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 64 #include <linux/filter.h> 65 #include <linux/rculist_nulls.h> 66 #include <linux/poll.h> 67 68 #include <linux/atomic.h> 69 #include <net/dst.h> 70 #include <net/checksum.h> 71 #include <net/tcp_states.h> 72 #include <linux/net_tstamp.h> 73 74 struct cgroup; 75 struct cgroup_subsys; 76 #ifdef CONFIG_NET 77 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss); 78 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg); 79 #else 80 static inline 81 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 82 { 83 return 0; 84 } 85 static inline 86 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 87 { 88 } 89 #endif 90 /* 91 * This structure really needs to be cleaned up. 92 * Most of it is for TCP, and not used by any of 93 * the other protocols. 94 */ 95 96 /* Define this to get the SOCK_DBG debugging facility. */ 97 #define SOCK_DEBUGGING 98 #ifdef SOCK_DEBUGGING 99 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 100 printk(KERN_DEBUG msg); } while (0) 101 #else 102 /* Validate arguments and do nothing */ 103 static inline __printf(2, 3) 104 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 105 { 106 } 107 #endif 108 109 /* This is the per-socket lock. The spinlock provides a synchronization 110 * between user contexts and software interrupt processing, whereas the 111 * mini-semaphore synchronizes multiple users amongst themselves. 112 */ 113 typedef struct { 114 spinlock_t slock; 115 int owned; 116 wait_queue_head_t wq; 117 /* 118 * We express the mutex-alike socket_lock semantics 119 * to the lock validator by explicitly managing 120 * the slock as a lock variant (in addition to 121 * the slock itself): 122 */ 123 #ifdef CONFIG_DEBUG_LOCK_ALLOC 124 struct lockdep_map dep_map; 125 #endif 126 } socket_lock_t; 127 128 struct sock; 129 struct proto; 130 struct net; 131 132 typedef __u32 __bitwise __portpair; 133 typedef __u64 __bitwise __addrpair; 134 135 /** 136 * struct sock_common - minimal network layer representation of sockets 137 * @skc_daddr: Foreign IPv4 addr 138 * @skc_rcv_saddr: Bound local IPv4 addr 139 * @skc_hash: hash value used with various protocol lookup tables 140 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 141 * @skc_dport: placeholder for inet_dport/tw_dport 142 * @skc_num: placeholder for inet_num/tw_num 143 * @skc_family: network address family 144 * @skc_state: Connection state 145 * @skc_reuse: %SO_REUSEADDR setting 146 * @skc_reuseport: %SO_REUSEPORT setting 147 * @skc_bound_dev_if: bound device index if != 0 148 * @skc_bind_node: bind hash linkage for various protocol lookup tables 149 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 150 * @skc_prot: protocol handlers inside a network family 151 * @skc_net: reference to the network namespace of this socket 152 * @skc_node: main hash linkage for various protocol lookup tables 153 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 154 * @skc_tx_queue_mapping: tx queue number for this connection 155 * @skc_flags: place holder for sk_flags 156 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 157 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 158 * @skc_incoming_cpu: record/match cpu processing incoming packets 159 * @skc_refcnt: reference count 160 * 161 * This is the minimal network layer representation of sockets, the header 162 * for struct sock and struct inet_timewait_sock. 163 */ 164 struct sock_common { 165 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 166 * address on 64bit arches : cf INET_MATCH() 167 */ 168 union { 169 __addrpair skc_addrpair; 170 struct { 171 __be32 skc_daddr; 172 __be32 skc_rcv_saddr; 173 }; 174 }; 175 union { 176 unsigned int skc_hash; 177 __u16 skc_u16hashes[2]; 178 }; 179 /* skc_dport && skc_num must be grouped as well */ 180 union { 181 __portpair skc_portpair; 182 struct { 183 __be16 skc_dport; 184 __u16 skc_num; 185 }; 186 }; 187 188 unsigned short skc_family; 189 volatile unsigned char skc_state; 190 unsigned char skc_reuse:4; 191 unsigned char skc_reuseport:1; 192 unsigned char skc_ipv6only:1; 193 unsigned char skc_net_refcnt:1; 194 int skc_bound_dev_if; 195 union { 196 struct hlist_node skc_bind_node; 197 struct hlist_nulls_node skc_portaddr_node; 198 }; 199 struct proto *skc_prot; 200 possible_net_t skc_net; 201 202 #if IS_ENABLED(CONFIG_IPV6) 203 struct in6_addr skc_v6_daddr; 204 struct in6_addr skc_v6_rcv_saddr; 205 #endif 206 207 atomic64_t skc_cookie; 208 209 /* following fields are padding to force 210 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 211 * assuming IPV6 is enabled. We use this padding differently 212 * for different kind of 'sockets' 213 */ 214 union { 215 unsigned long skc_flags; 216 struct sock *skc_listener; /* request_sock */ 217 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 218 }; 219 /* 220 * fields between dontcopy_begin/dontcopy_end 221 * are not copied in sock_copy() 222 */ 223 /* private: */ 224 int skc_dontcopy_begin[0]; 225 /* public: */ 226 union { 227 struct hlist_node skc_node; 228 struct hlist_nulls_node skc_nulls_node; 229 }; 230 int skc_tx_queue_mapping; 231 union { 232 int skc_incoming_cpu; 233 u32 skc_rcv_wnd; 234 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 235 }; 236 237 atomic_t skc_refcnt; 238 /* private: */ 239 int skc_dontcopy_end[0]; 240 union { 241 u32 skc_rxhash; 242 u32 skc_window_clamp; 243 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 244 }; 245 /* public: */ 246 }; 247 248 struct cg_proto; 249 /** 250 * struct sock - network layer representation of sockets 251 * @__sk_common: shared layout with inet_timewait_sock 252 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 253 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 254 * @sk_lock: synchronizer 255 * @sk_rcvbuf: size of receive buffer in bytes 256 * @sk_wq: sock wait queue and async head 257 * @sk_rx_dst: receive input route used by early demux 258 * @sk_dst_cache: destination cache 259 * @sk_policy: flow policy 260 * @sk_receive_queue: incoming packets 261 * @sk_wmem_alloc: transmit queue bytes committed 262 * @sk_write_queue: Packet sending queue 263 * @sk_omem_alloc: "o" is "option" or "other" 264 * @sk_wmem_queued: persistent queue size 265 * @sk_forward_alloc: space allocated forward 266 * @sk_napi_id: id of the last napi context to receive data for sk 267 * @sk_ll_usec: usecs to busypoll when there is no data 268 * @sk_allocation: allocation mode 269 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 270 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 271 * @sk_sndbuf: size of send buffer in bytes 272 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 273 * @sk_no_check_rx: allow zero checksum in RX packets 274 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 275 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 276 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 277 * @sk_gso_max_size: Maximum GSO segment size to build 278 * @sk_gso_max_segs: Maximum number of GSO segments 279 * @sk_lingertime: %SO_LINGER l_linger setting 280 * @sk_backlog: always used with the per-socket spinlock held 281 * @sk_callback_lock: used with the callbacks in the end of this struct 282 * @sk_error_queue: rarely used 283 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 284 * IPV6_ADDRFORM for instance) 285 * @sk_err: last error 286 * @sk_err_soft: errors that don't cause failure but are the cause of a 287 * persistent failure not just 'timed out' 288 * @sk_drops: raw/udp drops counter 289 * @sk_ack_backlog: current listen backlog 290 * @sk_max_ack_backlog: listen backlog set in listen() 291 * @sk_priority: %SO_PRIORITY setting 292 * @sk_type: socket type (%SOCK_STREAM, etc) 293 * @sk_protocol: which protocol this socket belongs in this network family 294 * @sk_peer_pid: &struct pid for this socket's peer 295 * @sk_peer_cred: %SO_PEERCRED setting 296 * @sk_rcvlowat: %SO_RCVLOWAT setting 297 * @sk_rcvtimeo: %SO_RCVTIMEO setting 298 * @sk_sndtimeo: %SO_SNDTIMEO setting 299 * @sk_txhash: computed flow hash for use on transmit 300 * @sk_filter: socket filtering instructions 301 * @sk_timer: sock cleanup timer 302 * @sk_stamp: time stamp of last packet received 303 * @sk_tsflags: SO_TIMESTAMPING socket options 304 * @sk_tskey: counter to disambiguate concurrent tstamp requests 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_cgrp: this socket's cgroup-specific proto data 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 */ 323 struct sock { 324 /* 325 * Now struct inet_timewait_sock also uses sock_common, so please just 326 * don't add nothing before this first member (__sk_common) --acme 327 */ 328 struct sock_common __sk_common; 329 #define sk_node __sk_common.skc_node 330 #define sk_nulls_node __sk_common.skc_nulls_node 331 #define sk_refcnt __sk_common.skc_refcnt 332 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 333 334 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 335 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 336 #define sk_hash __sk_common.skc_hash 337 #define sk_portpair __sk_common.skc_portpair 338 #define sk_num __sk_common.skc_num 339 #define sk_dport __sk_common.skc_dport 340 #define sk_addrpair __sk_common.skc_addrpair 341 #define sk_daddr __sk_common.skc_daddr 342 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 343 #define sk_family __sk_common.skc_family 344 #define sk_state __sk_common.skc_state 345 #define sk_reuse __sk_common.skc_reuse 346 #define sk_reuseport __sk_common.skc_reuseport 347 #define sk_ipv6only __sk_common.skc_ipv6only 348 #define sk_net_refcnt __sk_common.skc_net_refcnt 349 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 350 #define sk_bind_node __sk_common.skc_bind_node 351 #define sk_prot __sk_common.skc_prot 352 #define sk_net __sk_common.skc_net 353 #define sk_v6_daddr __sk_common.skc_v6_daddr 354 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 355 #define sk_cookie __sk_common.skc_cookie 356 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 357 #define sk_flags __sk_common.skc_flags 358 #define sk_rxhash __sk_common.skc_rxhash 359 360 socket_lock_t sk_lock; 361 struct sk_buff_head sk_receive_queue; 362 /* 363 * The backlog queue is special, it is always used with 364 * the per-socket spinlock held and requires low latency 365 * access. Therefore we special case it's implementation. 366 * Note : rmem_alloc is in this structure to fill a hole 367 * on 64bit arches, not because its logically part of 368 * backlog. 369 */ 370 struct { 371 atomic_t rmem_alloc; 372 int len; 373 struct sk_buff *head; 374 struct sk_buff *tail; 375 } sk_backlog; 376 #define sk_rmem_alloc sk_backlog.rmem_alloc 377 int sk_forward_alloc; 378 379 __u32 sk_txhash; 380 #ifdef CONFIG_NET_RX_BUSY_POLL 381 unsigned int sk_napi_id; 382 unsigned int sk_ll_usec; 383 #endif 384 atomic_t sk_drops; 385 int sk_rcvbuf; 386 387 struct sk_filter __rcu *sk_filter; 388 union { 389 struct socket_wq __rcu *sk_wq; 390 struct socket_wq *sk_wq_raw; 391 }; 392 #ifdef CONFIG_XFRM 393 struct xfrm_policy __rcu *sk_policy[2]; 394 #endif 395 struct dst_entry *sk_rx_dst; 396 struct dst_entry __rcu *sk_dst_cache; 397 /* Note: 32bit hole on 64bit arches */ 398 atomic_t sk_wmem_alloc; 399 atomic_t sk_omem_alloc; 400 int sk_sndbuf; 401 struct sk_buff_head sk_write_queue; 402 kmemcheck_bitfield_begin(flags); 403 unsigned int sk_shutdown : 2, 404 sk_no_check_tx : 1, 405 sk_no_check_rx : 1, 406 sk_userlocks : 4, 407 sk_protocol : 8, 408 sk_type : 16; 409 #define SK_PROTOCOL_MAX U8_MAX 410 kmemcheck_bitfield_end(flags); 411 int sk_wmem_queued; 412 gfp_t sk_allocation; 413 u32 sk_pacing_rate; /* bytes per second */ 414 u32 sk_max_pacing_rate; 415 netdev_features_t sk_route_caps; 416 netdev_features_t sk_route_nocaps; 417 int sk_gso_type; 418 unsigned int sk_gso_max_size; 419 u16 sk_gso_max_segs; 420 int sk_rcvlowat; 421 unsigned long sk_lingertime; 422 struct sk_buff_head sk_error_queue; 423 struct proto *sk_prot_creator; 424 rwlock_t sk_callback_lock; 425 int sk_err, 426 sk_err_soft; 427 u32 sk_ack_backlog; 428 u32 sk_max_ack_backlog; 429 __u32 sk_priority; 430 __u32 sk_mark; 431 struct pid *sk_peer_pid; 432 const struct cred *sk_peer_cred; 433 long sk_rcvtimeo; 434 long sk_sndtimeo; 435 struct timer_list sk_timer; 436 ktime_t sk_stamp; 437 u16 sk_tsflags; 438 u32 sk_tskey; 439 struct socket *sk_socket; 440 void *sk_user_data; 441 struct page_frag sk_frag; 442 struct sk_buff *sk_send_head; 443 __s32 sk_peek_off; 444 int sk_write_pending; 445 #ifdef CONFIG_SECURITY 446 void *sk_security; 447 #endif 448 struct sock_cgroup_data sk_cgrp_data; 449 struct cg_proto *sk_cgrp; 450 void (*sk_state_change)(struct sock *sk); 451 void (*sk_data_ready)(struct sock *sk); 452 void (*sk_write_space)(struct sock *sk); 453 void (*sk_error_report)(struct sock *sk); 454 int (*sk_backlog_rcv)(struct sock *sk, 455 struct sk_buff *skb); 456 void (*sk_destruct)(struct sock *sk); 457 struct sock_reuseport __rcu *sk_reuseport_cb; 458 }; 459 460 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 461 462 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 463 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 464 465 /* 466 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 467 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 468 * on a socket means that the socket will reuse everybody else's port 469 * without looking at the other's sk_reuse value. 470 */ 471 472 #define SK_NO_REUSE 0 473 #define SK_CAN_REUSE 1 474 #define SK_FORCE_REUSE 2 475 476 static inline int sk_peek_offset(struct sock *sk, int flags) 477 { 478 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 479 return sk->sk_peek_off; 480 else 481 return 0; 482 } 483 484 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 485 { 486 if (sk->sk_peek_off >= 0) { 487 if (sk->sk_peek_off >= val) 488 sk->sk_peek_off -= val; 489 else 490 sk->sk_peek_off = 0; 491 } 492 } 493 494 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 495 { 496 if (sk->sk_peek_off >= 0) 497 sk->sk_peek_off += val; 498 } 499 500 /* 501 * Hashed lists helper routines 502 */ 503 static inline struct sock *sk_entry(const struct hlist_node *node) 504 { 505 return hlist_entry(node, struct sock, sk_node); 506 } 507 508 static inline struct sock *__sk_head(const struct hlist_head *head) 509 { 510 return hlist_entry(head->first, struct sock, sk_node); 511 } 512 513 static inline struct sock *sk_head(const struct hlist_head *head) 514 { 515 return hlist_empty(head) ? NULL : __sk_head(head); 516 } 517 518 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 519 { 520 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 521 } 522 523 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 524 { 525 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 526 } 527 528 static inline struct sock *sk_next(const struct sock *sk) 529 { 530 return sk->sk_node.next ? 531 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 532 } 533 534 static inline struct sock *sk_nulls_next(const struct sock *sk) 535 { 536 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 537 hlist_nulls_entry(sk->sk_nulls_node.next, 538 struct sock, sk_nulls_node) : 539 NULL; 540 } 541 542 static inline bool sk_unhashed(const struct sock *sk) 543 { 544 return hlist_unhashed(&sk->sk_node); 545 } 546 547 static inline bool sk_hashed(const struct sock *sk) 548 { 549 return !sk_unhashed(sk); 550 } 551 552 static inline void sk_node_init(struct hlist_node *node) 553 { 554 node->pprev = NULL; 555 } 556 557 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 558 { 559 node->pprev = NULL; 560 } 561 562 static inline void __sk_del_node(struct sock *sk) 563 { 564 __hlist_del(&sk->sk_node); 565 } 566 567 /* NB: equivalent to hlist_del_init_rcu */ 568 static inline bool __sk_del_node_init(struct sock *sk) 569 { 570 if (sk_hashed(sk)) { 571 __sk_del_node(sk); 572 sk_node_init(&sk->sk_node); 573 return true; 574 } 575 return false; 576 } 577 578 /* Grab socket reference count. This operation is valid only 579 when sk is ALREADY grabbed f.e. it is found in hash table 580 or a list and the lookup is made under lock preventing hash table 581 modifications. 582 */ 583 584 static inline void sock_hold(struct sock *sk) 585 { 586 atomic_inc(&sk->sk_refcnt); 587 } 588 589 /* Ungrab socket in the context, which assumes that socket refcnt 590 cannot hit zero, f.e. it is true in context of any socketcall. 591 */ 592 static inline void __sock_put(struct sock *sk) 593 { 594 atomic_dec(&sk->sk_refcnt); 595 } 596 597 static inline bool sk_del_node_init(struct sock *sk) 598 { 599 bool rc = __sk_del_node_init(sk); 600 601 if (rc) { 602 /* paranoid for a while -acme */ 603 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 604 __sock_put(sk); 605 } 606 return rc; 607 } 608 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 609 610 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 611 { 612 if (sk_hashed(sk)) { 613 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 614 return true; 615 } 616 return false; 617 } 618 619 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 620 { 621 bool rc = __sk_nulls_del_node_init_rcu(sk); 622 623 if (rc) { 624 /* paranoid for a while -acme */ 625 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 626 __sock_put(sk); 627 } 628 return rc; 629 } 630 631 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 632 { 633 hlist_add_head(&sk->sk_node, list); 634 } 635 636 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 637 { 638 sock_hold(sk); 639 __sk_add_node(sk, list); 640 } 641 642 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 643 { 644 sock_hold(sk); 645 hlist_add_head_rcu(&sk->sk_node, list); 646 } 647 648 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 649 { 650 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 651 } 652 653 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 654 { 655 sock_hold(sk); 656 __sk_nulls_add_node_rcu(sk, list); 657 } 658 659 static inline void __sk_del_bind_node(struct sock *sk) 660 { 661 __hlist_del(&sk->sk_bind_node); 662 } 663 664 static inline void sk_add_bind_node(struct sock *sk, 665 struct hlist_head *list) 666 { 667 hlist_add_head(&sk->sk_bind_node, list); 668 } 669 670 #define sk_for_each(__sk, list) \ 671 hlist_for_each_entry(__sk, list, sk_node) 672 #define sk_for_each_rcu(__sk, list) \ 673 hlist_for_each_entry_rcu(__sk, list, sk_node) 674 #define sk_nulls_for_each(__sk, node, list) \ 675 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 676 #define sk_nulls_for_each_rcu(__sk, node, list) \ 677 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 678 #define sk_for_each_from(__sk) \ 679 hlist_for_each_entry_from(__sk, sk_node) 680 #define sk_nulls_for_each_from(__sk, node) \ 681 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 682 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 683 #define sk_for_each_safe(__sk, tmp, list) \ 684 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 685 #define sk_for_each_bound(__sk, list) \ 686 hlist_for_each_entry(__sk, list, sk_bind_node) 687 688 /** 689 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset 690 * @tpos: the type * to use as a loop cursor. 691 * @pos: the &struct hlist_node to use as a loop cursor. 692 * @head: the head for your list. 693 * @offset: offset of hlist_node within the struct. 694 * 695 */ 696 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \ 697 for (pos = (head)->first; \ 698 (!is_a_nulls(pos)) && \ 699 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 700 pos = pos->next) 701 702 static inline struct user_namespace *sk_user_ns(struct sock *sk) 703 { 704 /* Careful only use this in a context where these parameters 705 * can not change and must all be valid, such as recvmsg from 706 * userspace. 707 */ 708 return sk->sk_socket->file->f_cred->user_ns; 709 } 710 711 /* Sock flags */ 712 enum sock_flags { 713 SOCK_DEAD, 714 SOCK_DONE, 715 SOCK_URGINLINE, 716 SOCK_KEEPOPEN, 717 SOCK_LINGER, 718 SOCK_DESTROY, 719 SOCK_BROADCAST, 720 SOCK_TIMESTAMP, 721 SOCK_ZAPPED, 722 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 723 SOCK_DBG, /* %SO_DEBUG setting */ 724 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 725 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 726 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 727 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 728 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 729 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 730 SOCK_FASYNC, /* fasync() active */ 731 SOCK_RXQ_OVFL, 732 SOCK_ZEROCOPY, /* buffers from userspace */ 733 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 734 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 735 * Will use last 4 bytes of packet sent from 736 * user-space instead. 737 */ 738 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 739 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 740 }; 741 742 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 743 744 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 745 { 746 nsk->sk_flags = osk->sk_flags; 747 } 748 749 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 750 { 751 __set_bit(flag, &sk->sk_flags); 752 } 753 754 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 755 { 756 __clear_bit(flag, &sk->sk_flags); 757 } 758 759 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 760 { 761 return test_bit(flag, &sk->sk_flags); 762 } 763 764 #ifdef CONFIG_NET 765 extern struct static_key memalloc_socks; 766 static inline int sk_memalloc_socks(void) 767 { 768 return static_key_false(&memalloc_socks); 769 } 770 #else 771 772 static inline int sk_memalloc_socks(void) 773 { 774 return 0; 775 } 776 777 #endif 778 779 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 780 { 781 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 782 } 783 784 static inline void sk_acceptq_removed(struct sock *sk) 785 { 786 sk->sk_ack_backlog--; 787 } 788 789 static inline void sk_acceptq_added(struct sock *sk) 790 { 791 sk->sk_ack_backlog++; 792 } 793 794 static inline bool sk_acceptq_is_full(const struct sock *sk) 795 { 796 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 797 } 798 799 /* 800 * Compute minimal free write space needed to queue new packets. 801 */ 802 static inline int sk_stream_min_wspace(const struct sock *sk) 803 { 804 return sk->sk_wmem_queued >> 1; 805 } 806 807 static inline int sk_stream_wspace(const struct sock *sk) 808 { 809 return sk->sk_sndbuf - sk->sk_wmem_queued; 810 } 811 812 void sk_stream_write_space(struct sock *sk); 813 814 /* OOB backlog add */ 815 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 816 { 817 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 818 skb_dst_force_safe(skb); 819 820 if (!sk->sk_backlog.tail) 821 sk->sk_backlog.head = skb; 822 else 823 sk->sk_backlog.tail->next = skb; 824 825 sk->sk_backlog.tail = skb; 826 skb->next = NULL; 827 } 828 829 /* 830 * Take into account size of receive queue and backlog queue 831 * Do not take into account this skb truesize, 832 * to allow even a single big packet to come. 833 */ 834 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 835 { 836 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 837 838 return qsize > limit; 839 } 840 841 /* The per-socket spinlock must be held here. */ 842 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 843 unsigned int limit) 844 { 845 if (sk_rcvqueues_full(sk, limit)) 846 return -ENOBUFS; 847 848 /* 849 * If the skb was allocated from pfmemalloc reserves, only 850 * allow SOCK_MEMALLOC sockets to use it as this socket is 851 * helping free memory 852 */ 853 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 854 return -ENOMEM; 855 856 __sk_add_backlog(sk, skb); 857 sk->sk_backlog.len += skb->truesize; 858 return 0; 859 } 860 861 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 862 863 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 864 { 865 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 866 return __sk_backlog_rcv(sk, skb); 867 868 return sk->sk_backlog_rcv(sk, skb); 869 } 870 871 static inline void sk_incoming_cpu_update(struct sock *sk) 872 { 873 sk->sk_incoming_cpu = raw_smp_processor_id(); 874 } 875 876 static inline void sock_rps_record_flow_hash(__u32 hash) 877 { 878 #ifdef CONFIG_RPS 879 struct rps_sock_flow_table *sock_flow_table; 880 881 rcu_read_lock(); 882 sock_flow_table = rcu_dereference(rps_sock_flow_table); 883 rps_record_sock_flow(sock_flow_table, hash); 884 rcu_read_unlock(); 885 #endif 886 } 887 888 static inline void sock_rps_record_flow(const struct sock *sk) 889 { 890 #ifdef CONFIG_RPS 891 sock_rps_record_flow_hash(sk->sk_rxhash); 892 #endif 893 } 894 895 static inline void sock_rps_save_rxhash(struct sock *sk, 896 const struct sk_buff *skb) 897 { 898 #ifdef CONFIG_RPS 899 if (unlikely(sk->sk_rxhash != skb->hash)) 900 sk->sk_rxhash = skb->hash; 901 #endif 902 } 903 904 static inline void sock_rps_reset_rxhash(struct sock *sk) 905 { 906 #ifdef CONFIG_RPS 907 sk->sk_rxhash = 0; 908 #endif 909 } 910 911 #define sk_wait_event(__sk, __timeo, __condition) \ 912 ({ int __rc; \ 913 release_sock(__sk); \ 914 __rc = __condition; \ 915 if (!__rc) { \ 916 *(__timeo) = schedule_timeout(*(__timeo)); \ 917 } \ 918 sched_annotate_sleep(); \ 919 lock_sock(__sk); \ 920 __rc = __condition; \ 921 __rc; \ 922 }) 923 924 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 925 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 926 void sk_stream_wait_close(struct sock *sk, long timeo_p); 927 int sk_stream_error(struct sock *sk, int flags, int err); 928 void sk_stream_kill_queues(struct sock *sk); 929 void sk_set_memalloc(struct sock *sk); 930 void sk_clear_memalloc(struct sock *sk); 931 932 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 933 934 struct request_sock_ops; 935 struct timewait_sock_ops; 936 struct inet_hashinfo; 937 struct raw_hashinfo; 938 struct module; 939 940 /* 941 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 942 * un-modified. Special care is taken when initializing object to zero. 943 */ 944 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 945 { 946 if (offsetof(struct sock, sk_node.next) != 0) 947 memset(sk, 0, offsetof(struct sock, sk_node.next)); 948 memset(&sk->sk_node.pprev, 0, 949 size - offsetof(struct sock, sk_node.pprev)); 950 } 951 952 /* Networking protocol blocks we attach to sockets. 953 * socket layer -> transport layer interface 954 */ 955 struct proto { 956 void (*close)(struct sock *sk, 957 long timeout); 958 int (*connect)(struct sock *sk, 959 struct sockaddr *uaddr, 960 int addr_len); 961 int (*disconnect)(struct sock *sk, int flags); 962 963 struct sock * (*accept)(struct sock *sk, int flags, int *err); 964 965 int (*ioctl)(struct sock *sk, int cmd, 966 unsigned long arg); 967 int (*init)(struct sock *sk); 968 void (*destroy)(struct sock *sk); 969 void (*shutdown)(struct sock *sk, int how); 970 int (*setsockopt)(struct sock *sk, int level, 971 int optname, char __user *optval, 972 unsigned int optlen); 973 int (*getsockopt)(struct sock *sk, int level, 974 int optname, char __user *optval, 975 int __user *option); 976 #ifdef CONFIG_COMPAT 977 int (*compat_setsockopt)(struct sock *sk, 978 int level, 979 int optname, char __user *optval, 980 unsigned int optlen); 981 int (*compat_getsockopt)(struct sock *sk, 982 int level, 983 int optname, char __user *optval, 984 int __user *option); 985 int (*compat_ioctl)(struct sock *sk, 986 unsigned int cmd, unsigned long arg); 987 #endif 988 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 989 size_t len); 990 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 991 size_t len, int noblock, int flags, 992 int *addr_len); 993 int (*sendpage)(struct sock *sk, struct page *page, 994 int offset, size_t size, int flags); 995 int (*bind)(struct sock *sk, 996 struct sockaddr *uaddr, int addr_len); 997 998 int (*backlog_rcv) (struct sock *sk, 999 struct sk_buff *skb); 1000 1001 void (*release_cb)(struct sock *sk); 1002 1003 /* Keeping track of sk's, looking them up, and port selection methods. */ 1004 void (*hash)(struct sock *sk); 1005 void (*unhash)(struct sock *sk); 1006 void (*rehash)(struct sock *sk); 1007 int (*get_port)(struct sock *sk, unsigned short snum); 1008 void (*clear_sk)(struct sock *sk, int size); 1009 1010 /* Keeping track of sockets in use */ 1011 #ifdef CONFIG_PROC_FS 1012 unsigned int inuse_idx; 1013 #endif 1014 1015 bool (*stream_memory_free)(const struct sock *sk); 1016 /* Memory pressure */ 1017 void (*enter_memory_pressure)(struct sock *sk); 1018 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1019 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1020 /* 1021 * Pressure flag: try to collapse. 1022 * Technical note: it is used by multiple contexts non atomically. 1023 * All the __sk_mem_schedule() is of this nature: accounting 1024 * is strict, actions are advisory and have some latency. 1025 */ 1026 int *memory_pressure; 1027 long *sysctl_mem; 1028 int *sysctl_wmem; 1029 int *sysctl_rmem; 1030 int max_header; 1031 bool no_autobind; 1032 1033 struct kmem_cache *slab; 1034 unsigned int obj_size; 1035 int slab_flags; 1036 1037 struct percpu_counter *orphan_count; 1038 1039 struct request_sock_ops *rsk_prot; 1040 struct timewait_sock_ops *twsk_prot; 1041 1042 union { 1043 struct inet_hashinfo *hashinfo; 1044 struct udp_table *udp_table; 1045 struct raw_hashinfo *raw_hash; 1046 } h; 1047 1048 struct module *owner; 1049 1050 char name[32]; 1051 1052 struct list_head node; 1053 #ifdef SOCK_REFCNT_DEBUG 1054 atomic_t socks; 1055 #endif 1056 #ifdef CONFIG_MEMCG_KMEM 1057 /* 1058 * cgroup specific init/deinit functions. Called once for all 1059 * protocols that implement it, from cgroups populate function. 1060 * This function has to setup any files the protocol want to 1061 * appear in the kmem cgroup filesystem. 1062 */ 1063 int (*init_cgroup)(struct mem_cgroup *memcg, 1064 struct cgroup_subsys *ss); 1065 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1066 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1067 #endif 1068 int (*diag_destroy)(struct sock *sk, int err); 1069 }; 1070 1071 int proto_register(struct proto *prot, int alloc_slab); 1072 void proto_unregister(struct proto *prot); 1073 1074 #ifdef SOCK_REFCNT_DEBUG 1075 static inline void sk_refcnt_debug_inc(struct sock *sk) 1076 { 1077 atomic_inc(&sk->sk_prot->socks); 1078 } 1079 1080 static inline void sk_refcnt_debug_dec(struct sock *sk) 1081 { 1082 atomic_dec(&sk->sk_prot->socks); 1083 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1084 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1085 } 1086 1087 static inline void sk_refcnt_debug_release(const struct sock *sk) 1088 { 1089 if (atomic_read(&sk->sk_refcnt) != 1) 1090 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1091 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1092 } 1093 #else /* SOCK_REFCNT_DEBUG */ 1094 #define sk_refcnt_debug_inc(sk) do { } while (0) 1095 #define sk_refcnt_debug_dec(sk) do { } while (0) 1096 #define sk_refcnt_debug_release(sk) do { } while (0) 1097 #endif /* SOCK_REFCNT_DEBUG */ 1098 1099 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1100 extern struct static_key memcg_socket_limit_enabled; 1101 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1102 struct cg_proto *cg_proto) 1103 { 1104 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1105 } 1106 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1107 #else 1108 #define mem_cgroup_sockets_enabled 0 1109 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1110 struct cg_proto *cg_proto) 1111 { 1112 return NULL; 1113 } 1114 #endif 1115 1116 static inline bool sk_stream_memory_free(const struct sock *sk) 1117 { 1118 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1119 return false; 1120 1121 return sk->sk_prot->stream_memory_free ? 1122 sk->sk_prot->stream_memory_free(sk) : true; 1123 } 1124 1125 static inline bool sk_stream_is_writeable(const struct sock *sk) 1126 { 1127 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1128 sk_stream_memory_free(sk); 1129 } 1130 1131 1132 static inline bool sk_has_memory_pressure(const struct sock *sk) 1133 { 1134 return sk->sk_prot->memory_pressure != NULL; 1135 } 1136 1137 static inline bool sk_under_memory_pressure(const struct sock *sk) 1138 { 1139 if (!sk->sk_prot->memory_pressure) 1140 return false; 1141 1142 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1143 return !!sk->sk_cgrp->memory_pressure; 1144 1145 return !!*sk->sk_prot->memory_pressure; 1146 } 1147 1148 static inline void sk_leave_memory_pressure(struct sock *sk) 1149 { 1150 int *memory_pressure = sk->sk_prot->memory_pressure; 1151 1152 if (!memory_pressure) 1153 return; 1154 1155 if (*memory_pressure) 1156 *memory_pressure = 0; 1157 1158 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1159 struct cg_proto *cg_proto = sk->sk_cgrp; 1160 struct proto *prot = sk->sk_prot; 1161 1162 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1163 cg_proto->memory_pressure = 0; 1164 } 1165 1166 } 1167 1168 static inline void sk_enter_memory_pressure(struct sock *sk) 1169 { 1170 if (!sk->sk_prot->enter_memory_pressure) 1171 return; 1172 1173 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1174 struct cg_proto *cg_proto = sk->sk_cgrp; 1175 struct proto *prot = sk->sk_prot; 1176 1177 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1178 cg_proto->memory_pressure = 1; 1179 } 1180 1181 sk->sk_prot->enter_memory_pressure(sk); 1182 } 1183 1184 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1185 { 1186 long *prot = sk->sk_prot->sysctl_mem; 1187 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1188 prot = sk->sk_cgrp->sysctl_mem; 1189 return prot[index]; 1190 } 1191 1192 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1193 unsigned long amt, 1194 int *parent_status) 1195 { 1196 page_counter_charge(&prot->memory_allocated, amt); 1197 1198 if (page_counter_read(&prot->memory_allocated) > 1199 prot->memory_allocated.limit) 1200 *parent_status = OVER_LIMIT; 1201 } 1202 1203 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1204 unsigned long amt) 1205 { 1206 page_counter_uncharge(&prot->memory_allocated, amt); 1207 } 1208 1209 static inline long 1210 sk_memory_allocated(const struct sock *sk) 1211 { 1212 struct proto *prot = sk->sk_prot; 1213 1214 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1215 return page_counter_read(&sk->sk_cgrp->memory_allocated); 1216 1217 return atomic_long_read(prot->memory_allocated); 1218 } 1219 1220 static inline long 1221 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1222 { 1223 struct proto *prot = sk->sk_prot; 1224 1225 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1226 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1227 /* update the root cgroup regardless */ 1228 atomic_long_add_return(amt, prot->memory_allocated); 1229 return page_counter_read(&sk->sk_cgrp->memory_allocated); 1230 } 1231 1232 return atomic_long_add_return(amt, prot->memory_allocated); 1233 } 1234 1235 static inline void 1236 sk_memory_allocated_sub(struct sock *sk, int amt) 1237 { 1238 struct proto *prot = sk->sk_prot; 1239 1240 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1241 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1242 1243 atomic_long_sub(amt, prot->memory_allocated); 1244 } 1245 1246 static inline void sk_sockets_allocated_dec(struct sock *sk) 1247 { 1248 struct proto *prot = sk->sk_prot; 1249 1250 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1251 struct cg_proto *cg_proto = sk->sk_cgrp; 1252 1253 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1254 percpu_counter_dec(&cg_proto->sockets_allocated); 1255 } 1256 1257 percpu_counter_dec(prot->sockets_allocated); 1258 } 1259 1260 static inline void sk_sockets_allocated_inc(struct sock *sk) 1261 { 1262 struct proto *prot = sk->sk_prot; 1263 1264 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1265 struct cg_proto *cg_proto = sk->sk_cgrp; 1266 1267 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1268 percpu_counter_inc(&cg_proto->sockets_allocated); 1269 } 1270 1271 percpu_counter_inc(prot->sockets_allocated); 1272 } 1273 1274 static inline int 1275 sk_sockets_allocated_read_positive(struct sock *sk) 1276 { 1277 struct proto *prot = sk->sk_prot; 1278 1279 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1280 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated); 1281 1282 return percpu_counter_read_positive(prot->sockets_allocated); 1283 } 1284 1285 static inline int 1286 proto_sockets_allocated_sum_positive(struct proto *prot) 1287 { 1288 return percpu_counter_sum_positive(prot->sockets_allocated); 1289 } 1290 1291 static inline long 1292 proto_memory_allocated(struct proto *prot) 1293 { 1294 return atomic_long_read(prot->memory_allocated); 1295 } 1296 1297 static inline bool 1298 proto_memory_pressure(struct proto *prot) 1299 { 1300 if (!prot->memory_pressure) 1301 return false; 1302 return !!*prot->memory_pressure; 1303 } 1304 1305 1306 #ifdef CONFIG_PROC_FS 1307 /* Called with local bh disabled */ 1308 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1309 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1310 #else 1311 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1312 int inc) 1313 { 1314 } 1315 #endif 1316 1317 1318 /* With per-bucket locks this operation is not-atomic, so that 1319 * this version is not worse. 1320 */ 1321 static inline void __sk_prot_rehash(struct sock *sk) 1322 { 1323 sk->sk_prot->unhash(sk); 1324 sk->sk_prot->hash(sk); 1325 } 1326 1327 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1328 1329 /* About 10 seconds */ 1330 #define SOCK_DESTROY_TIME (10*HZ) 1331 1332 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1333 #define PROT_SOCK 1024 1334 1335 #define SHUTDOWN_MASK 3 1336 #define RCV_SHUTDOWN 1 1337 #define SEND_SHUTDOWN 2 1338 1339 #define SOCK_SNDBUF_LOCK 1 1340 #define SOCK_RCVBUF_LOCK 2 1341 #define SOCK_BINDADDR_LOCK 4 1342 #define SOCK_BINDPORT_LOCK 8 1343 1344 struct socket_alloc { 1345 struct socket socket; 1346 struct inode vfs_inode; 1347 }; 1348 1349 static inline struct socket *SOCKET_I(struct inode *inode) 1350 { 1351 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1352 } 1353 1354 static inline struct inode *SOCK_INODE(struct socket *socket) 1355 { 1356 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1357 } 1358 1359 /* 1360 * Functions for memory accounting 1361 */ 1362 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1363 void __sk_mem_reclaim(struct sock *sk, int amount); 1364 1365 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1366 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1367 #define SK_MEM_SEND 0 1368 #define SK_MEM_RECV 1 1369 1370 static inline int sk_mem_pages(int amt) 1371 { 1372 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1373 } 1374 1375 static inline bool sk_has_account(struct sock *sk) 1376 { 1377 /* return true if protocol supports memory accounting */ 1378 return !!sk->sk_prot->memory_allocated; 1379 } 1380 1381 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1382 { 1383 if (!sk_has_account(sk)) 1384 return true; 1385 return size <= sk->sk_forward_alloc || 1386 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1387 } 1388 1389 static inline bool 1390 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1391 { 1392 if (!sk_has_account(sk)) 1393 return true; 1394 return size<= sk->sk_forward_alloc || 1395 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1396 skb_pfmemalloc(skb); 1397 } 1398 1399 static inline void sk_mem_reclaim(struct sock *sk) 1400 { 1401 if (!sk_has_account(sk)) 1402 return; 1403 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1404 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1405 } 1406 1407 static inline void sk_mem_reclaim_partial(struct sock *sk) 1408 { 1409 if (!sk_has_account(sk)) 1410 return; 1411 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1412 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1413 } 1414 1415 static inline void sk_mem_charge(struct sock *sk, int size) 1416 { 1417 if (!sk_has_account(sk)) 1418 return; 1419 sk->sk_forward_alloc -= size; 1420 } 1421 1422 static inline void sk_mem_uncharge(struct sock *sk, int size) 1423 { 1424 if (!sk_has_account(sk)) 1425 return; 1426 sk->sk_forward_alloc += size; 1427 } 1428 1429 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1430 { 1431 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1432 sk->sk_wmem_queued -= skb->truesize; 1433 sk_mem_uncharge(sk, skb->truesize); 1434 __kfree_skb(skb); 1435 } 1436 1437 /* Used by processes to "lock" a socket state, so that 1438 * interrupts and bottom half handlers won't change it 1439 * from under us. It essentially blocks any incoming 1440 * packets, so that we won't get any new data or any 1441 * packets that change the state of the socket. 1442 * 1443 * While locked, BH processing will add new packets to 1444 * the backlog queue. This queue is processed by the 1445 * owner of the socket lock right before it is released. 1446 * 1447 * Since ~2.3.5 it is also exclusive sleep lock serializing 1448 * accesses from user process context. 1449 */ 1450 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1451 1452 static inline void sock_release_ownership(struct sock *sk) 1453 { 1454 sk->sk_lock.owned = 0; 1455 } 1456 1457 /* 1458 * Macro so as to not evaluate some arguments when 1459 * lockdep is not enabled. 1460 * 1461 * Mark both the sk_lock and the sk_lock.slock as a 1462 * per-address-family lock class. 1463 */ 1464 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1465 do { \ 1466 sk->sk_lock.owned = 0; \ 1467 init_waitqueue_head(&sk->sk_lock.wq); \ 1468 spin_lock_init(&(sk)->sk_lock.slock); \ 1469 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1470 sizeof((sk)->sk_lock)); \ 1471 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1472 (skey), (sname)); \ 1473 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1474 } while (0) 1475 1476 void lock_sock_nested(struct sock *sk, int subclass); 1477 1478 static inline void lock_sock(struct sock *sk) 1479 { 1480 lock_sock_nested(sk, 0); 1481 } 1482 1483 void release_sock(struct sock *sk); 1484 1485 /* BH context may only use the following locking interface. */ 1486 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1487 #define bh_lock_sock_nested(__sk) \ 1488 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1489 SINGLE_DEPTH_NESTING) 1490 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1491 1492 bool lock_sock_fast(struct sock *sk); 1493 /** 1494 * unlock_sock_fast - complement of lock_sock_fast 1495 * @sk: socket 1496 * @slow: slow mode 1497 * 1498 * fast unlock socket for user context. 1499 * If slow mode is on, we call regular release_sock() 1500 */ 1501 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1502 { 1503 if (slow) 1504 release_sock(sk); 1505 else 1506 spin_unlock_bh(&sk->sk_lock.slock); 1507 } 1508 1509 1510 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1511 struct proto *prot, int kern); 1512 void sk_free(struct sock *sk); 1513 void sk_destruct(struct sock *sk); 1514 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1515 1516 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1517 gfp_t priority); 1518 void sock_wfree(struct sk_buff *skb); 1519 void skb_orphan_partial(struct sk_buff *skb); 1520 void sock_rfree(struct sk_buff *skb); 1521 void sock_efree(struct sk_buff *skb); 1522 #ifdef CONFIG_INET 1523 void sock_edemux(struct sk_buff *skb); 1524 #else 1525 #define sock_edemux(skb) sock_efree(skb) 1526 #endif 1527 1528 int sock_setsockopt(struct socket *sock, int level, int op, 1529 char __user *optval, unsigned int optlen); 1530 1531 int sock_getsockopt(struct socket *sock, int level, int op, 1532 char __user *optval, int __user *optlen); 1533 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1534 int noblock, int *errcode); 1535 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1536 unsigned long data_len, int noblock, 1537 int *errcode, int max_page_order); 1538 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1539 void sock_kfree_s(struct sock *sk, void *mem, int size); 1540 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1541 void sk_send_sigurg(struct sock *sk); 1542 1543 struct sockcm_cookie { 1544 u32 mark; 1545 }; 1546 1547 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1548 struct sockcm_cookie *sockc); 1549 1550 /* 1551 * Functions to fill in entries in struct proto_ops when a protocol 1552 * does not implement a particular function. 1553 */ 1554 int sock_no_bind(struct socket *, struct sockaddr *, int); 1555 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1556 int sock_no_socketpair(struct socket *, struct socket *); 1557 int sock_no_accept(struct socket *, struct socket *, int); 1558 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1559 unsigned int sock_no_poll(struct file *, struct socket *, 1560 struct poll_table_struct *); 1561 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1562 int sock_no_listen(struct socket *, int); 1563 int sock_no_shutdown(struct socket *, int); 1564 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1565 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1566 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1567 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1568 int sock_no_mmap(struct file *file, struct socket *sock, 1569 struct vm_area_struct *vma); 1570 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1571 size_t size, int flags); 1572 1573 /* 1574 * Functions to fill in entries in struct proto_ops when a protocol 1575 * uses the inet style. 1576 */ 1577 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1578 char __user *optval, int __user *optlen); 1579 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1580 int flags); 1581 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1582 char __user *optval, unsigned int optlen); 1583 int compat_sock_common_getsockopt(struct socket *sock, int level, 1584 int optname, char __user *optval, int __user *optlen); 1585 int compat_sock_common_setsockopt(struct socket *sock, int level, 1586 int optname, char __user *optval, unsigned int optlen); 1587 1588 void sk_common_release(struct sock *sk); 1589 1590 /* 1591 * Default socket callbacks and setup code 1592 */ 1593 1594 /* Initialise core socket variables */ 1595 void sock_init_data(struct socket *sock, struct sock *sk); 1596 1597 /* 1598 * Socket reference counting postulates. 1599 * 1600 * * Each user of socket SHOULD hold a reference count. 1601 * * Each access point to socket (an hash table bucket, reference from a list, 1602 * running timer, skb in flight MUST hold a reference count. 1603 * * When reference count hits 0, it means it will never increase back. 1604 * * When reference count hits 0, it means that no references from 1605 * outside exist to this socket and current process on current CPU 1606 * is last user and may/should destroy this socket. 1607 * * sk_free is called from any context: process, BH, IRQ. When 1608 * it is called, socket has no references from outside -> sk_free 1609 * may release descendant resources allocated by the socket, but 1610 * to the time when it is called, socket is NOT referenced by any 1611 * hash tables, lists etc. 1612 * * Packets, delivered from outside (from network or from another process) 1613 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1614 * when they sit in queue. Otherwise, packets will leak to hole, when 1615 * socket is looked up by one cpu and unhasing is made by another CPU. 1616 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1617 * (leak to backlog). Packet socket does all the processing inside 1618 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1619 * use separate SMP lock, so that they are prone too. 1620 */ 1621 1622 /* Ungrab socket and destroy it, if it was the last reference. */ 1623 static inline void sock_put(struct sock *sk) 1624 { 1625 if (atomic_dec_and_test(&sk->sk_refcnt)) 1626 sk_free(sk); 1627 } 1628 /* Generic version of sock_put(), dealing with all sockets 1629 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1630 */ 1631 void sock_gen_put(struct sock *sk); 1632 1633 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested); 1634 1635 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1636 { 1637 sk->sk_tx_queue_mapping = tx_queue; 1638 } 1639 1640 static inline void sk_tx_queue_clear(struct sock *sk) 1641 { 1642 sk->sk_tx_queue_mapping = -1; 1643 } 1644 1645 static inline int sk_tx_queue_get(const struct sock *sk) 1646 { 1647 return sk ? sk->sk_tx_queue_mapping : -1; 1648 } 1649 1650 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1651 { 1652 sk_tx_queue_clear(sk); 1653 sk->sk_socket = sock; 1654 } 1655 1656 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1657 { 1658 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1659 return &rcu_dereference_raw(sk->sk_wq)->wait; 1660 } 1661 /* Detach socket from process context. 1662 * Announce socket dead, detach it from wait queue and inode. 1663 * Note that parent inode held reference count on this struct sock, 1664 * we do not release it in this function, because protocol 1665 * probably wants some additional cleanups or even continuing 1666 * to work with this socket (TCP). 1667 */ 1668 static inline void sock_orphan(struct sock *sk) 1669 { 1670 write_lock_bh(&sk->sk_callback_lock); 1671 sock_set_flag(sk, SOCK_DEAD); 1672 sk_set_socket(sk, NULL); 1673 sk->sk_wq = NULL; 1674 write_unlock_bh(&sk->sk_callback_lock); 1675 } 1676 1677 static inline void sock_graft(struct sock *sk, struct socket *parent) 1678 { 1679 write_lock_bh(&sk->sk_callback_lock); 1680 sk->sk_wq = parent->wq; 1681 parent->sk = sk; 1682 sk_set_socket(sk, parent); 1683 security_sock_graft(sk, parent); 1684 write_unlock_bh(&sk->sk_callback_lock); 1685 } 1686 1687 kuid_t sock_i_uid(struct sock *sk); 1688 unsigned long sock_i_ino(struct sock *sk); 1689 1690 static inline u32 net_tx_rndhash(void) 1691 { 1692 u32 v = prandom_u32(); 1693 1694 return v ?: 1; 1695 } 1696 1697 static inline void sk_set_txhash(struct sock *sk) 1698 { 1699 sk->sk_txhash = net_tx_rndhash(); 1700 } 1701 1702 static inline void sk_rethink_txhash(struct sock *sk) 1703 { 1704 if (sk->sk_txhash) 1705 sk_set_txhash(sk); 1706 } 1707 1708 static inline struct dst_entry * 1709 __sk_dst_get(struct sock *sk) 1710 { 1711 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1712 lockdep_is_held(&sk->sk_lock.slock)); 1713 } 1714 1715 static inline struct dst_entry * 1716 sk_dst_get(struct sock *sk) 1717 { 1718 struct dst_entry *dst; 1719 1720 rcu_read_lock(); 1721 dst = rcu_dereference(sk->sk_dst_cache); 1722 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1723 dst = NULL; 1724 rcu_read_unlock(); 1725 return dst; 1726 } 1727 1728 static inline void dst_negative_advice(struct sock *sk) 1729 { 1730 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1731 1732 sk_rethink_txhash(sk); 1733 1734 if (dst && dst->ops->negative_advice) { 1735 ndst = dst->ops->negative_advice(dst); 1736 1737 if (ndst != dst) { 1738 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1739 sk_tx_queue_clear(sk); 1740 } 1741 } 1742 } 1743 1744 static inline void 1745 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1746 { 1747 struct dst_entry *old_dst; 1748 1749 sk_tx_queue_clear(sk); 1750 /* 1751 * This can be called while sk is owned by the caller only, 1752 * with no state that can be checked in a rcu_dereference_check() cond 1753 */ 1754 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1755 rcu_assign_pointer(sk->sk_dst_cache, dst); 1756 dst_release(old_dst); 1757 } 1758 1759 static inline void 1760 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1761 { 1762 struct dst_entry *old_dst; 1763 1764 sk_tx_queue_clear(sk); 1765 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1766 dst_release(old_dst); 1767 } 1768 1769 static inline void 1770 __sk_dst_reset(struct sock *sk) 1771 { 1772 __sk_dst_set(sk, NULL); 1773 } 1774 1775 static inline void 1776 sk_dst_reset(struct sock *sk) 1777 { 1778 sk_dst_set(sk, NULL); 1779 } 1780 1781 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1782 1783 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1784 1785 bool sk_mc_loop(struct sock *sk); 1786 1787 static inline bool sk_can_gso(const struct sock *sk) 1788 { 1789 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1790 } 1791 1792 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1793 1794 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1795 { 1796 sk->sk_route_nocaps |= flags; 1797 sk->sk_route_caps &= ~flags; 1798 } 1799 1800 static inline bool sk_check_csum_caps(struct sock *sk) 1801 { 1802 return (sk->sk_route_caps & NETIF_F_HW_CSUM) || 1803 (sk->sk_family == PF_INET && 1804 (sk->sk_route_caps & NETIF_F_IP_CSUM)) || 1805 (sk->sk_family == PF_INET6 && 1806 (sk->sk_route_caps & NETIF_F_IPV6_CSUM)); 1807 } 1808 1809 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1810 struct iov_iter *from, char *to, 1811 int copy, int offset) 1812 { 1813 if (skb->ip_summed == CHECKSUM_NONE) { 1814 __wsum csum = 0; 1815 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy) 1816 return -EFAULT; 1817 skb->csum = csum_block_add(skb->csum, csum, offset); 1818 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1819 if (copy_from_iter_nocache(to, copy, from) != copy) 1820 return -EFAULT; 1821 } else if (copy_from_iter(to, copy, from) != copy) 1822 return -EFAULT; 1823 1824 return 0; 1825 } 1826 1827 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1828 struct iov_iter *from, int copy) 1829 { 1830 int err, offset = skb->len; 1831 1832 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1833 copy, offset); 1834 if (err) 1835 __skb_trim(skb, offset); 1836 1837 return err; 1838 } 1839 1840 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1841 struct sk_buff *skb, 1842 struct page *page, 1843 int off, int copy) 1844 { 1845 int err; 1846 1847 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1848 copy, skb->len); 1849 if (err) 1850 return err; 1851 1852 skb->len += copy; 1853 skb->data_len += copy; 1854 skb->truesize += copy; 1855 sk->sk_wmem_queued += copy; 1856 sk_mem_charge(sk, copy); 1857 return 0; 1858 } 1859 1860 /** 1861 * sk_wmem_alloc_get - returns write allocations 1862 * @sk: socket 1863 * 1864 * Returns sk_wmem_alloc minus initial offset of one 1865 */ 1866 static inline int sk_wmem_alloc_get(const struct sock *sk) 1867 { 1868 return atomic_read(&sk->sk_wmem_alloc) - 1; 1869 } 1870 1871 /** 1872 * sk_rmem_alloc_get - returns read allocations 1873 * @sk: socket 1874 * 1875 * Returns sk_rmem_alloc 1876 */ 1877 static inline int sk_rmem_alloc_get(const struct sock *sk) 1878 { 1879 return atomic_read(&sk->sk_rmem_alloc); 1880 } 1881 1882 /** 1883 * sk_has_allocations - check if allocations are outstanding 1884 * @sk: socket 1885 * 1886 * Returns true if socket has write or read allocations 1887 */ 1888 static inline bool sk_has_allocations(const struct sock *sk) 1889 { 1890 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1891 } 1892 1893 /** 1894 * skwq_has_sleeper - check if there are any waiting processes 1895 * @wq: struct socket_wq 1896 * 1897 * Returns true if socket_wq has waiting processes 1898 * 1899 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 1900 * barrier call. They were added due to the race found within the tcp code. 1901 * 1902 * Consider following tcp code paths: 1903 * 1904 * CPU1 CPU2 1905 * 1906 * sys_select receive packet 1907 * ... ... 1908 * __add_wait_queue update tp->rcv_nxt 1909 * ... ... 1910 * tp->rcv_nxt check sock_def_readable 1911 * ... { 1912 * schedule rcu_read_lock(); 1913 * wq = rcu_dereference(sk->sk_wq); 1914 * if (wq && waitqueue_active(&wq->wait)) 1915 * wake_up_interruptible(&wq->wait) 1916 * ... 1917 * } 1918 * 1919 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1920 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1921 * could then endup calling schedule and sleep forever if there are no more 1922 * data on the socket. 1923 * 1924 */ 1925 static inline bool skwq_has_sleeper(struct socket_wq *wq) 1926 { 1927 return wq && wq_has_sleeper(&wq->wait); 1928 } 1929 1930 /** 1931 * sock_poll_wait - place memory barrier behind the poll_wait call. 1932 * @filp: file 1933 * @wait_address: socket wait queue 1934 * @p: poll_table 1935 * 1936 * See the comments in the wq_has_sleeper function. 1937 */ 1938 static inline void sock_poll_wait(struct file *filp, 1939 wait_queue_head_t *wait_address, poll_table *p) 1940 { 1941 if (!poll_does_not_wait(p) && wait_address) { 1942 poll_wait(filp, wait_address, p); 1943 /* We need to be sure we are in sync with the 1944 * socket flags modification. 1945 * 1946 * This memory barrier is paired in the wq_has_sleeper. 1947 */ 1948 smp_mb(); 1949 } 1950 } 1951 1952 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1953 { 1954 if (sk->sk_txhash) { 1955 skb->l4_hash = 1; 1956 skb->hash = sk->sk_txhash; 1957 } 1958 } 1959 1960 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 1961 1962 /* 1963 * Queue a received datagram if it will fit. Stream and sequenced 1964 * protocols can't normally use this as they need to fit buffers in 1965 * and play with them. 1966 * 1967 * Inlined as it's very short and called for pretty much every 1968 * packet ever received. 1969 */ 1970 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1971 { 1972 skb_orphan(skb); 1973 skb->sk = sk; 1974 skb->destructor = sock_rfree; 1975 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1976 sk_mem_charge(sk, skb->truesize); 1977 } 1978 1979 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1980 unsigned long expires); 1981 1982 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1983 1984 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1985 1986 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1987 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 1988 1989 /* 1990 * Recover an error report and clear atomically 1991 */ 1992 1993 static inline int sock_error(struct sock *sk) 1994 { 1995 int err; 1996 if (likely(!sk->sk_err)) 1997 return 0; 1998 err = xchg(&sk->sk_err, 0); 1999 return -err; 2000 } 2001 2002 static inline unsigned long sock_wspace(struct sock *sk) 2003 { 2004 int amt = 0; 2005 2006 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2007 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2008 if (amt < 0) 2009 amt = 0; 2010 } 2011 return amt; 2012 } 2013 2014 /* Note: 2015 * We use sk->sk_wq_raw, from contexts knowing this 2016 * pointer is not NULL and cannot disappear/change. 2017 */ 2018 static inline void sk_set_bit(int nr, struct sock *sk) 2019 { 2020 set_bit(nr, &sk->sk_wq_raw->flags); 2021 } 2022 2023 static inline void sk_clear_bit(int nr, struct sock *sk) 2024 { 2025 clear_bit(nr, &sk->sk_wq_raw->flags); 2026 } 2027 2028 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2029 { 2030 if (sock_flag(sk, SOCK_FASYNC)) { 2031 rcu_read_lock(); 2032 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2033 rcu_read_unlock(); 2034 } 2035 } 2036 2037 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2038 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2039 * Note: for send buffers, TCP works better if we can build two skbs at 2040 * minimum. 2041 */ 2042 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2043 2044 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2045 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2046 2047 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2048 { 2049 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2050 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2051 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2052 } 2053 } 2054 2055 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2056 bool force_schedule); 2057 2058 /** 2059 * sk_page_frag - return an appropriate page_frag 2060 * @sk: socket 2061 * 2062 * If socket allocation mode allows current thread to sleep, it means its 2063 * safe to use the per task page_frag instead of the per socket one. 2064 */ 2065 static inline struct page_frag *sk_page_frag(struct sock *sk) 2066 { 2067 if (gfpflags_allow_blocking(sk->sk_allocation)) 2068 return ¤t->task_frag; 2069 2070 return &sk->sk_frag; 2071 } 2072 2073 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2074 2075 /* 2076 * Default write policy as shown to user space via poll/select/SIGIO 2077 */ 2078 static inline bool sock_writeable(const struct sock *sk) 2079 { 2080 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2081 } 2082 2083 static inline gfp_t gfp_any(void) 2084 { 2085 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2086 } 2087 2088 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2089 { 2090 return noblock ? 0 : sk->sk_rcvtimeo; 2091 } 2092 2093 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2094 { 2095 return noblock ? 0 : sk->sk_sndtimeo; 2096 } 2097 2098 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2099 { 2100 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2101 } 2102 2103 /* Alas, with timeout socket operations are not restartable. 2104 * Compare this to poll(). 2105 */ 2106 static inline int sock_intr_errno(long timeo) 2107 { 2108 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2109 } 2110 2111 struct sock_skb_cb { 2112 u32 dropcount; 2113 }; 2114 2115 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2116 * using skb->cb[] would keep using it directly and utilize its 2117 * alignement guarantee. 2118 */ 2119 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2120 sizeof(struct sock_skb_cb))) 2121 2122 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2123 SOCK_SKB_CB_OFFSET)) 2124 2125 #define sock_skb_cb_check_size(size) \ 2126 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2127 2128 static inline void 2129 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2130 { 2131 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops); 2132 } 2133 2134 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2135 struct sk_buff *skb); 2136 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2137 struct sk_buff *skb); 2138 2139 static inline void 2140 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2141 { 2142 ktime_t kt = skb->tstamp; 2143 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2144 2145 /* 2146 * generate control messages if 2147 * - receive time stamping in software requested 2148 * - software time stamp available and wanted 2149 * - hardware time stamps available and wanted 2150 */ 2151 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2152 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2153 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2154 (hwtstamps->hwtstamp.tv64 && 2155 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2156 __sock_recv_timestamp(msg, sk, skb); 2157 else 2158 sk->sk_stamp = kt; 2159 2160 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2161 __sock_recv_wifi_status(msg, sk, skb); 2162 } 2163 2164 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2165 struct sk_buff *skb); 2166 2167 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2168 struct sk_buff *skb) 2169 { 2170 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2171 (1UL << SOCK_RCVTSTAMP)) 2172 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2173 SOF_TIMESTAMPING_RAW_HARDWARE) 2174 2175 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2176 __sock_recv_ts_and_drops(msg, sk, skb); 2177 else 2178 sk->sk_stamp = skb->tstamp; 2179 } 2180 2181 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags); 2182 2183 /** 2184 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2185 * @sk: socket sending this packet 2186 * @tx_flags: completed with instructions for time stamping 2187 * 2188 * Note : callers should take care of initial *tx_flags value (usually 0) 2189 */ 2190 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags) 2191 { 2192 if (unlikely(sk->sk_tsflags)) 2193 __sock_tx_timestamp(sk, tx_flags); 2194 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2195 *tx_flags |= SKBTX_WIFI_STATUS; 2196 } 2197 2198 /** 2199 * sk_eat_skb - Release a skb if it is no longer needed 2200 * @sk: socket to eat this skb from 2201 * @skb: socket buffer to eat 2202 * 2203 * This routine must be called with interrupts disabled or with the socket 2204 * locked so that the sk_buff queue operation is ok. 2205 */ 2206 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2207 { 2208 __skb_unlink(skb, &sk->sk_receive_queue); 2209 __kfree_skb(skb); 2210 } 2211 2212 static inline 2213 struct net *sock_net(const struct sock *sk) 2214 { 2215 return read_pnet(&sk->sk_net); 2216 } 2217 2218 static inline 2219 void sock_net_set(struct sock *sk, struct net *net) 2220 { 2221 write_pnet(&sk->sk_net, net); 2222 } 2223 2224 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2225 { 2226 if (skb->sk) { 2227 struct sock *sk = skb->sk; 2228 2229 skb->destructor = NULL; 2230 skb->sk = NULL; 2231 return sk; 2232 } 2233 return NULL; 2234 } 2235 2236 /* This helper checks if a socket is a full socket, 2237 * ie _not_ a timewait or request socket. 2238 */ 2239 static inline bool sk_fullsock(const struct sock *sk) 2240 { 2241 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2242 } 2243 2244 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2245 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2246 */ 2247 static inline bool sk_listener(const struct sock *sk) 2248 { 2249 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2250 } 2251 2252 /** 2253 * sk_state_load - read sk->sk_state for lockless contexts 2254 * @sk: socket pointer 2255 * 2256 * Paired with sk_state_store(). Used in places we do not hold socket lock : 2257 * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... 2258 */ 2259 static inline int sk_state_load(const struct sock *sk) 2260 { 2261 return smp_load_acquire(&sk->sk_state); 2262 } 2263 2264 /** 2265 * sk_state_store - update sk->sk_state 2266 * @sk: socket pointer 2267 * @newstate: new state 2268 * 2269 * Paired with sk_state_load(). Should be used in contexts where 2270 * state change might impact lockless readers. 2271 */ 2272 static inline void sk_state_store(struct sock *sk, int newstate) 2273 { 2274 smp_store_release(&sk->sk_state, newstate); 2275 } 2276 2277 void sock_enable_timestamp(struct sock *sk, int flag); 2278 int sock_get_timestamp(struct sock *, struct timeval __user *); 2279 int sock_get_timestampns(struct sock *, struct timespec __user *); 2280 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2281 int type); 2282 2283 bool sk_ns_capable(const struct sock *sk, 2284 struct user_namespace *user_ns, int cap); 2285 bool sk_capable(const struct sock *sk, int cap); 2286 bool sk_net_capable(const struct sock *sk, int cap); 2287 2288 extern __u32 sysctl_wmem_max; 2289 extern __u32 sysctl_rmem_max; 2290 2291 extern int sysctl_tstamp_allow_data; 2292 extern int sysctl_optmem_max; 2293 2294 extern __u32 sysctl_wmem_default; 2295 extern __u32 sysctl_rmem_default; 2296 2297 #endif /* _SOCK_H */ 2298