 /* Validate arguments and do nothing */ 84 static inline __printf(2, 3) 85 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 86 { 87 } 88 #endif 89 90 /* This is the per-socket lock. The spinlock provides a synchronization 91 * between user contexts and software interrupt processing, whereas the 92 * mini-semaphore synchronizes multiple users amongst themselves. 93 */ 94 typedef struct { 95 spinlock_t slock; 96 int owned; 97 wait_queue_head_t wq; 98 /* 99 * We express the mutex-alike socket_lock semantics 100 * to the lock validator by explicitly managing 101 * the slock as a lock variant (in addition to 102 * the slock itself): 103 */ 104 #ifdef CONFIG_DEBUG_LOCK_ALLOC 105 struct lockdep_map dep_map; 106 #endif 107 } socket_lock_t; 108 109 struct sock; 110 struct proto; 111 struct net; 112 113 typedef __u32 __bitwise __portpair; 114 typedef __u64 __bitwise __addrpair; 115 116 /** 117 * struct sock_common - minimal network layer representation of sockets 118 * @skc_daddr: Foreign IPv4 addr 119 * @skc_rcv_saddr: Bound local IPv4 addr 120 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 121 * @skc_hash: hash value used with various protocol lookup tables 122 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 123 * @skc_dport: placeholder for inet_dport/tw_dport 124 * @skc_num: placeholder for inet_num/tw_num 125 * @skc_portpair: __u32 union of @skc_dport & @skc_num 126 * @skc_family: network address family 127 * @skc_state: Connection state 128 * @skc_reuse: %SO_REUSEADDR setting 129 * @skc_reuseport: %SO_REUSEPORT setting 130 * @skc_ipv6only: socket is IPV6 only 131 * @skc_net_refcnt: socket is using net ref counting 132 * @skc_bound_dev_if: bound device index if != 0 133 * @skc_bind_node: bind hash linkage for various protocol lookup tables 134 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 135 * @skc_prot: protocol handlers inside a network family 136 * @skc_net: reference to the network namespace of this socket 137 * @skc_v6_daddr: IPV6 destination address 138 * @skc_v6_rcv_saddr: IPV6 source address 139 * @skc_cookie: socket's cookie value 140 * @skc_node: main hash linkage for various protocol lookup tables 141 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 142 * @skc_tx_queue_mapping: tx queue number for this connection 143 * @skc_rx_queue_mapping: rx queue number for this connection 144 * @skc_flags: place holder for sk_flags 145 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 146 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 147 * @skc_listener: connection request listener socket (aka rsk_listener) 148 * [union with @skc_flags] 149 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 150 * [union with @skc_flags] 151 * @skc_incoming_cpu: record/match cpu processing incoming packets 152 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 153 * [union with @skc_incoming_cpu] 154 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 155 * [union with @skc_incoming_cpu] 156 * @skc_refcnt: reference count 157 * 158 * This is the minimal network layer representation of sockets, the header 159 * for struct sock and struct inet_timewait_sock. 160 */ 161 struct sock_common { 162 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 163 * address on 64bit arches : cf INET_MATCH() 164 */ 165 union { 166 __addrpair skc_addrpair; 167 struct { 168 __be32 skc_daddr; 169 __be32 skc_rcv_saddr; 170 }; 171 }; 172 union { 173 unsigned int skc_hash; 174 __u16 skc_u16hashes[2]; 175 }; 176 /* skc_dport && skc_num must be grouped as well */ 177 union { 178 __portpair skc_portpair; 179 struct { 180 __be16 skc_dport; 181 __u16 skc_num; 182 }; 183 }; 184 185 unsigned short skc_family; 186 volatile unsigned char skc_state; 187 unsigned char skc_reuse:4; 188 unsigned char skc_reuseport:1; 189 unsigned char skc_ipv6only:1; 190 unsigned char skc_net_refcnt:1; 191 int skc_bound_dev_if; 192 union { 193 struct hlist_node skc_bind_node; 194 struct hlist_node skc_portaddr_node; 195 }; 196 struct proto *skc_prot; 197 possible_net_t skc_net; 198 199 #if IS_ENABLED(CONFIG_IPV6) 200 struct in6_addr skc_v6_daddr; 201 struct in6_addr skc_v6_rcv_saddr; 202 #endif 203 204 atomic64_t skc_cookie; 205 206 /* following fields are padding to force 207 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 208 * assuming IPV6 is enabled. We use this padding differently 209 * for different kind of 'sockets' 210 */ 211 union { 212 unsigned long skc_flags; 213 struct sock *skc_listener; /* request_sock */ 214 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 215 }; 216 /* 217 * fields between dontcopy_begin/dontcopy_end 218 * are not copied in sock_copy() 219 */ 220 /* private: */ 221 int skc_dontcopy_begin[0]; 222 /* public: */ 223 union { 224 struct hlist_node skc_node; 225 struct hlist_nulls_node skc_nulls_node; 226 }; 227 unsigned short skc_tx_queue_mapping; 228 #ifdef CONFIG_XPS 229 unsigned short skc_rx_queue_mapping; 230 #endif 231 union { 232 int skc_incoming_cpu; 233 u32 skc_rcv_wnd; 234 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 235 }; 236 237 refcount_t skc_refcnt; 238 /* private: */ 239 int skc_dontcopy_end[0]; 240 union { 241 u32 skc_rxhash; 242 u32 skc_window_clamp; 243 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 244 }; 245 /* public: */ 246 }; 247 248 struct bpf_sk_storage; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_dst_cache: destination cache 261 * @sk_dst_pending_confirm: need to confirm neighbour 262 * @sk_policy: flow policy 263 * @sk_rx_skb_cache: cache copy of recently accessed RX skb 264 * @sk_receive_queue: incoming packets 265 * @sk_wmem_alloc: transmit queue bytes committed 266 * @sk_tsq_flags: TCP Small Queues flags 267 * @sk_write_queue: Packet sending queue 268 * @sk_omem_alloc: "o" is "option" or "other" 269 * @sk_wmem_queued: persistent queue size 270 * @sk_forward_alloc: space allocated forward 271 * @sk_napi_id: id of the last napi context to receive data for sk 272 * @sk_ll_usec: usecs to busypoll when there is no data 273 * @sk_allocation: allocation mode 274 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 275 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 276 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 277 * @sk_sndbuf: size of send buffer in bytes 278 * @__sk_flags_offset: empty field used to determine location of bitfield 279 * @sk_padding: unused element for alignment 280 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 281 * @sk_no_check_rx: allow zero checksum in RX packets 282 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 283 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 284 * @sk_route_forced_caps: static, forced route capabilities 285 * (set in tcp_init_sock()) 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_priority: %SO_PRIORITY setting 304 * @sk_type: socket type (%SOCK_STREAM, etc) 305 * @sk_protocol: which protocol this socket belongs in this network family 306 * @sk_peer_pid: &struct pid for this socket's peer 307 * @sk_peer_cred: %SO_PEERCRED setting 308 * @sk_rcvlowat: %SO_RCVLOWAT setting 309 * @sk_rcvtimeo: %SO_RCVTIMEO setting 310 * @sk_sndtimeo: %SO_SNDTIMEO setting 311 * @sk_txhash: computed flow hash for use on transmit 312 * @sk_filter: socket filtering instructions 313 * @sk_timer: sock cleanup timer 314 * @sk_stamp: time stamp of last packet received 315 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 316 * @sk_tsflags: SO_TIMESTAMPING socket options 317 * @sk_tskey: counter to disambiguate concurrent tstamp requests 318 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 319 * @sk_socket: Identd and reporting IO signals 320 * @sk_user_data: RPC layer private data 321 * @sk_frag: cached page frag 322 * @sk_peek_off: current peek_offset value 323 * @sk_send_head: front of stuff to transmit 324 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 325 * @sk_tx_skb_cache: cache copy of recently accessed TX skb 326 * @sk_security: used by security modules 327 * @sk_mark: generic packet mark 328 * @sk_cgrp_data: cgroup data for this cgroup 329 * @sk_memcg: this socket's memory cgroup association 330 * @sk_write_pending: a write to stream socket waits to start 331 * @sk_state_change: callback to indicate change in the state of the sock 332 * @sk_data_ready: callback to indicate there is data to be processed 333 * @sk_write_space: callback to indicate there is bf sending space available 334 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 335 * @sk_backlog_rcv: callback to process the backlog 336 * @sk_validate_xmit_skb: ptr to an optional validate function 337 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 338 * @sk_reuseport_cb: reuseport group container 339 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 340 * @sk_rcu: used during RCU grace period 341 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 342 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 343 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 344 * @sk_txtime_unused: unused txtime flags 345 */ 346 struct sock { 347 /* 348 * Now struct inet_timewait_sock also uses sock_common, so please just 349 * don't add nothing before this first member (__sk_common) --acme 350 */ 351 struct sock_common __sk_common; 352 #define sk_node __sk_common.skc_node 353 #define sk_nulls_node __sk_common.skc_nulls_node 354 #define sk_refcnt __sk_common.skc_refcnt 355 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 356 #ifdef CONFIG_XPS 357 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 358 #endif 359 360 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 361 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 362 #define sk_hash __sk_common.skc_hash 363 #define sk_portpair __sk_common.skc_portpair 364 #define sk_num __sk_common.skc_num 365 #define sk_dport __sk_common.skc_dport 366 #define sk_addrpair __sk_common.skc_addrpair 367 #define sk_daddr __sk_common.skc_daddr 368 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 369 #define sk_family __sk_common.skc_family 370 #define sk_state __sk_common.skc_state 371 #define sk_reuse __sk_common.skc_reuse 372 #define sk_reuseport __sk_common.skc_reuseport 373 #define sk_ipv6only __sk_common.skc_ipv6only 374 #define sk_net_refcnt __sk_common.skc_net_refcnt 375 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 376 #define sk_bind_node __sk_common.skc_bind_node 377 #define sk_prot __sk_common.skc_prot 378 #define sk_net __sk_common.skc_net 379 #define sk_v6_daddr __sk_common.skc_v6_daddr 380 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 381 #define sk_cookie __sk_common.skc_cookie 382 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 383 #define sk_flags __sk_common.skc_flags 384 #define sk_rxhash __sk_common.skc_rxhash 385 386 socket_lock_t sk_lock; 387 atomic_t sk_drops; 388 int sk_rcvlowat; 389 struct sk_buff_head sk_error_queue; 390 struct sk_buff *sk_rx_skb_cache; 391 struct sk_buff_head sk_receive_queue; 392 /* 393 * The backlog queue is special, it is always used with 394 * the per-socket spinlock held and requires low latency 395 * access. Therefore we special case it's implementation. 396 * Note : rmem_alloc is in this structure to fill a hole 397 * on 64bit arches, not because its logically part of 398 * backlog. 399 */ 400 struct { 401 atomic_t rmem_alloc; 402 int len; 403 struct sk_buff *head; 404 struct sk_buff *tail; 405 } sk_backlog; 406 #define sk_rmem_alloc sk_backlog.rmem_alloc 407 408 int sk_forward_alloc; 409 #ifdef CONFIG_NET_RX_BUSY_POLL 410 unsigned int sk_ll_usec; 411 /* ===== mostly read cache line ===== */ 412 unsigned int sk_napi_id; 413 #endif 414 int sk_rcvbuf; 415 416 struct sk_filter __rcu *sk_filter; 417 union { 418 struct socket_wq __rcu *sk_wq; 419 /* private: */ 420 struct socket_wq *sk_wq_raw; 421 /* public: */ 422 }; 423 #ifdef CONFIG_XFRM 424 struct xfrm_policy __rcu *sk_policy[2]; 425 #endif 426 struct dst_entry *sk_rx_dst; 427 struct dst_entry __rcu *sk_dst_cache; 428 atomic_t sk_omem_alloc; 429 int sk_sndbuf; 430 431 /* ===== cache line for TX ===== */ 432 int sk_wmem_queued; 433 refcount_t sk_wmem_alloc; 434 unsigned long sk_tsq_flags; 435 union { 436 struct sk_buff *sk_send_head; 437 struct rb_root tcp_rtx_queue; 438 }; 439 struct sk_buff *sk_tx_skb_cache; 440 struct sk_buff_head sk_write_queue; 441 __s32 sk_peek_off; 442 int sk_write_pending; 443 __u32 sk_dst_pending_confirm; 444 u32 sk_pacing_status; /* see enum sk_pacing */ 445 long sk_sndtimeo; 446 struct timer_list sk_timer; 447 __u32 sk_priority; 448 __u32 sk_mark; 449 unsigned long sk_pacing_rate; /* bytes per second */ 450 unsigned long sk_max_pacing_rate; 451 struct page_frag sk_frag; 452 netdev_features_t sk_route_caps; 453 netdev_features_t sk_route_nocaps; 454 netdev_features_t sk_route_forced_caps; 455 int sk_gso_type; 456 unsigned int sk_gso_max_size; 457 gfp_t sk_allocation; 458 __u32 sk_txhash; 459 460 /* 461 * Because of non atomicity rules, all 462 * changes are protected by socket lock. 463 */ 464 u8 sk_padding : 1, 465 sk_kern_sock : 1, 466 sk_no_check_tx : 1, 467 sk_no_check_rx : 1, 468 sk_userlocks : 4; 469 u8 sk_pacing_shift; 470 u16 sk_type; 471 u16 sk_protocol; 472 u16 sk_gso_max_segs; 473 unsigned long sk_lingertime; 474 struct proto *sk_prot_creator; 475 rwlock_t sk_callback_lock; 476 int sk_err, 477 sk_err_soft; 478 u32 sk_ack_backlog; 479 u32 sk_max_ack_backlog; 480 kuid_t sk_uid; 481 struct pid *sk_peer_pid; 482 const struct cred *sk_peer_cred; 483 long sk_rcvtimeo; 484 ktime_t sk_stamp; 485 #if BITS_PER_LONG==32 486 seqlock_t sk_stamp_seq; 487 #endif 488 u16 sk_tsflags; 489 u8 sk_shutdown; 490 u32 sk_tskey; 491 atomic_t sk_zckey; 492 493 u8 sk_clockid; 494 u8 sk_txtime_deadline_mode : 1, 495 sk_txtime_report_errors : 1, 496 sk_txtime_unused : 6; 497 498 struct socket *sk_socket; 499 void *sk_user_data; 500 #ifdef CONFIG_SECURITY 501 void *sk_security; 502 #endif 503 struct sock_cgroup_data sk_cgrp_data; 504 struct mem_cgroup *sk_memcg; 505 void (*sk_state_change)(struct sock *sk); 506 void (*sk_data_ready)(struct sock *sk); 507 void (*sk_write_space)(struct sock *sk); 508 void (*sk_error_report)(struct sock *sk); 509 int (*sk_backlog_rcv)(struct sock *sk, 510 struct sk_buff *skb); 511 #ifdef CONFIG_SOCK_VALIDATE_XMIT 512 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 513 struct net_device *dev, 514 struct sk_buff *skb); 515 #endif 516 void (*sk_destruct)(struct sock *sk); 517 struct sock_reuseport __rcu *sk_reuseport_cb; 518 #ifdef CONFIG_BPF_SYSCALL 519 struct bpf_sk_storage __rcu *sk_bpf_storage; 520 #endif 521 struct rcu_head sk_rcu; 522 }; 523 524 enum sk_pacing { 525 SK_PACING_NONE = 0, 526 SK_PACING_NEEDED = 1, 527 SK_PACING_FQ = 2, 528 }; 529 530 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 531 532 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 533 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 534 535 /* 536 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 537 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 538 * on a socket means that the socket will reuse everybody else's port 539 * without looking at the other's sk_reuse value. 540 */ 541 542 #define SK_NO_REUSE 0 543 #define SK_CAN_REUSE 1 544 #define SK_FORCE_REUSE 2 545 546 int sk_set_peek_off(struct sock *sk, int val); 547 548 static inline int sk_peek_offset(struct sock *sk, int flags) 549 { 550 if (unlikely(flags & MSG_PEEK)) { 551 return READ_ONCE(sk->sk_peek_off); 552 } 553 554 return 0; 555 } 556 557 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 558 { 559 s32 off = READ_ONCE(sk->sk_peek_off); 560 561 if (unlikely(off >= 0)) { 562 off = max_t(s32, off - val, 0); 563 WRITE_ONCE(sk->sk_peek_off, off); 564 } 565 } 566 567 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 568 { 569 sk_peek_offset_bwd(sk, -val); 570 } 571 572 /* 573 * Hashed lists helper routines 574 */ 575 static inline struct sock *sk_entry(const struct hlist_node *node) 576 { 577 return hlist_entry(node, struct sock, sk_node); 578 } 579 580 static inline struct sock *__sk_head(const struct hlist_head *head) 581 { 582 return hlist_entry(head->first, struct sock, sk_node); 583 } 584 585 static inline struct sock *sk_head(const struct hlist_head *head) 586 { 587 return hlist_empty(head) ? NULL : __sk_head(head); 588 } 589 590 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 591 { 592 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 593 } 594 595 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 596 { 597 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 598 } 599 600 static inline struct sock *sk_next(const struct sock *sk) 601 { 602 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 603 } 604 605 static inline struct sock *sk_nulls_next(const struct sock *sk) 606 { 607 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 608 hlist_nulls_entry(sk->sk_nulls_node.next, 609 struct sock, sk_nulls_node) : 610 NULL; 611 } 612 613 static inline bool sk_unhashed(const struct sock *sk) 614 { 615 return hlist_unhashed(&sk->sk_node); 616 } 617 618 static inline bool sk_hashed(const struct sock *sk) 619 { 620 return !sk_unhashed(sk); 621 } 622 623 static inline void sk_node_init(struct hlist_node *node) 624 { 625 node->pprev = NULL; 626 } 627 628 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 629 { 630 node->pprev = NULL; 631 } 632 633 static inline void __sk_del_node(struct sock *sk) 634 { 635 __hlist_del(&sk->sk_node); 636 } 637 638 /* NB: equivalent to hlist_del_init_rcu */ 639 static inline bool __sk_del_node_init(struct sock *sk) 640 { 641 if (sk_hashed(sk)) { 642 __sk_del_node(sk); 643 sk_node_init(&sk->sk_node); 644 return true; 645 } 646 return false; 647 } 648 649 /* Grab socket reference count. This operation is valid only 650 when sk is ALREADY grabbed f.e. it is found in hash table 651 or a list and the lookup is made under lock preventing hash table 652 modifications. 653 */ 654 655 static __always_inline void sock_hold(struct sock *sk) 656 { 657 refcount_inc(&sk->sk_refcnt); 658 } 659 660 /* Ungrab socket in the context, which assumes that socket refcnt 661 cannot hit zero, f.e. it is true in context of any socketcall. 662 */ 663 static __always_inline void __sock_put(struct sock *sk) 664 { 665 refcount_dec(&sk->sk_refcnt); 666 } 667 668 static inline bool sk_del_node_init(struct sock *sk) 669 { 670 bool rc = __sk_del_node_init(sk); 671 672 if (rc) { 673 /* paranoid for a while -acme */ 674 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 675 __sock_put(sk); 676 } 677 return rc; 678 } 679 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 680 681 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 682 { 683 if (sk_hashed(sk)) { 684 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 685 return true; 686 } 687 return false; 688 } 689 690 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 691 { 692 bool rc = __sk_nulls_del_node_init_rcu(sk); 693 694 if (rc) { 695 /* paranoid for a while -acme */ 696 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 697 __sock_put(sk); 698 } 699 return rc; 700 } 701 702 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 703 { 704 hlist_add_head(&sk->sk_node, list); 705 } 706 707 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 708 { 709 sock_hold(sk); 710 __sk_add_node(sk, list); 711 } 712 713 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 714 { 715 sock_hold(sk); 716 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 717 sk->sk_family == AF_INET6) 718 hlist_add_tail_rcu(&sk->sk_node, list); 719 else 720 hlist_add_head_rcu(&sk->sk_node, list); 721 } 722 723 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 724 { 725 sock_hold(sk); 726 hlist_add_tail_rcu(&sk->sk_node, list); 727 } 728 729 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 730 { 731 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 732 } 733 734 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 735 { 736 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 737 } 738 739 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 740 { 741 sock_hold(sk); 742 __sk_nulls_add_node_rcu(sk, list); 743 } 744 745 static inline void __sk_del_bind_node(struct sock *sk) 746 { 747 __hlist_del(&sk->sk_bind_node); 748 } 749 750 static inline void sk_add_bind_node(struct sock *sk, 751 struct hlist_head *list) 752 { 753 hlist_add_head(&sk->sk_bind_node, list); 754 } 755 756 #define sk_for_each(__sk, list) \ 757 hlist_for_each_entry(__sk, list, sk_node) 758 #define sk_for_each_rcu(__sk, list) \ 759 hlist_for_each_entry_rcu(__sk, list, sk_node) 760 #define sk_nulls_for_each(__sk, node, list) \ 761 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 762 #define sk_nulls_for_each_rcu(__sk, node, list) \ 763 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 764 #define sk_for_each_from(__sk) \ 765 hlist_for_each_entry_from(__sk, sk_node) 766 #define sk_nulls_for_each_from(__sk, node) \ 767 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 768 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 769 #define sk_for_each_safe(__sk, tmp, list) \ 770 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 771 #define sk_for_each_bound(__sk, list) \ 772 hlist_for_each_entry(__sk, list, sk_bind_node) 773 774 /** 775 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 776 * @tpos: the type * to use as a loop cursor. 777 * @pos: the &struct hlist_node to use as a loop cursor. 778 * @head: the head for your list. 779 * @offset: offset of hlist_node within the struct. 780 * 781 */ 782 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 783 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 784 pos != NULL && \ 785 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 786 pos = rcu_dereference(hlist_next_rcu(pos))) 787 788 static inline struct user_namespace *sk_user_ns(struct sock *sk) 789 { 790 /* Careful only use this in a context where these parameters 791 * can not change and must all be valid, such as recvmsg from 792 * userspace. 793 */ 794 return sk->sk_socket->file->f_cred->user_ns; 795 } 796 797 /* Sock flags */ 798 enum sock_flags { 799 SOCK_DEAD, 800 SOCK_DONE, 801 SOCK_URGINLINE, 802 SOCK_KEEPOPEN, 803 SOCK_LINGER, 804 SOCK_DESTROY, 805 SOCK_BROADCAST, 806 SOCK_TIMESTAMP, 807 SOCK_ZAPPED, 808 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 809 SOCK_DBG, /* %SO_DEBUG setting */ 810 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 811 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 812 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 813 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 814 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 815 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 816 SOCK_FASYNC, /* fasync() active */ 817 SOCK_RXQ_OVFL, 818 SOCK_ZEROCOPY, /* buffers from userspace */ 819 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 820 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 821 * Will use last 4 bytes of packet sent from 822 * user-space instead. 823 */ 824 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 825 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 826 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 827 SOCK_TXTIME, 828 SOCK_XDP, /* XDP is attached */ 829 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 830 }; 831 832 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 833 834 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 835 { 836 nsk->sk_flags = osk->sk_flags; 837 } 838 839 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 840 { 841 __set_bit(flag, &sk->sk_flags); 842 } 843 844 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 845 { 846 __clear_bit(flag, &sk->sk_flags); 847 } 848 849 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 850 { 851 return test_bit(flag, &sk->sk_flags); 852 } 853 854 #ifdef CONFIG_NET 855 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 856 static inline int sk_memalloc_socks(void) 857 { 858 return static_branch_unlikely(&memalloc_socks_key); 859 } 860 #else 861 862 static inline int sk_memalloc_socks(void) 863 { 864 return 0; 865 } 866 867 #endif 868 869 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 870 { 871 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 872 } 873 874 static inline void sk_acceptq_removed(struct sock *sk) 875 { 876 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 877 } 878 879 static inline void sk_acceptq_added(struct sock *sk) 880 { 881 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 882 } 883 884 static inline bool sk_acceptq_is_full(const struct sock *sk) 885 { 886 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 887 } 888 889 /* 890 * Compute minimal free write space needed to queue new packets. 891 */ 892 static inline int sk_stream_min_wspace(const struct sock *sk) 893 { 894 return READ_ONCE(sk->sk_wmem_queued) >> 1; 895 } 896 897 static inline int sk_stream_wspace(const struct sock *sk) 898 { 899 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 900 } 901 902 static inline void sk_wmem_queued_add(struct sock *sk, int val) 903 { 904 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 905 } 906 907 void sk_stream_write_space(struct sock *sk); 908 909 /* OOB backlog add */ 910 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 911 { 912 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 913 skb_dst_force(skb); 914 915 if (!sk->sk_backlog.tail) 916 WRITE_ONCE(sk->sk_backlog.head, skb); 917 else 918 sk->sk_backlog.tail->next = skb; 919 920 WRITE_ONCE(sk->sk_backlog.tail, skb); 921 skb->next = NULL; 922 } 923 924 /* 925 * Take into account size of receive queue and backlog queue 926 * Do not take into account this skb truesize, 927 * to allow even a single big packet to come. 928 */ 929 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 930 { 931 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 932 933 return qsize > limit; 934 } 935 936 /* The per-socket spinlock must be held here. */ 937 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 938 unsigned int limit) 939 { 940 if (sk_rcvqueues_full(sk, limit)) 941 return -ENOBUFS; 942 943 /* 944 * If the skb was allocated from pfmemalloc reserves, only 945 * allow SOCK_MEMALLOC sockets to use it as this socket is 946 * helping free memory 947 */ 948 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 949 return -ENOMEM; 950 951 __sk_add_backlog(sk, skb); 952 sk->sk_backlog.len += skb->truesize; 953 return 0; 954 } 955 956 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 957 958 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 959 { 960 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 961 return __sk_backlog_rcv(sk, skb); 962 963 return sk->sk_backlog_rcv(sk, skb); 964 } 965 966 static inline void sk_incoming_cpu_update(struct sock *sk) 967 { 968 int cpu = raw_smp_processor_id(); 969 970 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 971 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 972 } 973 974 static inline void sock_rps_record_flow_hash(__u32 hash) 975 { 976 #ifdef CONFIG_RPS 977 struct rps_sock_flow_table *sock_flow_table; 978 979 rcu_read_lock(); 980 sock_flow_table = rcu_dereference(rps_sock_flow_table); 981 rps_record_sock_flow(sock_flow_table, hash); 982 rcu_read_unlock(); 983 #endif 984 } 985 986 static inline void sock_rps_record_flow(const struct sock *sk) 987 { 988 #ifdef CONFIG_RPS 989 if (static_branch_unlikely(&rfs_needed)) { 990 /* Reading sk->sk_rxhash might incur an expensive cache line 991 * miss. 992 * 993 * TCP_ESTABLISHED does cover almost all states where RFS 994 * might be useful, and is cheaper [1] than testing : 995 * IPv4: inet_sk(sk)->inet_daddr 996 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 997 * OR an additional socket flag 998 * [1] : sk_state and sk_prot are in the same cache line. 999 */ 1000 if (sk->sk_state == TCP_ESTABLISHED) 1001 sock_rps_record_flow_hash(sk->sk_rxhash); 1002 } 1003 #endif 1004 } 1005 1006 static inline void sock_rps_save_rxhash(struct sock *sk, 1007 const struct sk_buff *skb) 1008 { 1009 #ifdef CONFIG_RPS 1010 if (unlikely(sk->sk_rxhash != skb->hash)) 1011 sk->sk_rxhash = skb->hash; 1012 #endif 1013 } 1014 1015 static inline void sock_rps_reset_rxhash(struct sock *sk) 1016 { 1017 #ifdef CONFIG_RPS 1018 sk->sk_rxhash = 0; 1019 #endif 1020 } 1021 1022 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1023 ({ int __rc; \ 1024 release_sock(__sk); \ 1025 __rc = __condition; \ 1026 if (!__rc) { \ 1027 *(__timeo) = wait_woken(__wait, \ 1028 TASK_INTERRUPTIBLE, \ 1029 *(__timeo)); \ 1030 } \ 1031 sched_annotate_sleep(); \ 1032 lock_sock(__sk); \ 1033 __rc = __condition; \ 1034 __rc; \ 1035 }) 1036 1037 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1038 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1039 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1040 int sk_stream_error(struct sock *sk, int flags, int err); 1041 void sk_stream_kill_queues(struct sock *sk); 1042 void sk_set_memalloc(struct sock *sk); 1043 void sk_clear_memalloc(struct sock *sk); 1044 1045 void __sk_flush_backlog(struct sock *sk); 1046 1047 static inline bool sk_flush_backlog(struct sock *sk) 1048 { 1049 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1050 __sk_flush_backlog(sk); 1051 return true; 1052 } 1053 return false; 1054 } 1055 1056 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1057 1058 struct request_sock_ops; 1059 struct timewait_sock_ops; 1060 struct inet_hashinfo; 1061 struct raw_hashinfo; 1062 struct smc_hashinfo; 1063 struct module; 1064 1065 /* 1066 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1067 * un-modified. Special care is taken when initializing object to zero. 1068 */ 1069 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1070 { 1071 if (offsetof(struct sock, sk_node.next) != 0) 1072 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1073 memset(&sk->sk_node.pprev, 0, 1074 size - offsetof(struct sock, sk_node.pprev)); 1075 } 1076 1077 /* Networking protocol blocks we attach to sockets. 1078 * socket layer -> transport layer interface 1079 */ 1080 struct proto { 1081 void (*close)(struct sock *sk, 1082 long timeout); 1083 int (*pre_connect)(struct sock *sk, 1084 struct sockaddr *uaddr, 1085 int addr_len); 1086 int (*connect)(struct sock *sk, 1087 struct sockaddr *uaddr, 1088 int addr_len); 1089 int (*disconnect)(struct sock *sk, int flags); 1090 1091 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1092 bool kern); 1093 1094 int (*ioctl)(struct sock *sk, int cmd, 1095 unsigned long arg); 1096 int (*init)(struct sock *sk); 1097 void (*destroy)(struct sock *sk); 1098 void (*shutdown)(struct sock *sk, int how); 1099 int (*setsockopt)(struct sock *sk, int level, 1100 int optname, char __user *optval, 1101 unsigned int optlen); 1102 int (*getsockopt)(struct sock *sk, int level, 1103 int optname, char __user *optval, 1104 int __user *option); 1105 void (*keepalive)(struct sock *sk, int valbool); 1106 #ifdef CONFIG_COMPAT 1107 int (*compat_setsockopt)(struct sock *sk, 1108 int level, 1109 int optname, char __user *optval, 1110 unsigned int optlen); 1111 int (*compat_getsockopt)(struct sock *sk, 1112 int level, 1113 int optname, char __user *optval, 1114 int __user *option); 1115 int (*compat_ioctl)(struct sock *sk, 1116 unsigned int cmd, unsigned long arg); 1117 #endif 1118 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1119 size_t len); 1120 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1121 size_t len, int noblock, int flags, 1122 int *addr_len); 1123 int (*sendpage)(struct sock *sk, struct page *page, 1124 int offset, size_t size, int flags); 1125 int (*bind)(struct sock *sk, 1126 struct sockaddr *uaddr, int addr_len); 1127 1128 int (*backlog_rcv) (struct sock *sk, 1129 struct sk_buff *skb); 1130 1131 void (*release_cb)(struct sock *sk); 1132 1133 /* Keeping track of sk's, looking them up, and port selection methods. */ 1134 int (*hash)(struct sock *sk); 1135 void (*unhash)(struct sock *sk); 1136 void (*rehash)(struct sock *sk); 1137 int (*get_port)(struct sock *sk, unsigned short snum); 1138 1139 /* Keeping track of sockets in use */ 1140 #ifdef CONFIG_PROC_FS 1141 unsigned int inuse_idx; 1142 #endif 1143 1144 bool (*stream_memory_free)(const struct sock *sk, int wake); 1145 bool (*stream_memory_read)(const struct sock *sk); 1146 /* Memory pressure */ 1147 void (*enter_memory_pressure)(struct sock *sk); 1148 void (*leave_memory_pressure)(struct sock *sk); 1149 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1150 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1151 /* 1152 * Pressure flag: try to collapse. 1153 * Technical note: it is used by multiple contexts non atomically. 1154 * All the __sk_mem_schedule() is of this nature: accounting 1155 * is strict, actions are advisory and have some latency. 1156 */ 1157 unsigned long *memory_pressure; 1158 long *sysctl_mem; 1159 1160 int *sysctl_wmem; 1161 int *sysctl_rmem; 1162 u32 sysctl_wmem_offset; 1163 u32 sysctl_rmem_offset; 1164 1165 int max_header; 1166 bool no_autobind; 1167 1168 struct kmem_cache *slab; 1169 unsigned int obj_size; 1170 slab_flags_t slab_flags; 1171 unsigned int useroffset; /* Usercopy region offset */ 1172 unsigned int usersize; /* Usercopy region size */ 1173 1174 struct percpu_counter *orphan_count; 1175 1176 struct request_sock_ops *rsk_prot; 1177 struct timewait_sock_ops *twsk_prot; 1178 1179 union { 1180 struct inet_hashinfo *hashinfo; 1181 struct udp_table *udp_table; 1182 struct raw_hashinfo *raw_hash; 1183 struct smc_hashinfo *smc_hash; 1184 } h; 1185 1186 struct module *owner; 1187 1188 char name[32]; 1189 1190 struct list_head node; 1191 #ifdef SOCK_REFCNT_DEBUG 1192 atomic_t socks; 1193 #endif 1194 int (*diag_destroy)(struct sock *sk, int err); 1195 } __randomize_layout; 1196 1197 int proto_register(struct proto *prot, int alloc_slab); 1198 void proto_unregister(struct proto *prot); 1199 int sock_load_diag_module(int family, int protocol); 1200 1201 #ifdef SOCK_REFCNT_DEBUG 1202 static inline void sk_refcnt_debug_inc(struct sock *sk) 1203 { 1204 atomic_inc(&sk->sk_prot->socks); 1205 } 1206 1207 static inline void sk_refcnt_debug_dec(struct sock *sk) 1208 { 1209 atomic_dec(&sk->sk_prot->socks); 1210 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1211 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1212 } 1213 1214 static inline void sk_refcnt_debug_release(const struct sock *sk) 1215 { 1216 if (refcount_read(&sk->sk_refcnt) != 1) 1217 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1218 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1219 } 1220 #else /* SOCK_REFCNT_DEBUG */ 1221 #define sk_refcnt_debug_inc(sk) do { } while (0) 1222 #define sk_refcnt_debug_dec(sk) do { } while (0) 1223 #define sk_refcnt_debug_release(sk) do { } while (0) 1224 #endif /* SOCK_REFCNT_DEBUG */ 1225 1226 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1227 { 1228 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1229 return false; 1230 1231 return sk->sk_prot->stream_memory_free ? 1232 sk->sk_prot->stream_memory_free(sk, wake) : true; 1233 } 1234 1235 static inline bool sk_stream_memory_free(const struct sock *sk) 1236 { 1237 return __sk_stream_memory_free(sk, 0); 1238 } 1239 1240 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1241 { 1242 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1243 __sk_stream_memory_free(sk, wake); 1244 } 1245 1246 static inline bool sk_stream_is_writeable(const struct sock *sk) 1247 { 1248 return __sk_stream_is_writeable(sk, 0); 1249 } 1250 1251 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1252 struct cgroup *ancestor) 1253 { 1254 #ifdef CONFIG_SOCK_CGROUP_DATA 1255 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1256 ancestor); 1257 #else 1258 return -ENOTSUPP; 1259 #endif 1260 } 1261 1262 static inline bool sk_has_memory_pressure(const struct sock *sk) 1263 { 1264 return sk->sk_prot->memory_pressure != NULL; 1265 } 1266 1267 static inline bool sk_under_memory_pressure(const struct sock *sk) 1268 { 1269 if (!sk->sk_prot->memory_pressure) 1270 return false; 1271 1272 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1273 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1274 return true; 1275 1276 return !!*sk->sk_prot->memory_pressure; 1277 } 1278 1279 static inline long 1280 sk_memory_allocated(const struct sock *sk) 1281 { 1282 return atomic_long_read(sk->sk_prot->memory_allocated); 1283 } 1284 1285 static inline long 1286 sk_memory_allocated_add(struct sock *sk, int amt) 1287 { 1288 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1289 } 1290 1291 static inline void 1292 sk_memory_allocated_sub(struct sock *sk, int amt) 1293 { 1294 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1295 } 1296 1297 static inline void sk_sockets_allocated_dec(struct sock *sk) 1298 { 1299 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1300 } 1301 1302 static inline void sk_sockets_allocated_inc(struct sock *sk) 1303 { 1304 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1305 } 1306 1307 static inline u64 1308 sk_sockets_allocated_read_positive(struct sock *sk) 1309 { 1310 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1311 } 1312 1313 static inline int 1314 proto_sockets_allocated_sum_positive(struct proto *prot) 1315 { 1316 return percpu_counter_sum_positive(prot->sockets_allocated); 1317 } 1318 1319 static inline long 1320 proto_memory_allocated(struct proto *prot) 1321 { 1322 return atomic_long_read(prot->memory_allocated); 1323 } 1324 1325 static inline bool 1326 proto_memory_pressure(struct proto *prot) 1327 { 1328 if (!prot->memory_pressure) 1329 return false; 1330 return !!*prot->memory_pressure; 1331 } 1332 1333 1334 #ifdef CONFIG_PROC_FS 1335 /* Called with local bh disabled */ 1336 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1337 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1338 int sock_inuse_get(struct net *net); 1339 #else 1340 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1341 int inc) 1342 { 1343 } 1344 #endif 1345 1346 1347 /* With per-bucket locks this operation is not-atomic, so that 1348 * this version is not worse. 1349 */ 1350 static inline int __sk_prot_rehash(struct sock *sk) 1351 { 1352 sk->sk_prot->unhash(sk); 1353 return sk->sk_prot->hash(sk); 1354 } 1355 1356 /* About 10 seconds */ 1357 #define SOCK_DESTROY_TIME (10*HZ) 1358 1359 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1360 #define PROT_SOCK 1024 1361 1362 #define SHUTDOWN_MASK 3 1363 #define RCV_SHUTDOWN 1 1364 #define SEND_SHUTDOWN 2 1365 1366 #define SOCK_SNDBUF_LOCK 1 1367 #define SOCK_RCVBUF_LOCK 2 1368 #define SOCK_BINDADDR_LOCK 4 1369 #define SOCK_BINDPORT_LOCK 8 1370 1371 struct socket_alloc { 1372 struct socket socket; 1373 struct inode vfs_inode; 1374 }; 1375 1376 static inline struct socket *SOCKET_I(struct inode *inode) 1377 { 1378 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1379 } 1380 1381 static inline struct inode *SOCK_INODE(struct socket *socket) 1382 { 1383 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1384 } 1385 1386 /* 1387 * Functions for memory accounting 1388 */ 1389 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1390 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1391 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1392 void __sk_mem_reclaim(struct sock *sk, int amount); 1393 1394 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1395 * do not necessarily have 16x time more memory than 4KB ones. 1396 */ 1397 #define SK_MEM_QUANTUM 4096 1398 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1399 #define SK_MEM_SEND 0 1400 #define SK_MEM_RECV 1 1401 1402 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1403 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1404 { 1405 long val = sk->sk_prot->sysctl_mem[index]; 1406 1407 #if PAGE_SIZE > SK_MEM_QUANTUM 1408 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1409 #elif PAGE_SIZE < SK_MEM_QUANTUM 1410 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1411 #endif 1412 return val; 1413 } 1414 1415 static inline int sk_mem_pages(int amt) 1416 { 1417 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1418 } 1419 1420 static inline bool sk_has_account(struct sock *sk) 1421 { 1422 /* return true if protocol supports memory accounting */ 1423 return !!sk->sk_prot->memory_allocated; 1424 } 1425 1426 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1427 { 1428 if (!sk_has_account(sk)) 1429 return true; 1430 return size <= sk->sk_forward_alloc || 1431 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1432 } 1433 1434 static inline bool 1435 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1436 { 1437 if (!sk_has_account(sk)) 1438 return true; 1439 return size<= sk->sk_forward_alloc || 1440 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1441 skb_pfmemalloc(skb); 1442 } 1443 1444 static inline void sk_mem_reclaim(struct sock *sk) 1445 { 1446 if (!sk_has_account(sk)) 1447 return; 1448 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1449 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1450 } 1451 1452 static inline void sk_mem_reclaim_partial(struct sock *sk) 1453 { 1454 if (!sk_has_account(sk)) 1455 return; 1456 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1457 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1458 } 1459 1460 static inline void sk_mem_charge(struct sock *sk, int size) 1461 { 1462 if (!sk_has_account(sk)) 1463 return; 1464 sk->sk_forward_alloc -= size; 1465 } 1466 1467 static inline void sk_mem_uncharge(struct sock *sk, int size) 1468 { 1469 if (!sk_has_account(sk)) 1470 return; 1471 sk->sk_forward_alloc += size; 1472 1473 /* Avoid a possible overflow. 1474 * TCP send queues can make this happen, if sk_mem_reclaim() 1475 * is not called and more than 2 GBytes are released at once. 1476 * 1477 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1478 * no need to hold that much forward allocation anyway. 1479 */ 1480 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1481 __sk_mem_reclaim(sk, 1 << 20); 1482 } 1483 1484 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key); 1485 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1486 { 1487 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1488 sk_wmem_queued_add(sk, -skb->truesize); 1489 sk_mem_uncharge(sk, skb->truesize); 1490 if (static_branch_unlikely(&tcp_tx_skb_cache_key) && 1491 !sk->sk_tx_skb_cache && !skb_cloned(skb)) { 1492 skb_ext_reset(skb); 1493 skb_zcopy_clear(skb, true); 1494 sk->sk_tx_skb_cache = skb; 1495 return; 1496 } 1497 __kfree_skb(skb); 1498 } 1499 1500 static inline void sock_release_ownership(struct sock *sk) 1501 { 1502 if (sk->sk_lock.owned) { 1503 sk->sk_lock.owned = 0; 1504 1505 /* The sk_lock has mutex_unlock() semantics: */ 1506 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1507 } 1508 } 1509 1510 /* 1511 * Macro so as to not evaluate some arguments when 1512 * lockdep is not enabled. 1513 * 1514 * Mark both the sk_lock and the sk_lock.slock as a 1515 * per-address-family lock class. 1516 */ 1517 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1518 do { \ 1519 sk->sk_lock.owned = 0; \ 1520 init_waitqueue_head(&sk->sk_lock.wq); \ 1521 spin_lock_init(&(sk)->sk_lock.slock); \ 1522 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1523 sizeof((sk)->sk_lock)); \ 1524 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1525 (skey), (sname)); \ 1526 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1527 } while (0) 1528 1529 #ifdef CONFIG_LOCKDEP 1530 static inline bool lockdep_sock_is_held(const struct sock *sk) 1531 { 1532 return lockdep_is_held(&sk->sk_lock) || 1533 lockdep_is_held(&sk->sk_lock.slock); 1534 } 1535 #endif 1536 1537 void lock_sock_nested(struct sock *sk, int subclass); 1538 1539 static inline void lock_sock(struct sock *sk) 1540 { 1541 lock_sock_nested(sk, 0); 1542 } 1543 1544 void __release_sock(struct sock *sk); 1545 void release_sock(struct sock *sk); 1546 1547 /* BH context may only use the following locking interface. */ 1548 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1549 #define bh_lock_sock_nested(__sk) \ 1550 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1551 SINGLE_DEPTH_NESTING) 1552 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1553 1554 bool lock_sock_fast(struct sock *sk); 1555 /** 1556 * unlock_sock_fast - complement of lock_sock_fast 1557 * @sk: socket 1558 * @slow: slow mode 1559 * 1560 * fast unlock socket for user context. 1561 * If slow mode is on, we call regular release_sock() 1562 */ 1563 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1564 { 1565 if (slow) 1566 release_sock(sk); 1567 else 1568 spin_unlock_bh(&sk->sk_lock.slock); 1569 } 1570 1571 /* Used by processes to "lock" a socket state, so that 1572 * interrupts and bottom half handlers won't change it 1573 * from under us. It essentially blocks any incoming 1574 * packets, so that we won't get any new data or any 1575 * packets that change the state of the socket. 1576 * 1577 * While locked, BH processing will add new packets to 1578 * the backlog queue. This queue is processed by the 1579 * owner of the socket lock right before it is released. 1580 * 1581 * Since ~2.3.5 it is also exclusive sleep lock serializing 1582 * accesses from user process context. 1583 */ 1584 1585 static inline void sock_owned_by_me(const struct sock *sk) 1586 { 1587 #ifdef CONFIG_LOCKDEP 1588 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1589 #endif 1590 } 1591 1592 static inline bool sock_owned_by_user(const struct sock *sk) 1593 { 1594 sock_owned_by_me(sk); 1595 return sk->sk_lock.owned; 1596 } 1597 1598 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1599 { 1600 return sk->sk_lock.owned; 1601 } 1602 1603 /* no reclassification while locks are held */ 1604 static inline bool sock_allow_reclassification(const struct sock *csk) 1605 { 1606 struct sock *sk = (struct sock *)csk; 1607 1608 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1609 } 1610 1611 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1612 struct proto *prot, int kern); 1613 void sk_free(struct sock *sk); 1614 void sk_destruct(struct sock *sk); 1615 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1616 void sk_free_unlock_clone(struct sock *sk); 1617 1618 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1619 gfp_t priority); 1620 void __sock_wfree(struct sk_buff *skb); 1621 void sock_wfree(struct sk_buff *skb); 1622 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1623 gfp_t priority); 1624 void skb_orphan_partial(struct sk_buff *skb); 1625 void sock_rfree(struct sk_buff *skb); 1626 void sock_efree(struct sk_buff *skb); 1627 #ifdef CONFIG_INET 1628 void sock_edemux(struct sk_buff *skb); 1629 #else 1630 #define sock_edemux sock_efree 1631 #endif 1632 1633 int sock_setsockopt(struct socket *sock, int level, int op, 1634 char __user *optval, unsigned int optlen); 1635 1636 int sock_getsockopt(struct socket *sock, int level, int op, 1637 char __user *optval, int __user *optlen); 1638 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1639 bool timeval, bool time32); 1640 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1641 int noblock, int *errcode); 1642 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1643 unsigned long data_len, int noblock, 1644 int *errcode, int max_page_order); 1645 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1646 void sock_kfree_s(struct sock *sk, void *mem, int size); 1647 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1648 void sk_send_sigurg(struct sock *sk); 1649 1650 struct sockcm_cookie { 1651 u64 transmit_time; 1652 u32 mark; 1653 u16 tsflags; 1654 }; 1655 1656 static inline void sockcm_init(struct sockcm_cookie *sockc, 1657 const struct sock *sk) 1658 { 1659 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1660 } 1661 1662 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1663 struct sockcm_cookie *sockc); 1664 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1665 struct sockcm_cookie *sockc); 1666 1667 /* 1668 * Functions to fill in entries in struct proto_ops when a protocol 1669 * does not implement a particular function. 1670 */ 1671 int sock_no_bind(struct socket *, struct sockaddr *, int); 1672 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1673 int sock_no_socketpair(struct socket *, struct socket *); 1674 int sock_no_accept(struct socket *, struct socket *, int, bool); 1675 int sock_no_getname(struct socket *, struct sockaddr *, int); 1676 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1677 int sock_no_listen(struct socket *, int); 1678 int sock_no_shutdown(struct socket *, int); 1679 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1680 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1681 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1682 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1683 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1684 int sock_no_mmap(struct file *file, struct socket *sock, 1685 struct vm_area_struct *vma); 1686 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1687 size_t size, int flags); 1688 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1689 int offset, size_t size, int flags); 1690 1691 /* 1692 * Functions to fill in entries in struct proto_ops when a protocol 1693 * uses the inet style. 1694 */ 1695 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1696 char __user *optval, int __user *optlen); 1697 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1698 int flags); 1699 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1700 char __user *optval, unsigned int optlen); 1701 int compat_sock_common_getsockopt(struct socket *sock, int level, 1702 int optname, char __user *optval, int __user *optlen); 1703 int compat_sock_common_setsockopt(struct socket *sock, int level, 1704 int optname, char __user *optval, unsigned int optlen); 1705 1706 void sk_common_release(struct sock *sk); 1707 1708 /* 1709 * Default socket callbacks and setup code 1710 */ 1711 1712 /* Initialise core socket variables */ 1713 void sock_init_data(struct socket *sock, struct sock *sk); 1714 1715 /* 1716 * Socket reference counting postulates. 1717 * 1718 * * Each user of socket SHOULD hold a reference count. 1719 * * Each access point to socket (an hash table bucket, reference from a list, 1720 * running timer, skb in flight MUST hold a reference count. 1721 * * When reference count hits 0, it means it will never increase back. 1722 * * When reference count hits 0, it means that no references from 1723 * outside exist to this socket and current process on current CPU 1724 * is last user and may/should destroy this socket. 1725 * * sk_free is called from any context: process, BH, IRQ. When 1726 * it is called, socket has no references from outside -> sk_free 1727 * may release descendant resources allocated by the socket, but 1728 * to the time when it is called, socket is NOT referenced by any 1729 * hash tables, lists etc. 1730 * * Packets, delivered from outside (from network or from another process) 1731 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1732 * when they sit in queue. Otherwise, packets will leak to hole, when 1733 * socket is looked up by one cpu and unhasing is made by another CPU. 1734 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1735 * (leak to backlog). Packet socket does all the processing inside 1736 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1737 * use separate SMP lock, so that they are prone too. 1738 */ 1739 1740 /* Ungrab socket and destroy it, if it was the last reference. */ 1741 static inline void sock_put(struct sock *sk) 1742 { 1743 if (refcount_dec_and_test(&sk->sk_refcnt)) 1744 sk_free(sk); 1745 } 1746 /* Generic version of sock_put(), dealing with all sockets 1747 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1748 */ 1749 void sock_gen_put(struct sock *sk); 1750 1751 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1752 unsigned int trim_cap, bool refcounted); 1753 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1754 const int nested) 1755 { 1756 return __sk_receive_skb(sk, skb, nested, 1, true); 1757 } 1758 1759 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1760 { 1761 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1762 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1763 return; 1764 sk->sk_tx_queue_mapping = tx_queue; 1765 } 1766 1767 #define NO_QUEUE_MAPPING USHRT_MAX 1768 1769 static inline void sk_tx_queue_clear(struct sock *sk) 1770 { 1771 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1772 } 1773 1774 static inline int sk_tx_queue_get(const struct sock *sk) 1775 { 1776 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1777 return sk->sk_tx_queue_mapping; 1778 1779 return -1; 1780 } 1781 1782 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1783 { 1784 #ifdef CONFIG_XPS 1785 if (skb_rx_queue_recorded(skb)) { 1786 u16 rx_queue = skb_get_rx_queue(skb); 1787 1788 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING)) 1789 return; 1790 1791 sk->sk_rx_queue_mapping = rx_queue; 1792 } 1793 #endif 1794 } 1795 1796 static inline void sk_rx_queue_clear(struct sock *sk) 1797 { 1798 #ifdef CONFIG_XPS 1799 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING; 1800 #endif 1801 } 1802 1803 #ifdef CONFIG_XPS 1804 static inline int sk_rx_queue_get(const struct sock *sk) 1805 { 1806 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING) 1807 return sk->sk_rx_queue_mapping; 1808 1809 return -1; 1810 } 1811 #endif 1812 1813 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1814 { 1815 sk_tx_queue_clear(sk); 1816 sk->sk_socket = sock; 1817 } 1818 1819 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1820 { 1821 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1822 return &rcu_dereference_raw(sk->sk_wq)->wait; 1823 } 1824 /* Detach socket from process context. 1825 * Announce socket dead, detach it from wait queue and inode. 1826 * Note that parent inode held reference count on this struct sock, 1827 * we do not release it in this function, because protocol 1828 * probably wants some additional cleanups or even continuing 1829 * to work with this socket (TCP). 1830 */ 1831 static inline void sock_orphan(struct sock *sk) 1832 { 1833 write_lock_bh(&sk->sk_callback_lock); 1834 sock_set_flag(sk, SOCK_DEAD); 1835 sk_set_socket(sk, NULL); 1836 sk->sk_wq = NULL; 1837 write_unlock_bh(&sk->sk_callback_lock); 1838 } 1839 1840 static inline void sock_graft(struct sock *sk, struct socket *parent) 1841 { 1842 WARN_ON(parent->sk); 1843 write_lock_bh(&sk->sk_callback_lock); 1844 rcu_assign_pointer(sk->sk_wq, &parent->wq); 1845 parent->sk = sk; 1846 sk_set_socket(sk, parent); 1847 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1848 security_sock_graft(sk, parent); 1849 write_unlock_bh(&sk->sk_callback_lock); 1850 } 1851 1852 kuid_t sock_i_uid(struct sock *sk); 1853 unsigned long sock_i_ino(struct sock *sk); 1854 1855 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1856 { 1857 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1858 } 1859 1860 static inline u32 net_tx_rndhash(void) 1861 { 1862 u32 v = prandom_u32(); 1863 1864 return v ?: 1; 1865 } 1866 1867 static inline void sk_set_txhash(struct sock *sk) 1868 { 1869 sk->sk_txhash = net_tx_rndhash(); 1870 } 1871 1872 static inline void sk_rethink_txhash(struct sock *sk) 1873 { 1874 if (sk->sk_txhash) 1875 sk_set_txhash(sk); 1876 } 1877 1878 static inline struct dst_entry * 1879 __sk_dst_get(struct sock *sk) 1880 { 1881 return rcu_dereference_check(sk->sk_dst_cache, 1882 lockdep_sock_is_held(sk)); 1883 } 1884 1885 static inline struct dst_entry * 1886 sk_dst_get(struct sock *sk) 1887 { 1888 struct dst_entry *dst; 1889 1890 rcu_read_lock(); 1891 dst = rcu_dereference(sk->sk_dst_cache); 1892 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1893 dst = NULL; 1894 rcu_read_unlock(); 1895 return dst; 1896 } 1897 1898 static inline void dst_negative_advice(struct sock *sk) 1899 { 1900 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1901 1902 sk_rethink_txhash(sk); 1903 1904 if (dst && dst->ops->negative_advice) { 1905 ndst = dst->ops->negative_advice(dst); 1906 1907 if (ndst != dst) { 1908 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1909 sk_tx_queue_clear(sk); 1910 sk->sk_dst_pending_confirm = 0; 1911 } 1912 } 1913 } 1914 1915 static inline void 1916 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1917 { 1918 struct dst_entry *old_dst; 1919 1920 sk_tx_queue_clear(sk); 1921 sk->sk_dst_pending_confirm = 0; 1922 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1923 lockdep_sock_is_held(sk)); 1924 rcu_assign_pointer(sk->sk_dst_cache, dst); 1925 dst_release(old_dst); 1926 } 1927 1928 static inline void 1929 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1930 { 1931 struct dst_entry *old_dst; 1932 1933 sk_tx_queue_clear(sk); 1934 sk->sk_dst_pending_confirm = 0; 1935 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1936 dst_release(old_dst); 1937 } 1938 1939 static inline void 1940 __sk_dst_reset(struct sock *sk) 1941 { 1942 __sk_dst_set(sk, NULL); 1943 } 1944 1945 static inline void 1946 sk_dst_reset(struct sock *sk) 1947 { 1948 sk_dst_set(sk, NULL); 1949 } 1950 1951 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1952 1953 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1954 1955 static inline void sk_dst_confirm(struct sock *sk) 1956 { 1957 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 1958 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 1959 } 1960 1961 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1962 { 1963 if (skb_get_dst_pending_confirm(skb)) { 1964 struct sock *sk = skb->sk; 1965 unsigned long now = jiffies; 1966 1967 /* avoid dirtying neighbour */ 1968 if (READ_ONCE(n->confirmed) != now) 1969 WRITE_ONCE(n->confirmed, now); 1970 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 1971 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 1972 } 1973 } 1974 1975 bool sk_mc_loop(struct sock *sk); 1976 1977 static inline bool sk_can_gso(const struct sock *sk) 1978 { 1979 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1980 } 1981 1982 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1983 1984 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1985 { 1986 sk->sk_route_nocaps |= flags; 1987 sk->sk_route_caps &= ~flags; 1988 } 1989 1990 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1991 struct iov_iter *from, char *to, 1992 int copy, int offset) 1993 { 1994 if (skb->ip_summed == CHECKSUM_NONE) { 1995 __wsum csum = 0; 1996 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1997 return -EFAULT; 1998 skb->csum = csum_block_add(skb->csum, csum, offset); 1999 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2000 if (!copy_from_iter_full_nocache(to, copy, from)) 2001 return -EFAULT; 2002 } else if (!copy_from_iter_full(to, copy, from)) 2003 return -EFAULT; 2004 2005 return 0; 2006 } 2007 2008 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2009 struct iov_iter *from, int copy) 2010 { 2011 int err, offset = skb->len; 2012 2013 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2014 copy, offset); 2015 if (err) 2016 __skb_trim(skb, offset); 2017 2018 return err; 2019 } 2020 2021 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2022 struct sk_buff *skb, 2023 struct page *page, 2024 int off, int copy) 2025 { 2026 int err; 2027 2028 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2029 copy, skb->len); 2030 if (err) 2031 return err; 2032 2033 skb->len += copy; 2034 skb->data_len += copy; 2035 skb->truesize += copy; 2036 sk_wmem_queued_add(sk, copy); 2037 sk_mem_charge(sk, copy); 2038 return 0; 2039 } 2040 2041 /** 2042 * sk_wmem_alloc_get - returns write allocations 2043 * @sk: socket 2044 * 2045 * Return: sk_wmem_alloc minus initial offset of one 2046 */ 2047 static inline int sk_wmem_alloc_get(const struct sock *sk) 2048 { 2049 return refcount_read(&sk->sk_wmem_alloc) - 1; 2050 } 2051 2052 /** 2053 * sk_rmem_alloc_get - returns read allocations 2054 * @sk: socket 2055 * 2056 * Return: sk_rmem_alloc 2057 */ 2058 static inline int sk_rmem_alloc_get(const struct sock *sk) 2059 { 2060 return atomic_read(&sk->sk_rmem_alloc); 2061 } 2062 2063 /** 2064 * sk_has_allocations - check if allocations are outstanding 2065 * @sk: socket 2066 * 2067 * Return: true if socket has write or read allocations 2068 */ 2069 static inline bool sk_has_allocations(const struct sock *sk) 2070 { 2071 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2072 } 2073 2074 /** 2075 * skwq_has_sleeper - check if there are any waiting processes 2076 * @wq: struct socket_wq 2077 * 2078 * Return: true if socket_wq has waiting processes 2079 * 2080 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2081 * barrier call. They were added due to the race found within the tcp code. 2082 * 2083 * Consider following tcp code paths:: 2084 * 2085 * CPU1 CPU2 2086 * sys_select receive packet 2087 * ... ... 2088 * __add_wait_queue update tp->rcv_nxt 2089 * ... ... 2090 * tp->rcv_nxt check sock_def_readable 2091 * ... { 2092 * schedule rcu_read_lock(); 2093 * wq = rcu_dereference(sk->sk_wq); 2094 * if (wq && waitqueue_active(&wq->wait)) 2095 * wake_up_interruptible(&wq->wait) 2096 * ... 2097 * } 2098 * 2099 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2100 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2101 * could then endup calling schedule and sleep forever if there are no more 2102 * data on the socket. 2103 * 2104 */ 2105 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2106 { 2107 return wq && wq_has_sleeper(&wq->wait); 2108 } 2109 2110 /** 2111 * sock_poll_wait - place memory barrier behind the poll_wait call. 2112 * @filp: file 2113 * @sock: socket to wait on 2114 * @p: poll_table 2115 * 2116 * See the comments in the wq_has_sleeper function. 2117 */ 2118 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2119 poll_table *p) 2120 { 2121 if (!poll_does_not_wait(p)) { 2122 poll_wait(filp, &sock->wq.wait, p); 2123 /* We need to be sure we are in sync with the 2124 * socket flags modification. 2125 * 2126 * This memory barrier is paired in the wq_has_sleeper. 2127 */ 2128 smp_mb(); 2129 } 2130 } 2131 2132 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2133 { 2134 if (sk->sk_txhash) { 2135 skb->l4_hash = 1; 2136 skb->hash = sk->sk_txhash; 2137 } 2138 } 2139 2140 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2141 2142 /* 2143 * Queue a received datagram if it will fit. Stream and sequenced 2144 * protocols can't normally use this as they need to fit buffers in 2145 * and play with them. 2146 * 2147 * Inlined as it's very short and called for pretty much every 2148 * packet ever received. 2149 */ 2150 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2151 { 2152 skb_orphan(skb); 2153 skb->sk = sk; 2154 skb->destructor = sock_rfree; 2155 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2156 sk_mem_charge(sk, skb->truesize); 2157 } 2158 2159 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2160 unsigned long expires); 2161 2162 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2163 2164 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2165 struct sk_buff *skb, unsigned int flags, 2166 void (*destructor)(struct sock *sk, 2167 struct sk_buff *skb)); 2168 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2169 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2170 2171 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2172 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2173 2174 /* 2175 * Recover an error report and clear atomically 2176 */ 2177 2178 static inline int sock_error(struct sock *sk) 2179 { 2180 int err; 2181 if (likely(!sk->sk_err)) 2182 return 0; 2183 err = xchg(&sk->sk_err, 0); 2184 return -err; 2185 } 2186 2187 static inline unsigned long sock_wspace(struct sock *sk) 2188 { 2189 int amt = 0; 2190 2191 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2192 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2193 if (amt < 0) 2194 amt = 0; 2195 } 2196 return amt; 2197 } 2198 2199 /* Note: 2200 * We use sk->sk_wq_raw, from contexts knowing this 2201 * pointer is not NULL and cannot disappear/change. 2202 */ 2203 static inline void sk_set_bit(int nr, struct sock *sk) 2204 { 2205 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2206 !sock_flag(sk, SOCK_FASYNC)) 2207 return; 2208 2209 set_bit(nr, &sk->sk_wq_raw->flags); 2210 } 2211 2212 static inline void sk_clear_bit(int nr, struct sock *sk) 2213 { 2214 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2215 !sock_flag(sk, SOCK_FASYNC)) 2216 return; 2217 2218 clear_bit(nr, &sk->sk_wq_raw->flags); 2219 } 2220 2221 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2222 { 2223 if (sock_flag(sk, SOCK_FASYNC)) { 2224 rcu_read_lock(); 2225 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2226 rcu_read_unlock(); 2227 } 2228 } 2229 2230 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2231 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2232 * Note: for send buffers, TCP works better if we can build two skbs at 2233 * minimum. 2234 */ 2235 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2236 2237 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2238 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2239 2240 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2241 { 2242 u32 val; 2243 2244 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2245 return; 2246 2247 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2248 2249 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2250 } 2251 2252 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2253 bool force_schedule); 2254 2255 /** 2256 * sk_page_frag - return an appropriate page_frag 2257 * @sk: socket 2258 * 2259 * Use the per task page_frag instead of the per socket one for 2260 * optimization when we know that we're in the normal context and owns 2261 * everything that's associated with %current. 2262 * 2263 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest 2264 * inside other socket operations and end up recursing into sk_page_frag() 2265 * while it's already in use. 2266 * 2267 * Return: a per task page_frag if context allows that, 2268 * otherwise a per socket one. 2269 */ 2270 static inline struct page_frag *sk_page_frag(struct sock *sk) 2271 { 2272 if (gfpflags_normal_context(sk->sk_allocation)) 2273 return ¤t->task_frag; 2274 2275 return &sk->sk_frag; 2276 } 2277 2278 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2279 2280 /* 2281 * Default write policy as shown to user space via poll/select/SIGIO 2282 */ 2283 static inline bool sock_writeable(const struct sock *sk) 2284 { 2285 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2286 } 2287 2288 static inline gfp_t gfp_any(void) 2289 { 2290 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2291 } 2292 2293 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2294 { 2295 return noblock ? 0 : sk->sk_rcvtimeo; 2296 } 2297 2298 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2299 { 2300 return noblock ? 0 : sk->sk_sndtimeo; 2301 } 2302 2303 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2304 { 2305 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2306 2307 return v ?: 1; 2308 } 2309 2310 /* Alas, with timeout socket operations are not restartable. 2311 * Compare this to poll(). 2312 */ 2313 static inline int sock_intr_errno(long timeo) 2314 { 2315 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2316 } 2317 2318 struct sock_skb_cb { 2319 u32 dropcount; 2320 }; 2321 2322 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2323 * using skb->cb[] would keep using it directly and utilize its 2324 * alignement guarantee. 2325 */ 2326 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2327 sizeof(struct sock_skb_cb))) 2328 2329 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2330 SOCK_SKB_CB_OFFSET)) 2331 2332 #define sock_skb_cb_check_size(size) \ 2333 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2334 2335 static inline void 2336 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2337 { 2338 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2339 atomic_read(&sk->sk_drops) : 0; 2340 } 2341 2342 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2343 { 2344 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2345 2346 atomic_add(segs, &sk->sk_drops); 2347 } 2348 2349 static inline ktime_t sock_read_timestamp(struct sock *sk) 2350 { 2351 #if BITS_PER_LONG==32 2352 unsigned int seq; 2353 ktime_t kt; 2354 2355 do { 2356 seq = read_seqbegin(&sk->sk_stamp_seq); 2357 kt = sk->sk_stamp; 2358 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2359 2360 return kt; 2361 #else 2362 return READ_ONCE(sk->sk_stamp); 2363 #endif 2364 } 2365 2366 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2367 { 2368 #if BITS_PER_LONG==32 2369 write_seqlock(&sk->sk_stamp_seq); 2370 sk->sk_stamp = kt; 2371 write_sequnlock(&sk->sk_stamp_seq); 2372 #else 2373 WRITE_ONCE(sk->sk_stamp, kt); 2374 #endif 2375 } 2376 2377 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2378 struct sk_buff *skb); 2379 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2380 struct sk_buff *skb); 2381 2382 static inline void 2383 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2384 { 2385 ktime_t kt = skb->tstamp; 2386 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2387 2388 /* 2389 * generate control messages if 2390 * - receive time stamping in software requested 2391 * - software time stamp available and wanted 2392 * - hardware time stamps available and wanted 2393 */ 2394 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2395 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2396 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2397 (hwtstamps->hwtstamp && 2398 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2399 __sock_recv_timestamp(msg, sk, skb); 2400 else 2401 sock_write_timestamp(sk, kt); 2402 2403 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2404 __sock_recv_wifi_status(msg, sk, skb); 2405 } 2406 2407 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2408 struct sk_buff *skb); 2409 2410 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2411 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2412 struct sk_buff *skb) 2413 { 2414 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2415 (1UL << SOCK_RCVTSTAMP)) 2416 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2417 SOF_TIMESTAMPING_RAW_HARDWARE) 2418 2419 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2420 __sock_recv_ts_and_drops(msg, sk, skb); 2421 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2422 sock_write_timestamp(sk, skb->tstamp); 2423 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2424 sock_write_timestamp(sk, 0); 2425 } 2426 2427 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2428 2429 /** 2430 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2431 * @sk: socket sending this packet 2432 * @tsflags: timestamping flags to use 2433 * @tx_flags: completed with instructions for time stamping 2434 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2435 * 2436 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2437 */ 2438 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2439 __u8 *tx_flags, __u32 *tskey) 2440 { 2441 if (unlikely(tsflags)) { 2442 __sock_tx_timestamp(tsflags, tx_flags); 2443 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2444 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2445 *tskey = sk->sk_tskey++; 2446 } 2447 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2448 *tx_flags |= SKBTX_WIFI_STATUS; 2449 } 2450 2451 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2452 __u8 *tx_flags) 2453 { 2454 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2455 } 2456 2457 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2458 { 2459 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2460 &skb_shinfo(skb)->tskey); 2461 } 2462 2463 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key); 2464 /** 2465 * sk_eat_skb - Release a skb if it is no longer needed 2466 * @sk: socket to eat this skb from 2467 * @skb: socket buffer to eat 2468 * 2469 * This routine must be called with interrupts disabled or with the socket 2470 * locked so that the sk_buff queue operation is ok. 2471 */ 2472 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2473 { 2474 __skb_unlink(skb, &sk->sk_receive_queue); 2475 if (static_branch_unlikely(&tcp_rx_skb_cache_key) && 2476 !sk->sk_rx_skb_cache) { 2477 sk->sk_rx_skb_cache = skb; 2478 skb_orphan(skb); 2479 return; 2480 } 2481 __kfree_skb(skb); 2482 } 2483 2484 static inline 2485 struct net *sock_net(const struct sock *sk) 2486 { 2487 return read_pnet(&sk->sk_net); 2488 } 2489 2490 static inline 2491 void sock_net_set(struct sock *sk, struct net *net) 2492 { 2493 write_pnet(&sk->sk_net, net); 2494 } 2495 2496 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2497 { 2498 if (skb->sk) { 2499 struct sock *sk = skb->sk; 2500 2501 skb->destructor = NULL; 2502 skb->sk = NULL; 2503 return sk; 2504 } 2505 return NULL; 2506 } 2507 2508 /* This helper checks if a socket is a full socket, 2509 * ie _not_ a timewait or request socket. 2510 */ 2511 static inline bool sk_fullsock(const struct sock *sk) 2512 { 2513 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2514 } 2515 2516 /* Checks if this SKB belongs to an HW offloaded socket 2517 * and whether any SW fallbacks are required based on dev. 2518 * Check decrypted mark in case skb_orphan() cleared socket. 2519 */ 2520 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2521 struct net_device *dev) 2522 { 2523 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2524 struct sock *sk = skb->sk; 2525 2526 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2527 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2528 #ifdef CONFIG_TLS_DEVICE 2529 } else if (unlikely(skb->decrypted)) { 2530 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2531 kfree_skb(skb); 2532 skb = NULL; 2533 #endif 2534 } 2535 #endif 2536 2537 return skb; 2538 } 2539 2540 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2541 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2542 */ 2543 static inline bool sk_listener(const struct sock *sk) 2544 { 2545 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2546 } 2547 2548 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2549 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2550 int type); 2551 2552 bool sk_ns_capable(const struct sock *sk, 2553 struct user_namespace *user_ns, int cap); 2554 bool sk_capable(const struct sock *sk, int cap); 2555 bool sk_net_capable(const struct sock *sk, int cap); 2556 2557 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2558 2559 /* Take into consideration the size of the struct sk_buff overhead in the 2560 * determination of these values, since that is non-constant across 2561 * platforms. This makes socket queueing behavior and performance 2562 * not depend upon such differences. 2563 */ 2564 #define _SK_MEM_PACKETS 256 2565 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2566 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2567 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2568 2569 extern __u32 sysctl_wmem_max; 2570 extern __u32 sysctl_rmem_max; 2571 2572 extern int sysctl_tstamp_allow_data; 2573 extern int sysctl_optmem_max; 2574 2575 extern __u32 sysctl_wmem_default; 2576 extern __u32 sysctl_rmem_default; 2577 2578 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2579 2580 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2581 { 2582 /* Does this proto have per netns sysctl_wmem ? */ 2583 if (proto->sysctl_wmem_offset) 2584 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2585 2586 return *proto->sysctl_wmem; 2587 } 2588 2589 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2590 { 2591 /* Does this proto have per netns sysctl_rmem ? */ 2592 if (proto->sysctl_rmem_offset) 2593 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2594 2595 return *proto->sysctl_rmem; 2596 } 2597 2598 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2599 * Some wifi drivers need to tweak it to get more chunks. 2600 * They can use this helper from their ndo_start_xmit() 2601 */ 2602 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2603 { 2604 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2605 return; 2606 WRITE_ONCE(sk->sk_pacing_shift, val); 2607 } 2608 2609 /* if a socket is bound to a device, check that the given device 2610 * index is either the same or that the socket is bound to an L3 2611 * master device and the given device index is also enslaved to 2612 * that L3 master 2613 */ 2614 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2615 { 2616 int mdif; 2617 2618 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2619 return true; 2620 2621 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2622 if (mdif && mdif == sk->sk_bound_dev_if) 2623 return true; 2624 2625 return false; 2626 } 2627 2628 void sock_def_readable(struct sock *sk); 2629 2630 #endif /* _SOCK_H */ 2631