1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * Definitions for the AF_INET socket handler. 7 * 8 * Version: @(#)sock.h 1.0.4 05/13/93 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Corey Minyard <wf-rch!minyard@relay.EU.net> 13 * Florian La Roche <flla@stud.uni-sb.de> 14 * 15 * Fixes: 16 * Alan Cox : Volatiles in skbuff pointers. See 17 * skbuff comments. May be overdone, 18 * better to prove they can be removed 19 * than the reverse. 20 * Alan Cox : Added a zapped field for tcp to note 21 * a socket is reset and must stay shut up 22 * Alan Cox : New fields for options 23 * Pauline Middelink : identd support 24 * Alan Cox : Eliminate low level recv/recvfrom 25 * David S. Miller : New socket lookup architecture. 26 * Steve Whitehouse: Default routines for sock_ops 27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 28 * protinfo be just a void pointer, as the 29 * protocol specific parts were moved to 30 * respective headers and ipv4/v6, etc now 31 * use private slabcaches for its socks 32 * Pedro Hortas : New flags field for socket options 33 * 34 * 35 * This program is free software; you can redistribute it and/or 36 * modify it under the terms of the GNU General Public License 37 * as published by the Free Software Foundation; either version 38 * 2 of the License, or (at your option) any later version. 39 */ 40 #ifndef _SOCK_H 41 #define _SOCK_H 42 43 #include <linux/hardirq.h> 44 #include <linux/kernel.h> 45 #include <linux/list.h> 46 #include <linux/list_nulls.h> 47 #include <linux/timer.h> 48 #include <linux/cache.h> 49 #include <linux/bitops.h> 50 #include <linux/lockdep.h> 51 #include <linux/netdevice.h> 52 #include <linux/skbuff.h> /* struct sk_buff */ 53 #include <linux/mm.h> 54 #include <linux/security.h> 55 #include <linux/slab.h> 56 #include <linux/uaccess.h> 57 #include <linux/page_counter.h> 58 #include <linux/memcontrol.h> 59 #include <linux/static_key.h> 60 #include <linux/sched.h> 61 #include <linux/wait.h> 62 #include <linux/cgroup-defs.h> 63 #include <linux/rbtree.h> 64 #include <linux/filter.h> 65 #include <linux/rculist_nulls.h> 66 #include <linux/poll.h> 67 68 #include <linux/atomic.h> 69 #include <linux/refcount.h> 70 #include <net/dst.h> 71 #include <net/checksum.h> 72 #include <net/tcp_states.h> 73 #include <linux/net_tstamp.h> 74 #include <net/smc.h> 75 #include <net/l3mdev.h> 76 77 /* 78 * This structure really needs to be cleaned up. 79 * Most of it is for TCP, and not used by any of 80 * the other protocols. 81 */ 82 83 /* Define this to get the SOCK_DBG debugging facility. */ 84 #define SOCK_DEBUGGING 85 #ifdef SOCK_DEBUGGING 86 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 87 printk(KERN_DEBUG msg); } while (0) 88 #else 89 /* Validate arguments and do nothing */ 90 static inline __printf(2, 3) 91 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 92 { 93 } 94 #endif 95 96 /* This is the per-socket lock. The spinlock provides a synchronization 97 * between user contexts and software interrupt processing, whereas the 98 * mini-semaphore synchronizes multiple users amongst themselves. 99 */ 100 typedef struct { 101 spinlock_t slock; 102 int owned; 103 wait_queue_head_t wq; 104 /* 105 * We express the mutex-alike socket_lock semantics 106 * to the lock validator by explicitly managing 107 * the slock as a lock variant (in addition to 108 * the slock itself): 109 */ 110 #ifdef CONFIG_DEBUG_LOCK_ALLOC 111 struct lockdep_map dep_map; 112 #endif 113 } socket_lock_t; 114 115 struct sock; 116 struct proto; 117 struct net; 118 119 typedef __u32 __bitwise __portpair; 120 typedef __u64 __bitwise __addrpair; 121 122 /** 123 * struct sock_common - minimal network layer representation of sockets 124 * @skc_daddr: Foreign IPv4 addr 125 * @skc_rcv_saddr: Bound local IPv4 addr 126 * @skc_hash: hash value used with various protocol lookup tables 127 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 128 * @skc_dport: placeholder for inet_dport/tw_dport 129 * @skc_num: placeholder for inet_num/tw_num 130 * @skc_family: network address family 131 * @skc_state: Connection state 132 * @skc_reuse: %SO_REUSEADDR setting 133 * @skc_reuseport: %SO_REUSEPORT setting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_node: main hash linkage for various protocol lookup tables 140 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 141 * @skc_tx_queue_mapping: tx queue number for this connection 142 * @skc_rx_queue_mapping: rx queue number for this connection 143 * @skc_flags: place holder for sk_flags 144 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 145 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 146 * @skc_incoming_cpu: record/match cpu processing incoming packets 147 * @skc_refcnt: reference count 148 * 149 * This is the minimal network layer representation of sockets, the header 150 * for struct sock and struct inet_timewait_sock. 151 */ 152 struct sock_common { 153 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 154 * address on 64bit arches : cf INET_MATCH() 155 */ 156 union { 157 __addrpair skc_addrpair; 158 struct { 159 __be32 skc_daddr; 160 __be32 skc_rcv_saddr; 161 }; 162 }; 163 union { 164 unsigned int skc_hash; 165 __u16 skc_u16hashes[2]; 166 }; 167 /* skc_dport && skc_num must be grouped as well */ 168 union { 169 __portpair skc_portpair; 170 struct { 171 __be16 skc_dport; 172 __u16 skc_num; 173 }; 174 }; 175 176 unsigned short skc_family; 177 volatile unsigned char skc_state; 178 unsigned char skc_reuse:4; 179 unsigned char skc_reuseport:1; 180 unsigned char skc_ipv6only:1; 181 unsigned char skc_net_refcnt:1; 182 int skc_bound_dev_if; 183 union { 184 struct hlist_node skc_bind_node; 185 struct hlist_node skc_portaddr_node; 186 }; 187 struct proto *skc_prot; 188 possible_net_t skc_net; 189 190 #if IS_ENABLED(CONFIG_IPV6) 191 struct in6_addr skc_v6_daddr; 192 struct in6_addr skc_v6_rcv_saddr; 193 #endif 194 195 atomic64_t skc_cookie; 196 197 /* following fields are padding to force 198 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 199 * assuming IPV6 is enabled. We use this padding differently 200 * for different kind of 'sockets' 201 */ 202 union { 203 unsigned long skc_flags; 204 struct sock *skc_listener; /* request_sock */ 205 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 206 }; 207 /* 208 * fields between dontcopy_begin/dontcopy_end 209 * are not copied in sock_copy() 210 */ 211 /* private: */ 212 int skc_dontcopy_begin[0]; 213 /* public: */ 214 union { 215 struct hlist_node skc_node; 216 struct hlist_nulls_node skc_nulls_node; 217 }; 218 unsigned short skc_tx_queue_mapping; 219 #ifdef CONFIG_XPS 220 unsigned short skc_rx_queue_mapping; 221 #endif 222 union { 223 int skc_incoming_cpu; 224 u32 skc_rcv_wnd; 225 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 226 }; 227 228 refcount_t skc_refcnt; 229 /* private: */ 230 int skc_dontcopy_end[0]; 231 union { 232 u32 skc_rxhash; 233 u32 skc_window_clamp; 234 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 235 }; 236 /* public: */ 237 }; 238 239 /** 240 * struct sock - network layer representation of sockets 241 * @__sk_common: shared layout with inet_timewait_sock 242 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 243 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 244 * @sk_lock: synchronizer 245 * @sk_kern_sock: True if sock is using kernel lock classes 246 * @sk_rcvbuf: size of receive buffer in bytes 247 * @sk_wq: sock wait queue and async head 248 * @sk_rx_dst: receive input route used by early demux 249 * @sk_dst_cache: destination cache 250 * @sk_dst_pending_confirm: need to confirm neighbour 251 * @sk_policy: flow policy 252 * @sk_receive_queue: incoming packets 253 * @sk_wmem_alloc: transmit queue bytes committed 254 * @sk_tsq_flags: TCP Small Queues flags 255 * @sk_write_queue: Packet sending queue 256 * @sk_omem_alloc: "o" is "option" or "other" 257 * @sk_wmem_queued: persistent queue size 258 * @sk_forward_alloc: space allocated forward 259 * @sk_napi_id: id of the last napi context to receive data for sk 260 * @sk_ll_usec: usecs to busypoll when there is no data 261 * @sk_allocation: allocation mode 262 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 263 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 264 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 265 * @sk_sndbuf: size of send buffer in bytes 266 * @__sk_flags_offset: empty field used to determine location of bitfield 267 * @sk_padding: unused element for alignment 268 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 269 * @sk_no_check_rx: allow zero checksum in RX packets 270 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 271 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 272 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 273 * @sk_gso_max_size: Maximum GSO segment size to build 274 * @sk_gso_max_segs: Maximum number of GSO segments 275 * @sk_pacing_shift: scaling factor for TCP Small Queues 276 * @sk_lingertime: %SO_LINGER l_linger setting 277 * @sk_backlog: always used with the per-socket spinlock held 278 * @sk_callback_lock: used with the callbacks in the end of this struct 279 * @sk_error_queue: rarely used 280 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 281 * IPV6_ADDRFORM for instance) 282 * @sk_err: last error 283 * @sk_err_soft: errors that don't cause failure but are the cause of a 284 * persistent failure not just 'timed out' 285 * @sk_drops: raw/udp drops counter 286 * @sk_ack_backlog: current listen backlog 287 * @sk_max_ack_backlog: listen backlog set in listen() 288 * @sk_uid: user id of owner 289 * @sk_priority: %SO_PRIORITY setting 290 * @sk_type: socket type (%SOCK_STREAM, etc) 291 * @sk_protocol: which protocol this socket belongs in this network family 292 * @sk_peer_pid: &struct pid for this socket's peer 293 * @sk_peer_cred: %SO_PEERCRED setting 294 * @sk_rcvlowat: %SO_RCVLOWAT setting 295 * @sk_rcvtimeo: %SO_RCVTIMEO setting 296 * @sk_sndtimeo: %SO_SNDTIMEO setting 297 * @sk_txhash: computed flow hash for use on transmit 298 * @sk_filter: socket filtering instructions 299 * @sk_timer: sock cleanup timer 300 * @sk_stamp: time stamp of last packet received 301 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 302 * @sk_tsflags: SO_TIMESTAMPING socket options 303 * @sk_tskey: counter to disambiguate concurrent tstamp requests 304 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 305 * @sk_socket: Identd and reporting IO signals 306 * @sk_user_data: RPC layer private data 307 * @sk_frag: cached page frag 308 * @sk_peek_off: current peek_offset value 309 * @sk_send_head: front of stuff to transmit 310 * @sk_security: used by security modules 311 * @sk_mark: generic packet mark 312 * @sk_cgrp_data: cgroup data for this cgroup 313 * @sk_memcg: this socket's memory cgroup association 314 * @sk_write_pending: a write to stream socket waits to start 315 * @sk_state_change: callback to indicate change in the state of the sock 316 * @sk_data_ready: callback to indicate there is data to be processed 317 * @sk_write_space: callback to indicate there is bf sending space available 318 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 319 * @sk_backlog_rcv: callback to process the backlog 320 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 321 * @sk_reuseport_cb: reuseport group container 322 * @sk_rcu: used during RCU grace period 323 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 324 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 325 * @sk_txtime_unused: unused txtime flags 326 */ 327 struct sock { 328 /* 329 * Now struct inet_timewait_sock also uses sock_common, so please just 330 * don't add nothing before this first member (__sk_common) --acme 331 */ 332 struct sock_common __sk_common; 333 #define sk_node __sk_common.skc_node 334 #define sk_nulls_node __sk_common.skc_nulls_node 335 #define sk_refcnt __sk_common.skc_refcnt 336 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 337 #ifdef CONFIG_XPS 338 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 339 #endif 340 341 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 342 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 343 #define sk_hash __sk_common.skc_hash 344 #define sk_portpair __sk_common.skc_portpair 345 #define sk_num __sk_common.skc_num 346 #define sk_dport __sk_common.skc_dport 347 #define sk_addrpair __sk_common.skc_addrpair 348 #define sk_daddr __sk_common.skc_daddr 349 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 350 #define sk_family __sk_common.skc_family 351 #define sk_state __sk_common.skc_state 352 #define sk_reuse __sk_common.skc_reuse 353 #define sk_reuseport __sk_common.skc_reuseport 354 #define sk_ipv6only __sk_common.skc_ipv6only 355 #define sk_net_refcnt __sk_common.skc_net_refcnt 356 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 357 #define sk_bind_node __sk_common.skc_bind_node 358 #define sk_prot __sk_common.skc_prot 359 #define sk_net __sk_common.skc_net 360 #define sk_v6_daddr __sk_common.skc_v6_daddr 361 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 362 #define sk_cookie __sk_common.skc_cookie 363 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 364 #define sk_flags __sk_common.skc_flags 365 #define sk_rxhash __sk_common.skc_rxhash 366 367 socket_lock_t sk_lock; 368 atomic_t sk_drops; 369 int sk_rcvlowat; 370 struct sk_buff_head sk_error_queue; 371 struct sk_buff_head sk_receive_queue; 372 /* 373 * The backlog queue is special, it is always used with 374 * the per-socket spinlock held and requires low latency 375 * access. Therefore we special case it's implementation. 376 * Note : rmem_alloc is in this structure to fill a hole 377 * on 64bit arches, not because its logically part of 378 * backlog. 379 */ 380 struct { 381 atomic_t rmem_alloc; 382 int len; 383 struct sk_buff *head; 384 struct sk_buff *tail; 385 } sk_backlog; 386 #define sk_rmem_alloc sk_backlog.rmem_alloc 387 388 int sk_forward_alloc; 389 #ifdef CONFIG_NET_RX_BUSY_POLL 390 unsigned int sk_ll_usec; 391 /* ===== mostly read cache line ===== */ 392 unsigned int sk_napi_id; 393 #endif 394 int sk_rcvbuf; 395 396 struct sk_filter __rcu *sk_filter; 397 union { 398 struct socket_wq __rcu *sk_wq; 399 struct socket_wq *sk_wq_raw; 400 }; 401 #ifdef CONFIG_XFRM 402 struct xfrm_policy __rcu *sk_policy[2]; 403 #endif 404 struct dst_entry *sk_rx_dst; 405 struct dst_entry __rcu *sk_dst_cache; 406 atomic_t sk_omem_alloc; 407 int sk_sndbuf; 408 409 /* ===== cache line for TX ===== */ 410 int sk_wmem_queued; 411 refcount_t sk_wmem_alloc; 412 unsigned long sk_tsq_flags; 413 union { 414 struct sk_buff *sk_send_head; 415 struct rb_root tcp_rtx_queue; 416 }; 417 struct sk_buff_head sk_write_queue; 418 __s32 sk_peek_off; 419 int sk_write_pending; 420 __u32 sk_dst_pending_confirm; 421 u32 sk_pacing_status; /* see enum sk_pacing */ 422 long sk_sndtimeo; 423 struct timer_list sk_timer; 424 __u32 sk_priority; 425 __u32 sk_mark; 426 unsigned long sk_pacing_rate; /* bytes per second */ 427 unsigned long sk_max_pacing_rate; 428 struct page_frag sk_frag; 429 netdev_features_t sk_route_caps; 430 netdev_features_t sk_route_nocaps; 431 netdev_features_t sk_route_forced_caps; 432 int sk_gso_type; 433 unsigned int sk_gso_max_size; 434 gfp_t sk_allocation; 435 __u32 sk_txhash; 436 437 /* 438 * Because of non atomicity rules, all 439 * changes are protected by socket lock. 440 */ 441 unsigned int __sk_flags_offset[0]; 442 #ifdef __BIG_ENDIAN_BITFIELD 443 #define SK_FL_PROTO_SHIFT 16 444 #define SK_FL_PROTO_MASK 0x00ff0000 445 446 #define SK_FL_TYPE_SHIFT 0 447 #define SK_FL_TYPE_MASK 0x0000ffff 448 #else 449 #define SK_FL_PROTO_SHIFT 8 450 #define SK_FL_PROTO_MASK 0x0000ff00 451 452 #define SK_FL_TYPE_SHIFT 16 453 #define SK_FL_TYPE_MASK 0xffff0000 454 #endif 455 456 unsigned int sk_padding : 1, 457 sk_kern_sock : 1, 458 sk_no_check_tx : 1, 459 sk_no_check_rx : 1, 460 sk_userlocks : 4, 461 sk_protocol : 8, 462 sk_type : 16; 463 #define SK_PROTOCOL_MAX U8_MAX 464 u16 sk_gso_max_segs; 465 u8 sk_pacing_shift; 466 unsigned long sk_lingertime; 467 struct proto *sk_prot_creator; 468 rwlock_t sk_callback_lock; 469 int sk_err, 470 sk_err_soft; 471 u32 sk_ack_backlog; 472 u32 sk_max_ack_backlog; 473 kuid_t sk_uid; 474 struct pid *sk_peer_pid; 475 const struct cred *sk_peer_cred; 476 long sk_rcvtimeo; 477 ktime_t sk_stamp; 478 #if BITS_PER_LONG==32 479 seqlock_t sk_stamp_seq; 480 #endif 481 u16 sk_tsflags; 482 u8 sk_shutdown; 483 u32 sk_tskey; 484 atomic_t sk_zckey; 485 486 u8 sk_clockid; 487 u8 sk_txtime_deadline_mode : 1, 488 sk_txtime_report_errors : 1, 489 sk_txtime_unused : 6; 490 491 struct socket *sk_socket; 492 void *sk_user_data; 493 #ifdef CONFIG_SECURITY 494 void *sk_security; 495 #endif 496 struct sock_cgroup_data sk_cgrp_data; 497 struct mem_cgroup *sk_memcg; 498 void (*sk_state_change)(struct sock *sk); 499 void (*sk_data_ready)(struct sock *sk); 500 void (*sk_write_space)(struct sock *sk); 501 void (*sk_error_report)(struct sock *sk); 502 int (*sk_backlog_rcv)(struct sock *sk, 503 struct sk_buff *skb); 504 #ifdef CONFIG_SOCK_VALIDATE_XMIT 505 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 506 struct net_device *dev, 507 struct sk_buff *skb); 508 #endif 509 void (*sk_destruct)(struct sock *sk); 510 struct sock_reuseport __rcu *sk_reuseport_cb; 511 struct rcu_head sk_rcu; 512 }; 513 514 enum sk_pacing { 515 SK_PACING_NONE = 0, 516 SK_PACING_NEEDED = 1, 517 SK_PACING_FQ = 2, 518 }; 519 520 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 521 522 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 523 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 524 525 /* 526 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 527 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 528 * on a socket means that the socket will reuse everybody else's port 529 * without looking at the other's sk_reuse value. 530 */ 531 532 #define SK_NO_REUSE 0 533 #define SK_CAN_REUSE 1 534 #define SK_FORCE_REUSE 2 535 536 int sk_set_peek_off(struct sock *sk, int val); 537 538 static inline int sk_peek_offset(struct sock *sk, int flags) 539 { 540 if (unlikely(flags & MSG_PEEK)) { 541 return READ_ONCE(sk->sk_peek_off); 542 } 543 544 return 0; 545 } 546 547 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 548 { 549 s32 off = READ_ONCE(sk->sk_peek_off); 550 551 if (unlikely(off >= 0)) { 552 off = max_t(s32, off - val, 0); 553 WRITE_ONCE(sk->sk_peek_off, off); 554 } 555 } 556 557 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 558 { 559 sk_peek_offset_bwd(sk, -val); 560 } 561 562 /* 563 * Hashed lists helper routines 564 */ 565 static inline struct sock *sk_entry(const struct hlist_node *node) 566 { 567 return hlist_entry(node, struct sock, sk_node); 568 } 569 570 static inline struct sock *__sk_head(const struct hlist_head *head) 571 { 572 return hlist_entry(head->first, struct sock, sk_node); 573 } 574 575 static inline struct sock *sk_head(const struct hlist_head *head) 576 { 577 return hlist_empty(head) ? NULL : __sk_head(head); 578 } 579 580 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 581 { 582 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 583 } 584 585 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 586 { 587 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 588 } 589 590 static inline struct sock *sk_next(const struct sock *sk) 591 { 592 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 593 } 594 595 static inline struct sock *sk_nulls_next(const struct sock *sk) 596 { 597 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 598 hlist_nulls_entry(sk->sk_nulls_node.next, 599 struct sock, sk_nulls_node) : 600 NULL; 601 } 602 603 static inline bool sk_unhashed(const struct sock *sk) 604 { 605 return hlist_unhashed(&sk->sk_node); 606 } 607 608 static inline bool sk_hashed(const struct sock *sk) 609 { 610 return !sk_unhashed(sk); 611 } 612 613 static inline void sk_node_init(struct hlist_node *node) 614 { 615 node->pprev = NULL; 616 } 617 618 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 619 { 620 node->pprev = NULL; 621 } 622 623 static inline void __sk_del_node(struct sock *sk) 624 { 625 __hlist_del(&sk->sk_node); 626 } 627 628 /* NB: equivalent to hlist_del_init_rcu */ 629 static inline bool __sk_del_node_init(struct sock *sk) 630 { 631 if (sk_hashed(sk)) { 632 __sk_del_node(sk); 633 sk_node_init(&sk->sk_node); 634 return true; 635 } 636 return false; 637 } 638 639 /* Grab socket reference count. This operation is valid only 640 when sk is ALREADY grabbed f.e. it is found in hash table 641 or a list and the lookup is made under lock preventing hash table 642 modifications. 643 */ 644 645 static __always_inline void sock_hold(struct sock *sk) 646 { 647 refcount_inc(&sk->sk_refcnt); 648 } 649 650 /* Ungrab socket in the context, which assumes that socket refcnt 651 cannot hit zero, f.e. it is true in context of any socketcall. 652 */ 653 static __always_inline void __sock_put(struct sock *sk) 654 { 655 refcount_dec(&sk->sk_refcnt); 656 } 657 658 static inline bool sk_del_node_init(struct sock *sk) 659 { 660 bool rc = __sk_del_node_init(sk); 661 662 if (rc) { 663 /* paranoid for a while -acme */ 664 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 665 __sock_put(sk); 666 } 667 return rc; 668 } 669 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 670 671 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 672 { 673 if (sk_hashed(sk)) { 674 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 675 return true; 676 } 677 return false; 678 } 679 680 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 681 { 682 bool rc = __sk_nulls_del_node_init_rcu(sk); 683 684 if (rc) { 685 /* paranoid for a while -acme */ 686 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 687 __sock_put(sk); 688 } 689 return rc; 690 } 691 692 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 693 { 694 hlist_add_head(&sk->sk_node, list); 695 } 696 697 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 698 { 699 sock_hold(sk); 700 __sk_add_node(sk, list); 701 } 702 703 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 704 { 705 sock_hold(sk); 706 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 707 sk->sk_family == AF_INET6) 708 hlist_add_tail_rcu(&sk->sk_node, list); 709 else 710 hlist_add_head_rcu(&sk->sk_node, list); 711 } 712 713 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 714 { 715 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 716 } 717 718 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 719 { 720 sock_hold(sk); 721 __sk_nulls_add_node_rcu(sk, list); 722 } 723 724 static inline void __sk_del_bind_node(struct sock *sk) 725 { 726 __hlist_del(&sk->sk_bind_node); 727 } 728 729 static inline void sk_add_bind_node(struct sock *sk, 730 struct hlist_head *list) 731 { 732 hlist_add_head(&sk->sk_bind_node, list); 733 } 734 735 #define sk_for_each(__sk, list) \ 736 hlist_for_each_entry(__sk, list, sk_node) 737 #define sk_for_each_rcu(__sk, list) \ 738 hlist_for_each_entry_rcu(__sk, list, sk_node) 739 #define sk_nulls_for_each(__sk, node, list) \ 740 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 741 #define sk_nulls_for_each_rcu(__sk, node, list) \ 742 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 743 #define sk_for_each_from(__sk) \ 744 hlist_for_each_entry_from(__sk, sk_node) 745 #define sk_nulls_for_each_from(__sk, node) \ 746 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 747 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 748 #define sk_for_each_safe(__sk, tmp, list) \ 749 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 750 #define sk_for_each_bound(__sk, list) \ 751 hlist_for_each_entry(__sk, list, sk_bind_node) 752 753 /** 754 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 755 * @tpos: the type * to use as a loop cursor. 756 * @pos: the &struct hlist_node to use as a loop cursor. 757 * @head: the head for your list. 758 * @offset: offset of hlist_node within the struct. 759 * 760 */ 761 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 762 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 763 pos != NULL && \ 764 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 765 pos = rcu_dereference(hlist_next_rcu(pos))) 766 767 static inline struct user_namespace *sk_user_ns(struct sock *sk) 768 { 769 /* Careful only use this in a context where these parameters 770 * can not change and must all be valid, such as recvmsg from 771 * userspace. 772 */ 773 return sk->sk_socket->file->f_cred->user_ns; 774 } 775 776 /* Sock flags */ 777 enum sock_flags { 778 SOCK_DEAD, 779 SOCK_DONE, 780 SOCK_URGINLINE, 781 SOCK_KEEPOPEN, 782 SOCK_LINGER, 783 SOCK_DESTROY, 784 SOCK_BROADCAST, 785 SOCK_TIMESTAMP, 786 SOCK_ZAPPED, 787 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 788 SOCK_DBG, /* %SO_DEBUG setting */ 789 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 790 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 791 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 792 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 793 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 794 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 795 SOCK_FASYNC, /* fasync() active */ 796 SOCK_RXQ_OVFL, 797 SOCK_ZEROCOPY, /* buffers from userspace */ 798 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 799 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 800 * Will use last 4 bytes of packet sent from 801 * user-space instead. 802 */ 803 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 804 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 805 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 806 SOCK_TXTIME, 807 SOCK_XDP, /* XDP is attached */ 808 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 809 }; 810 811 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 812 813 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 814 { 815 nsk->sk_flags = osk->sk_flags; 816 } 817 818 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 819 { 820 __set_bit(flag, &sk->sk_flags); 821 } 822 823 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 824 { 825 __clear_bit(flag, &sk->sk_flags); 826 } 827 828 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 829 { 830 return test_bit(flag, &sk->sk_flags); 831 } 832 833 #ifdef CONFIG_NET 834 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 835 static inline int sk_memalloc_socks(void) 836 { 837 return static_branch_unlikely(&memalloc_socks_key); 838 } 839 #else 840 841 static inline int sk_memalloc_socks(void) 842 { 843 return 0; 844 } 845 846 #endif 847 848 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 849 { 850 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 851 } 852 853 static inline void sk_acceptq_removed(struct sock *sk) 854 { 855 sk->sk_ack_backlog--; 856 } 857 858 static inline void sk_acceptq_added(struct sock *sk) 859 { 860 sk->sk_ack_backlog++; 861 } 862 863 static inline bool sk_acceptq_is_full(const struct sock *sk) 864 { 865 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 866 } 867 868 /* 869 * Compute minimal free write space needed to queue new packets. 870 */ 871 static inline int sk_stream_min_wspace(const struct sock *sk) 872 { 873 return sk->sk_wmem_queued >> 1; 874 } 875 876 static inline int sk_stream_wspace(const struct sock *sk) 877 { 878 return sk->sk_sndbuf - sk->sk_wmem_queued; 879 } 880 881 void sk_stream_write_space(struct sock *sk); 882 883 /* OOB backlog add */ 884 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 885 { 886 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 887 skb_dst_force(skb); 888 889 if (!sk->sk_backlog.tail) 890 sk->sk_backlog.head = skb; 891 else 892 sk->sk_backlog.tail->next = skb; 893 894 sk->sk_backlog.tail = skb; 895 skb->next = NULL; 896 } 897 898 /* 899 * Take into account size of receive queue and backlog queue 900 * Do not take into account this skb truesize, 901 * to allow even a single big packet to come. 902 */ 903 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 904 { 905 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 906 907 return qsize > limit; 908 } 909 910 /* The per-socket spinlock must be held here. */ 911 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 912 unsigned int limit) 913 { 914 if (sk_rcvqueues_full(sk, limit)) 915 return -ENOBUFS; 916 917 /* 918 * If the skb was allocated from pfmemalloc reserves, only 919 * allow SOCK_MEMALLOC sockets to use it as this socket is 920 * helping free memory 921 */ 922 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 923 return -ENOMEM; 924 925 __sk_add_backlog(sk, skb); 926 sk->sk_backlog.len += skb->truesize; 927 return 0; 928 } 929 930 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 931 932 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 933 { 934 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 935 return __sk_backlog_rcv(sk, skb); 936 937 return sk->sk_backlog_rcv(sk, skb); 938 } 939 940 static inline void sk_incoming_cpu_update(struct sock *sk) 941 { 942 int cpu = raw_smp_processor_id(); 943 944 if (unlikely(sk->sk_incoming_cpu != cpu)) 945 sk->sk_incoming_cpu = cpu; 946 } 947 948 static inline void sock_rps_record_flow_hash(__u32 hash) 949 { 950 #ifdef CONFIG_RPS 951 struct rps_sock_flow_table *sock_flow_table; 952 953 rcu_read_lock(); 954 sock_flow_table = rcu_dereference(rps_sock_flow_table); 955 rps_record_sock_flow(sock_flow_table, hash); 956 rcu_read_unlock(); 957 #endif 958 } 959 960 static inline void sock_rps_record_flow(const struct sock *sk) 961 { 962 #ifdef CONFIG_RPS 963 if (static_key_false(&rfs_needed)) { 964 /* Reading sk->sk_rxhash might incur an expensive cache line 965 * miss. 966 * 967 * TCP_ESTABLISHED does cover almost all states where RFS 968 * might be useful, and is cheaper [1] than testing : 969 * IPv4: inet_sk(sk)->inet_daddr 970 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 971 * OR an additional socket flag 972 * [1] : sk_state and sk_prot are in the same cache line. 973 */ 974 if (sk->sk_state == TCP_ESTABLISHED) 975 sock_rps_record_flow_hash(sk->sk_rxhash); 976 } 977 #endif 978 } 979 980 static inline void sock_rps_save_rxhash(struct sock *sk, 981 const struct sk_buff *skb) 982 { 983 #ifdef CONFIG_RPS 984 if (unlikely(sk->sk_rxhash != skb->hash)) 985 sk->sk_rxhash = skb->hash; 986 #endif 987 } 988 989 static inline void sock_rps_reset_rxhash(struct sock *sk) 990 { 991 #ifdef CONFIG_RPS 992 sk->sk_rxhash = 0; 993 #endif 994 } 995 996 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 997 ({ int __rc; \ 998 release_sock(__sk); \ 999 __rc = __condition; \ 1000 if (!__rc) { \ 1001 *(__timeo) = wait_woken(__wait, \ 1002 TASK_INTERRUPTIBLE, \ 1003 *(__timeo)); \ 1004 } \ 1005 sched_annotate_sleep(); \ 1006 lock_sock(__sk); \ 1007 __rc = __condition; \ 1008 __rc; \ 1009 }) 1010 1011 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1012 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1013 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1014 int sk_stream_error(struct sock *sk, int flags, int err); 1015 void sk_stream_kill_queues(struct sock *sk); 1016 void sk_set_memalloc(struct sock *sk); 1017 void sk_clear_memalloc(struct sock *sk); 1018 1019 void __sk_flush_backlog(struct sock *sk); 1020 1021 static inline bool sk_flush_backlog(struct sock *sk) 1022 { 1023 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1024 __sk_flush_backlog(sk); 1025 return true; 1026 } 1027 return false; 1028 } 1029 1030 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1031 1032 struct request_sock_ops; 1033 struct timewait_sock_ops; 1034 struct inet_hashinfo; 1035 struct raw_hashinfo; 1036 struct smc_hashinfo; 1037 struct module; 1038 1039 /* 1040 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1041 * un-modified. Special care is taken when initializing object to zero. 1042 */ 1043 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1044 { 1045 if (offsetof(struct sock, sk_node.next) != 0) 1046 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1047 memset(&sk->sk_node.pprev, 0, 1048 size - offsetof(struct sock, sk_node.pprev)); 1049 } 1050 1051 /* Networking protocol blocks we attach to sockets. 1052 * socket layer -> transport layer interface 1053 */ 1054 struct proto { 1055 void (*close)(struct sock *sk, 1056 long timeout); 1057 int (*pre_connect)(struct sock *sk, 1058 struct sockaddr *uaddr, 1059 int addr_len); 1060 int (*connect)(struct sock *sk, 1061 struct sockaddr *uaddr, 1062 int addr_len); 1063 int (*disconnect)(struct sock *sk, int flags); 1064 1065 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1066 bool kern); 1067 1068 int (*ioctl)(struct sock *sk, int cmd, 1069 unsigned long arg); 1070 int (*init)(struct sock *sk); 1071 void (*destroy)(struct sock *sk); 1072 void (*shutdown)(struct sock *sk, int how); 1073 int (*setsockopt)(struct sock *sk, int level, 1074 int optname, char __user *optval, 1075 unsigned int optlen); 1076 int (*getsockopt)(struct sock *sk, int level, 1077 int optname, char __user *optval, 1078 int __user *option); 1079 void (*keepalive)(struct sock *sk, int valbool); 1080 #ifdef CONFIG_COMPAT 1081 int (*compat_setsockopt)(struct sock *sk, 1082 int level, 1083 int optname, char __user *optval, 1084 unsigned int optlen); 1085 int (*compat_getsockopt)(struct sock *sk, 1086 int level, 1087 int optname, char __user *optval, 1088 int __user *option); 1089 int (*compat_ioctl)(struct sock *sk, 1090 unsigned int cmd, unsigned long arg); 1091 #endif 1092 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1093 size_t len); 1094 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1095 size_t len, int noblock, int flags, 1096 int *addr_len); 1097 int (*sendpage)(struct sock *sk, struct page *page, 1098 int offset, size_t size, int flags); 1099 int (*bind)(struct sock *sk, 1100 struct sockaddr *uaddr, int addr_len); 1101 1102 int (*backlog_rcv) (struct sock *sk, 1103 struct sk_buff *skb); 1104 1105 void (*release_cb)(struct sock *sk); 1106 1107 /* Keeping track of sk's, looking them up, and port selection methods. */ 1108 int (*hash)(struct sock *sk); 1109 void (*unhash)(struct sock *sk); 1110 void (*rehash)(struct sock *sk); 1111 int (*get_port)(struct sock *sk, unsigned short snum); 1112 1113 /* Keeping track of sockets in use */ 1114 #ifdef CONFIG_PROC_FS 1115 unsigned int inuse_idx; 1116 #endif 1117 1118 bool (*stream_memory_free)(const struct sock *sk, int wake); 1119 bool (*stream_memory_read)(const struct sock *sk); 1120 /* Memory pressure */ 1121 void (*enter_memory_pressure)(struct sock *sk); 1122 void (*leave_memory_pressure)(struct sock *sk); 1123 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1124 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1125 /* 1126 * Pressure flag: try to collapse. 1127 * Technical note: it is used by multiple contexts non atomically. 1128 * All the __sk_mem_schedule() is of this nature: accounting 1129 * is strict, actions are advisory and have some latency. 1130 */ 1131 unsigned long *memory_pressure; 1132 long *sysctl_mem; 1133 1134 int *sysctl_wmem; 1135 int *sysctl_rmem; 1136 u32 sysctl_wmem_offset; 1137 u32 sysctl_rmem_offset; 1138 1139 int max_header; 1140 bool no_autobind; 1141 1142 struct kmem_cache *slab; 1143 unsigned int obj_size; 1144 slab_flags_t slab_flags; 1145 unsigned int useroffset; /* Usercopy region offset */ 1146 unsigned int usersize; /* Usercopy region size */ 1147 1148 struct percpu_counter *orphan_count; 1149 1150 struct request_sock_ops *rsk_prot; 1151 struct timewait_sock_ops *twsk_prot; 1152 1153 union { 1154 struct inet_hashinfo *hashinfo; 1155 struct udp_table *udp_table; 1156 struct raw_hashinfo *raw_hash; 1157 struct smc_hashinfo *smc_hash; 1158 } h; 1159 1160 struct module *owner; 1161 1162 char name[32]; 1163 1164 struct list_head node; 1165 #ifdef SOCK_REFCNT_DEBUG 1166 atomic_t socks; 1167 #endif 1168 int (*diag_destroy)(struct sock *sk, int err); 1169 } __randomize_layout; 1170 1171 int proto_register(struct proto *prot, int alloc_slab); 1172 void proto_unregister(struct proto *prot); 1173 int sock_load_diag_module(int family, int protocol); 1174 1175 #ifdef SOCK_REFCNT_DEBUG 1176 static inline void sk_refcnt_debug_inc(struct sock *sk) 1177 { 1178 atomic_inc(&sk->sk_prot->socks); 1179 } 1180 1181 static inline void sk_refcnt_debug_dec(struct sock *sk) 1182 { 1183 atomic_dec(&sk->sk_prot->socks); 1184 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1185 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1186 } 1187 1188 static inline void sk_refcnt_debug_release(const struct sock *sk) 1189 { 1190 if (refcount_read(&sk->sk_refcnt) != 1) 1191 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1192 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1193 } 1194 #else /* SOCK_REFCNT_DEBUG */ 1195 #define sk_refcnt_debug_inc(sk) do { } while (0) 1196 #define sk_refcnt_debug_dec(sk) do { } while (0) 1197 #define sk_refcnt_debug_release(sk) do { } while (0) 1198 #endif /* SOCK_REFCNT_DEBUG */ 1199 1200 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1201 { 1202 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1203 return false; 1204 1205 return sk->sk_prot->stream_memory_free ? 1206 sk->sk_prot->stream_memory_free(sk, wake) : true; 1207 } 1208 1209 static inline bool sk_stream_memory_free(const struct sock *sk) 1210 { 1211 return __sk_stream_memory_free(sk, 0); 1212 } 1213 1214 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1215 { 1216 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1217 __sk_stream_memory_free(sk, wake); 1218 } 1219 1220 static inline bool sk_stream_is_writeable(const struct sock *sk) 1221 { 1222 return __sk_stream_is_writeable(sk, 0); 1223 } 1224 1225 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1226 struct cgroup *ancestor) 1227 { 1228 #ifdef CONFIG_SOCK_CGROUP_DATA 1229 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1230 ancestor); 1231 #else 1232 return -ENOTSUPP; 1233 #endif 1234 } 1235 1236 static inline bool sk_has_memory_pressure(const struct sock *sk) 1237 { 1238 return sk->sk_prot->memory_pressure != NULL; 1239 } 1240 1241 static inline bool sk_under_memory_pressure(const struct sock *sk) 1242 { 1243 if (!sk->sk_prot->memory_pressure) 1244 return false; 1245 1246 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1247 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1248 return true; 1249 1250 return !!*sk->sk_prot->memory_pressure; 1251 } 1252 1253 static inline long 1254 sk_memory_allocated(const struct sock *sk) 1255 { 1256 return atomic_long_read(sk->sk_prot->memory_allocated); 1257 } 1258 1259 static inline long 1260 sk_memory_allocated_add(struct sock *sk, int amt) 1261 { 1262 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1263 } 1264 1265 static inline void 1266 sk_memory_allocated_sub(struct sock *sk, int amt) 1267 { 1268 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1269 } 1270 1271 static inline void sk_sockets_allocated_dec(struct sock *sk) 1272 { 1273 percpu_counter_dec(sk->sk_prot->sockets_allocated); 1274 } 1275 1276 static inline void sk_sockets_allocated_inc(struct sock *sk) 1277 { 1278 percpu_counter_inc(sk->sk_prot->sockets_allocated); 1279 } 1280 1281 static inline u64 1282 sk_sockets_allocated_read_positive(struct sock *sk) 1283 { 1284 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1285 } 1286 1287 static inline int 1288 proto_sockets_allocated_sum_positive(struct proto *prot) 1289 { 1290 return percpu_counter_sum_positive(prot->sockets_allocated); 1291 } 1292 1293 static inline long 1294 proto_memory_allocated(struct proto *prot) 1295 { 1296 return atomic_long_read(prot->memory_allocated); 1297 } 1298 1299 static inline bool 1300 proto_memory_pressure(struct proto *prot) 1301 { 1302 if (!prot->memory_pressure) 1303 return false; 1304 return !!*prot->memory_pressure; 1305 } 1306 1307 1308 #ifdef CONFIG_PROC_FS 1309 /* Called with local bh disabled */ 1310 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1311 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1312 int sock_inuse_get(struct net *net); 1313 #else 1314 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1315 int inc) 1316 { 1317 } 1318 #endif 1319 1320 1321 /* With per-bucket locks this operation is not-atomic, so that 1322 * this version is not worse. 1323 */ 1324 static inline int __sk_prot_rehash(struct sock *sk) 1325 { 1326 sk->sk_prot->unhash(sk); 1327 return sk->sk_prot->hash(sk); 1328 } 1329 1330 /* About 10 seconds */ 1331 #define SOCK_DESTROY_TIME (10*HZ) 1332 1333 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1334 #define PROT_SOCK 1024 1335 1336 #define SHUTDOWN_MASK 3 1337 #define RCV_SHUTDOWN 1 1338 #define SEND_SHUTDOWN 2 1339 1340 #define SOCK_SNDBUF_LOCK 1 1341 #define SOCK_RCVBUF_LOCK 2 1342 #define SOCK_BINDADDR_LOCK 4 1343 #define SOCK_BINDPORT_LOCK 8 1344 1345 struct socket_alloc { 1346 struct socket socket; 1347 struct inode vfs_inode; 1348 }; 1349 1350 static inline struct socket *SOCKET_I(struct inode *inode) 1351 { 1352 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1353 } 1354 1355 static inline struct inode *SOCK_INODE(struct socket *socket) 1356 { 1357 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1358 } 1359 1360 /* 1361 * Functions for memory accounting 1362 */ 1363 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1364 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1365 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1366 void __sk_mem_reclaim(struct sock *sk, int amount); 1367 1368 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1369 * do not necessarily have 16x time more memory than 4KB ones. 1370 */ 1371 #define SK_MEM_QUANTUM 4096 1372 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1373 #define SK_MEM_SEND 0 1374 #define SK_MEM_RECV 1 1375 1376 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1377 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1378 { 1379 long val = sk->sk_prot->sysctl_mem[index]; 1380 1381 #if PAGE_SIZE > SK_MEM_QUANTUM 1382 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1383 #elif PAGE_SIZE < SK_MEM_QUANTUM 1384 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1385 #endif 1386 return val; 1387 } 1388 1389 static inline int sk_mem_pages(int amt) 1390 { 1391 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1392 } 1393 1394 static inline bool sk_has_account(struct sock *sk) 1395 { 1396 /* return true if protocol supports memory accounting */ 1397 return !!sk->sk_prot->memory_allocated; 1398 } 1399 1400 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1401 { 1402 if (!sk_has_account(sk)) 1403 return true; 1404 return size <= sk->sk_forward_alloc || 1405 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1406 } 1407 1408 static inline bool 1409 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1410 { 1411 if (!sk_has_account(sk)) 1412 return true; 1413 return size<= sk->sk_forward_alloc || 1414 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1415 skb_pfmemalloc(skb); 1416 } 1417 1418 static inline void sk_mem_reclaim(struct sock *sk) 1419 { 1420 if (!sk_has_account(sk)) 1421 return; 1422 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1423 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1424 } 1425 1426 static inline void sk_mem_reclaim_partial(struct sock *sk) 1427 { 1428 if (!sk_has_account(sk)) 1429 return; 1430 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1431 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1432 } 1433 1434 static inline void sk_mem_charge(struct sock *sk, int size) 1435 { 1436 if (!sk_has_account(sk)) 1437 return; 1438 sk->sk_forward_alloc -= size; 1439 } 1440 1441 static inline void sk_mem_uncharge(struct sock *sk, int size) 1442 { 1443 if (!sk_has_account(sk)) 1444 return; 1445 sk->sk_forward_alloc += size; 1446 1447 /* Avoid a possible overflow. 1448 * TCP send queues can make this happen, if sk_mem_reclaim() 1449 * is not called and more than 2 GBytes are released at once. 1450 * 1451 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1452 * no need to hold that much forward allocation anyway. 1453 */ 1454 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1455 __sk_mem_reclaim(sk, 1 << 20); 1456 } 1457 1458 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1459 { 1460 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1461 sk->sk_wmem_queued -= skb->truesize; 1462 sk_mem_uncharge(sk, skb->truesize); 1463 __kfree_skb(skb); 1464 } 1465 1466 static inline void sock_release_ownership(struct sock *sk) 1467 { 1468 if (sk->sk_lock.owned) { 1469 sk->sk_lock.owned = 0; 1470 1471 /* The sk_lock has mutex_unlock() semantics: */ 1472 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_); 1473 } 1474 } 1475 1476 /* 1477 * Macro so as to not evaluate some arguments when 1478 * lockdep is not enabled. 1479 * 1480 * Mark both the sk_lock and the sk_lock.slock as a 1481 * per-address-family lock class. 1482 */ 1483 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1484 do { \ 1485 sk->sk_lock.owned = 0; \ 1486 init_waitqueue_head(&sk->sk_lock.wq); \ 1487 spin_lock_init(&(sk)->sk_lock.slock); \ 1488 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1489 sizeof((sk)->sk_lock)); \ 1490 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1491 (skey), (sname)); \ 1492 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1493 } while (0) 1494 1495 #ifdef CONFIG_LOCKDEP 1496 static inline bool lockdep_sock_is_held(const struct sock *sk) 1497 { 1498 return lockdep_is_held(&sk->sk_lock) || 1499 lockdep_is_held(&sk->sk_lock.slock); 1500 } 1501 #endif 1502 1503 void lock_sock_nested(struct sock *sk, int subclass); 1504 1505 static inline void lock_sock(struct sock *sk) 1506 { 1507 lock_sock_nested(sk, 0); 1508 } 1509 1510 void __release_sock(struct sock *sk); 1511 void release_sock(struct sock *sk); 1512 1513 /* BH context may only use the following locking interface. */ 1514 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1515 #define bh_lock_sock_nested(__sk) \ 1516 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1517 SINGLE_DEPTH_NESTING) 1518 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1519 1520 bool lock_sock_fast(struct sock *sk); 1521 /** 1522 * unlock_sock_fast - complement of lock_sock_fast 1523 * @sk: socket 1524 * @slow: slow mode 1525 * 1526 * fast unlock socket for user context. 1527 * If slow mode is on, we call regular release_sock() 1528 */ 1529 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1530 { 1531 if (slow) 1532 release_sock(sk); 1533 else 1534 spin_unlock_bh(&sk->sk_lock.slock); 1535 } 1536 1537 /* Used by processes to "lock" a socket state, so that 1538 * interrupts and bottom half handlers won't change it 1539 * from under us. It essentially blocks any incoming 1540 * packets, so that we won't get any new data or any 1541 * packets that change the state of the socket. 1542 * 1543 * While locked, BH processing will add new packets to 1544 * the backlog queue. This queue is processed by the 1545 * owner of the socket lock right before it is released. 1546 * 1547 * Since ~2.3.5 it is also exclusive sleep lock serializing 1548 * accesses from user process context. 1549 */ 1550 1551 static inline void sock_owned_by_me(const struct sock *sk) 1552 { 1553 #ifdef CONFIG_LOCKDEP 1554 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1555 #endif 1556 } 1557 1558 static inline bool sock_owned_by_user(const struct sock *sk) 1559 { 1560 sock_owned_by_me(sk); 1561 return sk->sk_lock.owned; 1562 } 1563 1564 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1565 { 1566 return sk->sk_lock.owned; 1567 } 1568 1569 /* no reclassification while locks are held */ 1570 static inline bool sock_allow_reclassification(const struct sock *csk) 1571 { 1572 struct sock *sk = (struct sock *)csk; 1573 1574 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1575 } 1576 1577 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1578 struct proto *prot, int kern); 1579 void sk_free(struct sock *sk); 1580 void sk_destruct(struct sock *sk); 1581 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1582 void sk_free_unlock_clone(struct sock *sk); 1583 1584 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1585 gfp_t priority); 1586 void __sock_wfree(struct sk_buff *skb); 1587 void sock_wfree(struct sk_buff *skb); 1588 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1589 gfp_t priority); 1590 void skb_orphan_partial(struct sk_buff *skb); 1591 void sock_rfree(struct sk_buff *skb); 1592 void sock_efree(struct sk_buff *skb); 1593 #ifdef CONFIG_INET 1594 void sock_edemux(struct sk_buff *skb); 1595 #else 1596 #define sock_edemux sock_efree 1597 #endif 1598 1599 int sock_setsockopt(struct socket *sock, int level, int op, 1600 char __user *optval, unsigned int optlen); 1601 1602 int sock_getsockopt(struct socket *sock, int level, int op, 1603 char __user *optval, int __user *optlen); 1604 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1605 int noblock, int *errcode); 1606 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1607 unsigned long data_len, int noblock, 1608 int *errcode, int max_page_order); 1609 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1610 void sock_kfree_s(struct sock *sk, void *mem, int size); 1611 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1612 void sk_send_sigurg(struct sock *sk); 1613 1614 struct sockcm_cookie { 1615 u64 transmit_time; 1616 u32 mark; 1617 u16 tsflags; 1618 }; 1619 1620 static inline void sockcm_init(struct sockcm_cookie *sockc, 1621 const struct sock *sk) 1622 { 1623 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1624 } 1625 1626 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1627 struct sockcm_cookie *sockc); 1628 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1629 struct sockcm_cookie *sockc); 1630 1631 /* 1632 * Functions to fill in entries in struct proto_ops when a protocol 1633 * does not implement a particular function. 1634 */ 1635 int sock_no_bind(struct socket *, struct sockaddr *, int); 1636 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1637 int sock_no_socketpair(struct socket *, struct socket *); 1638 int sock_no_accept(struct socket *, struct socket *, int, bool); 1639 int sock_no_getname(struct socket *, struct sockaddr *, int); 1640 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1641 int sock_no_listen(struct socket *, int); 1642 int sock_no_shutdown(struct socket *, int); 1643 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1644 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1645 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1646 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1647 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1648 int sock_no_mmap(struct file *file, struct socket *sock, 1649 struct vm_area_struct *vma); 1650 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1651 size_t size, int flags); 1652 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1653 int offset, size_t size, int flags); 1654 1655 /* 1656 * Functions to fill in entries in struct proto_ops when a protocol 1657 * uses the inet style. 1658 */ 1659 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1660 char __user *optval, int __user *optlen); 1661 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1662 int flags); 1663 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1664 char __user *optval, unsigned int optlen); 1665 int compat_sock_common_getsockopt(struct socket *sock, int level, 1666 int optname, char __user *optval, int __user *optlen); 1667 int compat_sock_common_setsockopt(struct socket *sock, int level, 1668 int optname, char __user *optval, unsigned int optlen); 1669 1670 void sk_common_release(struct sock *sk); 1671 1672 /* 1673 * Default socket callbacks and setup code 1674 */ 1675 1676 /* Initialise core socket variables */ 1677 void sock_init_data(struct socket *sock, struct sock *sk); 1678 1679 /* 1680 * Socket reference counting postulates. 1681 * 1682 * * Each user of socket SHOULD hold a reference count. 1683 * * Each access point to socket (an hash table bucket, reference from a list, 1684 * running timer, skb in flight MUST hold a reference count. 1685 * * When reference count hits 0, it means it will never increase back. 1686 * * When reference count hits 0, it means that no references from 1687 * outside exist to this socket and current process on current CPU 1688 * is last user and may/should destroy this socket. 1689 * * sk_free is called from any context: process, BH, IRQ. When 1690 * it is called, socket has no references from outside -> sk_free 1691 * may release descendant resources allocated by the socket, but 1692 * to the time when it is called, socket is NOT referenced by any 1693 * hash tables, lists etc. 1694 * * Packets, delivered from outside (from network or from another process) 1695 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1696 * when they sit in queue. Otherwise, packets will leak to hole, when 1697 * socket is looked up by one cpu and unhasing is made by another CPU. 1698 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1699 * (leak to backlog). Packet socket does all the processing inside 1700 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1701 * use separate SMP lock, so that they are prone too. 1702 */ 1703 1704 /* Ungrab socket and destroy it, if it was the last reference. */ 1705 static inline void sock_put(struct sock *sk) 1706 { 1707 if (refcount_dec_and_test(&sk->sk_refcnt)) 1708 sk_free(sk); 1709 } 1710 /* Generic version of sock_put(), dealing with all sockets 1711 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1712 */ 1713 void sock_gen_put(struct sock *sk); 1714 1715 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1716 unsigned int trim_cap, bool refcounted); 1717 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1718 const int nested) 1719 { 1720 return __sk_receive_skb(sk, skb, nested, 1, true); 1721 } 1722 1723 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1724 { 1725 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1726 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1727 return; 1728 sk->sk_tx_queue_mapping = tx_queue; 1729 } 1730 1731 #define NO_QUEUE_MAPPING USHRT_MAX 1732 1733 static inline void sk_tx_queue_clear(struct sock *sk) 1734 { 1735 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1736 } 1737 1738 static inline int sk_tx_queue_get(const struct sock *sk) 1739 { 1740 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1741 return sk->sk_tx_queue_mapping; 1742 1743 return -1; 1744 } 1745 1746 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1747 { 1748 #ifdef CONFIG_XPS 1749 if (skb_rx_queue_recorded(skb)) { 1750 u16 rx_queue = skb_get_rx_queue(skb); 1751 1752 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING)) 1753 return; 1754 1755 sk->sk_rx_queue_mapping = rx_queue; 1756 } 1757 #endif 1758 } 1759 1760 static inline void sk_rx_queue_clear(struct sock *sk) 1761 { 1762 #ifdef CONFIG_XPS 1763 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING; 1764 #endif 1765 } 1766 1767 #ifdef CONFIG_XPS 1768 static inline int sk_rx_queue_get(const struct sock *sk) 1769 { 1770 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING) 1771 return sk->sk_rx_queue_mapping; 1772 1773 return -1; 1774 } 1775 #endif 1776 1777 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1778 { 1779 sk_tx_queue_clear(sk); 1780 sk->sk_socket = sock; 1781 } 1782 1783 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1784 { 1785 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1786 return &rcu_dereference_raw(sk->sk_wq)->wait; 1787 } 1788 /* Detach socket from process context. 1789 * Announce socket dead, detach it from wait queue and inode. 1790 * Note that parent inode held reference count on this struct sock, 1791 * we do not release it in this function, because protocol 1792 * probably wants some additional cleanups or even continuing 1793 * to work with this socket (TCP). 1794 */ 1795 static inline void sock_orphan(struct sock *sk) 1796 { 1797 write_lock_bh(&sk->sk_callback_lock); 1798 sock_set_flag(sk, SOCK_DEAD); 1799 sk_set_socket(sk, NULL); 1800 sk->sk_wq = NULL; 1801 write_unlock_bh(&sk->sk_callback_lock); 1802 } 1803 1804 static inline void sock_graft(struct sock *sk, struct socket *parent) 1805 { 1806 WARN_ON(parent->sk); 1807 write_lock_bh(&sk->sk_callback_lock); 1808 rcu_assign_pointer(sk->sk_wq, parent->wq); 1809 parent->sk = sk; 1810 sk_set_socket(sk, parent); 1811 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1812 security_sock_graft(sk, parent); 1813 write_unlock_bh(&sk->sk_callback_lock); 1814 } 1815 1816 kuid_t sock_i_uid(struct sock *sk); 1817 unsigned long sock_i_ino(struct sock *sk); 1818 1819 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1820 { 1821 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1822 } 1823 1824 static inline u32 net_tx_rndhash(void) 1825 { 1826 u32 v = prandom_u32(); 1827 1828 return v ?: 1; 1829 } 1830 1831 static inline void sk_set_txhash(struct sock *sk) 1832 { 1833 sk->sk_txhash = net_tx_rndhash(); 1834 } 1835 1836 static inline void sk_rethink_txhash(struct sock *sk) 1837 { 1838 if (sk->sk_txhash) 1839 sk_set_txhash(sk); 1840 } 1841 1842 static inline struct dst_entry * 1843 __sk_dst_get(struct sock *sk) 1844 { 1845 return rcu_dereference_check(sk->sk_dst_cache, 1846 lockdep_sock_is_held(sk)); 1847 } 1848 1849 static inline struct dst_entry * 1850 sk_dst_get(struct sock *sk) 1851 { 1852 struct dst_entry *dst; 1853 1854 rcu_read_lock(); 1855 dst = rcu_dereference(sk->sk_dst_cache); 1856 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1857 dst = NULL; 1858 rcu_read_unlock(); 1859 return dst; 1860 } 1861 1862 static inline void dst_negative_advice(struct sock *sk) 1863 { 1864 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1865 1866 sk_rethink_txhash(sk); 1867 1868 if (dst && dst->ops->negative_advice) { 1869 ndst = dst->ops->negative_advice(dst); 1870 1871 if (ndst != dst) { 1872 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1873 sk_tx_queue_clear(sk); 1874 sk->sk_dst_pending_confirm = 0; 1875 } 1876 } 1877 } 1878 1879 static inline void 1880 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1881 { 1882 struct dst_entry *old_dst; 1883 1884 sk_tx_queue_clear(sk); 1885 sk->sk_dst_pending_confirm = 0; 1886 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 1887 lockdep_sock_is_held(sk)); 1888 rcu_assign_pointer(sk->sk_dst_cache, dst); 1889 dst_release(old_dst); 1890 } 1891 1892 static inline void 1893 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1894 { 1895 struct dst_entry *old_dst; 1896 1897 sk_tx_queue_clear(sk); 1898 sk->sk_dst_pending_confirm = 0; 1899 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1900 dst_release(old_dst); 1901 } 1902 1903 static inline void 1904 __sk_dst_reset(struct sock *sk) 1905 { 1906 __sk_dst_set(sk, NULL); 1907 } 1908 1909 static inline void 1910 sk_dst_reset(struct sock *sk) 1911 { 1912 sk_dst_set(sk, NULL); 1913 } 1914 1915 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1916 1917 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1918 1919 static inline void sk_dst_confirm(struct sock *sk) 1920 { 1921 if (!sk->sk_dst_pending_confirm) 1922 sk->sk_dst_pending_confirm = 1; 1923 } 1924 1925 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 1926 { 1927 if (skb_get_dst_pending_confirm(skb)) { 1928 struct sock *sk = skb->sk; 1929 unsigned long now = jiffies; 1930 1931 /* avoid dirtying neighbour */ 1932 if (n->confirmed != now) 1933 n->confirmed = now; 1934 if (sk && sk->sk_dst_pending_confirm) 1935 sk->sk_dst_pending_confirm = 0; 1936 } 1937 } 1938 1939 bool sk_mc_loop(struct sock *sk); 1940 1941 static inline bool sk_can_gso(const struct sock *sk) 1942 { 1943 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1944 } 1945 1946 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1947 1948 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1949 { 1950 sk->sk_route_nocaps |= flags; 1951 sk->sk_route_caps &= ~flags; 1952 } 1953 1954 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1955 struct iov_iter *from, char *to, 1956 int copy, int offset) 1957 { 1958 if (skb->ip_summed == CHECKSUM_NONE) { 1959 __wsum csum = 0; 1960 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 1961 return -EFAULT; 1962 skb->csum = csum_block_add(skb->csum, csum, offset); 1963 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1964 if (!copy_from_iter_full_nocache(to, copy, from)) 1965 return -EFAULT; 1966 } else if (!copy_from_iter_full(to, copy, from)) 1967 return -EFAULT; 1968 1969 return 0; 1970 } 1971 1972 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1973 struct iov_iter *from, int copy) 1974 { 1975 int err, offset = skb->len; 1976 1977 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1978 copy, offset); 1979 if (err) 1980 __skb_trim(skb, offset); 1981 1982 return err; 1983 } 1984 1985 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1986 struct sk_buff *skb, 1987 struct page *page, 1988 int off, int copy) 1989 { 1990 int err; 1991 1992 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1993 copy, skb->len); 1994 if (err) 1995 return err; 1996 1997 skb->len += copy; 1998 skb->data_len += copy; 1999 skb->truesize += copy; 2000 sk->sk_wmem_queued += copy; 2001 sk_mem_charge(sk, copy); 2002 return 0; 2003 } 2004 2005 /** 2006 * sk_wmem_alloc_get - returns write allocations 2007 * @sk: socket 2008 * 2009 * Returns sk_wmem_alloc minus initial offset of one 2010 */ 2011 static inline int sk_wmem_alloc_get(const struct sock *sk) 2012 { 2013 return refcount_read(&sk->sk_wmem_alloc) - 1; 2014 } 2015 2016 /** 2017 * sk_rmem_alloc_get - returns read allocations 2018 * @sk: socket 2019 * 2020 * Returns sk_rmem_alloc 2021 */ 2022 static inline int sk_rmem_alloc_get(const struct sock *sk) 2023 { 2024 return atomic_read(&sk->sk_rmem_alloc); 2025 } 2026 2027 /** 2028 * sk_has_allocations - check if allocations are outstanding 2029 * @sk: socket 2030 * 2031 * Returns true if socket has write or read allocations 2032 */ 2033 static inline bool sk_has_allocations(const struct sock *sk) 2034 { 2035 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2036 } 2037 2038 /** 2039 * skwq_has_sleeper - check if there are any waiting processes 2040 * @wq: struct socket_wq 2041 * 2042 * Returns true if socket_wq has waiting processes 2043 * 2044 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2045 * barrier call. They were added due to the race found within the tcp code. 2046 * 2047 * Consider following tcp code paths:: 2048 * 2049 * CPU1 CPU2 2050 * sys_select receive packet 2051 * ... ... 2052 * __add_wait_queue update tp->rcv_nxt 2053 * ... ... 2054 * tp->rcv_nxt check sock_def_readable 2055 * ... { 2056 * schedule rcu_read_lock(); 2057 * wq = rcu_dereference(sk->sk_wq); 2058 * if (wq && waitqueue_active(&wq->wait)) 2059 * wake_up_interruptible(&wq->wait) 2060 * ... 2061 * } 2062 * 2063 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2064 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2065 * could then endup calling schedule and sleep forever if there are no more 2066 * data on the socket. 2067 * 2068 */ 2069 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2070 { 2071 return wq && wq_has_sleeper(&wq->wait); 2072 } 2073 2074 /** 2075 * sock_poll_wait - place memory barrier behind the poll_wait call. 2076 * @filp: file 2077 * @sock: socket to wait on 2078 * @p: poll_table 2079 * 2080 * See the comments in the wq_has_sleeper function. 2081 * 2082 * Do not derive sock from filp->private_data here. An SMC socket establishes 2083 * an internal TCP socket that is used in the fallback case. All socket 2084 * operations on the SMC socket are then forwarded to the TCP socket. In case of 2085 * poll, the filp->private_data pointer references the SMC socket because the 2086 * TCP socket has no file assigned. 2087 */ 2088 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2089 poll_table *p) 2090 { 2091 if (!poll_does_not_wait(p)) { 2092 poll_wait(filp, &sock->wq->wait, p); 2093 /* We need to be sure we are in sync with the 2094 * socket flags modification. 2095 * 2096 * This memory barrier is paired in the wq_has_sleeper. 2097 */ 2098 smp_mb(); 2099 } 2100 } 2101 2102 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2103 { 2104 if (sk->sk_txhash) { 2105 skb->l4_hash = 1; 2106 skb->hash = sk->sk_txhash; 2107 } 2108 } 2109 2110 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2111 2112 /* 2113 * Queue a received datagram if it will fit. Stream and sequenced 2114 * protocols can't normally use this as they need to fit buffers in 2115 * and play with them. 2116 * 2117 * Inlined as it's very short and called for pretty much every 2118 * packet ever received. 2119 */ 2120 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2121 { 2122 skb_orphan(skb); 2123 skb->sk = sk; 2124 skb->destructor = sock_rfree; 2125 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2126 sk_mem_charge(sk, skb->truesize); 2127 } 2128 2129 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2130 unsigned long expires); 2131 2132 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2133 2134 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2135 struct sk_buff *skb, unsigned int flags, 2136 void (*destructor)(struct sock *sk, 2137 struct sk_buff *skb)); 2138 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2139 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2140 2141 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2142 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2143 2144 /* 2145 * Recover an error report and clear atomically 2146 */ 2147 2148 static inline int sock_error(struct sock *sk) 2149 { 2150 int err; 2151 if (likely(!sk->sk_err)) 2152 return 0; 2153 err = xchg(&sk->sk_err, 0); 2154 return -err; 2155 } 2156 2157 static inline unsigned long sock_wspace(struct sock *sk) 2158 { 2159 int amt = 0; 2160 2161 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2162 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2163 if (amt < 0) 2164 amt = 0; 2165 } 2166 return amt; 2167 } 2168 2169 /* Note: 2170 * We use sk->sk_wq_raw, from contexts knowing this 2171 * pointer is not NULL and cannot disappear/change. 2172 */ 2173 static inline void sk_set_bit(int nr, struct sock *sk) 2174 { 2175 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2176 !sock_flag(sk, SOCK_FASYNC)) 2177 return; 2178 2179 set_bit(nr, &sk->sk_wq_raw->flags); 2180 } 2181 2182 static inline void sk_clear_bit(int nr, struct sock *sk) 2183 { 2184 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2185 !sock_flag(sk, SOCK_FASYNC)) 2186 return; 2187 2188 clear_bit(nr, &sk->sk_wq_raw->flags); 2189 } 2190 2191 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2192 { 2193 if (sock_flag(sk, SOCK_FASYNC)) { 2194 rcu_read_lock(); 2195 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2196 rcu_read_unlock(); 2197 } 2198 } 2199 2200 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2201 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2202 * Note: for send buffers, TCP works better if we can build two skbs at 2203 * minimum. 2204 */ 2205 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2206 2207 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2208 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2209 2210 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2211 { 2212 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2213 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2214 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2215 } 2216 } 2217 2218 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2219 bool force_schedule); 2220 2221 /** 2222 * sk_page_frag - return an appropriate page_frag 2223 * @sk: socket 2224 * 2225 * If socket allocation mode allows current thread to sleep, it means its 2226 * safe to use the per task page_frag instead of the per socket one. 2227 */ 2228 static inline struct page_frag *sk_page_frag(struct sock *sk) 2229 { 2230 if (gfpflags_allow_blocking(sk->sk_allocation)) 2231 return ¤t->task_frag; 2232 2233 return &sk->sk_frag; 2234 } 2235 2236 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2237 2238 /* 2239 * Default write policy as shown to user space via poll/select/SIGIO 2240 */ 2241 static inline bool sock_writeable(const struct sock *sk) 2242 { 2243 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2244 } 2245 2246 static inline gfp_t gfp_any(void) 2247 { 2248 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2249 } 2250 2251 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2252 { 2253 return noblock ? 0 : sk->sk_rcvtimeo; 2254 } 2255 2256 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2257 { 2258 return noblock ? 0 : sk->sk_sndtimeo; 2259 } 2260 2261 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2262 { 2263 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2264 } 2265 2266 /* Alas, with timeout socket operations are not restartable. 2267 * Compare this to poll(). 2268 */ 2269 static inline int sock_intr_errno(long timeo) 2270 { 2271 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2272 } 2273 2274 struct sock_skb_cb { 2275 u32 dropcount; 2276 }; 2277 2278 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2279 * using skb->cb[] would keep using it directly and utilize its 2280 * alignement guarantee. 2281 */ 2282 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2283 sizeof(struct sock_skb_cb))) 2284 2285 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2286 SOCK_SKB_CB_OFFSET)) 2287 2288 #define sock_skb_cb_check_size(size) \ 2289 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2290 2291 static inline void 2292 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2293 { 2294 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2295 atomic_read(&sk->sk_drops) : 0; 2296 } 2297 2298 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2299 { 2300 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2301 2302 atomic_add(segs, &sk->sk_drops); 2303 } 2304 2305 static inline ktime_t sock_read_timestamp(struct sock *sk) 2306 { 2307 #if BITS_PER_LONG==32 2308 unsigned int seq; 2309 ktime_t kt; 2310 2311 do { 2312 seq = read_seqbegin(&sk->sk_stamp_seq); 2313 kt = sk->sk_stamp; 2314 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2315 2316 return kt; 2317 #else 2318 return sk->sk_stamp; 2319 #endif 2320 } 2321 2322 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2323 { 2324 #if BITS_PER_LONG==32 2325 write_seqlock(&sk->sk_stamp_seq); 2326 sk->sk_stamp = kt; 2327 write_sequnlock(&sk->sk_stamp_seq); 2328 #else 2329 sk->sk_stamp = kt; 2330 #endif 2331 } 2332 2333 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2334 struct sk_buff *skb); 2335 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2336 struct sk_buff *skb); 2337 2338 static inline void 2339 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2340 { 2341 ktime_t kt = skb->tstamp; 2342 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2343 2344 /* 2345 * generate control messages if 2346 * - receive time stamping in software requested 2347 * - software time stamp available and wanted 2348 * - hardware time stamps available and wanted 2349 */ 2350 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2351 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2352 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2353 (hwtstamps->hwtstamp && 2354 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2355 __sock_recv_timestamp(msg, sk, skb); 2356 else 2357 sock_write_timestamp(sk, kt); 2358 2359 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2360 __sock_recv_wifi_status(msg, sk, skb); 2361 } 2362 2363 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2364 struct sk_buff *skb); 2365 2366 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2367 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2368 struct sk_buff *skb) 2369 { 2370 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2371 (1UL << SOCK_RCVTSTAMP)) 2372 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2373 SOF_TIMESTAMPING_RAW_HARDWARE) 2374 2375 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2376 __sock_recv_ts_and_drops(msg, sk, skb); 2377 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2378 sock_write_timestamp(sk, skb->tstamp); 2379 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2380 sock_write_timestamp(sk, 0); 2381 } 2382 2383 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2384 2385 /** 2386 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2387 * @sk: socket sending this packet 2388 * @tsflags: timestamping flags to use 2389 * @tx_flags: completed with instructions for time stamping 2390 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2391 * 2392 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2393 */ 2394 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2395 __u8 *tx_flags, __u32 *tskey) 2396 { 2397 if (unlikely(tsflags)) { 2398 __sock_tx_timestamp(tsflags, tx_flags); 2399 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2400 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2401 *tskey = sk->sk_tskey++; 2402 } 2403 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2404 *tx_flags |= SKBTX_WIFI_STATUS; 2405 } 2406 2407 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2408 __u8 *tx_flags) 2409 { 2410 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2411 } 2412 2413 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2414 { 2415 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2416 &skb_shinfo(skb)->tskey); 2417 } 2418 2419 /** 2420 * sk_eat_skb - Release a skb if it is no longer needed 2421 * @sk: socket to eat this skb from 2422 * @skb: socket buffer to eat 2423 * 2424 * This routine must be called with interrupts disabled or with the socket 2425 * locked so that the sk_buff queue operation is ok. 2426 */ 2427 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2428 { 2429 __skb_unlink(skb, &sk->sk_receive_queue); 2430 __kfree_skb(skb); 2431 } 2432 2433 static inline 2434 struct net *sock_net(const struct sock *sk) 2435 { 2436 return read_pnet(&sk->sk_net); 2437 } 2438 2439 static inline 2440 void sock_net_set(struct sock *sk, struct net *net) 2441 { 2442 write_pnet(&sk->sk_net, net); 2443 } 2444 2445 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2446 { 2447 if (skb->sk) { 2448 struct sock *sk = skb->sk; 2449 2450 skb->destructor = NULL; 2451 skb->sk = NULL; 2452 return sk; 2453 } 2454 return NULL; 2455 } 2456 2457 /* This helper checks if a socket is a full socket, 2458 * ie _not_ a timewait or request socket. 2459 */ 2460 static inline bool sk_fullsock(const struct sock *sk) 2461 { 2462 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2463 } 2464 2465 /* Checks if this SKB belongs to an HW offloaded socket 2466 * and whether any SW fallbacks are required based on dev. 2467 */ 2468 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2469 struct net_device *dev) 2470 { 2471 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2472 struct sock *sk = skb->sk; 2473 2474 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) 2475 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2476 #endif 2477 2478 return skb; 2479 } 2480 2481 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2482 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2483 */ 2484 static inline bool sk_listener(const struct sock *sk) 2485 { 2486 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2487 } 2488 2489 void sock_enable_timestamp(struct sock *sk, int flag); 2490 int sock_get_timestamp(struct sock *, struct timeval __user *); 2491 int sock_get_timestampns(struct sock *, struct timespec __user *); 2492 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2493 int type); 2494 2495 bool sk_ns_capable(const struct sock *sk, 2496 struct user_namespace *user_ns, int cap); 2497 bool sk_capable(const struct sock *sk, int cap); 2498 bool sk_net_capable(const struct sock *sk, int cap); 2499 2500 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2501 2502 /* Take into consideration the size of the struct sk_buff overhead in the 2503 * determination of these values, since that is non-constant across 2504 * platforms. This makes socket queueing behavior and performance 2505 * not depend upon such differences. 2506 */ 2507 #define _SK_MEM_PACKETS 256 2508 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2509 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2510 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2511 2512 extern __u32 sysctl_wmem_max; 2513 extern __u32 sysctl_rmem_max; 2514 2515 extern int sysctl_tstamp_allow_data; 2516 extern int sysctl_optmem_max; 2517 2518 extern __u32 sysctl_wmem_default; 2519 extern __u32 sysctl_rmem_default; 2520 2521 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2522 { 2523 /* Does this proto have per netns sysctl_wmem ? */ 2524 if (proto->sysctl_wmem_offset) 2525 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2526 2527 return *proto->sysctl_wmem; 2528 } 2529 2530 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2531 { 2532 /* Does this proto have per netns sysctl_rmem ? */ 2533 if (proto->sysctl_rmem_offset) 2534 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2535 2536 return *proto->sysctl_rmem; 2537 } 2538 2539 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2540 * Some wifi drivers need to tweak it to get more chunks. 2541 * They can use this helper from their ndo_start_xmit() 2542 */ 2543 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2544 { 2545 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val) 2546 return; 2547 sk->sk_pacing_shift = val; 2548 } 2549 2550 /* if a socket is bound to a device, check that the given device 2551 * index is either the same or that the socket is bound to an L3 2552 * master device and the given device index is also enslaved to 2553 * that L3 master 2554 */ 2555 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2556 { 2557 int mdif; 2558 2559 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2560 return true; 2561 2562 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2563 if (mdif && mdif == sk->sk_bound_dev_if) 2564 return true; 2565 2566 return false; 2567 } 2568 2569 #endif /* _SOCK_H */ 2570