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() and INET_TW_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 * fields between dontcopy_begin/dontcopy_end 196 * are not copied in sock_copy() 197 */ 198 /* private: */ 199 int skc_dontcopy_begin[0]; 200 /* public: */ 201 union { 202 struct hlist_node skc_node; 203 struct hlist_nulls_node skc_nulls_node; 204 }; 205 int skc_tx_queue_mapping; 206 atomic_t skc_refcnt; 207 /* private: */ 208 int skc_dontcopy_end[0]; 209 /* public: */ 210 }; 211 212 struct cg_proto; 213 /** 214 * struct sock - network layer representation of sockets 215 * @__sk_common: shared layout with inet_timewait_sock 216 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 217 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 218 * @sk_lock: synchronizer 219 * @sk_rcvbuf: size of receive buffer in bytes 220 * @sk_wq: sock wait queue and async head 221 * @sk_rx_dst: receive input route used by early tcp demux 222 * @sk_dst_cache: destination cache 223 * @sk_dst_lock: destination cache lock 224 * @sk_policy: flow policy 225 * @sk_receive_queue: incoming packets 226 * @sk_wmem_alloc: transmit queue bytes committed 227 * @sk_write_queue: Packet sending queue 228 * @sk_async_wait_queue: DMA copied packets 229 * @sk_omem_alloc: "o" is "option" or "other" 230 * @sk_wmem_queued: persistent queue size 231 * @sk_forward_alloc: space allocated forward 232 * @sk_napi_id: id of the last napi context to receive data for sk 233 * @sk_ll_usec: usecs to busypoll when there is no data 234 * @sk_allocation: allocation mode 235 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 236 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 237 * @sk_sndbuf: size of send buffer in bytes 238 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 239 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 240 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets 241 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 242 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 243 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 244 * @sk_gso_max_size: Maximum GSO segment size to build 245 * @sk_gso_max_segs: Maximum number of GSO segments 246 * @sk_lingertime: %SO_LINGER l_linger setting 247 * @sk_backlog: always used with the per-socket spinlock held 248 * @sk_callback_lock: used with the callbacks in the end of this struct 249 * @sk_error_queue: rarely used 250 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 251 * IPV6_ADDRFORM for instance) 252 * @sk_err: last error 253 * @sk_err_soft: errors that don't cause failure but are the cause of a 254 * persistent failure not just 'timed out' 255 * @sk_drops: raw/udp drops counter 256 * @sk_ack_backlog: current listen backlog 257 * @sk_max_ack_backlog: listen backlog set in listen() 258 * @sk_priority: %SO_PRIORITY setting 259 * @sk_cgrp_prioidx: socket group's priority map index 260 * @sk_type: socket type (%SOCK_STREAM, etc) 261 * @sk_protocol: which protocol this socket belongs in this network family 262 * @sk_peer_pid: &struct pid for this socket's peer 263 * @sk_peer_cred: %SO_PEERCRED setting 264 * @sk_rcvlowat: %SO_RCVLOWAT setting 265 * @sk_rcvtimeo: %SO_RCVTIMEO setting 266 * @sk_sndtimeo: %SO_SNDTIMEO setting 267 * @sk_rxhash: flow hash received from netif layer 268 * @sk_filter: socket filtering instructions 269 * @sk_protinfo: private area, net family specific, when not using slab 270 * @sk_timer: sock cleanup timer 271 * @sk_stamp: time stamp of last packet received 272 * @sk_socket: Identd and reporting IO signals 273 * @sk_user_data: RPC layer private data 274 * @sk_frag: cached page frag 275 * @sk_peek_off: current peek_offset value 276 * @sk_send_head: front of stuff to transmit 277 * @sk_security: used by security modules 278 * @sk_mark: generic packet mark 279 * @sk_classid: this socket's cgroup classid 280 * @sk_cgrp: this socket's cgroup-specific proto data 281 * @sk_write_pending: a write to stream socket waits to start 282 * @sk_state_change: callback to indicate change in the state of the sock 283 * @sk_data_ready: callback to indicate there is data to be processed 284 * @sk_write_space: callback to indicate there is bf sending space available 285 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 286 * @sk_backlog_rcv: callback to process the backlog 287 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 288 */ 289 struct sock { 290 /* 291 * Now struct inet_timewait_sock also uses sock_common, so please just 292 * don't add nothing before this first member (__sk_common) --acme 293 */ 294 struct sock_common __sk_common; 295 #define sk_node __sk_common.skc_node 296 #define sk_nulls_node __sk_common.skc_nulls_node 297 #define sk_refcnt __sk_common.skc_refcnt 298 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 299 300 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 301 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 302 #define sk_hash __sk_common.skc_hash 303 #define sk_family __sk_common.skc_family 304 #define sk_state __sk_common.skc_state 305 #define sk_reuse __sk_common.skc_reuse 306 #define sk_reuseport __sk_common.skc_reuseport 307 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 308 #define sk_bind_node __sk_common.skc_bind_node 309 #define sk_prot __sk_common.skc_prot 310 #define sk_net __sk_common.skc_net 311 socket_lock_t sk_lock; 312 struct sk_buff_head sk_receive_queue; 313 /* 314 * The backlog queue is special, it is always used with 315 * the per-socket spinlock held and requires low latency 316 * access. Therefore we special case it's implementation. 317 * Note : rmem_alloc is in this structure to fill a hole 318 * on 64bit arches, not because its logically part of 319 * backlog. 320 */ 321 struct { 322 atomic_t rmem_alloc; 323 int len; 324 struct sk_buff *head; 325 struct sk_buff *tail; 326 } sk_backlog; 327 #define sk_rmem_alloc sk_backlog.rmem_alloc 328 int sk_forward_alloc; 329 #ifdef CONFIG_RPS 330 __u32 sk_rxhash; 331 #endif 332 #ifdef CONFIG_NET_RX_BUSY_POLL 333 unsigned int sk_napi_id; 334 unsigned int sk_ll_usec; 335 #endif 336 atomic_t sk_drops; 337 int sk_rcvbuf; 338 339 struct sk_filter __rcu *sk_filter; 340 struct socket_wq __rcu *sk_wq; 341 342 #ifdef CONFIG_NET_DMA 343 struct sk_buff_head sk_async_wait_queue; 344 #endif 345 346 #ifdef CONFIG_XFRM 347 struct xfrm_policy *sk_policy[2]; 348 #endif 349 unsigned long sk_flags; 350 struct dst_entry *sk_rx_dst; 351 struct dst_entry __rcu *sk_dst_cache; 352 spinlock_t sk_dst_lock; 353 atomic_t sk_wmem_alloc; 354 atomic_t sk_omem_alloc; 355 int sk_sndbuf; 356 struct sk_buff_head sk_write_queue; 357 kmemcheck_bitfield_begin(flags); 358 unsigned int sk_shutdown : 2, 359 sk_no_check : 2, 360 sk_userlocks : 4, 361 sk_protocol : 8, 362 sk_type : 16; 363 kmemcheck_bitfield_end(flags); 364 int sk_wmem_queued; 365 gfp_t sk_allocation; 366 u32 sk_pacing_rate; /* bytes per second */ 367 u32 sk_max_pacing_rate; 368 netdev_features_t sk_route_caps; 369 netdev_features_t sk_route_nocaps; 370 int sk_gso_type; 371 unsigned int sk_gso_max_size; 372 u16 sk_gso_max_segs; 373 int sk_rcvlowat; 374 unsigned long sk_lingertime; 375 struct sk_buff_head sk_error_queue; 376 struct proto *sk_prot_creator; 377 rwlock_t sk_callback_lock; 378 int sk_err, 379 sk_err_soft; 380 unsigned short sk_ack_backlog; 381 unsigned short sk_max_ack_backlog; 382 __u32 sk_priority; 383 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) 384 __u32 sk_cgrp_prioidx; 385 #endif 386 struct pid *sk_peer_pid; 387 const struct cred *sk_peer_cred; 388 long sk_rcvtimeo; 389 long sk_sndtimeo; 390 void *sk_protinfo; 391 struct timer_list sk_timer; 392 ktime_t sk_stamp; 393 struct socket *sk_socket; 394 void *sk_user_data; 395 struct page_frag sk_frag; 396 struct sk_buff *sk_send_head; 397 __s32 sk_peek_off; 398 int sk_write_pending; 399 #ifdef CONFIG_SECURITY 400 void *sk_security; 401 #endif 402 __u32 sk_mark; 403 u32 sk_classid; 404 struct cg_proto *sk_cgrp; 405 void (*sk_state_change)(struct sock *sk); 406 void (*sk_data_ready)(struct sock *sk, int bytes); 407 void (*sk_write_space)(struct sock *sk); 408 void (*sk_error_report)(struct sock *sk); 409 int (*sk_backlog_rcv)(struct sock *sk, 410 struct sk_buff *skb); 411 void (*sk_destruct)(struct sock *sk); 412 }; 413 414 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 415 416 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk))) 417 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr) 418 419 /* 420 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 421 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 422 * on a socket means that the socket will reuse everybody else's port 423 * without looking at the other's sk_reuse value. 424 */ 425 426 #define SK_NO_REUSE 0 427 #define SK_CAN_REUSE 1 428 #define SK_FORCE_REUSE 2 429 430 static inline int sk_peek_offset(struct sock *sk, int flags) 431 { 432 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 433 return sk->sk_peek_off; 434 else 435 return 0; 436 } 437 438 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 439 { 440 if (sk->sk_peek_off >= 0) { 441 if (sk->sk_peek_off >= val) 442 sk->sk_peek_off -= val; 443 else 444 sk->sk_peek_off = 0; 445 } 446 } 447 448 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 449 { 450 if (sk->sk_peek_off >= 0) 451 sk->sk_peek_off += val; 452 } 453 454 /* 455 * Hashed lists helper routines 456 */ 457 static inline struct sock *sk_entry(const struct hlist_node *node) 458 { 459 return hlist_entry(node, struct sock, sk_node); 460 } 461 462 static inline struct sock *__sk_head(const struct hlist_head *head) 463 { 464 return hlist_entry(head->first, struct sock, sk_node); 465 } 466 467 static inline struct sock *sk_head(const struct hlist_head *head) 468 { 469 return hlist_empty(head) ? NULL : __sk_head(head); 470 } 471 472 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 473 { 474 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 475 } 476 477 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 478 { 479 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 480 } 481 482 static inline struct sock *sk_next(const struct sock *sk) 483 { 484 return sk->sk_node.next ? 485 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 486 } 487 488 static inline struct sock *sk_nulls_next(const struct sock *sk) 489 { 490 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 491 hlist_nulls_entry(sk->sk_nulls_node.next, 492 struct sock, sk_nulls_node) : 493 NULL; 494 } 495 496 static inline bool sk_unhashed(const struct sock *sk) 497 { 498 return hlist_unhashed(&sk->sk_node); 499 } 500 501 static inline bool sk_hashed(const struct sock *sk) 502 { 503 return !sk_unhashed(sk); 504 } 505 506 static inline void sk_node_init(struct hlist_node *node) 507 { 508 node->pprev = NULL; 509 } 510 511 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 512 { 513 node->pprev = NULL; 514 } 515 516 static inline void __sk_del_node(struct sock *sk) 517 { 518 __hlist_del(&sk->sk_node); 519 } 520 521 /* NB: equivalent to hlist_del_init_rcu */ 522 static inline bool __sk_del_node_init(struct sock *sk) 523 { 524 if (sk_hashed(sk)) { 525 __sk_del_node(sk); 526 sk_node_init(&sk->sk_node); 527 return true; 528 } 529 return false; 530 } 531 532 /* Grab socket reference count. This operation is valid only 533 when sk is ALREADY grabbed f.e. it is found in hash table 534 or a list and the lookup is made under lock preventing hash table 535 modifications. 536 */ 537 538 static inline void sock_hold(struct sock *sk) 539 { 540 atomic_inc(&sk->sk_refcnt); 541 } 542 543 /* Ungrab socket in the context, which assumes that socket refcnt 544 cannot hit zero, f.e. it is true in context of any socketcall. 545 */ 546 static inline void __sock_put(struct sock *sk) 547 { 548 atomic_dec(&sk->sk_refcnt); 549 } 550 551 static inline bool sk_del_node_init(struct sock *sk) 552 { 553 bool rc = __sk_del_node_init(sk); 554 555 if (rc) { 556 /* paranoid for a while -acme */ 557 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 558 __sock_put(sk); 559 } 560 return rc; 561 } 562 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 563 564 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 565 { 566 if (sk_hashed(sk)) { 567 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 568 return true; 569 } 570 return false; 571 } 572 573 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 574 { 575 bool rc = __sk_nulls_del_node_init_rcu(sk); 576 577 if (rc) { 578 /* paranoid for a while -acme */ 579 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 580 __sock_put(sk); 581 } 582 return rc; 583 } 584 585 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 586 { 587 hlist_add_head(&sk->sk_node, list); 588 } 589 590 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 591 { 592 sock_hold(sk); 593 __sk_add_node(sk, list); 594 } 595 596 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 597 { 598 sock_hold(sk); 599 hlist_add_head_rcu(&sk->sk_node, list); 600 } 601 602 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 603 { 604 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 605 } 606 607 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 608 { 609 sock_hold(sk); 610 __sk_nulls_add_node_rcu(sk, list); 611 } 612 613 static inline void __sk_del_bind_node(struct sock *sk) 614 { 615 __hlist_del(&sk->sk_bind_node); 616 } 617 618 static inline void sk_add_bind_node(struct sock *sk, 619 struct hlist_head *list) 620 { 621 hlist_add_head(&sk->sk_bind_node, list); 622 } 623 624 #define sk_for_each(__sk, list) \ 625 hlist_for_each_entry(__sk, list, sk_node) 626 #define sk_for_each_rcu(__sk, list) \ 627 hlist_for_each_entry_rcu(__sk, list, sk_node) 628 #define sk_nulls_for_each(__sk, node, list) \ 629 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 630 #define sk_nulls_for_each_rcu(__sk, node, list) \ 631 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 632 #define sk_for_each_from(__sk) \ 633 hlist_for_each_entry_from(__sk, sk_node) 634 #define sk_nulls_for_each_from(__sk, node) \ 635 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 636 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 637 #define sk_for_each_safe(__sk, tmp, list) \ 638 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 639 #define sk_for_each_bound(__sk, list) \ 640 hlist_for_each_entry(__sk, list, sk_bind_node) 641 642 static inline struct user_namespace *sk_user_ns(struct sock *sk) 643 { 644 /* Careful only use this in a context where these parameters 645 * can not change and must all be valid, such as recvmsg from 646 * userspace. 647 */ 648 return sk->sk_socket->file->f_cred->user_ns; 649 } 650 651 /* Sock flags */ 652 enum sock_flags { 653 SOCK_DEAD, 654 SOCK_DONE, 655 SOCK_URGINLINE, 656 SOCK_KEEPOPEN, 657 SOCK_LINGER, 658 SOCK_DESTROY, 659 SOCK_BROADCAST, 660 SOCK_TIMESTAMP, 661 SOCK_ZAPPED, 662 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 663 SOCK_DBG, /* %SO_DEBUG setting */ 664 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 665 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 666 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 667 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 668 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 669 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 670 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 671 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 672 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 673 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 674 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 675 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 676 SOCK_FASYNC, /* fasync() active */ 677 SOCK_RXQ_OVFL, 678 SOCK_ZEROCOPY, /* buffers from userspace */ 679 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 680 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 681 * Will use last 4 bytes of packet sent from 682 * user-space instead. 683 */ 684 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 685 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 686 }; 687 688 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 689 { 690 nsk->sk_flags = osk->sk_flags; 691 } 692 693 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 694 { 695 __set_bit(flag, &sk->sk_flags); 696 } 697 698 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 699 { 700 __clear_bit(flag, &sk->sk_flags); 701 } 702 703 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 704 { 705 return test_bit(flag, &sk->sk_flags); 706 } 707 708 #ifdef CONFIG_NET 709 extern struct static_key memalloc_socks; 710 static inline int sk_memalloc_socks(void) 711 { 712 return static_key_false(&memalloc_socks); 713 } 714 #else 715 716 static inline int sk_memalloc_socks(void) 717 { 718 return 0; 719 } 720 721 #endif 722 723 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 724 { 725 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 726 } 727 728 static inline void sk_acceptq_removed(struct sock *sk) 729 { 730 sk->sk_ack_backlog--; 731 } 732 733 static inline void sk_acceptq_added(struct sock *sk) 734 { 735 sk->sk_ack_backlog++; 736 } 737 738 static inline bool sk_acceptq_is_full(const struct sock *sk) 739 { 740 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 741 } 742 743 /* 744 * Compute minimal free write space needed to queue new packets. 745 */ 746 static inline int sk_stream_min_wspace(const struct sock *sk) 747 { 748 return sk->sk_wmem_queued >> 1; 749 } 750 751 static inline int sk_stream_wspace(const struct sock *sk) 752 { 753 return sk->sk_sndbuf - sk->sk_wmem_queued; 754 } 755 756 void sk_stream_write_space(struct sock *sk); 757 758 /* OOB backlog add */ 759 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 760 { 761 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 762 skb_dst_force(skb); 763 764 if (!sk->sk_backlog.tail) 765 sk->sk_backlog.head = skb; 766 else 767 sk->sk_backlog.tail->next = skb; 768 769 sk->sk_backlog.tail = skb; 770 skb->next = NULL; 771 } 772 773 /* 774 * Take into account size of receive queue and backlog queue 775 * Do not take into account this skb truesize, 776 * to allow even a single big packet to come. 777 */ 778 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb, 779 unsigned int limit) 780 { 781 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 782 783 return qsize > limit; 784 } 785 786 /* The per-socket spinlock must be held here. */ 787 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 788 unsigned int limit) 789 { 790 if (sk_rcvqueues_full(sk, skb, limit)) 791 return -ENOBUFS; 792 793 __sk_add_backlog(sk, skb); 794 sk->sk_backlog.len += skb->truesize; 795 return 0; 796 } 797 798 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 799 800 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 801 { 802 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 803 return __sk_backlog_rcv(sk, skb); 804 805 return sk->sk_backlog_rcv(sk, skb); 806 } 807 808 static inline void sock_rps_record_flow(const struct sock *sk) 809 { 810 #ifdef CONFIG_RPS 811 struct rps_sock_flow_table *sock_flow_table; 812 813 rcu_read_lock(); 814 sock_flow_table = rcu_dereference(rps_sock_flow_table); 815 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 816 rcu_read_unlock(); 817 #endif 818 } 819 820 static inline void sock_rps_reset_flow(const struct sock *sk) 821 { 822 #ifdef CONFIG_RPS 823 struct rps_sock_flow_table *sock_flow_table; 824 825 rcu_read_lock(); 826 sock_flow_table = rcu_dereference(rps_sock_flow_table); 827 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 828 rcu_read_unlock(); 829 #endif 830 } 831 832 static inline void sock_rps_save_rxhash(struct sock *sk, 833 const struct sk_buff *skb) 834 { 835 #ifdef CONFIG_RPS 836 if (unlikely(sk->sk_rxhash != skb->rxhash)) { 837 sock_rps_reset_flow(sk); 838 sk->sk_rxhash = skb->rxhash; 839 } 840 #endif 841 } 842 843 static inline void sock_rps_reset_rxhash(struct sock *sk) 844 { 845 #ifdef CONFIG_RPS 846 sock_rps_reset_flow(sk); 847 sk->sk_rxhash = 0; 848 #endif 849 } 850 851 #define sk_wait_event(__sk, __timeo, __condition) \ 852 ({ int __rc; \ 853 release_sock(__sk); \ 854 __rc = __condition; \ 855 if (!__rc) { \ 856 *(__timeo) = schedule_timeout(*(__timeo)); \ 857 } \ 858 lock_sock(__sk); \ 859 __rc = __condition; \ 860 __rc; \ 861 }) 862 863 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 864 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 865 void sk_stream_wait_close(struct sock *sk, long timeo_p); 866 int sk_stream_error(struct sock *sk, int flags, int err); 867 void sk_stream_kill_queues(struct sock *sk); 868 void sk_set_memalloc(struct sock *sk); 869 void sk_clear_memalloc(struct sock *sk); 870 871 int sk_wait_data(struct sock *sk, long *timeo); 872 873 struct request_sock_ops; 874 struct timewait_sock_ops; 875 struct inet_hashinfo; 876 struct raw_hashinfo; 877 struct module; 878 879 /* 880 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 881 * un-modified. Special care is taken when initializing object to zero. 882 */ 883 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 884 { 885 if (offsetof(struct sock, sk_node.next) != 0) 886 memset(sk, 0, offsetof(struct sock, sk_node.next)); 887 memset(&sk->sk_node.pprev, 0, 888 size - offsetof(struct sock, sk_node.pprev)); 889 } 890 891 /* Networking protocol blocks we attach to sockets. 892 * socket layer -> transport layer interface 893 * transport -> network interface is defined by struct inet_proto 894 */ 895 struct proto { 896 void (*close)(struct sock *sk, 897 long timeout); 898 int (*connect)(struct sock *sk, 899 struct sockaddr *uaddr, 900 int addr_len); 901 int (*disconnect)(struct sock *sk, int flags); 902 903 struct sock * (*accept)(struct sock *sk, int flags, int *err); 904 905 int (*ioctl)(struct sock *sk, int cmd, 906 unsigned long arg); 907 int (*init)(struct sock *sk); 908 void (*destroy)(struct sock *sk); 909 void (*shutdown)(struct sock *sk, int how); 910 int (*setsockopt)(struct sock *sk, int level, 911 int optname, char __user *optval, 912 unsigned int optlen); 913 int (*getsockopt)(struct sock *sk, int level, 914 int optname, char __user *optval, 915 int __user *option); 916 #ifdef CONFIG_COMPAT 917 int (*compat_setsockopt)(struct sock *sk, 918 int level, 919 int optname, char __user *optval, 920 unsigned int optlen); 921 int (*compat_getsockopt)(struct sock *sk, 922 int level, 923 int optname, char __user *optval, 924 int __user *option); 925 int (*compat_ioctl)(struct sock *sk, 926 unsigned int cmd, unsigned long arg); 927 #endif 928 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 929 struct msghdr *msg, size_t len); 930 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 931 struct msghdr *msg, 932 size_t len, int noblock, int flags, 933 int *addr_len); 934 int (*sendpage)(struct sock *sk, struct page *page, 935 int offset, size_t size, int flags); 936 int (*bind)(struct sock *sk, 937 struct sockaddr *uaddr, int addr_len); 938 939 int (*backlog_rcv) (struct sock *sk, 940 struct sk_buff *skb); 941 942 void (*release_cb)(struct sock *sk); 943 void (*mtu_reduced)(struct sock *sk); 944 945 /* Keeping track of sk's, looking them up, and port selection methods. */ 946 void (*hash)(struct sock *sk); 947 void (*unhash)(struct sock *sk); 948 void (*rehash)(struct sock *sk); 949 int (*get_port)(struct sock *sk, unsigned short snum); 950 void (*clear_sk)(struct sock *sk, int size); 951 952 /* Keeping track of sockets in use */ 953 #ifdef CONFIG_PROC_FS 954 unsigned int inuse_idx; 955 #endif 956 957 bool (*stream_memory_free)(const struct sock *sk); 958 /* Memory pressure */ 959 void (*enter_memory_pressure)(struct sock *sk); 960 atomic_long_t *memory_allocated; /* Current allocated memory. */ 961 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 962 /* 963 * Pressure flag: try to collapse. 964 * Technical note: it is used by multiple contexts non atomically. 965 * All the __sk_mem_schedule() is of this nature: accounting 966 * is strict, actions are advisory and have some latency. 967 */ 968 int *memory_pressure; 969 long *sysctl_mem; 970 int *sysctl_wmem; 971 int *sysctl_rmem; 972 int max_header; 973 bool no_autobind; 974 975 struct kmem_cache *slab; 976 unsigned int obj_size; 977 int slab_flags; 978 979 struct percpu_counter *orphan_count; 980 981 struct request_sock_ops *rsk_prot; 982 struct timewait_sock_ops *twsk_prot; 983 984 union { 985 struct inet_hashinfo *hashinfo; 986 struct udp_table *udp_table; 987 struct raw_hashinfo *raw_hash; 988 } h; 989 990 struct module *owner; 991 992 char name[32]; 993 994 struct list_head node; 995 #ifdef SOCK_REFCNT_DEBUG 996 atomic_t socks; 997 #endif 998 #ifdef CONFIG_MEMCG_KMEM 999 /* 1000 * cgroup specific init/deinit functions. Called once for all 1001 * protocols that implement it, from cgroups populate function. 1002 * This function has to setup any files the protocol want to 1003 * appear in the kmem cgroup filesystem. 1004 */ 1005 int (*init_cgroup)(struct mem_cgroup *memcg, 1006 struct cgroup_subsys *ss); 1007 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1008 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1009 #endif 1010 }; 1011 1012 /* 1013 * Bits in struct cg_proto.flags 1014 */ 1015 enum cg_proto_flags { 1016 /* Currently active and new sockets should be assigned to cgroups */ 1017 MEMCG_SOCK_ACTIVE, 1018 /* It was ever activated; we must disarm static keys on destruction */ 1019 MEMCG_SOCK_ACTIVATED, 1020 }; 1021 1022 struct cg_proto { 1023 void (*enter_memory_pressure)(struct sock *sk); 1024 struct res_counter *memory_allocated; /* Current allocated memory. */ 1025 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1026 int *memory_pressure; 1027 long *sysctl_mem; 1028 unsigned long flags; 1029 /* 1030 * memcg field is used to find which memcg we belong directly 1031 * Each memcg struct can hold more than one cg_proto, so container_of 1032 * won't really cut. 1033 * 1034 * The elegant solution would be having an inverse function to 1035 * proto_cgroup in struct proto, but that means polluting the structure 1036 * for everybody, instead of just for memcg users. 1037 */ 1038 struct mem_cgroup *memcg; 1039 }; 1040 1041 int proto_register(struct proto *prot, int alloc_slab); 1042 void proto_unregister(struct proto *prot); 1043 1044 static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1045 { 1046 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1047 } 1048 1049 static inline bool memcg_proto_activated(struct cg_proto *cg_proto) 1050 { 1051 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags); 1052 } 1053 1054 #ifdef SOCK_REFCNT_DEBUG 1055 static inline void sk_refcnt_debug_inc(struct sock *sk) 1056 { 1057 atomic_inc(&sk->sk_prot->socks); 1058 } 1059 1060 static inline void sk_refcnt_debug_dec(struct sock *sk) 1061 { 1062 atomic_dec(&sk->sk_prot->socks); 1063 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1064 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1065 } 1066 1067 static inline void sk_refcnt_debug_release(const struct sock *sk) 1068 { 1069 if (atomic_read(&sk->sk_refcnt) != 1) 1070 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1071 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1072 } 1073 #else /* SOCK_REFCNT_DEBUG */ 1074 #define sk_refcnt_debug_inc(sk) do { } while (0) 1075 #define sk_refcnt_debug_dec(sk) do { } while (0) 1076 #define sk_refcnt_debug_release(sk) do { } while (0) 1077 #endif /* SOCK_REFCNT_DEBUG */ 1078 1079 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1080 extern struct static_key memcg_socket_limit_enabled; 1081 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1082 struct cg_proto *cg_proto) 1083 { 1084 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1085 } 1086 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1087 #else 1088 #define mem_cgroup_sockets_enabled 0 1089 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1090 struct cg_proto *cg_proto) 1091 { 1092 return NULL; 1093 } 1094 #endif 1095 1096 static inline bool sk_stream_memory_free(const struct sock *sk) 1097 { 1098 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1099 return false; 1100 1101 return sk->sk_prot->stream_memory_free ? 1102 sk->sk_prot->stream_memory_free(sk) : true; 1103 } 1104 1105 static inline bool sk_stream_is_writeable(const struct sock *sk) 1106 { 1107 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1108 sk_stream_memory_free(sk); 1109 } 1110 1111 1112 static inline bool sk_has_memory_pressure(const struct sock *sk) 1113 { 1114 return sk->sk_prot->memory_pressure != NULL; 1115 } 1116 1117 static inline bool sk_under_memory_pressure(const struct sock *sk) 1118 { 1119 if (!sk->sk_prot->memory_pressure) 1120 return false; 1121 1122 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1123 return !!*sk->sk_cgrp->memory_pressure; 1124 1125 return !!*sk->sk_prot->memory_pressure; 1126 } 1127 1128 static inline void sk_leave_memory_pressure(struct sock *sk) 1129 { 1130 int *memory_pressure = sk->sk_prot->memory_pressure; 1131 1132 if (!memory_pressure) 1133 return; 1134 1135 if (*memory_pressure) 1136 *memory_pressure = 0; 1137 1138 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1139 struct cg_proto *cg_proto = sk->sk_cgrp; 1140 struct proto *prot = sk->sk_prot; 1141 1142 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1143 if (*cg_proto->memory_pressure) 1144 *cg_proto->memory_pressure = 0; 1145 } 1146 1147 } 1148 1149 static inline void sk_enter_memory_pressure(struct sock *sk) 1150 { 1151 if (!sk->sk_prot->enter_memory_pressure) 1152 return; 1153 1154 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1155 struct cg_proto *cg_proto = sk->sk_cgrp; 1156 struct proto *prot = sk->sk_prot; 1157 1158 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1159 cg_proto->enter_memory_pressure(sk); 1160 } 1161 1162 sk->sk_prot->enter_memory_pressure(sk); 1163 } 1164 1165 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1166 { 1167 long *prot = sk->sk_prot->sysctl_mem; 1168 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1169 prot = sk->sk_cgrp->sysctl_mem; 1170 return prot[index]; 1171 } 1172 1173 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1174 unsigned long amt, 1175 int *parent_status) 1176 { 1177 struct res_counter *fail; 1178 int ret; 1179 1180 ret = res_counter_charge_nofail(prot->memory_allocated, 1181 amt << PAGE_SHIFT, &fail); 1182 if (ret < 0) 1183 *parent_status = OVER_LIMIT; 1184 } 1185 1186 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1187 unsigned long amt) 1188 { 1189 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT); 1190 } 1191 1192 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot) 1193 { 1194 u64 ret; 1195 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE); 1196 return ret >> PAGE_SHIFT; 1197 } 1198 1199 static inline long 1200 sk_memory_allocated(const struct sock *sk) 1201 { 1202 struct proto *prot = sk->sk_prot; 1203 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1204 return memcg_memory_allocated_read(sk->sk_cgrp); 1205 1206 return atomic_long_read(prot->memory_allocated); 1207 } 1208 1209 static inline long 1210 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1211 { 1212 struct proto *prot = sk->sk_prot; 1213 1214 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1215 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1216 /* update the root cgroup regardless */ 1217 atomic_long_add_return(amt, prot->memory_allocated); 1218 return memcg_memory_allocated_read(sk->sk_cgrp); 1219 } 1220 1221 return atomic_long_add_return(amt, prot->memory_allocated); 1222 } 1223 1224 static inline void 1225 sk_memory_allocated_sub(struct sock *sk, int amt) 1226 { 1227 struct proto *prot = sk->sk_prot; 1228 1229 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1230 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1231 1232 atomic_long_sub(amt, prot->memory_allocated); 1233 } 1234 1235 static inline void sk_sockets_allocated_dec(struct sock *sk) 1236 { 1237 struct proto *prot = sk->sk_prot; 1238 1239 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1240 struct cg_proto *cg_proto = sk->sk_cgrp; 1241 1242 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1243 percpu_counter_dec(cg_proto->sockets_allocated); 1244 } 1245 1246 percpu_counter_dec(prot->sockets_allocated); 1247 } 1248 1249 static inline void sk_sockets_allocated_inc(struct sock *sk) 1250 { 1251 struct proto *prot = sk->sk_prot; 1252 1253 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1254 struct cg_proto *cg_proto = sk->sk_cgrp; 1255 1256 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1257 percpu_counter_inc(cg_proto->sockets_allocated); 1258 } 1259 1260 percpu_counter_inc(prot->sockets_allocated); 1261 } 1262 1263 static inline int 1264 sk_sockets_allocated_read_positive(struct sock *sk) 1265 { 1266 struct proto *prot = sk->sk_prot; 1267 1268 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1269 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated); 1270 1271 return percpu_counter_read_positive(prot->sockets_allocated); 1272 } 1273 1274 static inline int 1275 proto_sockets_allocated_sum_positive(struct proto *prot) 1276 { 1277 return percpu_counter_sum_positive(prot->sockets_allocated); 1278 } 1279 1280 static inline long 1281 proto_memory_allocated(struct proto *prot) 1282 { 1283 return atomic_long_read(prot->memory_allocated); 1284 } 1285 1286 static inline bool 1287 proto_memory_pressure(struct proto *prot) 1288 { 1289 if (!prot->memory_pressure) 1290 return false; 1291 return !!*prot->memory_pressure; 1292 } 1293 1294 1295 #ifdef CONFIG_PROC_FS 1296 /* Called with local bh disabled */ 1297 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1298 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1299 #else 1300 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1301 int inc) 1302 { 1303 } 1304 #endif 1305 1306 1307 /* With per-bucket locks this operation is not-atomic, so that 1308 * this version is not worse. 1309 */ 1310 static inline void __sk_prot_rehash(struct sock *sk) 1311 { 1312 sk->sk_prot->unhash(sk); 1313 sk->sk_prot->hash(sk); 1314 } 1315 1316 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1317 1318 /* About 10 seconds */ 1319 #define SOCK_DESTROY_TIME (10*HZ) 1320 1321 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1322 #define PROT_SOCK 1024 1323 1324 #define SHUTDOWN_MASK 3 1325 #define RCV_SHUTDOWN 1 1326 #define SEND_SHUTDOWN 2 1327 1328 #define SOCK_SNDBUF_LOCK 1 1329 #define SOCK_RCVBUF_LOCK 2 1330 #define SOCK_BINDADDR_LOCK 4 1331 #define SOCK_BINDPORT_LOCK 8 1332 1333 /* sock_iocb: used to kick off async processing of socket ios */ 1334 struct sock_iocb { 1335 struct list_head list; 1336 1337 int flags; 1338 int size; 1339 struct socket *sock; 1340 struct sock *sk; 1341 struct scm_cookie *scm; 1342 struct msghdr *msg, async_msg; 1343 struct kiocb *kiocb; 1344 }; 1345 1346 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 1347 { 1348 return (struct sock_iocb *)iocb->private; 1349 } 1350 1351 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 1352 { 1353 return si->kiocb; 1354 } 1355 1356 struct socket_alloc { 1357 struct socket socket; 1358 struct inode vfs_inode; 1359 }; 1360 1361 static inline struct socket *SOCKET_I(struct inode *inode) 1362 { 1363 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1364 } 1365 1366 static inline struct inode *SOCK_INODE(struct socket *socket) 1367 { 1368 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1369 } 1370 1371 /* 1372 * Functions for memory accounting 1373 */ 1374 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1375 void __sk_mem_reclaim(struct sock *sk); 1376 1377 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1378 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1379 #define SK_MEM_SEND 0 1380 #define SK_MEM_RECV 1 1381 1382 static inline int sk_mem_pages(int amt) 1383 { 1384 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1385 } 1386 1387 static inline bool sk_has_account(struct sock *sk) 1388 { 1389 /* return true if protocol supports memory accounting */ 1390 return !!sk->sk_prot->memory_allocated; 1391 } 1392 1393 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1394 { 1395 if (!sk_has_account(sk)) 1396 return true; 1397 return size <= sk->sk_forward_alloc || 1398 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1399 } 1400 1401 static inline bool 1402 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1403 { 1404 if (!sk_has_account(sk)) 1405 return true; 1406 return size<= sk->sk_forward_alloc || 1407 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1408 skb_pfmemalloc(skb); 1409 } 1410 1411 static inline void sk_mem_reclaim(struct sock *sk) 1412 { 1413 if (!sk_has_account(sk)) 1414 return; 1415 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1416 __sk_mem_reclaim(sk); 1417 } 1418 1419 static inline void sk_mem_reclaim_partial(struct sock *sk) 1420 { 1421 if (!sk_has_account(sk)) 1422 return; 1423 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1424 __sk_mem_reclaim(sk); 1425 } 1426 1427 static inline void sk_mem_charge(struct sock *sk, int size) 1428 { 1429 if (!sk_has_account(sk)) 1430 return; 1431 sk->sk_forward_alloc -= size; 1432 } 1433 1434 static inline void sk_mem_uncharge(struct sock *sk, int size) 1435 { 1436 if (!sk_has_account(sk)) 1437 return; 1438 sk->sk_forward_alloc += size; 1439 } 1440 1441 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1442 { 1443 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1444 sk->sk_wmem_queued -= skb->truesize; 1445 sk_mem_uncharge(sk, skb->truesize); 1446 __kfree_skb(skb); 1447 } 1448 1449 /* Used by processes to "lock" a socket state, so that 1450 * interrupts and bottom half handlers won't change it 1451 * from under us. It essentially blocks any incoming 1452 * packets, so that we won't get any new data or any 1453 * packets that change the state of the socket. 1454 * 1455 * While locked, BH processing will add new packets to 1456 * the backlog queue. This queue is processed by the 1457 * owner of the socket lock right before it is released. 1458 * 1459 * Since ~2.3.5 it is also exclusive sleep lock serializing 1460 * accesses from user process context. 1461 */ 1462 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 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); 1519 void sk_free(struct sock *sk); 1520 void sk_release_kernel(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 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, 1526 gfp_t priority); 1527 void sock_wfree(struct sk_buff *skb); 1528 void skb_orphan_partial(struct sk_buff *skb); 1529 void sock_rfree(struct sk_buff *skb); 1530 void sock_edemux(struct sk_buff *skb); 1531 1532 int sock_setsockopt(struct socket *sock, int level, int op, 1533 char __user *optval, unsigned int optlen); 1534 1535 int sock_getsockopt(struct socket *sock, int level, int op, 1536 char __user *optval, int __user *optlen); 1537 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1538 int noblock, int *errcode); 1539 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1540 unsigned long data_len, int noblock, 1541 int *errcode, int max_page_order); 1542 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1543 void sock_kfree_s(struct sock *sk, void *mem, int size); 1544 void sk_send_sigurg(struct sock *sk); 1545 1546 /* 1547 * Functions to fill in entries in struct proto_ops when a protocol 1548 * does not implement a particular function. 1549 */ 1550 int sock_no_bind(struct socket *, struct sockaddr *, int); 1551 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1552 int sock_no_socketpair(struct socket *, struct socket *); 1553 int sock_no_accept(struct socket *, struct socket *, int); 1554 int sock_no_getname(struct socket *, struct sockaddr *, int *, int); 1555 unsigned int sock_no_poll(struct file *, struct socket *, 1556 struct poll_table_struct *); 1557 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1558 int sock_no_listen(struct socket *, int); 1559 int sock_no_shutdown(struct socket *, int); 1560 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *); 1561 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int); 1562 int sock_no_sendmsg(struct kiocb *, struct socket *, struct msghdr *, size_t); 1563 int sock_no_recvmsg(struct kiocb *, struct socket *, struct msghdr *, size_t, 1564 int); 1565 int sock_no_mmap(struct file *file, struct socket *sock, 1566 struct vm_area_struct *vma); 1567 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1568 size_t size, int flags); 1569 1570 /* 1571 * Functions to fill in entries in struct proto_ops when a protocol 1572 * uses the inet style. 1573 */ 1574 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1575 char __user *optval, int __user *optlen); 1576 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1577 struct msghdr *msg, size_t size, int flags); 1578 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1579 char __user *optval, unsigned int optlen); 1580 int compat_sock_common_getsockopt(struct socket *sock, int level, 1581 int optname, char __user *optval, int __user *optlen); 1582 int compat_sock_common_setsockopt(struct socket *sock, int level, 1583 int optname, char __user *optval, unsigned int optlen); 1584 1585 void sk_common_release(struct sock *sk); 1586 1587 /* 1588 * Default socket callbacks and setup code 1589 */ 1590 1591 /* Initialise core socket variables */ 1592 void sock_init_data(struct socket *sock, struct sock *sk); 1593 1594 void sk_filter_release_rcu(struct rcu_head *rcu); 1595 1596 /** 1597 * sk_filter_release - release a socket filter 1598 * @fp: filter to remove 1599 * 1600 * Remove a filter from a socket and release its resources. 1601 */ 1602 1603 static inline void sk_filter_release(struct sk_filter *fp) 1604 { 1605 if (atomic_dec_and_test(&fp->refcnt)) 1606 call_rcu(&fp->rcu, sk_filter_release_rcu); 1607 } 1608 1609 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1610 { 1611 unsigned int size = sk_filter_len(fp); 1612 1613 atomic_sub(size, &sk->sk_omem_alloc); 1614 sk_filter_release(fp); 1615 } 1616 1617 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1618 { 1619 atomic_inc(&fp->refcnt); 1620 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1621 } 1622 1623 /* 1624 * Socket reference counting postulates. 1625 * 1626 * * Each user of socket SHOULD hold a reference count. 1627 * * Each access point to socket (an hash table bucket, reference from a list, 1628 * running timer, skb in flight MUST hold a reference count. 1629 * * When reference count hits 0, it means it will never increase back. 1630 * * When reference count hits 0, it means that no references from 1631 * outside exist to this socket and current process on current CPU 1632 * is last user and may/should destroy this socket. 1633 * * sk_free is called from any context: process, BH, IRQ. When 1634 * it is called, socket has no references from outside -> sk_free 1635 * may release descendant resources allocated by the socket, but 1636 * to the time when it is called, socket is NOT referenced by any 1637 * hash tables, lists etc. 1638 * * Packets, delivered from outside (from network or from another process) 1639 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1640 * when they sit in queue. Otherwise, packets will leak to hole, when 1641 * socket is looked up by one cpu and unhasing is made by another CPU. 1642 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1643 * (leak to backlog). Packet socket does all the processing inside 1644 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1645 * use separate SMP lock, so that they are prone too. 1646 */ 1647 1648 /* Ungrab socket and destroy it, if it was the last reference. */ 1649 static inline void sock_put(struct sock *sk) 1650 { 1651 if (atomic_dec_and_test(&sk->sk_refcnt)) 1652 sk_free(sk); 1653 } 1654 1655 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested); 1656 1657 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1658 { 1659 sk->sk_tx_queue_mapping = tx_queue; 1660 } 1661 1662 static inline void sk_tx_queue_clear(struct sock *sk) 1663 { 1664 sk->sk_tx_queue_mapping = -1; 1665 } 1666 1667 static inline int sk_tx_queue_get(const struct sock *sk) 1668 { 1669 return sk ? sk->sk_tx_queue_mapping : -1; 1670 } 1671 1672 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1673 { 1674 sk_tx_queue_clear(sk); 1675 sk->sk_socket = sock; 1676 } 1677 1678 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1679 { 1680 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1681 return &rcu_dereference_raw(sk->sk_wq)->wait; 1682 } 1683 /* Detach socket from process context. 1684 * Announce socket dead, detach it from wait queue and inode. 1685 * Note that parent inode held reference count on this struct sock, 1686 * we do not release it in this function, because protocol 1687 * probably wants some additional cleanups or even continuing 1688 * to work with this socket (TCP). 1689 */ 1690 static inline void sock_orphan(struct sock *sk) 1691 { 1692 write_lock_bh(&sk->sk_callback_lock); 1693 sock_set_flag(sk, SOCK_DEAD); 1694 sk_set_socket(sk, NULL); 1695 sk->sk_wq = NULL; 1696 write_unlock_bh(&sk->sk_callback_lock); 1697 } 1698 1699 static inline void sock_graft(struct sock *sk, struct socket *parent) 1700 { 1701 write_lock_bh(&sk->sk_callback_lock); 1702 sk->sk_wq = parent->wq; 1703 parent->sk = sk; 1704 sk_set_socket(sk, parent); 1705 security_sock_graft(sk, parent); 1706 write_unlock_bh(&sk->sk_callback_lock); 1707 } 1708 1709 kuid_t sock_i_uid(struct sock *sk); 1710 unsigned long sock_i_ino(struct sock *sk); 1711 1712 static inline struct dst_entry * 1713 __sk_dst_get(struct sock *sk) 1714 { 1715 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1716 lockdep_is_held(&sk->sk_lock.slock)); 1717 } 1718 1719 static inline struct dst_entry * 1720 sk_dst_get(struct sock *sk) 1721 { 1722 struct dst_entry *dst; 1723 1724 rcu_read_lock(); 1725 dst = rcu_dereference(sk->sk_dst_cache); 1726 if (dst) 1727 dst_hold(dst); 1728 rcu_read_unlock(); 1729 return dst; 1730 } 1731 1732 void sk_reset_txq(struct sock *sk); 1733 1734 static inline void dst_negative_advice(struct sock *sk) 1735 { 1736 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1737 1738 if (dst && dst->ops->negative_advice) { 1739 ndst = dst->ops->negative_advice(dst); 1740 1741 if (ndst != dst) { 1742 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1743 sk_reset_txq(sk); 1744 } 1745 } 1746 } 1747 1748 static inline void 1749 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1750 { 1751 struct dst_entry *old_dst; 1752 1753 sk_tx_queue_clear(sk); 1754 /* 1755 * This can be called while sk is owned by the caller only, 1756 * with no state that can be checked in a rcu_dereference_check() cond 1757 */ 1758 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1759 rcu_assign_pointer(sk->sk_dst_cache, dst); 1760 dst_release(old_dst); 1761 } 1762 1763 static inline void 1764 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1765 { 1766 spin_lock(&sk->sk_dst_lock); 1767 __sk_dst_set(sk, dst); 1768 spin_unlock(&sk->sk_dst_lock); 1769 } 1770 1771 static inline void 1772 __sk_dst_reset(struct sock *sk) 1773 { 1774 __sk_dst_set(sk, NULL); 1775 } 1776 1777 static inline void 1778 sk_dst_reset(struct sock *sk) 1779 { 1780 spin_lock(&sk->sk_dst_lock); 1781 __sk_dst_reset(sk); 1782 spin_unlock(&sk->sk_dst_lock); 1783 } 1784 1785 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1786 1787 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1788 1789 static inline bool sk_can_gso(const struct sock *sk) 1790 { 1791 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1792 } 1793 1794 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1795 1796 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1797 { 1798 sk->sk_route_nocaps |= flags; 1799 sk->sk_route_caps &= ~flags; 1800 } 1801 1802 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1803 char __user *from, char *to, 1804 int copy, int offset) 1805 { 1806 if (skb->ip_summed == CHECKSUM_NONE) { 1807 int err = 0; 1808 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1809 if (err) 1810 return err; 1811 skb->csum = csum_block_add(skb->csum, csum, offset); 1812 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1813 if (!access_ok(VERIFY_READ, from, copy) || 1814 __copy_from_user_nocache(to, from, copy)) 1815 return -EFAULT; 1816 } else if (copy_from_user(to, from, copy)) 1817 return -EFAULT; 1818 1819 return 0; 1820 } 1821 1822 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1823 char __user *from, int copy) 1824 { 1825 int err, offset = skb->len; 1826 1827 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1828 copy, offset); 1829 if (err) 1830 __skb_trim(skb, offset); 1831 1832 return err; 1833 } 1834 1835 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from, 1836 struct sk_buff *skb, 1837 struct page *page, 1838 int off, int copy) 1839 { 1840 int err; 1841 1842 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1843 copy, skb->len); 1844 if (err) 1845 return err; 1846 1847 skb->len += copy; 1848 skb->data_len += copy; 1849 skb->truesize += copy; 1850 sk->sk_wmem_queued += copy; 1851 sk_mem_charge(sk, copy); 1852 return 0; 1853 } 1854 1855 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1856 struct sk_buff *skb, struct page *page, 1857 int off, int copy) 1858 { 1859 if (skb->ip_summed == CHECKSUM_NONE) { 1860 int err = 0; 1861 __wsum csum = csum_and_copy_from_user(from, 1862 page_address(page) + off, 1863 copy, 0, &err); 1864 if (err) 1865 return err; 1866 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1867 } else if (copy_from_user(page_address(page) + off, from, copy)) 1868 return -EFAULT; 1869 1870 skb->len += copy; 1871 skb->data_len += copy; 1872 skb->truesize += copy; 1873 sk->sk_wmem_queued += copy; 1874 sk_mem_charge(sk, copy); 1875 return 0; 1876 } 1877 1878 /** 1879 * sk_wmem_alloc_get - returns write allocations 1880 * @sk: socket 1881 * 1882 * Returns sk_wmem_alloc minus initial offset of one 1883 */ 1884 static inline int sk_wmem_alloc_get(const struct sock *sk) 1885 { 1886 return atomic_read(&sk->sk_wmem_alloc) - 1; 1887 } 1888 1889 /** 1890 * sk_rmem_alloc_get - returns read allocations 1891 * @sk: socket 1892 * 1893 * Returns sk_rmem_alloc 1894 */ 1895 static inline int sk_rmem_alloc_get(const struct sock *sk) 1896 { 1897 return atomic_read(&sk->sk_rmem_alloc); 1898 } 1899 1900 /** 1901 * sk_has_allocations - check if allocations are outstanding 1902 * @sk: socket 1903 * 1904 * Returns true if socket has write or read allocations 1905 */ 1906 static inline bool sk_has_allocations(const struct sock *sk) 1907 { 1908 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1909 } 1910 1911 /** 1912 * wq_has_sleeper - check if there are any waiting processes 1913 * @wq: struct socket_wq 1914 * 1915 * Returns true if socket_wq has waiting processes 1916 * 1917 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1918 * barrier call. They were added due to the race found within the tcp code. 1919 * 1920 * Consider following tcp code paths: 1921 * 1922 * CPU1 CPU2 1923 * 1924 * sys_select receive packet 1925 * ... ... 1926 * __add_wait_queue update tp->rcv_nxt 1927 * ... ... 1928 * tp->rcv_nxt check sock_def_readable 1929 * ... { 1930 * schedule rcu_read_lock(); 1931 * wq = rcu_dereference(sk->sk_wq); 1932 * if (wq && waitqueue_active(&wq->wait)) 1933 * wake_up_interruptible(&wq->wait) 1934 * ... 1935 * } 1936 * 1937 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1938 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1939 * could then endup calling schedule and sleep forever if there are no more 1940 * data on the socket. 1941 * 1942 */ 1943 static inline bool wq_has_sleeper(struct socket_wq *wq) 1944 { 1945 /* We need to be sure we are in sync with the 1946 * add_wait_queue modifications to the wait queue. 1947 * 1948 * This memory barrier is paired in the sock_poll_wait. 1949 */ 1950 smp_mb(); 1951 return wq && waitqueue_active(&wq->wait); 1952 } 1953 1954 /** 1955 * sock_poll_wait - place memory barrier behind the poll_wait call. 1956 * @filp: file 1957 * @wait_address: socket wait queue 1958 * @p: poll_table 1959 * 1960 * See the comments in the wq_has_sleeper function. 1961 */ 1962 static inline void sock_poll_wait(struct file *filp, 1963 wait_queue_head_t *wait_address, poll_table *p) 1964 { 1965 if (!poll_does_not_wait(p) && wait_address) { 1966 poll_wait(filp, wait_address, p); 1967 /* We need to be sure we are in sync with the 1968 * socket flags modification. 1969 * 1970 * This memory barrier is paired in the wq_has_sleeper. 1971 */ 1972 smp_mb(); 1973 } 1974 } 1975 1976 /* 1977 * Queue a received datagram if it will fit. Stream and sequenced 1978 * protocols can't normally use this as they need to fit buffers in 1979 * and play with them. 1980 * 1981 * Inlined as it's very short and called for pretty much every 1982 * packet ever received. 1983 */ 1984 1985 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1986 { 1987 skb_orphan(skb); 1988 skb->sk = sk; 1989 skb->destructor = sock_wfree; 1990 /* 1991 * We used to take a refcount on sk, but following operation 1992 * is enough to guarantee sk_free() wont free this sock until 1993 * all in-flight packets are completed 1994 */ 1995 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1996 } 1997 1998 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1999 { 2000 skb_orphan(skb); 2001 skb->sk = sk; 2002 skb->destructor = sock_rfree; 2003 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2004 sk_mem_charge(sk, skb->truesize); 2005 } 2006 2007 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2008 unsigned long expires); 2009 2010 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2011 2012 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2013 2014 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2015 2016 /* 2017 * Recover an error report and clear atomically 2018 */ 2019 2020 static inline int sock_error(struct sock *sk) 2021 { 2022 int err; 2023 if (likely(!sk->sk_err)) 2024 return 0; 2025 err = xchg(&sk->sk_err, 0); 2026 return -err; 2027 } 2028 2029 static inline unsigned long sock_wspace(struct sock *sk) 2030 { 2031 int amt = 0; 2032 2033 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2034 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2035 if (amt < 0) 2036 amt = 0; 2037 } 2038 return amt; 2039 } 2040 2041 static inline void sk_wake_async(struct sock *sk, int how, int band) 2042 { 2043 if (sock_flag(sk, SOCK_FASYNC)) 2044 sock_wake_async(sk->sk_socket, how, band); 2045 } 2046 2047 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2048 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2049 * Note: for send buffers, TCP works better if we can build two skbs at 2050 * minimum. 2051 */ 2052 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2053 2054 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2055 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2056 2057 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2058 { 2059 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2060 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2061 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2062 } 2063 } 2064 2065 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 2066 2067 /** 2068 * sk_page_frag - return an appropriate page_frag 2069 * @sk: socket 2070 * 2071 * If socket allocation mode allows current thread to sleep, it means its 2072 * safe to use the per task page_frag instead of the per socket one. 2073 */ 2074 static inline struct page_frag *sk_page_frag(struct sock *sk) 2075 { 2076 if (sk->sk_allocation & __GFP_WAIT) 2077 return ¤t->task_frag; 2078 2079 return &sk->sk_frag; 2080 } 2081 2082 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2083 2084 /* 2085 * Default write policy as shown to user space via poll/select/SIGIO 2086 */ 2087 static inline bool sock_writeable(const struct sock *sk) 2088 { 2089 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2090 } 2091 2092 static inline gfp_t gfp_any(void) 2093 { 2094 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2095 } 2096 2097 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2098 { 2099 return noblock ? 0 : sk->sk_rcvtimeo; 2100 } 2101 2102 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2103 { 2104 return noblock ? 0 : sk->sk_sndtimeo; 2105 } 2106 2107 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2108 { 2109 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2110 } 2111 2112 /* Alas, with timeout socket operations are not restartable. 2113 * Compare this to poll(). 2114 */ 2115 static inline int sock_intr_errno(long timeo) 2116 { 2117 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2118 } 2119 2120 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2121 struct sk_buff *skb); 2122 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2123 struct sk_buff *skb); 2124 2125 static inline void 2126 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2127 { 2128 ktime_t kt = skb->tstamp; 2129 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2130 2131 /* 2132 * generate control messages if 2133 * - receive time stamping in software requested (SOCK_RCVTSTAMP 2134 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 2135 * - software time stamp available and wanted 2136 * (SOCK_TIMESTAMPING_SOFTWARE) 2137 * - hardware time stamps available and wanted 2138 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 2139 * SOCK_TIMESTAMPING_RAW_HARDWARE) 2140 */ 2141 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2142 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 2143 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 2144 (hwtstamps->hwtstamp.tv64 && 2145 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 2146 (hwtstamps->syststamp.tv64 && 2147 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 2148 __sock_recv_timestamp(msg, sk, skb); 2149 else 2150 sk->sk_stamp = kt; 2151 2152 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2153 __sock_recv_wifi_status(msg, sk, skb); 2154 } 2155 2156 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2157 struct sk_buff *skb); 2158 2159 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2160 struct sk_buff *skb) 2161 { 2162 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2163 (1UL << SOCK_RCVTSTAMP) | \ 2164 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 2165 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 2166 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 2167 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 2168 2169 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 2170 __sock_recv_ts_and_drops(msg, sk, skb); 2171 else 2172 sk->sk_stamp = skb->tstamp; 2173 } 2174 2175 /** 2176 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2177 * @sk: socket sending this packet 2178 * @tx_flags: filled with instructions for time stamping 2179 * 2180 * Currently only depends on SOCK_TIMESTAMPING* flags. 2181 */ 2182 void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 2183 2184 /** 2185 * sk_eat_skb - Release a skb if it is no longer needed 2186 * @sk: socket to eat this skb from 2187 * @skb: socket buffer to eat 2188 * @copied_early: flag indicating whether DMA operations copied this data early 2189 * 2190 * This routine must be called with interrupts disabled or with the socket 2191 * locked so that the sk_buff queue operation is ok. 2192 */ 2193 #ifdef CONFIG_NET_DMA 2194 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2195 { 2196 __skb_unlink(skb, &sk->sk_receive_queue); 2197 if (!copied_early) 2198 __kfree_skb(skb); 2199 else 2200 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 2201 } 2202 #else 2203 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2204 { 2205 __skb_unlink(skb, &sk->sk_receive_queue); 2206 __kfree_skb(skb); 2207 } 2208 #endif 2209 2210 static inline 2211 struct net *sock_net(const struct sock *sk) 2212 { 2213 return read_pnet(&sk->sk_net); 2214 } 2215 2216 static inline 2217 void sock_net_set(struct sock *sk, struct net *net) 2218 { 2219 write_pnet(&sk->sk_net, net); 2220 } 2221 2222 /* 2223 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2224 * They should not hold a reference to a namespace in order to allow 2225 * to stop it. 2226 * Sockets after sk_change_net should be released using sk_release_kernel 2227 */ 2228 static inline void sk_change_net(struct sock *sk, struct net *net) 2229 { 2230 put_net(sock_net(sk)); 2231 sock_net_set(sk, hold_net(net)); 2232 } 2233 2234 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2235 { 2236 if (skb->sk) { 2237 struct sock *sk = skb->sk; 2238 2239 skb->destructor = NULL; 2240 skb->sk = NULL; 2241 return sk; 2242 } 2243 return NULL; 2244 } 2245 2246 void sock_enable_timestamp(struct sock *sk, int flag); 2247 int sock_get_timestamp(struct sock *, struct timeval __user *); 2248 int sock_get_timestampns(struct sock *, struct timespec __user *); 2249 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2250 int type); 2251 2252 /* 2253 * Enable debug/info messages 2254 */ 2255 extern int net_msg_warn; 2256 #define NETDEBUG(fmt, args...) \ 2257 do { if (net_msg_warn) printk(fmt,##args); } while (0) 2258 2259 #define LIMIT_NETDEBUG(fmt, args...) \ 2260 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 2261 2262 extern __u32 sysctl_wmem_max; 2263 extern __u32 sysctl_rmem_max; 2264 2265 extern int sysctl_optmem_max; 2266 2267 extern __u32 sysctl_wmem_default; 2268 extern __u32 sysctl_rmem_default; 2269 2270 #endif /* _SOCK_H */ 2271