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