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 62 #include <linux/filter.h> 63 #include <linux/rculist_nulls.h> 64 #include <linux/poll.h> 65 66 #include <linux/atomic.h> 67 #include <net/dst.h> 68 #include <net/checksum.h> 69 #include <net/tcp_states.h> 70 #include <linux/net_tstamp.h> 71 72 struct cgroup; 73 struct cgroup_subsys; 74 #ifdef CONFIG_NET 75 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss); 76 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg); 77 #else 78 static inline 79 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 80 { 81 return 0; 82 } 83 static inline 84 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 85 { 86 } 87 #endif 88 /* 89 * This structure really needs to be cleaned up. 90 * Most of it is for TCP, and not used by any of 91 * the other protocols. 92 */ 93 94 /* Define this to get the SOCK_DBG debugging facility. */ 95 #define SOCK_DEBUGGING 96 #ifdef SOCK_DEBUGGING 97 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 98 printk(KERN_DEBUG msg); } while (0) 99 #else 100 /* Validate arguments and do nothing */ 101 static inline __printf(2, 3) 102 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 103 { 104 } 105 #endif 106 107 /* This is the per-socket lock. The spinlock provides a synchronization 108 * between user contexts and software interrupt processing, whereas the 109 * mini-semaphore synchronizes multiple users amongst themselves. 110 */ 111 typedef struct { 112 spinlock_t slock; 113 int owned; 114 wait_queue_head_t wq; 115 /* 116 * We express the mutex-alike socket_lock semantics 117 * to the lock validator by explicitly managing 118 * the slock as a lock variant (in addition to 119 * the slock itself): 120 */ 121 #ifdef CONFIG_DEBUG_LOCK_ALLOC 122 struct lockdep_map dep_map; 123 #endif 124 } socket_lock_t; 125 126 struct sock; 127 struct proto; 128 struct net; 129 130 typedef __u32 __bitwise __portpair; 131 typedef __u64 __bitwise __addrpair; 132 133 /** 134 * struct sock_common - minimal network layer representation of sockets 135 * @skc_daddr: Foreign IPv4 addr 136 * @skc_rcv_saddr: Bound local IPv4 addr 137 * @skc_hash: hash value used with various protocol lookup tables 138 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 139 * @skc_dport: placeholder for inet_dport/tw_dport 140 * @skc_num: placeholder for inet_num/tw_num 141 * @skc_family: network address family 142 * @skc_state: Connection state 143 * @skc_reuse: %SO_REUSEADDR setting 144 * @skc_reuseport: %SO_REUSEPORT setting 145 * @skc_bound_dev_if: bound device index if != 0 146 * @skc_bind_node: bind hash linkage for various protocol lookup tables 147 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 148 * @skc_prot: protocol handlers inside a network family 149 * @skc_net: reference to the network namespace of this socket 150 * @skc_node: main hash linkage for various protocol lookup tables 151 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 152 * @skc_tx_queue_mapping: tx queue number for this connection 153 * @skc_refcnt: reference count 154 * 155 * This is the minimal network layer representation of sockets, the header 156 * for struct sock and struct inet_timewait_sock. 157 */ 158 struct sock_common { 159 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 160 * address on 64bit arches : cf INET_MATCH() 161 */ 162 union { 163 __addrpair skc_addrpair; 164 struct { 165 __be32 skc_daddr; 166 __be32 skc_rcv_saddr; 167 }; 168 }; 169 union { 170 unsigned int skc_hash; 171 __u16 skc_u16hashes[2]; 172 }; 173 /* skc_dport && skc_num must be grouped as well */ 174 union { 175 __portpair skc_portpair; 176 struct { 177 __be16 skc_dport; 178 __u16 skc_num; 179 }; 180 }; 181 182 unsigned short skc_family; 183 volatile unsigned char skc_state; 184 unsigned char skc_reuse:4; 185 unsigned char skc_reuseport:1; 186 unsigned char skc_ipv6only:1; 187 unsigned char skc_net_refcnt:1; 188 int skc_bound_dev_if; 189 union { 190 struct hlist_node skc_bind_node; 191 struct hlist_nulls_node skc_portaddr_node; 192 }; 193 struct proto *skc_prot; 194 possible_net_t skc_net; 195 196 #if IS_ENABLED(CONFIG_IPV6) 197 struct in6_addr skc_v6_daddr; 198 struct in6_addr skc_v6_rcv_saddr; 199 #endif 200 201 atomic64_t skc_cookie; 202 203 /* 204 * fields between dontcopy_begin/dontcopy_end 205 * are not copied in sock_copy() 206 */ 207 /* private: */ 208 int skc_dontcopy_begin[0]; 209 /* public: */ 210 union { 211 struct hlist_node skc_node; 212 struct hlist_nulls_node skc_nulls_node; 213 }; 214 int skc_tx_queue_mapping; 215 atomic_t skc_refcnt; 216 /* private: */ 217 int skc_dontcopy_end[0]; 218 /* public: */ 219 }; 220 221 struct cg_proto; 222 /** 223 * struct sock - network layer representation of sockets 224 * @__sk_common: shared layout with inet_timewait_sock 225 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 226 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 227 * @sk_lock: synchronizer 228 * @sk_rcvbuf: size of receive buffer in bytes 229 * @sk_wq: sock wait queue and async head 230 * @sk_rx_dst: receive input route used by early demux 231 * @sk_dst_cache: destination cache 232 * @sk_dst_lock: destination cache lock 233 * @sk_policy: flow policy 234 * @sk_receive_queue: incoming packets 235 * @sk_wmem_alloc: transmit queue bytes committed 236 * @sk_write_queue: Packet sending queue 237 * @sk_omem_alloc: "o" is "option" or "other" 238 * @sk_wmem_queued: persistent queue size 239 * @sk_forward_alloc: space allocated forward 240 * @sk_napi_id: id of the last napi context to receive data for sk 241 * @sk_ll_usec: usecs to busypoll when there is no data 242 * @sk_allocation: allocation mode 243 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 244 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 245 * @sk_sndbuf: size of send buffer in bytes 246 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 247 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 248 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 249 * @sk_no_check_rx: allow zero checksum in RX packets 250 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 251 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 252 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 253 * @sk_gso_max_size: Maximum GSO segment size to build 254 * @sk_gso_max_segs: Maximum number of GSO segments 255 * @sk_lingertime: %SO_LINGER l_linger setting 256 * @sk_backlog: always used with the per-socket spinlock held 257 * @sk_callback_lock: used with the callbacks in the end of this struct 258 * @sk_error_queue: rarely used 259 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 260 * IPV6_ADDRFORM for instance) 261 * @sk_err: last error 262 * @sk_err_soft: errors that don't cause failure but are the cause of a 263 * persistent failure not just 'timed out' 264 * @sk_drops: raw/udp drops counter 265 * @sk_ack_backlog: current listen backlog 266 * @sk_max_ack_backlog: listen backlog set in listen() 267 * @sk_priority: %SO_PRIORITY setting 268 * @sk_cgrp_prioidx: socket group's priority map index 269 * @sk_type: socket type (%SOCK_STREAM, etc) 270 * @sk_protocol: which protocol this socket belongs in this network family 271 * @sk_peer_pid: &struct pid for this socket's peer 272 * @sk_peer_cred: %SO_PEERCRED setting 273 * @sk_rcvlowat: %SO_RCVLOWAT setting 274 * @sk_rcvtimeo: %SO_RCVTIMEO setting 275 * @sk_sndtimeo: %SO_SNDTIMEO setting 276 * @sk_rxhash: flow hash received from netif layer 277 * @sk_incoming_cpu: record cpu processing incoming packets 278 * @sk_txhash: computed flow hash for use on transmit 279 * @sk_filter: socket filtering instructions 280 * @sk_protinfo: private area, net family specific, when not using slab 281 * @sk_timer: sock cleanup timer 282 * @sk_stamp: time stamp of last packet received 283 * @sk_tsflags: SO_TIMESTAMPING socket options 284 * @sk_tskey: counter to disambiguate concurrent tstamp requests 285 * @sk_socket: Identd and reporting IO signals 286 * @sk_user_data: RPC layer private data 287 * @sk_frag: cached page frag 288 * @sk_peek_off: current peek_offset value 289 * @sk_send_head: front of stuff to transmit 290 * @sk_security: used by security modules 291 * @sk_mark: generic packet mark 292 * @sk_classid: this socket's cgroup classid 293 * @sk_cgrp: this socket's cgroup-specific proto data 294 * @sk_write_pending: a write to stream socket waits to start 295 * @sk_state_change: callback to indicate change in the state of the sock 296 * @sk_data_ready: callback to indicate there is data to be processed 297 * @sk_write_space: callback to indicate there is bf sending space available 298 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 299 * @sk_backlog_rcv: callback to process the backlog 300 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 301 */ 302 struct sock { 303 /* 304 * Now struct inet_timewait_sock also uses sock_common, so please just 305 * don't add nothing before this first member (__sk_common) --acme 306 */ 307 struct sock_common __sk_common; 308 #define sk_node __sk_common.skc_node 309 #define sk_nulls_node __sk_common.skc_nulls_node 310 #define sk_refcnt __sk_common.skc_refcnt 311 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 312 313 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 314 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 315 #define sk_hash __sk_common.skc_hash 316 #define sk_portpair __sk_common.skc_portpair 317 #define sk_num __sk_common.skc_num 318 #define sk_dport __sk_common.skc_dport 319 #define sk_addrpair __sk_common.skc_addrpair 320 #define sk_daddr __sk_common.skc_daddr 321 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 322 #define sk_family __sk_common.skc_family 323 #define sk_state __sk_common.skc_state 324 #define sk_reuse __sk_common.skc_reuse 325 #define sk_reuseport __sk_common.skc_reuseport 326 #define sk_ipv6only __sk_common.skc_ipv6only 327 #define sk_net_refcnt __sk_common.skc_net_refcnt 328 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 329 #define sk_bind_node __sk_common.skc_bind_node 330 #define sk_prot __sk_common.skc_prot 331 #define sk_net __sk_common.skc_net 332 #define sk_v6_daddr __sk_common.skc_v6_daddr 333 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 334 #define sk_cookie __sk_common.skc_cookie 335 336 socket_lock_t sk_lock; 337 struct sk_buff_head sk_receive_queue; 338 /* 339 * The backlog queue is special, it is always used with 340 * the per-socket spinlock held and requires low latency 341 * access. Therefore we special case it's implementation. 342 * Note : rmem_alloc is in this structure to fill a hole 343 * on 64bit arches, not because its logically part of 344 * backlog. 345 */ 346 struct { 347 atomic_t rmem_alloc; 348 int len; 349 struct sk_buff *head; 350 struct sk_buff *tail; 351 } sk_backlog; 352 #define sk_rmem_alloc sk_backlog.rmem_alloc 353 int sk_forward_alloc; 354 #ifdef CONFIG_RPS 355 __u32 sk_rxhash; 356 #endif 357 u16 sk_incoming_cpu; 358 /* 16bit hole 359 * Warned : sk_incoming_cpu can be set from softirq, 360 * Do not use this hole without fully understanding possible issues. 361 */ 362 363 __u32 sk_txhash; 364 #ifdef CONFIG_NET_RX_BUSY_POLL 365 unsigned int sk_napi_id; 366 unsigned int sk_ll_usec; 367 #endif 368 atomic_t sk_drops; 369 int sk_rcvbuf; 370 371 struct sk_filter __rcu *sk_filter; 372 struct socket_wq __rcu *sk_wq; 373 374 #ifdef CONFIG_XFRM 375 struct xfrm_policy *sk_policy[2]; 376 #endif 377 unsigned long sk_flags; 378 struct dst_entry *sk_rx_dst; 379 struct dst_entry __rcu *sk_dst_cache; 380 spinlock_t sk_dst_lock; 381 atomic_t sk_wmem_alloc; 382 atomic_t sk_omem_alloc; 383 int sk_sndbuf; 384 struct sk_buff_head sk_write_queue; 385 kmemcheck_bitfield_begin(flags); 386 unsigned int sk_shutdown : 2, 387 sk_no_check_tx : 1, 388 sk_no_check_rx : 1, 389 sk_userlocks : 4, 390 sk_protocol : 8, 391 sk_type : 16; 392 kmemcheck_bitfield_end(flags); 393 int sk_wmem_queued; 394 gfp_t sk_allocation; 395 u32 sk_pacing_rate; /* bytes per second */ 396 u32 sk_max_pacing_rate; 397 netdev_features_t sk_route_caps; 398 netdev_features_t sk_route_nocaps; 399 int sk_gso_type; 400 unsigned int sk_gso_max_size; 401 u16 sk_gso_max_segs; 402 int sk_rcvlowat; 403 unsigned long sk_lingertime; 404 struct sk_buff_head sk_error_queue; 405 struct proto *sk_prot_creator; 406 rwlock_t sk_callback_lock; 407 int sk_err, 408 sk_err_soft; 409 u32 sk_ack_backlog; 410 u32 sk_max_ack_backlog; 411 __u32 sk_priority; 412 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 413 __u32 sk_cgrp_prioidx; 414 #endif 415 struct pid *sk_peer_pid; 416 const struct cred *sk_peer_cred; 417 long sk_rcvtimeo; 418 long sk_sndtimeo; 419 void *sk_protinfo; 420 struct timer_list sk_timer; 421 ktime_t sk_stamp; 422 u16 sk_tsflags; 423 u32 sk_tskey; 424 struct socket *sk_socket; 425 void *sk_user_data; 426 struct page_frag sk_frag; 427 struct sk_buff *sk_send_head; 428 __s32 sk_peek_off; 429 int sk_write_pending; 430 #ifdef CONFIG_SECURITY 431 void *sk_security; 432 #endif 433 __u32 sk_mark; 434 u32 sk_classid; 435 struct cg_proto *sk_cgrp; 436 void (*sk_state_change)(struct sock *sk); 437 void (*sk_data_ready)(struct sock *sk); 438 void (*sk_write_space)(struct sock *sk); 439 void (*sk_error_report)(struct sock *sk); 440 int (*sk_backlog_rcv)(struct sock *sk, 441 struct sk_buff *skb); 442 void (*sk_destruct)(struct sock *sk); 443 }; 444 445 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 446 447 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 448 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 449 450 /* 451 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 452 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 453 * on a socket means that the socket will reuse everybody else's port 454 * without looking at the other's sk_reuse value. 455 */ 456 457 #define SK_NO_REUSE 0 458 #define SK_CAN_REUSE 1 459 #define SK_FORCE_REUSE 2 460 461 static inline int sk_peek_offset(struct sock *sk, int flags) 462 { 463 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 464 return sk->sk_peek_off; 465 else 466 return 0; 467 } 468 469 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 470 { 471 if (sk->sk_peek_off >= 0) { 472 if (sk->sk_peek_off >= val) 473 sk->sk_peek_off -= val; 474 else 475 sk->sk_peek_off = 0; 476 } 477 } 478 479 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 480 { 481 if (sk->sk_peek_off >= 0) 482 sk->sk_peek_off += val; 483 } 484 485 /* 486 * Hashed lists helper routines 487 */ 488 static inline struct sock *sk_entry(const struct hlist_node *node) 489 { 490 return hlist_entry(node, struct sock, sk_node); 491 } 492 493 static inline struct sock *__sk_head(const struct hlist_head *head) 494 { 495 return hlist_entry(head->first, struct sock, sk_node); 496 } 497 498 static inline struct sock *sk_head(const struct hlist_head *head) 499 { 500 return hlist_empty(head) ? NULL : __sk_head(head); 501 } 502 503 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 504 { 505 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 506 } 507 508 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 509 { 510 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 511 } 512 513 static inline struct sock *sk_next(const struct sock *sk) 514 { 515 return sk->sk_node.next ? 516 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 517 } 518 519 static inline struct sock *sk_nulls_next(const struct sock *sk) 520 { 521 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 522 hlist_nulls_entry(sk->sk_nulls_node.next, 523 struct sock, sk_nulls_node) : 524 NULL; 525 } 526 527 static inline bool sk_unhashed(const struct sock *sk) 528 { 529 return hlist_unhashed(&sk->sk_node); 530 } 531 532 static inline bool sk_hashed(const struct sock *sk) 533 { 534 return !sk_unhashed(sk); 535 } 536 537 static inline void sk_node_init(struct hlist_node *node) 538 { 539 node->pprev = NULL; 540 } 541 542 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 543 { 544 node->pprev = NULL; 545 } 546 547 static inline void __sk_del_node(struct sock *sk) 548 { 549 __hlist_del(&sk->sk_node); 550 } 551 552 /* NB: equivalent to hlist_del_init_rcu */ 553 static inline bool __sk_del_node_init(struct sock *sk) 554 { 555 if (sk_hashed(sk)) { 556 __sk_del_node(sk); 557 sk_node_init(&sk->sk_node); 558 return true; 559 } 560 return false; 561 } 562 563 /* Grab socket reference count. This operation is valid only 564 when sk is ALREADY grabbed f.e. it is found in hash table 565 or a list and the lookup is made under lock preventing hash table 566 modifications. 567 */ 568 569 static inline void sock_hold(struct sock *sk) 570 { 571 atomic_inc(&sk->sk_refcnt); 572 } 573 574 /* Ungrab socket in the context, which assumes that socket refcnt 575 cannot hit zero, f.e. it is true in context of any socketcall. 576 */ 577 static inline void __sock_put(struct sock *sk) 578 { 579 atomic_dec(&sk->sk_refcnt); 580 } 581 582 static inline bool sk_del_node_init(struct sock *sk) 583 { 584 bool rc = __sk_del_node_init(sk); 585 586 if (rc) { 587 /* paranoid for a while -acme */ 588 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 589 __sock_put(sk); 590 } 591 return rc; 592 } 593 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 594 595 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 596 { 597 if (sk_hashed(sk)) { 598 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 599 return true; 600 } 601 return false; 602 } 603 604 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 605 { 606 bool rc = __sk_nulls_del_node_init_rcu(sk); 607 608 if (rc) { 609 /* paranoid for a while -acme */ 610 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 611 __sock_put(sk); 612 } 613 return rc; 614 } 615 616 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 617 { 618 hlist_add_head(&sk->sk_node, list); 619 } 620 621 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 622 { 623 sock_hold(sk); 624 __sk_add_node(sk, list); 625 } 626 627 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 628 { 629 sock_hold(sk); 630 hlist_add_head_rcu(&sk->sk_node, list); 631 } 632 633 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 634 { 635 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 636 } 637 638 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 639 { 640 sock_hold(sk); 641 __sk_nulls_add_node_rcu(sk, list); 642 } 643 644 static inline void __sk_del_bind_node(struct sock *sk) 645 { 646 __hlist_del(&sk->sk_bind_node); 647 } 648 649 static inline void sk_add_bind_node(struct sock *sk, 650 struct hlist_head *list) 651 { 652 hlist_add_head(&sk->sk_bind_node, list); 653 } 654 655 #define sk_for_each(__sk, list) \ 656 hlist_for_each_entry(__sk, list, sk_node) 657 #define sk_for_each_rcu(__sk, list) \ 658 hlist_for_each_entry_rcu(__sk, list, sk_node) 659 #define sk_nulls_for_each(__sk, node, list) \ 660 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 661 #define sk_nulls_for_each_rcu(__sk, node, list) \ 662 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 663 #define sk_for_each_from(__sk) \ 664 hlist_for_each_entry_from(__sk, sk_node) 665 #define sk_nulls_for_each_from(__sk, node) \ 666 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 667 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 668 #define sk_for_each_safe(__sk, tmp, list) \ 669 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 670 #define sk_for_each_bound(__sk, list) \ 671 hlist_for_each_entry(__sk, list, sk_bind_node) 672 673 /** 674 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset 675 * @tpos: the type * to use as a loop cursor. 676 * @pos: the &struct hlist_node to use as a loop cursor. 677 * @head: the head for your list. 678 * @offset: offset of hlist_node within the struct. 679 * 680 */ 681 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \ 682 for (pos = (head)->first; \ 683 (!is_a_nulls(pos)) && \ 684 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 685 pos = pos->next) 686 687 static inline struct user_namespace *sk_user_ns(struct sock *sk) 688 { 689 /* Careful only use this in a context where these parameters 690 * can not change and must all be valid, such as recvmsg from 691 * userspace. 692 */ 693 return sk->sk_socket->file->f_cred->user_ns; 694 } 695 696 /* Sock flags */ 697 enum sock_flags { 698 SOCK_DEAD, 699 SOCK_DONE, 700 SOCK_URGINLINE, 701 SOCK_KEEPOPEN, 702 SOCK_LINGER, 703 SOCK_DESTROY, 704 SOCK_BROADCAST, 705 SOCK_TIMESTAMP, 706 SOCK_ZAPPED, 707 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 708 SOCK_DBG, /* %SO_DEBUG setting */ 709 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 710 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 711 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 712 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 713 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 714 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 715 SOCK_FASYNC, /* fasync() active */ 716 SOCK_RXQ_OVFL, 717 SOCK_ZEROCOPY, /* buffers from userspace */ 718 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 719 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 720 * Will use last 4 bytes of packet sent from 721 * user-space instead. 722 */ 723 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 724 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 725 }; 726 727 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 728 { 729 nsk->sk_flags = osk->sk_flags; 730 } 731 732 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 733 { 734 __set_bit(flag, &sk->sk_flags); 735 } 736 737 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 738 { 739 __clear_bit(flag, &sk->sk_flags); 740 } 741 742 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 743 { 744 return test_bit(flag, &sk->sk_flags); 745 } 746 747 #ifdef CONFIG_NET 748 extern struct static_key memalloc_socks; 749 static inline int sk_memalloc_socks(void) 750 { 751 return static_key_false(&memalloc_socks); 752 } 753 #else 754 755 static inline int sk_memalloc_socks(void) 756 { 757 return 0; 758 } 759 760 #endif 761 762 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 763 { 764 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 765 } 766 767 static inline void sk_acceptq_removed(struct sock *sk) 768 { 769 sk->sk_ack_backlog--; 770 } 771 772 static inline void sk_acceptq_added(struct sock *sk) 773 { 774 sk->sk_ack_backlog++; 775 } 776 777 static inline bool sk_acceptq_is_full(const struct sock *sk) 778 { 779 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 780 } 781 782 /* 783 * Compute minimal free write space needed to queue new packets. 784 */ 785 static inline int sk_stream_min_wspace(const struct sock *sk) 786 { 787 return sk->sk_wmem_queued >> 1; 788 } 789 790 static inline int sk_stream_wspace(const struct sock *sk) 791 { 792 return sk->sk_sndbuf - sk->sk_wmem_queued; 793 } 794 795 void sk_stream_write_space(struct sock *sk); 796 797 /* OOB backlog add */ 798 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 799 { 800 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 801 skb_dst_force(skb); 802 803 if (!sk->sk_backlog.tail) 804 sk->sk_backlog.head = skb; 805 else 806 sk->sk_backlog.tail->next = skb; 807 808 sk->sk_backlog.tail = skb; 809 skb->next = NULL; 810 } 811 812 /* 813 * Take into account size of receive queue and backlog queue 814 * Do not take into account this skb truesize, 815 * to allow even a single big packet to come. 816 */ 817 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 818 { 819 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 820 821 return qsize > limit; 822 } 823 824 /* The per-socket spinlock must be held here. */ 825 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 826 unsigned int limit) 827 { 828 if (sk_rcvqueues_full(sk, limit)) 829 return -ENOBUFS; 830 831 __sk_add_backlog(sk, skb); 832 sk->sk_backlog.len += skb->truesize; 833 return 0; 834 } 835 836 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 837 838 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 839 { 840 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 841 return __sk_backlog_rcv(sk, skb); 842 843 return sk->sk_backlog_rcv(sk, skb); 844 } 845 846 static inline void sk_incoming_cpu_update(struct sock *sk) 847 { 848 sk->sk_incoming_cpu = raw_smp_processor_id(); 849 } 850 851 static inline void sock_rps_record_flow_hash(__u32 hash) 852 { 853 #ifdef CONFIG_RPS 854 struct rps_sock_flow_table *sock_flow_table; 855 856 rcu_read_lock(); 857 sock_flow_table = rcu_dereference(rps_sock_flow_table); 858 rps_record_sock_flow(sock_flow_table, hash); 859 rcu_read_unlock(); 860 #endif 861 } 862 863 static inline void sock_rps_record_flow(const struct sock *sk) 864 { 865 #ifdef CONFIG_RPS 866 sock_rps_record_flow_hash(sk->sk_rxhash); 867 #endif 868 } 869 870 static inline void sock_rps_save_rxhash(struct sock *sk, 871 const struct sk_buff *skb) 872 { 873 #ifdef CONFIG_RPS 874 if (unlikely(sk->sk_rxhash != skb->hash)) 875 sk->sk_rxhash = skb->hash; 876 #endif 877 } 878 879 static inline void sock_rps_reset_rxhash(struct sock *sk) 880 { 881 #ifdef CONFIG_RPS 882 sk->sk_rxhash = 0; 883 #endif 884 } 885 886 #define sk_wait_event(__sk, __timeo, __condition) \ 887 ({ int __rc; \ 888 release_sock(__sk); \ 889 __rc = __condition; \ 890 if (!__rc) { \ 891 *(__timeo) = schedule_timeout(*(__timeo)); \ 892 } \ 893 sched_annotate_sleep(); \ 894 lock_sock(__sk); \ 895 __rc = __condition; \ 896 __rc; \ 897 }) 898 899 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 900 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 901 void sk_stream_wait_close(struct sock *sk, long timeo_p); 902 int sk_stream_error(struct sock *sk, int flags, int err); 903 void sk_stream_kill_queues(struct sock *sk); 904 void sk_set_memalloc(struct sock *sk); 905 void sk_clear_memalloc(struct sock *sk); 906 907 int sk_wait_data(struct sock *sk, long *timeo); 908 909 struct request_sock_ops; 910 struct timewait_sock_ops; 911 struct inet_hashinfo; 912 struct raw_hashinfo; 913 struct module; 914 915 /* 916 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 917 * un-modified. Special care is taken when initializing object to zero. 918 */ 919 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 920 { 921 if (offsetof(struct sock, sk_node.next) != 0) 922 memset(sk, 0, offsetof(struct sock, sk_node.next)); 923 memset(&sk->sk_node.pprev, 0, 924 size - offsetof(struct sock, sk_node.pprev)); 925 } 926 927 /* Networking protocol blocks we attach to sockets. 928 * socket layer -> transport layer interface 929 */ 930 struct proto { 931 void (*close)(struct sock *sk, 932 long timeout); 933 int (*connect)(struct sock *sk, 934 struct sockaddr *uaddr, 935 int addr_len); 936 int (*disconnect)(struct sock *sk, int flags); 937 938 struct sock * (*accept)(struct sock *sk, int flags, int *err); 939 940 int (*ioctl)(struct sock *sk, int cmd, 941 unsigned long arg); 942 int (*init)(struct sock *sk); 943 void (*destroy)(struct sock *sk); 944 void (*shutdown)(struct sock *sk, int how); 945 int (*setsockopt)(struct sock *sk, int level, 946 int optname, char __user *optval, 947 unsigned int optlen); 948 int (*getsockopt)(struct sock *sk, int level, 949 int optname, char __user *optval, 950 int __user *option); 951 #ifdef CONFIG_COMPAT 952 int (*compat_setsockopt)(struct sock *sk, 953 int level, 954 int optname, char __user *optval, 955 unsigned int optlen); 956 int (*compat_getsockopt)(struct sock *sk, 957 int level, 958 int optname, char __user *optval, 959 int __user *option); 960 int (*compat_ioctl)(struct sock *sk, 961 unsigned int cmd, unsigned long arg); 962 #endif 963 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 964 size_t len); 965 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 966 size_t len, int noblock, int flags, 967 int *addr_len); 968 int (*sendpage)(struct sock *sk, struct page *page, 969 int offset, size_t size, int flags); 970 int (*bind)(struct sock *sk, 971 struct sockaddr *uaddr, int addr_len); 972 973 int (*backlog_rcv) (struct sock *sk, 974 struct sk_buff *skb); 975 976 void (*release_cb)(struct sock *sk); 977 978 /* Keeping track of sk's, looking them up, and port selection methods. */ 979 void (*hash)(struct sock *sk); 980 void (*unhash)(struct sock *sk); 981 void (*rehash)(struct sock *sk); 982 int (*get_port)(struct sock *sk, unsigned short snum); 983 void (*clear_sk)(struct sock *sk, int size); 984 985 /* Keeping track of sockets in use */ 986 #ifdef CONFIG_PROC_FS 987 unsigned int inuse_idx; 988 #endif 989 990 bool (*stream_memory_free)(const struct sock *sk); 991 /* Memory pressure */ 992 void (*enter_memory_pressure)(struct sock *sk); 993 atomic_long_t *memory_allocated; /* Current allocated memory. */ 994 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 995 /* 996 * Pressure flag: try to collapse. 997 * Technical note: it is used by multiple contexts non atomically. 998 * All the __sk_mem_schedule() is of this nature: accounting 999 * is strict, actions are advisory and have some latency. 1000 */ 1001 int *memory_pressure; 1002 long *sysctl_mem; 1003 int *sysctl_wmem; 1004 int *sysctl_rmem; 1005 int max_header; 1006 bool no_autobind; 1007 1008 struct kmem_cache *slab; 1009 unsigned int obj_size; 1010 int slab_flags; 1011 1012 struct percpu_counter *orphan_count; 1013 1014 struct request_sock_ops *rsk_prot; 1015 struct timewait_sock_ops *twsk_prot; 1016 1017 union { 1018 struct inet_hashinfo *hashinfo; 1019 struct udp_table *udp_table; 1020 struct raw_hashinfo *raw_hash; 1021 } h; 1022 1023 struct module *owner; 1024 1025 char name[32]; 1026 1027 struct list_head node; 1028 #ifdef SOCK_REFCNT_DEBUG 1029 atomic_t socks; 1030 #endif 1031 #ifdef CONFIG_MEMCG_KMEM 1032 /* 1033 * cgroup specific init/deinit functions. Called once for all 1034 * protocols that implement it, from cgroups populate function. 1035 * This function has to setup any files the protocol want to 1036 * appear in the kmem cgroup filesystem. 1037 */ 1038 int (*init_cgroup)(struct mem_cgroup *memcg, 1039 struct cgroup_subsys *ss); 1040 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1041 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1042 #endif 1043 }; 1044 1045 /* 1046 * Bits in struct cg_proto.flags 1047 */ 1048 enum cg_proto_flags { 1049 /* Currently active and new sockets should be assigned to cgroups */ 1050 MEMCG_SOCK_ACTIVE, 1051 /* It was ever activated; we must disarm static keys on destruction */ 1052 MEMCG_SOCK_ACTIVATED, 1053 }; 1054 1055 struct cg_proto { 1056 struct page_counter memory_allocated; /* Current allocated memory. */ 1057 struct percpu_counter sockets_allocated; /* Current number of sockets. */ 1058 int memory_pressure; 1059 long sysctl_mem[3]; 1060 unsigned long flags; 1061 /* 1062 * memcg field is used to find which memcg we belong directly 1063 * Each memcg struct can hold more than one cg_proto, so container_of 1064 * won't really cut. 1065 * 1066 * The elegant solution would be having an inverse function to 1067 * proto_cgroup in struct proto, but that means polluting the structure 1068 * for everybody, instead of just for memcg users. 1069 */ 1070 struct mem_cgroup *memcg; 1071 }; 1072 1073 int proto_register(struct proto *prot, int alloc_slab); 1074 void proto_unregister(struct proto *prot); 1075 1076 static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1077 { 1078 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1079 } 1080 1081 #ifdef SOCK_REFCNT_DEBUG 1082 static inline void sk_refcnt_debug_inc(struct sock *sk) 1083 { 1084 atomic_inc(&sk->sk_prot->socks); 1085 } 1086 1087 static inline void sk_refcnt_debug_dec(struct sock *sk) 1088 { 1089 atomic_dec(&sk->sk_prot->socks); 1090 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1091 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1092 } 1093 1094 static inline void sk_refcnt_debug_release(const struct sock *sk) 1095 { 1096 if (atomic_read(&sk->sk_refcnt) != 1) 1097 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1098 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1099 } 1100 #else /* SOCK_REFCNT_DEBUG */ 1101 #define sk_refcnt_debug_inc(sk) do { } while (0) 1102 #define sk_refcnt_debug_dec(sk) do { } while (0) 1103 #define sk_refcnt_debug_release(sk) do { } while (0) 1104 #endif /* SOCK_REFCNT_DEBUG */ 1105 1106 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1107 extern struct static_key memcg_socket_limit_enabled; 1108 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1109 struct cg_proto *cg_proto) 1110 { 1111 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1112 } 1113 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1114 #else 1115 #define mem_cgroup_sockets_enabled 0 1116 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1117 struct cg_proto *cg_proto) 1118 { 1119 return NULL; 1120 } 1121 #endif 1122 1123 static inline bool sk_stream_memory_free(const struct sock *sk) 1124 { 1125 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1126 return false; 1127 1128 return sk->sk_prot->stream_memory_free ? 1129 sk->sk_prot->stream_memory_free(sk) : true; 1130 } 1131 1132 static inline bool sk_stream_is_writeable(const struct sock *sk) 1133 { 1134 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1135 sk_stream_memory_free(sk); 1136 } 1137 1138 1139 static inline bool sk_has_memory_pressure(const struct sock *sk) 1140 { 1141 return sk->sk_prot->memory_pressure != NULL; 1142 } 1143 1144 static inline bool sk_under_memory_pressure(const struct sock *sk) 1145 { 1146 if (!sk->sk_prot->memory_pressure) 1147 return false; 1148 1149 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1150 return !!sk->sk_cgrp->memory_pressure; 1151 1152 return !!*sk->sk_prot->memory_pressure; 1153 } 1154 1155 static inline void sk_leave_memory_pressure(struct sock *sk) 1156 { 1157 int *memory_pressure = sk->sk_prot->memory_pressure; 1158 1159 if (!memory_pressure) 1160 return; 1161 1162 if (*memory_pressure) 1163 *memory_pressure = 0; 1164 1165 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1166 struct cg_proto *cg_proto = sk->sk_cgrp; 1167 struct proto *prot = sk->sk_prot; 1168 1169 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1170 cg_proto->memory_pressure = 0; 1171 } 1172 1173 } 1174 1175 static inline void sk_enter_memory_pressure(struct sock *sk) 1176 { 1177 if (!sk->sk_prot->enter_memory_pressure) 1178 return; 1179 1180 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1181 struct cg_proto *cg_proto = sk->sk_cgrp; 1182 struct proto *prot = sk->sk_prot; 1183 1184 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1185 cg_proto->memory_pressure = 1; 1186 } 1187 1188 sk->sk_prot->enter_memory_pressure(sk); 1189 } 1190 1191 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1192 { 1193 long *prot = sk->sk_prot->sysctl_mem; 1194 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1195 prot = sk->sk_cgrp->sysctl_mem; 1196 return prot[index]; 1197 } 1198 1199 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1200 unsigned long amt, 1201 int *parent_status) 1202 { 1203 page_counter_charge(&prot->memory_allocated, amt); 1204 1205 if (page_counter_read(&prot->memory_allocated) > 1206 prot->memory_allocated.limit) 1207 *parent_status = OVER_LIMIT; 1208 } 1209 1210 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1211 unsigned long amt) 1212 { 1213 page_counter_uncharge(&prot->memory_allocated, amt); 1214 } 1215 1216 static inline long 1217 sk_memory_allocated(const struct sock *sk) 1218 { 1219 struct proto *prot = sk->sk_prot; 1220 1221 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1222 return page_counter_read(&sk->sk_cgrp->memory_allocated); 1223 1224 return atomic_long_read(prot->memory_allocated); 1225 } 1226 1227 static inline long 1228 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1229 { 1230 struct proto *prot = sk->sk_prot; 1231 1232 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1233 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1234 /* update the root cgroup regardless */ 1235 atomic_long_add_return(amt, prot->memory_allocated); 1236 return page_counter_read(&sk->sk_cgrp->memory_allocated); 1237 } 1238 1239 return atomic_long_add_return(amt, prot->memory_allocated); 1240 } 1241 1242 static inline void 1243 sk_memory_allocated_sub(struct sock *sk, int amt) 1244 { 1245 struct proto *prot = sk->sk_prot; 1246 1247 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1248 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1249 1250 atomic_long_sub(amt, prot->memory_allocated); 1251 } 1252 1253 static inline void sk_sockets_allocated_dec(struct sock *sk) 1254 { 1255 struct proto *prot = sk->sk_prot; 1256 1257 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1258 struct cg_proto *cg_proto = sk->sk_cgrp; 1259 1260 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1261 percpu_counter_dec(&cg_proto->sockets_allocated); 1262 } 1263 1264 percpu_counter_dec(prot->sockets_allocated); 1265 } 1266 1267 static inline void sk_sockets_allocated_inc(struct sock *sk) 1268 { 1269 struct proto *prot = sk->sk_prot; 1270 1271 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1272 struct cg_proto *cg_proto = sk->sk_cgrp; 1273 1274 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1275 percpu_counter_inc(&cg_proto->sockets_allocated); 1276 } 1277 1278 percpu_counter_inc(prot->sockets_allocated); 1279 } 1280 1281 static inline int 1282 sk_sockets_allocated_read_positive(struct sock *sk) 1283 { 1284 struct proto *prot = sk->sk_prot; 1285 1286 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1287 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated); 1288 1289 return percpu_counter_read_positive(prot->sockets_allocated); 1290 } 1291 1292 static inline int 1293 proto_sockets_allocated_sum_positive(struct proto *prot) 1294 { 1295 return percpu_counter_sum_positive(prot->sockets_allocated); 1296 } 1297 1298 static inline long 1299 proto_memory_allocated(struct proto *prot) 1300 { 1301 return atomic_long_read(prot->memory_allocated); 1302 } 1303 1304 static inline bool 1305 proto_memory_pressure(struct proto *prot) 1306 { 1307 if (!prot->memory_pressure) 1308 return false; 1309 return !!*prot->memory_pressure; 1310 } 1311 1312 1313 #ifdef CONFIG_PROC_FS 1314 /* Called with local bh disabled */ 1315 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1316 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1317 #else 1318 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1319 int inc) 1320 { 1321 } 1322 #endif 1323 1324 1325 /* With per-bucket locks this operation is not-atomic, so that 1326 * this version is not worse. 1327 */ 1328 static inline void __sk_prot_rehash(struct sock *sk) 1329 { 1330 sk->sk_prot->unhash(sk); 1331 sk->sk_prot->hash(sk); 1332 } 1333 1334 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1335 1336 /* About 10 seconds */ 1337 #define SOCK_DESTROY_TIME (10*HZ) 1338 1339 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1340 #define PROT_SOCK 1024 1341 1342 #define SHUTDOWN_MASK 3 1343 #define RCV_SHUTDOWN 1 1344 #define SEND_SHUTDOWN 2 1345 1346 #define SOCK_SNDBUF_LOCK 1 1347 #define SOCK_RCVBUF_LOCK 2 1348 #define SOCK_BINDADDR_LOCK 4 1349 #define SOCK_BINDPORT_LOCK 8 1350 1351 struct socket_alloc { 1352 struct socket socket; 1353 struct inode vfs_inode; 1354 }; 1355 1356 static inline struct socket *SOCKET_I(struct inode *inode) 1357 { 1358 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1359 } 1360 1361 static inline struct inode *SOCK_INODE(struct socket *socket) 1362 { 1363 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1364 } 1365 1366 /* 1367 * Functions for memory accounting 1368 */ 1369 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1370 void __sk_mem_reclaim(struct sock *sk, int amount); 1371 1372 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1373 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1374 #define SK_MEM_SEND 0 1375 #define SK_MEM_RECV 1 1376 1377 static inline int sk_mem_pages(int amt) 1378 { 1379 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1380 } 1381 1382 static inline bool sk_has_account(struct sock *sk) 1383 { 1384 /* return true if protocol supports memory accounting */ 1385 return !!sk->sk_prot->memory_allocated; 1386 } 1387 1388 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1389 { 1390 if (!sk_has_account(sk)) 1391 return true; 1392 return size <= sk->sk_forward_alloc || 1393 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1394 } 1395 1396 static inline bool 1397 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1398 { 1399 if (!sk_has_account(sk)) 1400 return true; 1401 return size<= sk->sk_forward_alloc || 1402 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1403 skb_pfmemalloc(skb); 1404 } 1405 1406 static inline void sk_mem_reclaim(struct sock *sk) 1407 { 1408 if (!sk_has_account(sk)) 1409 return; 1410 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1411 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1412 } 1413 1414 static inline void sk_mem_reclaim_partial(struct sock *sk) 1415 { 1416 if (!sk_has_account(sk)) 1417 return; 1418 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1419 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1420 } 1421 1422 static inline void sk_mem_charge(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_mem_uncharge(struct sock *sk, int size) 1430 { 1431 if (!sk_has_account(sk)) 1432 return; 1433 sk->sk_forward_alloc += size; 1434 } 1435 1436 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1437 { 1438 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1439 sk->sk_wmem_queued -= skb->truesize; 1440 sk_mem_uncharge(sk, skb->truesize); 1441 __kfree_skb(skb); 1442 } 1443 1444 /* Used by processes to "lock" a socket state, so that 1445 * interrupts and bottom half handlers won't change it 1446 * from under us. It essentially blocks any incoming 1447 * packets, so that we won't get any new data or any 1448 * packets that change the state of the socket. 1449 * 1450 * While locked, BH processing will add new packets to 1451 * the backlog queue. This queue is processed by the 1452 * owner of the socket lock right before it is released. 1453 * 1454 * Since ~2.3.5 it is also exclusive sleep lock serializing 1455 * accesses from user process context. 1456 */ 1457 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1458 1459 static inline void sock_release_ownership(struct sock *sk) 1460 { 1461 sk->sk_lock.owned = 0; 1462 } 1463 1464 /* 1465 * Macro so as to not evaluate some arguments when 1466 * lockdep is not enabled. 1467 * 1468 * Mark both the sk_lock and the sk_lock.slock as a 1469 * per-address-family lock class. 1470 */ 1471 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1472 do { \ 1473 sk->sk_lock.owned = 0; \ 1474 init_waitqueue_head(&sk->sk_lock.wq); \ 1475 spin_lock_init(&(sk)->sk_lock.slock); \ 1476 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1477 sizeof((sk)->sk_lock)); \ 1478 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1479 (skey), (sname)); \ 1480 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1481 } while (0) 1482 1483 void lock_sock_nested(struct sock *sk, int subclass); 1484 1485 static inline void lock_sock(struct sock *sk) 1486 { 1487 lock_sock_nested(sk, 0); 1488 } 1489 1490 void release_sock(struct sock *sk); 1491 1492 /* BH context may only use the following locking interface. */ 1493 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1494 #define bh_lock_sock_nested(__sk) \ 1495 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1496 SINGLE_DEPTH_NESTING) 1497 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1498 1499 bool lock_sock_fast(struct sock *sk); 1500 /** 1501 * unlock_sock_fast - complement of lock_sock_fast 1502 * @sk: socket 1503 * @slow: slow mode 1504 * 1505 * fast unlock socket for user context. 1506 * If slow mode is on, we call regular release_sock() 1507 */ 1508 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1509 { 1510 if (slow) 1511 release_sock(sk); 1512 else 1513 spin_unlock_bh(&sk->sk_lock.slock); 1514 } 1515 1516 1517 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1518 struct proto *prot, int kern); 1519 void sk_free(struct sock *sk); 1520 void sk_destruct(struct sock *sk); 1521 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1522 1523 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1524 gfp_t priority); 1525 void sock_wfree(struct sk_buff *skb); 1526 void skb_orphan_partial(struct sk_buff *skb); 1527 void sock_rfree(struct sk_buff *skb); 1528 void sock_efree(struct sk_buff *skb); 1529 #ifdef CONFIG_INET 1530 void sock_edemux(struct sk_buff *skb); 1531 #else 1532 #define sock_edemux(skb) sock_efree(skb) 1533 #endif 1534 1535 int sock_setsockopt(struct socket *sock, int level, int op, 1536 char __user *optval, unsigned int optlen); 1537 1538 int sock_getsockopt(struct socket *sock, int level, int op, 1539 char __user *optval, int __user *optlen); 1540 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1541 int noblock, int *errcode); 1542 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1543 unsigned long data_len, int noblock, 1544 int *errcode, int max_page_order); 1545 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1546 void sock_kfree_s(struct sock *sk, void *mem, int size); 1547 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1548 void sk_send_sigurg(struct sock *sk); 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 struct dst_entry * 1691 __sk_dst_get(struct sock *sk) 1692 { 1693 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1694 lockdep_is_held(&sk->sk_lock.slock)); 1695 } 1696 1697 static inline struct dst_entry * 1698 sk_dst_get(struct sock *sk) 1699 { 1700 struct dst_entry *dst; 1701 1702 rcu_read_lock(); 1703 dst = rcu_dereference(sk->sk_dst_cache); 1704 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 1705 dst = NULL; 1706 rcu_read_unlock(); 1707 return dst; 1708 } 1709 1710 static inline void dst_negative_advice(struct sock *sk) 1711 { 1712 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1713 1714 if (dst && dst->ops->negative_advice) { 1715 ndst = dst->ops->negative_advice(dst); 1716 1717 if (ndst != dst) { 1718 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1719 sk_tx_queue_clear(sk); 1720 } 1721 } 1722 } 1723 1724 static inline void 1725 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1726 { 1727 struct dst_entry *old_dst; 1728 1729 sk_tx_queue_clear(sk); 1730 /* 1731 * This can be called while sk is owned by the caller only, 1732 * with no state that can be checked in a rcu_dereference_check() cond 1733 */ 1734 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1735 rcu_assign_pointer(sk->sk_dst_cache, dst); 1736 dst_release(old_dst); 1737 } 1738 1739 static inline void 1740 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1741 { 1742 struct dst_entry *old_dst; 1743 1744 sk_tx_queue_clear(sk); 1745 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 1746 dst_release(old_dst); 1747 } 1748 1749 static inline void 1750 __sk_dst_reset(struct sock *sk) 1751 { 1752 __sk_dst_set(sk, NULL); 1753 } 1754 1755 static inline void 1756 sk_dst_reset(struct sock *sk) 1757 { 1758 sk_dst_set(sk, NULL); 1759 } 1760 1761 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1762 1763 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1764 1765 bool sk_mc_loop(struct sock *sk); 1766 1767 static inline bool sk_can_gso(const struct sock *sk) 1768 { 1769 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1770 } 1771 1772 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1773 1774 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1775 { 1776 sk->sk_route_nocaps |= flags; 1777 sk->sk_route_caps &= ~flags; 1778 } 1779 1780 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1781 struct iov_iter *from, char *to, 1782 int copy, int offset) 1783 { 1784 if (skb->ip_summed == CHECKSUM_NONE) { 1785 __wsum csum = 0; 1786 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy) 1787 return -EFAULT; 1788 skb->csum = csum_block_add(skb->csum, csum, offset); 1789 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1790 if (copy_from_iter_nocache(to, copy, from) != copy) 1791 return -EFAULT; 1792 } else if (copy_from_iter(to, copy, from) != copy) 1793 return -EFAULT; 1794 1795 return 0; 1796 } 1797 1798 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1799 struct iov_iter *from, int copy) 1800 { 1801 int err, offset = skb->len; 1802 1803 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1804 copy, offset); 1805 if (err) 1806 __skb_trim(skb, offset); 1807 1808 return err; 1809 } 1810 1811 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 1812 struct sk_buff *skb, 1813 struct page *page, 1814 int off, int copy) 1815 { 1816 int err; 1817 1818 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1819 copy, skb->len); 1820 if (err) 1821 return err; 1822 1823 skb->len += copy; 1824 skb->data_len += copy; 1825 skb->truesize += copy; 1826 sk->sk_wmem_queued += copy; 1827 sk_mem_charge(sk, copy); 1828 return 0; 1829 } 1830 1831 /** 1832 * sk_wmem_alloc_get - returns write allocations 1833 * @sk: socket 1834 * 1835 * Returns sk_wmem_alloc minus initial offset of one 1836 */ 1837 static inline int sk_wmem_alloc_get(const struct sock *sk) 1838 { 1839 return atomic_read(&sk->sk_wmem_alloc) - 1; 1840 } 1841 1842 /** 1843 * sk_rmem_alloc_get - returns read allocations 1844 * @sk: socket 1845 * 1846 * Returns sk_rmem_alloc 1847 */ 1848 static inline int sk_rmem_alloc_get(const struct sock *sk) 1849 { 1850 return atomic_read(&sk->sk_rmem_alloc); 1851 } 1852 1853 /** 1854 * sk_has_allocations - check if allocations are outstanding 1855 * @sk: socket 1856 * 1857 * Returns true if socket has write or read allocations 1858 */ 1859 static inline bool sk_has_allocations(const struct sock *sk) 1860 { 1861 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1862 } 1863 1864 /** 1865 * wq_has_sleeper - check if there are any waiting processes 1866 * @wq: struct socket_wq 1867 * 1868 * Returns true if socket_wq has waiting processes 1869 * 1870 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1871 * barrier call. They were added due to the race found within the tcp code. 1872 * 1873 * Consider following tcp code paths: 1874 * 1875 * CPU1 CPU2 1876 * 1877 * sys_select receive packet 1878 * ... ... 1879 * __add_wait_queue update tp->rcv_nxt 1880 * ... ... 1881 * tp->rcv_nxt check sock_def_readable 1882 * ... { 1883 * schedule rcu_read_lock(); 1884 * wq = rcu_dereference(sk->sk_wq); 1885 * if (wq && waitqueue_active(&wq->wait)) 1886 * wake_up_interruptible(&wq->wait) 1887 * ... 1888 * } 1889 * 1890 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1891 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1892 * could then endup calling schedule and sleep forever if there are no more 1893 * data on the socket. 1894 * 1895 */ 1896 static inline bool wq_has_sleeper(struct socket_wq *wq) 1897 { 1898 /* We need to be sure we are in sync with the 1899 * add_wait_queue modifications to the wait queue. 1900 * 1901 * This memory barrier is paired in the sock_poll_wait. 1902 */ 1903 smp_mb(); 1904 return wq && waitqueue_active(&wq->wait); 1905 } 1906 1907 /** 1908 * sock_poll_wait - place memory barrier behind the poll_wait call. 1909 * @filp: file 1910 * @wait_address: socket wait queue 1911 * @p: poll_table 1912 * 1913 * See the comments in the wq_has_sleeper function. 1914 */ 1915 static inline void sock_poll_wait(struct file *filp, 1916 wait_queue_head_t *wait_address, poll_table *p) 1917 { 1918 if (!poll_does_not_wait(p) && wait_address) { 1919 poll_wait(filp, wait_address, p); 1920 /* We need to be sure we are in sync with the 1921 * socket flags modification. 1922 * 1923 * This memory barrier is paired in the wq_has_sleeper. 1924 */ 1925 smp_mb(); 1926 } 1927 } 1928 1929 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 1930 { 1931 if (sk->sk_txhash) { 1932 skb->l4_hash = 1; 1933 skb->hash = sk->sk_txhash; 1934 } 1935 } 1936 1937 /* 1938 * Queue a received datagram if it will fit. Stream and sequenced 1939 * protocols can't normally use this as they need to fit buffers in 1940 * and play with them. 1941 * 1942 * Inlined as it's very short and called for pretty much every 1943 * packet ever received. 1944 */ 1945 1946 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1947 { 1948 skb_orphan(skb); 1949 skb->sk = sk; 1950 skb->destructor = sock_wfree; 1951 skb_set_hash_from_sk(skb, sk); 1952 /* 1953 * We used to take a refcount on sk, but following operation 1954 * is enough to guarantee sk_free() wont free this sock until 1955 * all in-flight packets are completed 1956 */ 1957 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1958 } 1959 1960 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1961 { 1962 skb_orphan(skb); 1963 skb->sk = sk; 1964 skb->destructor = sock_rfree; 1965 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1966 sk_mem_charge(sk, skb->truesize); 1967 } 1968 1969 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 1970 unsigned long expires); 1971 1972 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 1973 1974 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1975 1976 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1977 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 1978 1979 /* 1980 * Recover an error report and clear atomically 1981 */ 1982 1983 static inline int sock_error(struct sock *sk) 1984 { 1985 int err; 1986 if (likely(!sk->sk_err)) 1987 return 0; 1988 err = xchg(&sk->sk_err, 0); 1989 return -err; 1990 } 1991 1992 static inline unsigned long sock_wspace(struct sock *sk) 1993 { 1994 int amt = 0; 1995 1996 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1997 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 1998 if (amt < 0) 1999 amt = 0; 2000 } 2001 return amt; 2002 } 2003 2004 static inline void sk_wake_async(struct sock *sk, int how, int band) 2005 { 2006 if (sock_flag(sk, SOCK_FASYNC)) 2007 sock_wake_async(sk->sk_socket, how, band); 2008 } 2009 2010 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2011 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2012 * Note: for send buffers, TCP works better if we can build two skbs at 2013 * minimum. 2014 */ 2015 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2016 2017 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2018 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2019 2020 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2021 { 2022 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2023 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2024 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2025 } 2026 } 2027 2028 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2029 bool force_schedule); 2030 2031 /** 2032 * sk_page_frag - return an appropriate page_frag 2033 * @sk: socket 2034 * 2035 * If socket allocation mode allows current thread to sleep, it means its 2036 * safe to use the per task page_frag instead of the per socket one. 2037 */ 2038 static inline struct page_frag *sk_page_frag(struct sock *sk) 2039 { 2040 if (sk->sk_allocation & __GFP_WAIT) 2041 return ¤t->task_frag; 2042 2043 return &sk->sk_frag; 2044 } 2045 2046 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2047 2048 /* 2049 * Default write policy as shown to user space via poll/select/SIGIO 2050 */ 2051 static inline bool sock_writeable(const struct sock *sk) 2052 { 2053 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2054 } 2055 2056 static inline gfp_t gfp_any(void) 2057 { 2058 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2059 } 2060 2061 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2062 { 2063 return noblock ? 0 : sk->sk_rcvtimeo; 2064 } 2065 2066 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2067 { 2068 return noblock ? 0 : sk->sk_sndtimeo; 2069 } 2070 2071 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2072 { 2073 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2074 } 2075 2076 /* Alas, with timeout socket operations are not restartable. 2077 * Compare this to poll(). 2078 */ 2079 static inline int sock_intr_errno(long timeo) 2080 { 2081 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2082 } 2083 2084 struct sock_skb_cb { 2085 u32 dropcount; 2086 }; 2087 2088 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2089 * using skb->cb[] would keep using it directly and utilize its 2090 * alignement guarantee. 2091 */ 2092 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \ 2093 sizeof(struct sock_skb_cb))) 2094 2095 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2096 SOCK_SKB_CB_OFFSET)) 2097 2098 #define sock_skb_cb_check_size(size) \ 2099 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2100 2101 static inline void 2102 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2103 { 2104 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops); 2105 } 2106 2107 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2108 struct sk_buff *skb); 2109 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2110 struct sk_buff *skb); 2111 2112 static inline void 2113 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2114 { 2115 ktime_t kt = skb->tstamp; 2116 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2117 2118 /* 2119 * generate control messages if 2120 * - receive time stamping in software requested 2121 * - software time stamp available and wanted 2122 * - hardware time stamps available and wanted 2123 */ 2124 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2125 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2126 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2127 (hwtstamps->hwtstamp.tv64 && 2128 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2129 __sock_recv_timestamp(msg, sk, skb); 2130 else 2131 sk->sk_stamp = kt; 2132 2133 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2134 __sock_recv_wifi_status(msg, sk, skb); 2135 } 2136 2137 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2138 struct sk_buff *skb); 2139 2140 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2141 struct sk_buff *skb) 2142 { 2143 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2144 (1UL << SOCK_RCVTSTAMP)) 2145 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2146 SOF_TIMESTAMPING_RAW_HARDWARE) 2147 2148 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2149 __sock_recv_ts_and_drops(msg, sk, skb); 2150 else 2151 sk->sk_stamp = skb->tstamp; 2152 } 2153 2154 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags); 2155 2156 /** 2157 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2158 * @sk: socket sending this packet 2159 * @tx_flags: completed with instructions for time stamping 2160 * 2161 * Note : callers should take care of initial *tx_flags value (usually 0) 2162 */ 2163 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags) 2164 { 2165 if (unlikely(sk->sk_tsflags)) 2166 __sock_tx_timestamp(sk, tx_flags); 2167 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2168 *tx_flags |= SKBTX_WIFI_STATUS; 2169 } 2170 2171 /** 2172 * sk_eat_skb - Release a skb if it is no longer needed 2173 * @sk: socket to eat this skb from 2174 * @skb: socket buffer to eat 2175 * 2176 * This routine must be called with interrupts disabled or with the socket 2177 * locked so that the sk_buff queue operation is ok. 2178 */ 2179 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2180 { 2181 __skb_unlink(skb, &sk->sk_receive_queue); 2182 __kfree_skb(skb); 2183 } 2184 2185 static inline 2186 struct net *sock_net(const struct sock *sk) 2187 { 2188 return read_pnet(&sk->sk_net); 2189 } 2190 2191 static inline 2192 void sock_net_set(struct sock *sk, struct net *net) 2193 { 2194 write_pnet(&sk->sk_net, net); 2195 } 2196 2197 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2198 { 2199 if (skb->sk) { 2200 struct sock *sk = skb->sk; 2201 2202 skb->destructor = NULL; 2203 skb->sk = NULL; 2204 return sk; 2205 } 2206 return NULL; 2207 } 2208 2209 /* This helper checks if a socket is a full socket, 2210 * ie _not_ a timewait or request socket. 2211 */ 2212 static inline bool sk_fullsock(const struct sock *sk) 2213 { 2214 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2215 } 2216 2217 void sock_enable_timestamp(struct sock *sk, int flag); 2218 int sock_get_timestamp(struct sock *, struct timeval __user *); 2219 int sock_get_timestampns(struct sock *, struct timespec __user *); 2220 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2221 int type); 2222 2223 bool sk_ns_capable(const struct sock *sk, 2224 struct user_namespace *user_ns, int cap); 2225 bool sk_capable(const struct sock *sk, int cap); 2226 bool sk_net_capable(const struct sock *sk, int cap); 2227 2228 extern __u32 sysctl_wmem_max; 2229 extern __u32 sysctl_rmem_max; 2230 2231 extern int sysctl_tstamp_allow_data; 2232 extern int sysctl_optmem_max; 2233 2234 extern __u32 sysctl_wmem_default; 2235 extern __u32 sysctl_rmem_default; 2236 2237 #endif /* _SOCK_H */ 2238