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/memcontrol.h> 58 #include <linux/res_counter.h> 59 #include <linux/static_key.h> 60 #include <linux/aio.h> 61 #include <linux/sched.h> 62 63 #include <linux/filter.h> 64 #include <linux/rculist_nulls.h> 65 #include <linux/poll.h> 66 67 #include <linux/atomic.h> 68 #include <net/dst.h> 69 #include <net/checksum.h> 70 71 struct cgroup; 72 struct cgroup_subsys; 73 #ifdef CONFIG_NET 74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss); 75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg); 76 #else 77 static inline 78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss) 79 { 80 return 0; 81 } 82 static inline 83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg) 84 { 85 } 86 #endif 87 /* 88 * This structure really needs to be cleaned up. 89 * Most of it is for TCP, and not used by any of 90 * the other protocols. 91 */ 92 93 /* Define this to get the SOCK_DBG debugging facility. */ 94 #define SOCK_DEBUGGING 95 #ifdef SOCK_DEBUGGING 96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 97 printk(KERN_DEBUG msg); } while (0) 98 #else 99 /* Validate arguments and do nothing */ 100 static inline __printf(2, 3) 101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 102 { 103 } 104 #endif 105 106 /* This is the per-socket lock. The spinlock provides a synchronization 107 * between user contexts and software interrupt processing, whereas the 108 * mini-semaphore synchronizes multiple users amongst themselves. 109 */ 110 typedef struct { 111 spinlock_t slock; 112 int owned; 113 wait_queue_head_t wq; 114 /* 115 * We express the mutex-alike socket_lock semantics 116 * to the lock validator by explicitly managing 117 * the slock as a lock variant (in addition to 118 * the slock itself): 119 */ 120 #ifdef CONFIG_DEBUG_LOCK_ALLOC 121 struct lockdep_map dep_map; 122 #endif 123 } socket_lock_t; 124 125 struct sock; 126 struct proto; 127 struct net; 128 129 typedef __u32 __bitwise __portpair; 130 typedef __u64 __bitwise __addrpair; 131 132 /** 133 * struct sock_common - minimal network layer representation of sockets 134 * @skc_daddr: Foreign IPv4 addr 135 * @skc_rcv_saddr: Bound local IPv4 addr 136 * @skc_hash: hash value used with various protocol lookup tables 137 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 138 * @skc_dport: placeholder for inet_dport/tw_dport 139 * @skc_num: placeholder for inet_num/tw_num 140 * @skc_family: network address family 141 * @skc_state: Connection state 142 * @skc_reuse: %SO_REUSEADDR setting 143 * @skc_reuseport: %SO_REUSEPORT setting 144 * @skc_bound_dev_if: bound device index if != 0 145 * @skc_bind_node: bind hash linkage for various protocol lookup tables 146 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 147 * @skc_prot: protocol handlers inside a network family 148 * @skc_net: reference to the network namespace of this socket 149 * @skc_node: main hash linkage for various protocol lookup tables 150 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 151 * @skc_tx_queue_mapping: tx queue number for this connection 152 * @skc_refcnt: reference count 153 * 154 * This is the minimal network layer representation of sockets, the header 155 * for struct sock and struct inet_timewait_sock. 156 */ 157 struct sock_common { 158 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 159 * address on 64bit arches : cf INET_MATCH() 160 */ 161 union { 162 __addrpair skc_addrpair; 163 struct { 164 __be32 skc_daddr; 165 __be32 skc_rcv_saddr; 166 }; 167 }; 168 union { 169 unsigned int skc_hash; 170 __u16 skc_u16hashes[2]; 171 }; 172 /* skc_dport && skc_num must be grouped as well */ 173 union { 174 __portpair skc_portpair; 175 struct { 176 __be16 skc_dport; 177 __u16 skc_num; 178 }; 179 }; 180 181 unsigned short skc_family; 182 volatile unsigned char skc_state; 183 unsigned char skc_reuse:4; 184 unsigned char skc_reuseport:4; 185 int skc_bound_dev_if; 186 union { 187 struct hlist_node skc_bind_node; 188 struct hlist_nulls_node skc_portaddr_node; 189 }; 190 struct proto *skc_prot; 191 #ifdef CONFIG_NET_NS 192 struct net *skc_net; 193 #endif 194 195 #if IS_ENABLED(CONFIG_IPV6) 196 struct in6_addr skc_v6_daddr; 197 struct in6_addr skc_v6_rcv_saddr; 198 #endif 199 200 /* 201 * fields between dontcopy_begin/dontcopy_end 202 * are not copied in sock_copy() 203 */ 204 /* private: */ 205 int skc_dontcopy_begin[0]; 206 /* public: */ 207 union { 208 struct hlist_node skc_node; 209 struct hlist_nulls_node skc_nulls_node; 210 }; 211 int skc_tx_queue_mapping; 212 atomic_t skc_refcnt; 213 /* private: */ 214 int skc_dontcopy_end[0]; 215 /* public: */ 216 }; 217 218 struct cg_proto; 219 /** 220 * struct sock - network layer representation of sockets 221 * @__sk_common: shared layout with inet_timewait_sock 222 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 223 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 224 * @sk_lock: synchronizer 225 * @sk_rcvbuf: size of receive buffer in bytes 226 * @sk_wq: sock wait queue and async head 227 * @sk_rx_dst: receive input route used by early demux 228 * @sk_dst_cache: destination cache 229 * @sk_dst_lock: destination cache lock 230 * @sk_policy: flow policy 231 * @sk_receive_queue: incoming packets 232 * @sk_wmem_alloc: transmit queue bytes committed 233 * @sk_write_queue: Packet sending queue 234 * @sk_async_wait_queue: DMA copied packets 235 * @sk_omem_alloc: "o" is "option" or "other" 236 * @sk_wmem_queued: persistent queue size 237 * @sk_forward_alloc: space allocated forward 238 * @sk_napi_id: id of the last napi context to receive data for sk 239 * @sk_ll_usec: usecs to busypoll when there is no data 240 * @sk_allocation: allocation mode 241 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 242 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 243 * @sk_sndbuf: size of send buffer in bytes 244 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 245 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 246 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 247 * @sk_no_check_rx: allow zero checksum in RX packets 248 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 249 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 250 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 251 * @sk_gso_max_size: Maximum GSO segment size to build 252 * @sk_gso_max_segs: Maximum number of GSO segments 253 * @sk_lingertime: %SO_LINGER l_linger setting 254 * @sk_backlog: always used with the per-socket spinlock held 255 * @sk_callback_lock: used with the callbacks in the end of this struct 256 * @sk_error_queue: rarely used 257 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 258 * IPV6_ADDRFORM for instance) 259 * @sk_err: last error 260 * @sk_err_soft: errors that don't cause failure but are the cause of a 261 * persistent failure not just 'timed out' 262 * @sk_drops: raw/udp drops counter 263 * @sk_ack_backlog: current listen backlog 264 * @sk_max_ack_backlog: listen backlog set in listen() 265 * @sk_priority: %SO_PRIORITY setting 266 * @sk_cgrp_prioidx: socket group's priority map index 267 * @sk_type: socket type (%SOCK_STREAM, etc) 268 * @sk_protocol: which protocol this socket belongs in this network family 269 * @sk_peer_pid: &struct pid for this socket's peer 270 * @sk_peer_cred: %SO_PEERCRED setting 271 * @sk_rcvlowat: %SO_RCVLOWAT setting 272 * @sk_rcvtimeo: %SO_RCVTIMEO setting 273 * @sk_sndtimeo: %SO_SNDTIMEO setting 274 * @sk_rxhash: flow hash received from netif layer 275 * @sk_filter: socket filtering instructions 276 * @sk_protinfo: private area, net family specific, when not using slab 277 * @sk_timer: sock cleanup timer 278 * @sk_stamp: time stamp of last packet received 279 * @sk_socket: Identd and reporting IO signals 280 * @sk_user_data: RPC layer private data 281 * @sk_frag: cached page frag 282 * @sk_peek_off: current peek_offset value 283 * @sk_send_head: front of stuff to transmit 284 * @sk_security: used by security modules 285 * @sk_mark: generic packet mark 286 * @sk_classid: this socket's cgroup classid 287 * @sk_cgrp: this socket's cgroup-specific proto data 288 * @sk_write_pending: a write to stream socket waits to start 289 * @sk_state_change: callback to indicate change in the state of the sock 290 * @sk_data_ready: callback to indicate there is data to be processed 291 * @sk_write_space: callback to indicate there is bf sending space available 292 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 293 * @sk_backlog_rcv: callback to process the backlog 294 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 295 */ 296 struct sock { 297 /* 298 * Now struct inet_timewait_sock also uses sock_common, so please just 299 * don't add nothing before this first member (__sk_common) --acme 300 */ 301 struct sock_common __sk_common; 302 #define sk_node __sk_common.skc_node 303 #define sk_nulls_node __sk_common.skc_nulls_node 304 #define sk_refcnt __sk_common.skc_refcnt 305 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 306 307 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 308 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 309 #define sk_hash __sk_common.skc_hash 310 #define sk_portpair __sk_common.skc_portpair 311 #define sk_num __sk_common.skc_num 312 #define sk_dport __sk_common.skc_dport 313 #define sk_addrpair __sk_common.skc_addrpair 314 #define sk_daddr __sk_common.skc_daddr 315 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 316 #define sk_family __sk_common.skc_family 317 #define sk_state __sk_common.skc_state 318 #define sk_reuse __sk_common.skc_reuse 319 #define sk_reuseport __sk_common.skc_reuseport 320 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 321 #define sk_bind_node __sk_common.skc_bind_node 322 #define sk_prot __sk_common.skc_prot 323 #define sk_net __sk_common.skc_net 324 #define sk_v6_daddr __sk_common.skc_v6_daddr 325 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 326 327 socket_lock_t sk_lock; 328 struct sk_buff_head sk_receive_queue; 329 /* 330 * The backlog queue is special, it is always used with 331 * the per-socket spinlock held and requires low latency 332 * access. Therefore we special case it's implementation. 333 * Note : rmem_alloc is in this structure to fill a hole 334 * on 64bit arches, not because its logically part of 335 * backlog. 336 */ 337 struct { 338 atomic_t rmem_alloc; 339 int len; 340 struct sk_buff *head; 341 struct sk_buff *tail; 342 } sk_backlog; 343 #define sk_rmem_alloc sk_backlog.rmem_alloc 344 int sk_forward_alloc; 345 #ifdef CONFIG_RPS 346 __u32 sk_rxhash; 347 #endif 348 #ifdef CONFIG_NET_RX_BUSY_POLL 349 unsigned int sk_napi_id; 350 unsigned int sk_ll_usec; 351 #endif 352 atomic_t sk_drops; 353 int sk_rcvbuf; 354 355 struct sk_filter __rcu *sk_filter; 356 struct socket_wq __rcu *sk_wq; 357 358 #ifdef CONFIG_NET_DMA 359 struct sk_buff_head sk_async_wait_queue; 360 #endif 361 362 #ifdef CONFIG_XFRM 363 struct xfrm_policy *sk_policy[2]; 364 #endif 365 unsigned long sk_flags; 366 struct dst_entry *sk_rx_dst; 367 struct dst_entry __rcu *sk_dst_cache; 368 spinlock_t sk_dst_lock; 369 atomic_t sk_wmem_alloc; 370 atomic_t sk_omem_alloc; 371 int sk_sndbuf; 372 struct sk_buff_head sk_write_queue; 373 kmemcheck_bitfield_begin(flags); 374 unsigned int sk_shutdown : 2, 375 sk_no_check_tx : 1, 376 sk_no_check_rx : 1, 377 sk_userlocks : 4, 378 sk_protocol : 8, 379 sk_type : 16; 380 kmemcheck_bitfield_end(flags); 381 int sk_wmem_queued; 382 gfp_t sk_allocation; 383 u32 sk_pacing_rate; /* bytes per second */ 384 u32 sk_max_pacing_rate; 385 netdev_features_t sk_route_caps; 386 netdev_features_t sk_route_nocaps; 387 int sk_gso_type; 388 unsigned int sk_gso_max_size; 389 u16 sk_gso_max_segs; 390 int sk_rcvlowat; 391 unsigned long sk_lingertime; 392 struct sk_buff_head sk_error_queue; 393 struct proto *sk_prot_creator; 394 rwlock_t sk_callback_lock; 395 int sk_err, 396 sk_err_soft; 397 unsigned short sk_ack_backlog; 398 unsigned short sk_max_ack_backlog; 399 __u32 sk_priority; 400 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) 401 __u32 sk_cgrp_prioidx; 402 #endif 403 struct pid *sk_peer_pid; 404 const struct cred *sk_peer_cred; 405 long sk_rcvtimeo; 406 long sk_sndtimeo; 407 void *sk_protinfo; 408 struct timer_list sk_timer; 409 ktime_t sk_stamp; 410 struct socket *sk_socket; 411 void *sk_user_data; 412 struct page_frag sk_frag; 413 struct sk_buff *sk_send_head; 414 __s32 sk_peek_off; 415 int sk_write_pending; 416 #ifdef CONFIG_SECURITY 417 void *sk_security; 418 #endif 419 __u32 sk_mark; 420 u32 sk_classid; 421 struct cg_proto *sk_cgrp; 422 void (*sk_state_change)(struct sock *sk); 423 void (*sk_data_ready)(struct sock *sk); 424 void (*sk_write_space)(struct sock *sk); 425 void (*sk_error_report)(struct sock *sk); 426 int (*sk_backlog_rcv)(struct sock *sk, 427 struct sk_buff *skb); 428 void (*sk_destruct)(struct sock *sk); 429 }; 430 431 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 432 433 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 434 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 435 436 /* 437 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 438 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 439 * on a socket means that the socket will reuse everybody else's port 440 * without looking at the other's sk_reuse value. 441 */ 442 443 #define SK_NO_REUSE 0 444 #define SK_CAN_REUSE 1 445 #define SK_FORCE_REUSE 2 446 447 static inline int sk_peek_offset(struct sock *sk, int flags) 448 { 449 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 450 return sk->sk_peek_off; 451 else 452 return 0; 453 } 454 455 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 456 { 457 if (sk->sk_peek_off >= 0) { 458 if (sk->sk_peek_off >= val) 459 sk->sk_peek_off -= val; 460 else 461 sk->sk_peek_off = 0; 462 } 463 } 464 465 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 466 { 467 if (sk->sk_peek_off >= 0) 468 sk->sk_peek_off += val; 469 } 470 471 /* 472 * Hashed lists helper routines 473 */ 474 static inline struct sock *sk_entry(const struct hlist_node *node) 475 { 476 return hlist_entry(node, struct sock, sk_node); 477 } 478 479 static inline struct sock *__sk_head(const struct hlist_head *head) 480 { 481 return hlist_entry(head->first, struct sock, sk_node); 482 } 483 484 static inline struct sock *sk_head(const struct hlist_head *head) 485 { 486 return hlist_empty(head) ? NULL : __sk_head(head); 487 } 488 489 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 490 { 491 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 492 } 493 494 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 495 { 496 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 497 } 498 499 static inline struct sock *sk_next(const struct sock *sk) 500 { 501 return sk->sk_node.next ? 502 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 503 } 504 505 static inline struct sock *sk_nulls_next(const struct sock *sk) 506 { 507 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 508 hlist_nulls_entry(sk->sk_nulls_node.next, 509 struct sock, sk_nulls_node) : 510 NULL; 511 } 512 513 static inline bool sk_unhashed(const struct sock *sk) 514 { 515 return hlist_unhashed(&sk->sk_node); 516 } 517 518 static inline bool sk_hashed(const struct sock *sk) 519 { 520 return !sk_unhashed(sk); 521 } 522 523 static inline void sk_node_init(struct hlist_node *node) 524 { 525 node->pprev = NULL; 526 } 527 528 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 529 { 530 node->pprev = NULL; 531 } 532 533 static inline void __sk_del_node(struct sock *sk) 534 { 535 __hlist_del(&sk->sk_node); 536 } 537 538 /* NB: equivalent to hlist_del_init_rcu */ 539 static inline bool __sk_del_node_init(struct sock *sk) 540 { 541 if (sk_hashed(sk)) { 542 __sk_del_node(sk); 543 sk_node_init(&sk->sk_node); 544 return true; 545 } 546 return false; 547 } 548 549 /* Grab socket reference count. This operation is valid only 550 when sk is ALREADY grabbed f.e. it is found in hash table 551 or a list and the lookup is made under lock preventing hash table 552 modifications. 553 */ 554 555 static inline void sock_hold(struct sock *sk) 556 { 557 atomic_inc(&sk->sk_refcnt); 558 } 559 560 /* Ungrab socket in the context, which assumes that socket refcnt 561 cannot hit zero, f.e. it is true in context of any socketcall. 562 */ 563 static inline void __sock_put(struct sock *sk) 564 { 565 atomic_dec(&sk->sk_refcnt); 566 } 567 568 static inline bool sk_del_node_init(struct sock *sk) 569 { 570 bool rc = __sk_del_node_init(sk); 571 572 if (rc) { 573 /* paranoid for a while -acme */ 574 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 575 __sock_put(sk); 576 } 577 return rc; 578 } 579 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 580 581 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 582 { 583 if (sk_hashed(sk)) { 584 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 585 return true; 586 } 587 return false; 588 } 589 590 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 591 { 592 bool rc = __sk_nulls_del_node_init_rcu(sk); 593 594 if (rc) { 595 /* paranoid for a while -acme */ 596 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 597 __sock_put(sk); 598 } 599 return rc; 600 } 601 602 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 603 { 604 hlist_add_head(&sk->sk_node, list); 605 } 606 607 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 608 { 609 sock_hold(sk); 610 __sk_add_node(sk, list); 611 } 612 613 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 614 { 615 sock_hold(sk); 616 hlist_add_head_rcu(&sk->sk_node, list); 617 } 618 619 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 620 { 621 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 622 } 623 624 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 625 { 626 sock_hold(sk); 627 __sk_nulls_add_node_rcu(sk, list); 628 } 629 630 static inline void __sk_del_bind_node(struct sock *sk) 631 { 632 __hlist_del(&sk->sk_bind_node); 633 } 634 635 static inline void sk_add_bind_node(struct sock *sk, 636 struct hlist_head *list) 637 { 638 hlist_add_head(&sk->sk_bind_node, list); 639 } 640 641 #define sk_for_each(__sk, list) \ 642 hlist_for_each_entry(__sk, list, sk_node) 643 #define sk_for_each_rcu(__sk, list) \ 644 hlist_for_each_entry_rcu(__sk, list, sk_node) 645 #define sk_nulls_for_each(__sk, node, list) \ 646 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 647 #define sk_nulls_for_each_rcu(__sk, node, list) \ 648 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 649 #define sk_for_each_from(__sk) \ 650 hlist_for_each_entry_from(__sk, sk_node) 651 #define sk_nulls_for_each_from(__sk, node) \ 652 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 653 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 654 #define sk_for_each_safe(__sk, tmp, list) \ 655 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 656 #define sk_for_each_bound(__sk, list) \ 657 hlist_for_each_entry(__sk, list, sk_bind_node) 658 659 static inline struct user_namespace *sk_user_ns(struct sock *sk) 660 { 661 /* Careful only use this in a context where these parameters 662 * can not change and must all be valid, such as recvmsg from 663 * userspace. 664 */ 665 return sk->sk_socket->file->f_cred->user_ns; 666 } 667 668 /* Sock flags */ 669 enum sock_flags { 670 SOCK_DEAD, 671 SOCK_DONE, 672 SOCK_URGINLINE, 673 SOCK_KEEPOPEN, 674 SOCK_LINGER, 675 SOCK_DESTROY, 676 SOCK_BROADCAST, 677 SOCK_TIMESTAMP, 678 SOCK_ZAPPED, 679 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 680 SOCK_DBG, /* %SO_DEBUG setting */ 681 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 682 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 683 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 684 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 685 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 686 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 687 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 688 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 689 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 690 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 691 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 692 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 693 SOCK_FASYNC, /* fasync() active */ 694 SOCK_RXQ_OVFL, 695 SOCK_ZEROCOPY, /* buffers from userspace */ 696 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 697 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 698 * Will use last 4 bytes of packet sent from 699 * user-space instead. 700 */ 701 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 702 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 703 }; 704 705 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 706 { 707 nsk->sk_flags = osk->sk_flags; 708 } 709 710 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 711 { 712 __set_bit(flag, &sk->sk_flags); 713 } 714 715 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 716 { 717 __clear_bit(flag, &sk->sk_flags); 718 } 719 720 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 721 { 722 return test_bit(flag, &sk->sk_flags); 723 } 724 725 #ifdef CONFIG_NET 726 extern struct static_key memalloc_socks; 727 static inline int sk_memalloc_socks(void) 728 { 729 return static_key_false(&memalloc_socks); 730 } 731 #else 732 733 static inline int sk_memalloc_socks(void) 734 { 735 return 0; 736 } 737 738 #endif 739 740 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 741 { 742 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 743 } 744 745 static inline void sk_acceptq_removed(struct sock *sk) 746 { 747 sk->sk_ack_backlog--; 748 } 749 750 static inline void sk_acceptq_added(struct sock *sk) 751 { 752 sk->sk_ack_backlog++; 753 } 754 755 static inline bool sk_acceptq_is_full(const struct sock *sk) 756 { 757 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 758 } 759 760 /* 761 * Compute minimal free write space needed to queue new packets. 762 */ 763 static inline int sk_stream_min_wspace(const struct sock *sk) 764 { 765 return sk->sk_wmem_queued >> 1; 766 } 767 768 static inline int sk_stream_wspace(const struct sock *sk) 769 { 770 return sk->sk_sndbuf - sk->sk_wmem_queued; 771 } 772 773 void sk_stream_write_space(struct sock *sk); 774 775 /* OOB backlog add */ 776 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 777 { 778 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 779 skb_dst_force(skb); 780 781 if (!sk->sk_backlog.tail) 782 sk->sk_backlog.head = skb; 783 else 784 sk->sk_backlog.tail->next = skb; 785 786 sk->sk_backlog.tail = skb; 787 skb->next = NULL; 788 } 789 790 /* 791 * Take into account size of receive queue and backlog queue 792 * Do not take into account this skb truesize, 793 * to allow even a single big packet to come. 794 */ 795 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb, 796 unsigned int limit) 797 { 798 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 799 800 return qsize > limit; 801 } 802 803 /* The per-socket spinlock must be held here. */ 804 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 805 unsigned int limit) 806 { 807 if (sk_rcvqueues_full(sk, skb, limit)) 808 return -ENOBUFS; 809 810 __sk_add_backlog(sk, skb); 811 sk->sk_backlog.len += skb->truesize; 812 return 0; 813 } 814 815 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 816 817 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 818 { 819 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 820 return __sk_backlog_rcv(sk, skb); 821 822 return sk->sk_backlog_rcv(sk, skb); 823 } 824 825 static inline void sock_rps_record_flow_hash(__u32 hash) 826 { 827 #ifdef CONFIG_RPS 828 struct rps_sock_flow_table *sock_flow_table; 829 830 rcu_read_lock(); 831 sock_flow_table = rcu_dereference(rps_sock_flow_table); 832 rps_record_sock_flow(sock_flow_table, hash); 833 rcu_read_unlock(); 834 #endif 835 } 836 837 static inline void sock_rps_reset_flow_hash(__u32 hash) 838 { 839 #ifdef CONFIG_RPS 840 struct rps_sock_flow_table *sock_flow_table; 841 842 rcu_read_lock(); 843 sock_flow_table = rcu_dereference(rps_sock_flow_table); 844 rps_reset_sock_flow(sock_flow_table, hash); 845 rcu_read_unlock(); 846 #endif 847 } 848 849 static inline void sock_rps_record_flow(const struct sock *sk) 850 { 851 #ifdef CONFIG_RPS 852 sock_rps_record_flow_hash(sk->sk_rxhash); 853 #endif 854 } 855 856 static inline void sock_rps_reset_flow(const struct sock *sk) 857 { 858 #ifdef CONFIG_RPS 859 sock_rps_reset_flow_hash(sk->sk_rxhash); 860 #endif 861 } 862 863 static inline void sock_rps_save_rxhash(struct sock *sk, 864 const struct sk_buff *skb) 865 { 866 #ifdef CONFIG_RPS 867 if (unlikely(sk->sk_rxhash != skb->hash)) { 868 sock_rps_reset_flow(sk); 869 sk->sk_rxhash = skb->hash; 870 } 871 #endif 872 } 873 874 static inline void sock_rps_reset_rxhash(struct sock *sk) 875 { 876 #ifdef CONFIG_RPS 877 sock_rps_reset_flow(sk); 878 sk->sk_rxhash = 0; 879 #endif 880 } 881 882 #define sk_wait_event(__sk, __timeo, __condition) \ 883 ({ int __rc; \ 884 release_sock(__sk); \ 885 __rc = __condition; \ 886 if (!__rc) { \ 887 *(__timeo) = schedule_timeout(*(__timeo)); \ 888 } \ 889 lock_sock(__sk); \ 890 __rc = __condition; \ 891 __rc; \ 892 }) 893 894 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 895 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 896 void sk_stream_wait_close(struct sock *sk, long timeo_p); 897 int sk_stream_error(struct sock *sk, int flags, int err); 898 void sk_stream_kill_queues(struct sock *sk); 899 void sk_set_memalloc(struct sock *sk); 900 void sk_clear_memalloc(struct sock *sk); 901 902 int sk_wait_data(struct sock *sk, long *timeo); 903 904 struct request_sock_ops; 905 struct timewait_sock_ops; 906 struct inet_hashinfo; 907 struct raw_hashinfo; 908 struct module; 909 910 /* 911 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 912 * un-modified. Special care is taken when initializing object to zero. 913 */ 914 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 915 { 916 if (offsetof(struct sock, sk_node.next) != 0) 917 memset(sk, 0, offsetof(struct sock, sk_node.next)); 918 memset(&sk->sk_node.pprev, 0, 919 size - offsetof(struct sock, sk_node.pprev)); 920 } 921 922 /* Networking protocol blocks we attach to sockets. 923 * socket layer -> transport layer interface 924 * transport -> network interface is defined by struct inet_proto 925 */ 926 struct proto { 927 void (*close)(struct sock *sk, 928 long timeout); 929 int (*connect)(struct sock *sk, 930 struct sockaddr *uaddr, 931 int addr_len); 932 int (*disconnect)(struct sock *sk, int flags); 933 934 struct sock * (*accept)(struct sock *sk, int flags, int *err); 935 936 int (*ioctl)(struct sock *sk, int cmd, 937 unsigned long arg); 938 int (*init)(struct sock *sk); 939 void (*destroy)(struct sock *sk); 940 void (*shutdown)(struct sock *sk, int how); 941 int (*setsockopt)(struct sock *sk, int level, 942 int optname, char __user *optval, 943 unsigned int optlen); 944 int (*getsockopt)(struct sock *sk, int level, 945 int optname, char __user *optval, 946 int __user *option); 947 #ifdef CONFIG_COMPAT 948 int (*compat_setsockopt)(struct sock *sk, 949 int level, 950 int optname, char __user *optval, 951 unsigned int optlen); 952 int (*compat_getsockopt)(struct sock *sk, 953 int level, 954 int optname, char __user *optval, 955 int __user *option); 956 int (*compat_ioctl)(struct sock *sk, 957 unsigned int cmd, unsigned long arg); 958 #endif 959 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 960 struct msghdr *msg, size_t len); 961 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 962 struct msghdr *msg, 963 size_t len, int noblock, int flags, 964 int *addr_len); 965 int (*sendpage)(struct sock *sk, struct page *page, 966 int offset, size_t size, int flags); 967 int (*bind)(struct sock *sk, 968 struct sockaddr *uaddr, int addr_len); 969 970 int (*backlog_rcv) (struct sock *sk, 971 struct sk_buff *skb); 972 973 void (*release_cb)(struct sock *sk); 974 void (*mtu_reduced)(struct sock *sk); 975 976 /* Keeping track of sk's, looking them up, and port selection methods. */ 977 void (*hash)(struct sock *sk); 978 void (*unhash)(struct sock *sk); 979 void (*rehash)(struct sock *sk); 980 int (*get_port)(struct sock *sk, unsigned short snum); 981 void (*clear_sk)(struct sock *sk, int size); 982 983 /* Keeping track of sockets in use */ 984 #ifdef CONFIG_PROC_FS 985 unsigned int inuse_idx; 986 #endif 987 988 bool (*stream_memory_free)(const struct sock *sk); 989 /* Memory pressure */ 990 void (*enter_memory_pressure)(struct sock *sk); 991 atomic_long_t *memory_allocated; /* Current allocated memory. */ 992 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 993 /* 994 * Pressure flag: try to collapse. 995 * Technical note: it is used by multiple contexts non atomically. 996 * All the __sk_mem_schedule() is of this nature: accounting 997 * is strict, actions are advisory and have some latency. 998 */ 999 int *memory_pressure; 1000 long *sysctl_mem; 1001 int *sysctl_wmem; 1002 int *sysctl_rmem; 1003 int max_header; 1004 bool no_autobind; 1005 1006 struct kmem_cache *slab; 1007 unsigned int obj_size; 1008 int slab_flags; 1009 1010 struct percpu_counter *orphan_count; 1011 1012 struct request_sock_ops *rsk_prot; 1013 struct timewait_sock_ops *twsk_prot; 1014 1015 union { 1016 struct inet_hashinfo *hashinfo; 1017 struct udp_table *udp_table; 1018 struct raw_hashinfo *raw_hash; 1019 } h; 1020 1021 struct module *owner; 1022 1023 char name[32]; 1024 1025 struct list_head node; 1026 #ifdef SOCK_REFCNT_DEBUG 1027 atomic_t socks; 1028 #endif 1029 #ifdef CONFIG_MEMCG_KMEM 1030 /* 1031 * cgroup specific init/deinit functions. Called once for all 1032 * protocols that implement it, from cgroups populate function. 1033 * This function has to setup any files the protocol want to 1034 * appear in the kmem cgroup filesystem. 1035 */ 1036 int (*init_cgroup)(struct mem_cgroup *memcg, 1037 struct cgroup_subsys *ss); 1038 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1039 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1040 #endif 1041 }; 1042 1043 /* 1044 * Bits in struct cg_proto.flags 1045 */ 1046 enum cg_proto_flags { 1047 /* Currently active and new sockets should be assigned to cgroups */ 1048 MEMCG_SOCK_ACTIVE, 1049 /* It was ever activated; we must disarm static keys on destruction */ 1050 MEMCG_SOCK_ACTIVATED, 1051 }; 1052 1053 struct cg_proto { 1054 struct res_counter memory_allocated; /* Current allocated memory. */ 1055 struct percpu_counter sockets_allocated; /* Current number of sockets. */ 1056 int memory_pressure; 1057 long sysctl_mem[3]; 1058 unsigned long flags; 1059 /* 1060 * memcg field is used to find which memcg we belong directly 1061 * Each memcg struct can hold more than one cg_proto, so container_of 1062 * won't really cut. 1063 * 1064 * The elegant solution would be having an inverse function to 1065 * proto_cgroup in struct proto, but that means polluting the structure 1066 * for everybody, instead of just for memcg users. 1067 */ 1068 struct mem_cgroup *memcg; 1069 }; 1070 1071 int proto_register(struct proto *prot, int alloc_slab); 1072 void proto_unregister(struct proto *prot); 1073 1074 static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1075 { 1076 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1077 } 1078 1079 static inline bool memcg_proto_activated(struct cg_proto *cg_proto) 1080 { 1081 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags); 1082 } 1083 1084 #ifdef SOCK_REFCNT_DEBUG 1085 static inline void sk_refcnt_debug_inc(struct sock *sk) 1086 { 1087 atomic_inc(&sk->sk_prot->socks); 1088 } 1089 1090 static inline void sk_refcnt_debug_dec(struct sock *sk) 1091 { 1092 atomic_dec(&sk->sk_prot->socks); 1093 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1094 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1095 } 1096 1097 static inline void sk_refcnt_debug_release(const struct sock *sk) 1098 { 1099 if (atomic_read(&sk->sk_refcnt) != 1) 1100 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1101 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1102 } 1103 #else /* SOCK_REFCNT_DEBUG */ 1104 #define sk_refcnt_debug_inc(sk) do { } while (0) 1105 #define sk_refcnt_debug_dec(sk) do { } while (0) 1106 #define sk_refcnt_debug_release(sk) do { } while (0) 1107 #endif /* SOCK_REFCNT_DEBUG */ 1108 1109 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1110 extern struct static_key memcg_socket_limit_enabled; 1111 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1112 struct cg_proto *cg_proto) 1113 { 1114 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1115 } 1116 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1117 #else 1118 #define mem_cgroup_sockets_enabled 0 1119 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1120 struct cg_proto *cg_proto) 1121 { 1122 return NULL; 1123 } 1124 #endif 1125 1126 static inline bool sk_stream_memory_free(const struct sock *sk) 1127 { 1128 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1129 return false; 1130 1131 return sk->sk_prot->stream_memory_free ? 1132 sk->sk_prot->stream_memory_free(sk) : true; 1133 } 1134 1135 static inline bool sk_stream_is_writeable(const struct sock *sk) 1136 { 1137 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1138 sk_stream_memory_free(sk); 1139 } 1140 1141 1142 static inline bool sk_has_memory_pressure(const struct sock *sk) 1143 { 1144 return sk->sk_prot->memory_pressure != NULL; 1145 } 1146 1147 static inline bool sk_under_memory_pressure(const struct sock *sk) 1148 { 1149 if (!sk->sk_prot->memory_pressure) 1150 return false; 1151 1152 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1153 return !!sk->sk_cgrp->memory_pressure; 1154 1155 return !!*sk->sk_prot->memory_pressure; 1156 } 1157 1158 static inline void sk_leave_memory_pressure(struct sock *sk) 1159 { 1160 int *memory_pressure = sk->sk_prot->memory_pressure; 1161 1162 if (!memory_pressure) 1163 return; 1164 1165 if (*memory_pressure) 1166 *memory_pressure = 0; 1167 1168 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1169 struct cg_proto *cg_proto = sk->sk_cgrp; 1170 struct proto *prot = sk->sk_prot; 1171 1172 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1173 cg_proto->memory_pressure = 0; 1174 } 1175 1176 } 1177 1178 static inline void sk_enter_memory_pressure(struct sock *sk) 1179 { 1180 if (!sk->sk_prot->enter_memory_pressure) 1181 return; 1182 1183 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1184 struct cg_proto *cg_proto = sk->sk_cgrp; 1185 struct proto *prot = sk->sk_prot; 1186 1187 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1188 cg_proto->memory_pressure = 1; 1189 } 1190 1191 sk->sk_prot->enter_memory_pressure(sk); 1192 } 1193 1194 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1195 { 1196 long *prot = sk->sk_prot->sysctl_mem; 1197 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1198 prot = sk->sk_cgrp->sysctl_mem; 1199 return prot[index]; 1200 } 1201 1202 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1203 unsigned long amt, 1204 int *parent_status) 1205 { 1206 struct res_counter *fail; 1207 int ret; 1208 1209 ret = res_counter_charge_nofail(&prot->memory_allocated, 1210 amt << PAGE_SHIFT, &fail); 1211 if (ret < 0) 1212 *parent_status = OVER_LIMIT; 1213 } 1214 1215 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1216 unsigned long amt) 1217 { 1218 res_counter_uncharge(&prot->memory_allocated, amt << PAGE_SHIFT); 1219 } 1220 1221 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot) 1222 { 1223 u64 ret; 1224 ret = res_counter_read_u64(&prot->memory_allocated, RES_USAGE); 1225 return ret >> PAGE_SHIFT; 1226 } 1227 1228 static inline long 1229 sk_memory_allocated(const struct sock *sk) 1230 { 1231 struct proto *prot = sk->sk_prot; 1232 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1233 return memcg_memory_allocated_read(sk->sk_cgrp); 1234 1235 return atomic_long_read(prot->memory_allocated); 1236 } 1237 1238 static inline long 1239 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1240 { 1241 struct proto *prot = sk->sk_prot; 1242 1243 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1244 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1245 /* update the root cgroup regardless */ 1246 atomic_long_add_return(amt, prot->memory_allocated); 1247 return memcg_memory_allocated_read(sk->sk_cgrp); 1248 } 1249 1250 return atomic_long_add_return(amt, prot->memory_allocated); 1251 } 1252 1253 static inline void 1254 sk_memory_allocated_sub(struct sock *sk, int amt) 1255 { 1256 struct proto *prot = sk->sk_prot; 1257 1258 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1259 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1260 1261 atomic_long_sub(amt, prot->memory_allocated); 1262 } 1263 1264 static inline void sk_sockets_allocated_dec(struct sock *sk) 1265 { 1266 struct proto *prot = sk->sk_prot; 1267 1268 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1269 struct cg_proto *cg_proto = sk->sk_cgrp; 1270 1271 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1272 percpu_counter_dec(&cg_proto->sockets_allocated); 1273 } 1274 1275 percpu_counter_dec(prot->sockets_allocated); 1276 } 1277 1278 static inline void sk_sockets_allocated_inc(struct sock *sk) 1279 { 1280 struct proto *prot = sk->sk_prot; 1281 1282 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1283 struct cg_proto *cg_proto = sk->sk_cgrp; 1284 1285 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1286 percpu_counter_inc(&cg_proto->sockets_allocated); 1287 } 1288 1289 percpu_counter_inc(prot->sockets_allocated); 1290 } 1291 1292 static inline int 1293 sk_sockets_allocated_read_positive(struct sock *sk) 1294 { 1295 struct proto *prot = sk->sk_prot; 1296 1297 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1298 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated); 1299 1300 return percpu_counter_read_positive(prot->sockets_allocated); 1301 } 1302 1303 static inline int 1304 proto_sockets_allocated_sum_positive(struct proto *prot) 1305 { 1306 return percpu_counter_sum_positive(prot->sockets_allocated); 1307 } 1308 1309 static inline long 1310 proto_memory_allocated(struct proto *prot) 1311 { 1312 return atomic_long_read(prot->memory_allocated); 1313 } 1314 1315 static inline bool 1316 proto_memory_pressure(struct proto *prot) 1317 { 1318 if (!prot->memory_pressure) 1319 return false; 1320 return !!*prot->memory_pressure; 1321 } 1322 1323 1324 #ifdef CONFIG_PROC_FS 1325 /* Called with local bh disabled */ 1326 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1327 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1328 #else 1329 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1330 int inc) 1331 { 1332 } 1333 #endif 1334 1335 1336 /* With per-bucket locks this operation is not-atomic, so that 1337 * this version is not worse. 1338 */ 1339 static inline void __sk_prot_rehash(struct sock *sk) 1340 { 1341 sk->sk_prot->unhash(sk); 1342 sk->sk_prot->hash(sk); 1343 } 1344 1345 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1346 1347 /* About 10 seconds */ 1348 #define SOCK_DESTROY_TIME (10*HZ) 1349 1350 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1351 #define PROT_SOCK 1024 1352 1353 #define SHUTDOWN_MASK 3 1354 #define RCV_SHUTDOWN 1 1355 #define SEND_SHUTDOWN 2 1356 1357 #define SOCK_SNDBUF_LOCK 1 1358 #define SOCK_RCVBUF_LOCK 2 1359 #define SOCK_BINDADDR_LOCK 4 1360 #define SOCK_BINDPORT_LOCK 8 1361 1362 /* sock_iocb: used to kick off async processing of socket ios */ 1363 struct sock_iocb { 1364 struct list_head list; 1365 1366 int flags; 1367 int size; 1368 struct socket *sock; 1369 struct sock *sk; 1370 struct scm_cookie *scm; 1371 struct msghdr *msg, async_msg; 1372 struct kiocb *kiocb; 1373 }; 1374 1375 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 1376 { 1377 return (struct sock_iocb *)iocb->private; 1378 } 1379 1380 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 1381 { 1382 return si->kiocb; 1383 } 1384 1385 struct socket_alloc { 1386 struct socket socket; 1387 struct inode vfs_inode; 1388 }; 1389 1390 static inline struct socket *SOCKET_I(struct inode *inode) 1391 { 1392 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1393 } 1394 1395 static inline struct inode *SOCK_INODE(struct socket *socket) 1396 { 1397 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1398 } 1399 1400 /* 1401 * Functions for memory accounting 1402 */ 1403 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1404 void __sk_mem_reclaim(struct sock *sk); 1405 1406 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1407 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1408 #define SK_MEM_SEND 0 1409 #define SK_MEM_RECV 1 1410 1411 static inline int sk_mem_pages(int amt) 1412 { 1413 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1414 } 1415 1416 static inline bool sk_has_account(struct sock *sk) 1417 { 1418 /* return true if protocol supports memory accounting */ 1419 return !!sk->sk_prot->memory_allocated; 1420 } 1421 1422 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1423 { 1424 if (!sk_has_account(sk)) 1425 return true; 1426 return size <= sk->sk_forward_alloc || 1427 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1428 } 1429 1430 static inline bool 1431 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1432 { 1433 if (!sk_has_account(sk)) 1434 return true; 1435 return size<= sk->sk_forward_alloc || 1436 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1437 skb_pfmemalloc(skb); 1438 } 1439 1440 static inline void sk_mem_reclaim(struct sock *sk) 1441 { 1442 if (!sk_has_account(sk)) 1443 return; 1444 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1445 __sk_mem_reclaim(sk); 1446 } 1447 1448 static inline void sk_mem_reclaim_partial(struct sock *sk) 1449 { 1450 if (!sk_has_account(sk)) 1451 return; 1452 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1453 __sk_mem_reclaim(sk); 1454 } 1455 1456 static inline void sk_mem_charge(struct sock *sk, int size) 1457 { 1458 if (!sk_has_account(sk)) 1459 return; 1460 sk->sk_forward_alloc -= size; 1461 } 1462 1463 static inline void sk_mem_uncharge(struct sock *sk, int size) 1464 { 1465 if (!sk_has_account(sk)) 1466 return; 1467 sk->sk_forward_alloc += size; 1468 } 1469 1470 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1471 { 1472 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1473 sk->sk_wmem_queued -= skb->truesize; 1474 sk_mem_uncharge(sk, skb->truesize); 1475 __kfree_skb(skb); 1476 } 1477 1478 /* Used by processes to "lock" a socket state, so that 1479 * interrupts and bottom half handlers won't change it 1480 * from under us. It essentially blocks any incoming 1481 * packets, so that we won't get any new data or any 1482 * packets that change the state of the socket. 1483 * 1484 * While locked, BH processing will add new packets to 1485 * the backlog queue. This queue is processed by the 1486 * owner of the socket lock right before it is released. 1487 * 1488 * Since ~2.3.5 it is also exclusive sleep lock serializing 1489 * accesses from user process context. 1490 */ 1491 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1492 1493 static inline void sock_release_ownership(struct sock *sk) 1494 { 1495 sk->sk_lock.owned = 0; 1496 } 1497 1498 /* 1499 * Macro so as to not evaluate some arguments when 1500 * lockdep is not enabled. 1501 * 1502 * Mark both the sk_lock and the sk_lock.slock as a 1503 * per-address-family lock class. 1504 */ 1505 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1506 do { \ 1507 sk->sk_lock.owned = 0; \ 1508 init_waitqueue_head(&sk->sk_lock.wq); \ 1509 spin_lock_init(&(sk)->sk_lock.slock); \ 1510 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1511 sizeof((sk)->sk_lock)); \ 1512 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1513 (skey), (sname)); \ 1514 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1515 } while (0) 1516 1517 void lock_sock_nested(struct sock *sk, int subclass); 1518 1519 static inline void lock_sock(struct sock *sk) 1520 { 1521 lock_sock_nested(sk, 0); 1522 } 1523 1524 void release_sock(struct sock *sk); 1525 1526 /* BH context may only use the following locking interface. */ 1527 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1528 #define bh_lock_sock_nested(__sk) \ 1529 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1530 SINGLE_DEPTH_NESTING) 1531 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1532 1533 bool lock_sock_fast(struct sock *sk); 1534 /** 1535 * unlock_sock_fast - complement of lock_sock_fast 1536 * @sk: socket 1537 * @slow: slow mode 1538 * 1539 * fast unlock socket for user context. 1540 * If slow mode is on, we call regular release_sock() 1541 */ 1542 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1543 { 1544 if (slow) 1545 release_sock(sk); 1546 else 1547 spin_unlock_bh(&sk->sk_lock.slock); 1548 } 1549 1550 1551 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1552 struct proto *prot); 1553 void sk_free(struct sock *sk); 1554 void sk_release_kernel(struct sock *sk); 1555 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1556 1557 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1558 gfp_t priority); 1559 void sock_wfree(struct sk_buff *skb); 1560 void skb_orphan_partial(struct sk_buff *skb); 1561 void sock_rfree(struct sk_buff *skb); 1562 void sock_edemux(struct sk_buff *skb); 1563 1564 int sock_setsockopt(struct socket *sock, int level, int op, 1565 char __user *optval, unsigned int optlen); 1566 1567 int sock_getsockopt(struct socket *sock, int level, int op, 1568 char __user *optval, int __user *optlen); 1569 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1570 int noblock, int *errcode); 1571 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1572 unsigned long data_len, int noblock, 1573 int *errcode, int max_page_order); 1574 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1575 void sock_kfree_s(struct sock *sk, void *mem, int size); 1576 void sk_send_sigurg(struct sock *sk); 1577 1578 /* 1579 * Functions to fill in entries in struct proto_ops when a protocol 1580 * does not implement a particular function. 1581 */ 1582 int sock_no_bind(struct socket *, struct sockaddr *, int); 1583 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1584 int sock_no_socketpair(struct socket *, struct socket *); 1585 int sock_no_accept(struct socket *, struct socket *, int); 1586 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1587 unsigned int sock_no_poll(struct file *, struct socket *, 1588 struct poll_table_struct *); 1589 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1590 int sock_no_listen(struct socket *, int); 1591 int sock_no_shutdown(struct socket *, int); 1592 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1593 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1594 int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t); 1595 int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t, 1596 int); 1597 int sock_no_mmap(struct file *file, struct socket *sock, 1598 struct vm_area_struct *vma); 1599 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1600 size_t size, int flags); 1601 1602 /* 1603 * Functions to fill in entries in struct proto_ops when a protocol 1604 * uses the inet style. 1605 */ 1606 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1607 char __user *optval, int __user *optlen); 1608 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1609 struct msghdr *msg, size_t size, int flags); 1610 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1611 char __user *optval, unsigned int optlen); 1612 int compat_sock_common_getsockopt(struct socket *sock, int level, 1613 int optname, char __user *optval, int __user *optlen); 1614 int compat_sock_common_setsockopt(struct socket *sock, int level, 1615 int optname, char __user *optval, unsigned int optlen); 1616 1617 void sk_common_release(struct sock *sk); 1618 1619 /* 1620 * Default socket callbacks and setup code 1621 */ 1622 1623 /* Initialise core socket variables */ 1624 void sock_init_data(struct socket *sock, struct sock *sk); 1625 1626 /* 1627 * Socket reference counting postulates. 1628 * 1629 * * Each user of socket SHOULD hold a reference count. 1630 * * Each access point to socket (an hash table bucket, reference from a list, 1631 * running timer, skb in flight MUST hold a reference count. 1632 * * When reference count hits 0, it means it will never increase back. 1633 * * When reference count hits 0, it means that no references from 1634 * outside exist to this socket and current process on current CPU 1635 * is last user and may/should destroy this socket. 1636 * * sk_free is called from any context: process, BH, IRQ. When 1637 * it is called, socket has no references from outside -> sk_free 1638 * may release descendant resources allocated by the socket, but 1639 * to the time when it is called, socket is NOT referenced by any 1640 * hash tables, lists etc. 1641 * * Packets, delivered from outside (from network or from another process) 1642 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1643 * when they sit in queue. Otherwise, packets will leak to hole, when 1644 * socket is looked up by one cpu and unhasing is made by another CPU. 1645 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1646 * (leak to backlog). Packet socket does all the processing inside 1647 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1648 * use separate SMP lock, so that they are prone too. 1649 */ 1650 1651 /* Ungrab socket and destroy it, if it was the last reference. */ 1652 static inline void sock_put(struct sock *sk) 1653 { 1654 if (atomic_dec_and_test(&sk->sk_refcnt)) 1655 sk_free(sk); 1656 } 1657 /* Generic version of sock_put(), dealing with all sockets 1658 * (TCP_TIMEWAIT, ESTABLISHED...) 1659 */ 1660 void sock_gen_put(struct sock *sk); 1661 1662 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested); 1663 1664 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1665 { 1666 sk->sk_tx_queue_mapping = tx_queue; 1667 } 1668 1669 static inline void sk_tx_queue_clear(struct sock *sk) 1670 { 1671 sk->sk_tx_queue_mapping = -1; 1672 } 1673 1674 static inline int sk_tx_queue_get(const struct sock *sk) 1675 { 1676 return sk ? sk->sk_tx_queue_mapping : -1; 1677 } 1678 1679 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1680 { 1681 sk_tx_queue_clear(sk); 1682 sk->sk_socket = sock; 1683 } 1684 1685 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1686 { 1687 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1688 return &rcu_dereference_raw(sk->sk_wq)->wait; 1689 } 1690 /* Detach socket from process context. 1691 * Announce socket dead, detach it from wait queue and inode. 1692 * Note that parent inode held reference count on this struct sock, 1693 * we do not release it in this function, because protocol 1694 * probably wants some additional cleanups or even continuing 1695 * to work with this socket (TCP). 1696 */ 1697 static inline void sock_orphan(struct sock *sk) 1698 { 1699 write_lock_bh(&sk->sk_callback_lock); 1700 sock_set_flag(sk, SOCK_DEAD); 1701 sk_set_socket(sk, NULL); 1702 sk->sk_wq = NULL; 1703 write_unlock_bh(&sk->sk_callback_lock); 1704 } 1705 1706 static inline void sock_graft(struct sock *sk, struct socket *parent) 1707 { 1708 write_lock_bh(&sk->sk_callback_lock); 1709 sk->sk_wq = parent->wq; 1710 parent->sk = sk; 1711 sk_set_socket(sk, parent); 1712 security_sock_graft(sk, parent); 1713 write_unlock_bh(&sk->sk_callback_lock); 1714 } 1715 1716 kuid_t sock_i_uid(struct sock *sk); 1717 unsigned long sock_i_ino(struct sock *sk); 1718 1719 static inline struct dst_entry * 1720 __sk_dst_get(struct sock *sk) 1721 { 1722 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1723 lockdep_is_held(&sk->sk_lock.slock)); 1724 } 1725 1726 static inline struct dst_entry * 1727 sk_dst_get(struct sock *sk) 1728 { 1729 struct dst_entry *dst; 1730 1731 rcu_read_lock(); 1732 dst = rcu_dereference(sk->sk_dst_cache); 1733 if (dst) 1734 dst_hold(dst); 1735 rcu_read_unlock(); 1736 return dst; 1737 } 1738 1739 static inline void dst_negative_advice(struct sock *sk) 1740 { 1741 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1742 1743 if (dst && dst->ops->negative_advice) { 1744 ndst = dst->ops->negative_advice(dst); 1745 1746 if (ndst != dst) { 1747 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1748 sk_tx_queue_clear(sk); 1749 } 1750 } 1751 } 1752 1753 static inline void 1754 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1755 { 1756 struct dst_entry *old_dst; 1757 1758 sk_tx_queue_clear(sk); 1759 /* 1760 * This can be called while sk is owned by the caller only, 1761 * with no state that can be checked in a rcu_dereference_check() cond 1762 */ 1763 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1764 rcu_assign_pointer(sk->sk_dst_cache, dst); 1765 dst_release(old_dst); 1766 } 1767 1768 static inline void 1769 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1770 { 1771 spin_lock(&sk->sk_dst_lock); 1772 __sk_dst_set(sk, dst); 1773 spin_unlock(&sk->sk_dst_lock); 1774 } 1775 1776 static inline void 1777 __sk_dst_reset(struct sock *sk) 1778 { 1779 __sk_dst_set(sk, NULL); 1780 } 1781 1782 static inline void 1783 sk_dst_reset(struct sock *sk) 1784 { 1785 spin_lock(&sk->sk_dst_lock); 1786 __sk_dst_reset(sk); 1787 spin_unlock(&sk->sk_dst_lock); 1788 } 1789 1790 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1791 1792 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1793 1794 static inline bool sk_can_gso(const struct sock *sk) 1795 { 1796 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1797 } 1798 1799 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1800 1801 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1802 { 1803 sk->sk_route_nocaps |= flags; 1804 sk->sk_route_caps &= ~flags; 1805 } 1806 1807 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1808 char __user *from, char *to, 1809 int copy, int offset) 1810 { 1811 if (skb->ip_summed == CHECKSUM_NONE) { 1812 int err = 0; 1813 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1814 if (err) 1815 return err; 1816 skb->csum = csum_block_add(skb->csum, csum, offset); 1817 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1818 if (!access_ok(VERIFY_READ, from, copy) || 1819 __copy_from_user_nocache(to, from, copy)) 1820 return -EFAULT; 1821 } else if (copy_from_user(to, 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 char __user *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, char __user *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 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1861 struct sk_buff *skb, struct page *page, 1862 int off, int copy) 1863 { 1864 if (skb->ip_summed == CHECKSUM_NONE) { 1865 int err = 0; 1866 __wsum csum = csum_and_copy_from_user(from, 1867 page_address(page) + off, 1868 copy, 0, &err); 1869 if (err) 1870 return err; 1871 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1872 } else if (copy_from_user(page_address(page) + off, from, copy)) 1873 return -EFAULT; 1874 1875 skb->len += copy; 1876 skb->data_len += copy; 1877 skb->truesize += copy; 1878 sk->sk_wmem_queued += copy; 1879 sk_mem_charge(sk, copy); 1880 return 0; 1881 } 1882 1883 /** 1884 * sk_wmem_alloc_get - returns write allocations 1885 * @sk: socket 1886 * 1887 * Returns sk_wmem_alloc minus initial offset of one 1888 */ 1889 static inline int sk_wmem_alloc_get(const struct sock *sk) 1890 { 1891 return atomic_read(&sk->sk_wmem_alloc) - 1; 1892 } 1893 1894 /** 1895 * sk_rmem_alloc_get - returns read allocations 1896 * @sk: socket 1897 * 1898 * Returns sk_rmem_alloc 1899 */ 1900 static inline int sk_rmem_alloc_get(const struct sock *sk) 1901 { 1902 return atomic_read(&sk->sk_rmem_alloc); 1903 } 1904 1905 /** 1906 * sk_has_allocations - check if allocations are outstanding 1907 * @sk: socket 1908 * 1909 * Returns true if socket has write or read allocations 1910 */ 1911 static inline bool sk_has_allocations(const struct sock *sk) 1912 { 1913 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1914 } 1915 1916 /** 1917 * wq_has_sleeper - check if there are any waiting processes 1918 * @wq: struct socket_wq 1919 * 1920 * Returns true if socket_wq has waiting processes 1921 * 1922 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1923 * barrier call. They were added due to the race found within the tcp code. 1924 * 1925 * Consider following tcp code paths: 1926 * 1927 * CPU1 CPU2 1928 * 1929 * sys_select receive packet 1930 * ... ... 1931 * __add_wait_queue update tp->rcv_nxt 1932 * ... ... 1933 * tp->rcv_nxt check sock_def_readable 1934 * ... { 1935 * schedule rcu_read_lock(); 1936 * wq = rcu_dereference(sk->sk_wq); 1937 * if (wq && waitqueue_active(&wq->wait)) 1938 * wake_up_interruptible(&wq->wait) 1939 * ... 1940 * } 1941 * 1942 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1943 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1944 * could then endup calling schedule and sleep forever if there are no more 1945 * data on the socket. 1946 * 1947 */ 1948 static inline bool wq_has_sleeper(struct socket_wq *wq) 1949 { 1950 /* We need to be sure we are in sync with the 1951 * add_wait_queue modifications to the wait queue. 1952 * 1953 * This memory barrier is paired in the sock_poll_wait. 1954 */ 1955 smp_mb(); 1956 return wq && waitqueue_active(&wq->wait); 1957 } 1958 1959 /** 1960 * sock_poll_wait - place memory barrier behind the poll_wait call. 1961 * @filp: file 1962 * @wait_address: socket wait queue 1963 * @p: poll_table 1964 * 1965 * See the comments in the wq_has_sleeper function. 1966 */ 1967 static inline void sock_poll_wait(struct file *filp, 1968 wait_queue_head_t *wait_address, poll_table *p) 1969 { 1970 if (!poll_does_not_wait(p) && wait_address) { 1971 poll_wait(filp, wait_address, p); 1972 /* We need to be sure we are in sync with the 1973 * socket flags modification. 1974 * 1975 * This memory barrier is paired in the wq_has_sleeper. 1976 */ 1977 smp_mb(); 1978 } 1979 } 1980 1981 /* 1982 * Queue a received datagram if it will fit. Stream and sequenced 1983 * protocols can't normally use this as they need to fit buffers in 1984 * and play with them. 1985 * 1986 * Inlined as it's very short and called for pretty much every 1987 * packet ever received. 1988 */ 1989 1990 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1991 { 1992 skb_orphan(skb); 1993 skb->sk = sk; 1994 skb->destructor = sock_wfree; 1995 /* 1996 * We used to take a refcount on sk, but following operation 1997 * is enough to guarantee sk_free() wont free this sock until 1998 * all in-flight packets are completed 1999 */ 2000 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 2001 } 2002 2003 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2004 { 2005 skb_orphan(skb); 2006 skb->sk = sk; 2007 skb->destructor = sock_rfree; 2008 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2009 sk_mem_charge(sk, skb->truesize); 2010 } 2011 2012 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2013 unsigned long expires); 2014 2015 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2016 2017 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2018 2019 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2020 2021 /* 2022 * Recover an error report and clear atomically 2023 */ 2024 2025 static inline int sock_error(struct sock *sk) 2026 { 2027 int err; 2028 if (likely(!sk->sk_err)) 2029 return 0; 2030 err = xchg(&sk->sk_err, 0); 2031 return -err; 2032 } 2033 2034 static inline unsigned long sock_wspace(struct sock *sk) 2035 { 2036 int amt = 0; 2037 2038 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2039 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2040 if (amt < 0) 2041 amt = 0; 2042 } 2043 return amt; 2044 } 2045 2046 static inline void sk_wake_async(struct sock *sk, int how, int band) 2047 { 2048 if (sock_flag(sk, SOCK_FASYNC)) 2049 sock_wake_async(sk->sk_socket, how, band); 2050 } 2051 2052 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2053 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2054 * Note: for send buffers, TCP works better if we can build two skbs at 2055 * minimum. 2056 */ 2057 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2058 2059 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2060 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2061 2062 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2063 { 2064 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2065 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2066 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2067 } 2068 } 2069 2070 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 2071 2072 /** 2073 * sk_page_frag - return an appropriate page_frag 2074 * @sk: socket 2075 * 2076 * If socket allocation mode allows current thread to sleep, it means its 2077 * safe to use the per task page_frag instead of the per socket one. 2078 */ 2079 static inline struct page_frag *sk_page_frag(struct sock *sk) 2080 { 2081 if (sk->sk_allocation & __GFP_WAIT) 2082 return ¤t->task_frag; 2083 2084 return &sk->sk_frag; 2085 } 2086 2087 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2088 2089 /* 2090 * Default write policy as shown to user space via poll/select/SIGIO 2091 */ 2092 static inline bool sock_writeable(const struct sock *sk) 2093 { 2094 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2095 } 2096 2097 static inline gfp_t gfp_any(void) 2098 { 2099 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2100 } 2101 2102 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2103 { 2104 return noblock ? 0 : sk->sk_rcvtimeo; 2105 } 2106 2107 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2108 { 2109 return noblock ? 0 : sk->sk_sndtimeo; 2110 } 2111 2112 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2113 { 2114 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2115 } 2116 2117 /* Alas, with timeout socket operations are not restartable. 2118 * Compare this to poll(). 2119 */ 2120 static inline int sock_intr_errno(long timeo) 2121 { 2122 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2123 } 2124 2125 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2126 struct sk_buff *skb); 2127 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2128 struct sk_buff *skb); 2129 2130 static inline void 2131 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2132 { 2133 ktime_t kt = skb->tstamp; 2134 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2135 2136 /* 2137 * generate control messages if 2138 * - receive time stamping in software requested (SOCK_RCVTSTAMP 2139 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 2140 * - software time stamp available and wanted 2141 * (SOCK_TIMESTAMPING_SOFTWARE) 2142 * - hardware time stamps available and wanted 2143 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 2144 * SOCK_TIMESTAMPING_RAW_HARDWARE) 2145 */ 2146 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2147 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 2148 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 2149 (hwtstamps->hwtstamp.tv64 && 2150 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 2151 (hwtstamps->syststamp.tv64 && 2152 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 2153 __sock_recv_timestamp(msg, sk, skb); 2154 else 2155 sk->sk_stamp = kt; 2156 2157 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2158 __sock_recv_wifi_status(msg, sk, skb); 2159 } 2160 2161 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2162 struct sk_buff *skb); 2163 2164 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2165 struct sk_buff *skb) 2166 { 2167 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2168 (1UL << SOCK_RCVTSTAMP) | \ 2169 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 2170 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 2171 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 2172 2173 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 2174 __sock_recv_ts_and_drops(msg, sk, skb); 2175 else 2176 sk->sk_stamp = skb->tstamp; 2177 } 2178 2179 /** 2180 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2181 * @sk: socket sending this packet 2182 * @tx_flags: filled with instructions for time stamping 2183 * 2184 * Currently only depends on SOCK_TIMESTAMPING* flags. 2185 */ 2186 void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 2187 2188 /** 2189 * sk_eat_skb - Release a skb if it is no longer needed 2190 * @sk: socket to eat this skb from 2191 * @skb: socket buffer to eat 2192 * @copied_early: flag indicating whether DMA operations copied this data early 2193 * 2194 * This routine must be called with interrupts disabled or with the socket 2195 * locked so that the sk_buff queue operation is ok. 2196 */ 2197 #ifdef CONFIG_NET_DMA 2198 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2199 { 2200 __skb_unlink(skb, &sk->sk_receive_queue); 2201 if (!copied_early) 2202 __kfree_skb(skb); 2203 else 2204 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 2205 } 2206 #else 2207 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2208 { 2209 __skb_unlink(skb, &sk->sk_receive_queue); 2210 __kfree_skb(skb); 2211 } 2212 #endif 2213 2214 static inline 2215 struct net *sock_net(const struct sock *sk) 2216 { 2217 return read_pnet(&sk->sk_net); 2218 } 2219 2220 static inline 2221 void sock_net_set(struct sock *sk, struct net *net) 2222 { 2223 write_pnet(&sk->sk_net, net); 2224 } 2225 2226 /* 2227 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2228 * They should not hold a reference to a namespace in order to allow 2229 * to stop it. 2230 * Sockets after sk_change_net should be released using sk_release_kernel 2231 */ 2232 static inline void sk_change_net(struct sock *sk, struct net *net) 2233 { 2234 struct net *current_net = sock_net(sk); 2235 2236 if (!net_eq(current_net, net)) { 2237 put_net(current_net); 2238 sock_net_set(sk, hold_net(net)); 2239 } 2240 } 2241 2242 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2243 { 2244 if (skb->sk) { 2245 struct sock *sk = skb->sk; 2246 2247 skb->destructor = NULL; 2248 skb->sk = NULL; 2249 return sk; 2250 } 2251 return NULL; 2252 } 2253 2254 void sock_enable_timestamp(struct sock *sk, int flag); 2255 int sock_get_timestamp(struct sock *, struct timeval __user *); 2256 int sock_get_timestampns(struct sock *, struct timespec __user *); 2257 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2258 int type); 2259 2260 bool sk_ns_capable(const struct sock *sk, 2261 struct user_namespace *user_ns, int cap); 2262 bool sk_capable(const struct sock *sk, int cap); 2263 bool sk_net_capable(const struct sock *sk, int cap); 2264 2265 /* 2266 * Enable debug/info messages 2267 */ 2268 extern int net_msg_warn; 2269 #define NETDEBUG(fmt, args...) \ 2270 do { if (net_msg_warn) printk(fmt,##args); } while (0) 2271 2272 #define LIMIT_NETDEBUG(fmt, args...) \ 2273 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 2274 2275 extern __u32 sysctl_wmem_max; 2276 extern __u32 sysctl_rmem_max; 2277 2278 extern int sysctl_optmem_max; 2279 2280 extern __u32 sysctl_wmem_default; 2281 extern __u32 sysctl_rmem_default; 2282 2283 #endif /* _SOCK_H */ 2284