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_sndbuf: size of send buffer in bytes 236 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 237 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 238 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets 239 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 240 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 241 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 242 * @sk_gso_max_size: Maximum GSO segment size to build 243 * @sk_gso_max_segs: Maximum number of GSO segments 244 * @sk_lingertime: %SO_LINGER l_linger setting 245 * @sk_backlog: always used with the per-socket spinlock held 246 * @sk_callback_lock: used with the callbacks in the end of this struct 247 * @sk_error_queue: rarely used 248 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 249 * IPV6_ADDRFORM for instance) 250 * @sk_err: last error 251 * @sk_err_soft: errors that don't cause failure but are the cause of a 252 * persistent failure not just 'timed out' 253 * @sk_drops: raw/udp drops counter 254 * @sk_ack_backlog: current listen backlog 255 * @sk_max_ack_backlog: listen backlog set in listen() 256 * @sk_priority: %SO_PRIORITY setting 257 * @sk_cgrp_prioidx: socket group's priority map index 258 * @sk_type: socket type (%SOCK_STREAM, etc) 259 * @sk_protocol: which protocol this socket belongs in this network family 260 * @sk_peer_pid: &struct pid for this socket's peer 261 * @sk_peer_cred: %SO_PEERCRED setting 262 * @sk_rcvlowat: %SO_RCVLOWAT setting 263 * @sk_rcvtimeo: %SO_RCVTIMEO setting 264 * @sk_sndtimeo: %SO_SNDTIMEO setting 265 * @sk_rxhash: flow hash received from netif layer 266 * @sk_filter: socket filtering instructions 267 * @sk_protinfo: private area, net family specific, when not using slab 268 * @sk_timer: sock cleanup timer 269 * @sk_stamp: time stamp of last packet received 270 * @sk_socket: Identd and reporting IO signals 271 * @sk_user_data: RPC layer private data 272 * @sk_frag: cached page frag 273 * @sk_peek_off: current peek_offset value 274 * @sk_send_head: front of stuff to transmit 275 * @sk_security: used by security modules 276 * @sk_mark: generic packet mark 277 * @sk_classid: this socket's cgroup classid 278 * @sk_cgrp: this socket's cgroup-specific proto data 279 * @sk_write_pending: a write to stream socket waits to start 280 * @sk_state_change: callback to indicate change in the state of the sock 281 * @sk_data_ready: callback to indicate there is data to be processed 282 * @sk_write_space: callback to indicate there is bf sending space available 283 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 284 * @sk_backlog_rcv: callback to process the backlog 285 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 286 */ 287 struct sock { 288 /* 289 * Now struct inet_timewait_sock also uses sock_common, so please just 290 * don't add nothing before this first member (__sk_common) --acme 291 */ 292 struct sock_common __sk_common; 293 #define sk_node __sk_common.skc_node 294 #define sk_nulls_node __sk_common.skc_nulls_node 295 #define sk_refcnt __sk_common.skc_refcnt 296 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 297 298 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 299 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 300 #define sk_hash __sk_common.skc_hash 301 #define sk_family __sk_common.skc_family 302 #define sk_state __sk_common.skc_state 303 #define sk_reuse __sk_common.skc_reuse 304 #define sk_reuseport __sk_common.skc_reuseport 305 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 306 #define sk_bind_node __sk_common.skc_bind_node 307 #define sk_prot __sk_common.skc_prot 308 #define sk_net __sk_common.skc_net 309 socket_lock_t sk_lock; 310 struct sk_buff_head sk_receive_queue; 311 /* 312 * The backlog queue is special, it is always used with 313 * the per-socket spinlock held and requires low latency 314 * access. Therefore we special case it's implementation. 315 * Note : rmem_alloc is in this structure to fill a hole 316 * on 64bit arches, not because its logically part of 317 * backlog. 318 */ 319 struct { 320 atomic_t rmem_alloc; 321 int len; 322 struct sk_buff *head; 323 struct sk_buff *tail; 324 } sk_backlog; 325 #define sk_rmem_alloc sk_backlog.rmem_alloc 326 int sk_forward_alloc; 327 #ifdef CONFIG_RPS 328 __u32 sk_rxhash; 329 #endif 330 #ifdef CONFIG_NET_LL_RX_POLL 331 unsigned int sk_napi_id; 332 unsigned int sk_ll_usec; 333 #endif 334 atomic_t sk_drops; 335 int sk_rcvbuf; 336 337 struct sk_filter __rcu *sk_filter; 338 struct socket_wq __rcu *sk_wq; 339 340 #ifdef CONFIG_NET_DMA 341 struct sk_buff_head sk_async_wait_queue; 342 #endif 343 344 #ifdef CONFIG_XFRM 345 struct xfrm_policy *sk_policy[2]; 346 #endif 347 unsigned long sk_flags; 348 struct dst_entry *sk_rx_dst; 349 struct dst_entry __rcu *sk_dst_cache; 350 spinlock_t sk_dst_lock; 351 atomic_t sk_wmem_alloc; 352 atomic_t sk_omem_alloc; 353 int sk_sndbuf; 354 struct sk_buff_head sk_write_queue; 355 kmemcheck_bitfield_begin(flags); 356 unsigned int sk_shutdown : 2, 357 sk_no_check : 2, 358 sk_userlocks : 4, 359 sk_protocol : 8, 360 sk_type : 16; 361 kmemcheck_bitfield_end(flags); 362 int sk_wmem_queued; 363 gfp_t sk_allocation; 364 netdev_features_t sk_route_caps; 365 netdev_features_t sk_route_nocaps; 366 int sk_gso_type; 367 unsigned int sk_gso_max_size; 368 u16 sk_gso_max_segs; 369 int sk_rcvlowat; 370 unsigned long sk_lingertime; 371 struct sk_buff_head sk_error_queue; 372 struct proto *sk_prot_creator; 373 rwlock_t sk_callback_lock; 374 int sk_err, 375 sk_err_soft; 376 unsigned short sk_ack_backlog; 377 unsigned short sk_max_ack_backlog; 378 __u32 sk_priority; 379 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) 380 __u32 sk_cgrp_prioidx; 381 #endif 382 struct pid *sk_peer_pid; 383 const struct cred *sk_peer_cred; 384 long sk_rcvtimeo; 385 long sk_sndtimeo; 386 void *sk_protinfo; 387 struct timer_list sk_timer; 388 ktime_t sk_stamp; 389 struct socket *sk_socket; 390 void *sk_user_data; 391 struct page_frag sk_frag; 392 struct sk_buff *sk_send_head; 393 __s32 sk_peek_off; 394 int sk_write_pending; 395 #ifdef CONFIG_SECURITY 396 void *sk_security; 397 #endif 398 __u32 sk_mark; 399 u32 sk_classid; 400 struct cg_proto *sk_cgrp; 401 void (*sk_state_change)(struct sock *sk); 402 void (*sk_data_ready)(struct sock *sk, int bytes); 403 void (*sk_write_space)(struct sock *sk); 404 void (*sk_error_report)(struct sock *sk); 405 int (*sk_backlog_rcv)(struct sock *sk, 406 struct sk_buff *skb); 407 void (*sk_destruct)(struct sock *sk); 408 }; 409 410 /* 411 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 412 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 413 * on a socket means that the socket will reuse everybody else's port 414 * without looking at the other's sk_reuse value. 415 */ 416 417 #define SK_NO_REUSE 0 418 #define SK_CAN_REUSE 1 419 #define SK_FORCE_REUSE 2 420 421 static inline int sk_peek_offset(struct sock *sk, int flags) 422 { 423 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 424 return sk->sk_peek_off; 425 else 426 return 0; 427 } 428 429 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 430 { 431 if (sk->sk_peek_off >= 0) { 432 if (sk->sk_peek_off >= val) 433 sk->sk_peek_off -= val; 434 else 435 sk->sk_peek_off = 0; 436 } 437 } 438 439 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 440 { 441 if (sk->sk_peek_off >= 0) 442 sk->sk_peek_off += val; 443 } 444 445 /* 446 * Hashed lists helper routines 447 */ 448 static inline struct sock *sk_entry(const struct hlist_node *node) 449 { 450 return hlist_entry(node, struct sock, sk_node); 451 } 452 453 static inline struct sock *__sk_head(const struct hlist_head *head) 454 { 455 return hlist_entry(head->first, struct sock, sk_node); 456 } 457 458 static inline struct sock *sk_head(const struct hlist_head *head) 459 { 460 return hlist_empty(head) ? NULL : __sk_head(head); 461 } 462 463 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 464 { 465 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 466 } 467 468 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 469 { 470 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 471 } 472 473 static inline struct sock *sk_next(const struct sock *sk) 474 { 475 return sk->sk_node.next ? 476 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 477 } 478 479 static inline struct sock *sk_nulls_next(const struct sock *sk) 480 { 481 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 482 hlist_nulls_entry(sk->sk_nulls_node.next, 483 struct sock, sk_nulls_node) : 484 NULL; 485 } 486 487 static inline bool sk_unhashed(const struct sock *sk) 488 { 489 return hlist_unhashed(&sk->sk_node); 490 } 491 492 static inline bool sk_hashed(const struct sock *sk) 493 { 494 return !sk_unhashed(sk); 495 } 496 497 static inline void sk_node_init(struct hlist_node *node) 498 { 499 node->pprev = NULL; 500 } 501 502 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 503 { 504 node->pprev = NULL; 505 } 506 507 static inline void __sk_del_node(struct sock *sk) 508 { 509 __hlist_del(&sk->sk_node); 510 } 511 512 /* NB: equivalent to hlist_del_init_rcu */ 513 static inline bool __sk_del_node_init(struct sock *sk) 514 { 515 if (sk_hashed(sk)) { 516 __sk_del_node(sk); 517 sk_node_init(&sk->sk_node); 518 return true; 519 } 520 return false; 521 } 522 523 /* Grab socket reference count. This operation is valid only 524 when sk is ALREADY grabbed f.e. it is found in hash table 525 or a list and the lookup is made under lock preventing hash table 526 modifications. 527 */ 528 529 static inline void sock_hold(struct sock *sk) 530 { 531 atomic_inc(&sk->sk_refcnt); 532 } 533 534 /* Ungrab socket in the context, which assumes that socket refcnt 535 cannot hit zero, f.e. it is true in context of any socketcall. 536 */ 537 static inline void __sock_put(struct sock *sk) 538 { 539 atomic_dec(&sk->sk_refcnt); 540 } 541 542 static inline bool sk_del_node_init(struct sock *sk) 543 { 544 bool rc = __sk_del_node_init(sk); 545 546 if (rc) { 547 /* paranoid for a while -acme */ 548 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 549 __sock_put(sk); 550 } 551 return rc; 552 } 553 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 554 555 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 556 { 557 if (sk_hashed(sk)) { 558 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 559 return true; 560 } 561 return false; 562 } 563 564 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 565 { 566 bool rc = __sk_nulls_del_node_init_rcu(sk); 567 568 if (rc) { 569 /* paranoid for a while -acme */ 570 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 571 __sock_put(sk); 572 } 573 return rc; 574 } 575 576 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 577 { 578 hlist_add_head(&sk->sk_node, list); 579 } 580 581 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 582 { 583 sock_hold(sk); 584 __sk_add_node(sk, list); 585 } 586 587 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 588 { 589 sock_hold(sk); 590 hlist_add_head_rcu(&sk->sk_node, list); 591 } 592 593 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 594 { 595 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 596 } 597 598 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 599 { 600 sock_hold(sk); 601 __sk_nulls_add_node_rcu(sk, list); 602 } 603 604 static inline void __sk_del_bind_node(struct sock *sk) 605 { 606 __hlist_del(&sk->sk_bind_node); 607 } 608 609 static inline void sk_add_bind_node(struct sock *sk, 610 struct hlist_head *list) 611 { 612 hlist_add_head(&sk->sk_bind_node, list); 613 } 614 615 #define sk_for_each(__sk, list) \ 616 hlist_for_each_entry(__sk, list, sk_node) 617 #define sk_for_each_rcu(__sk, list) \ 618 hlist_for_each_entry_rcu(__sk, list, sk_node) 619 #define sk_nulls_for_each(__sk, node, list) \ 620 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 621 #define sk_nulls_for_each_rcu(__sk, node, list) \ 622 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 623 #define sk_for_each_from(__sk) \ 624 hlist_for_each_entry_from(__sk, sk_node) 625 #define sk_nulls_for_each_from(__sk, node) \ 626 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 627 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 628 #define sk_for_each_safe(__sk, tmp, list) \ 629 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 630 #define sk_for_each_bound(__sk, list) \ 631 hlist_for_each_entry(__sk, list, sk_bind_node) 632 633 static inline struct user_namespace *sk_user_ns(struct sock *sk) 634 { 635 /* Careful only use this in a context where these parameters 636 * can not change and must all be valid, such as recvmsg from 637 * userspace. 638 */ 639 return sk->sk_socket->file->f_cred->user_ns; 640 } 641 642 /* Sock flags */ 643 enum sock_flags { 644 SOCK_DEAD, 645 SOCK_DONE, 646 SOCK_URGINLINE, 647 SOCK_KEEPOPEN, 648 SOCK_LINGER, 649 SOCK_DESTROY, 650 SOCK_BROADCAST, 651 SOCK_TIMESTAMP, 652 SOCK_ZAPPED, 653 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 654 SOCK_DBG, /* %SO_DEBUG setting */ 655 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 656 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 657 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 658 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 659 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 660 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 661 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 662 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 663 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 664 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 665 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 666 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 667 SOCK_FASYNC, /* fasync() active */ 668 SOCK_RXQ_OVFL, 669 SOCK_ZEROCOPY, /* buffers from userspace */ 670 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 671 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 672 * Will use last 4 bytes of packet sent from 673 * user-space instead. 674 */ 675 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 676 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 677 }; 678 679 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 680 { 681 nsk->sk_flags = osk->sk_flags; 682 } 683 684 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 685 { 686 __set_bit(flag, &sk->sk_flags); 687 } 688 689 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 690 { 691 __clear_bit(flag, &sk->sk_flags); 692 } 693 694 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 695 { 696 return test_bit(flag, &sk->sk_flags); 697 } 698 699 #ifdef CONFIG_NET 700 extern struct static_key memalloc_socks; 701 static inline int sk_memalloc_socks(void) 702 { 703 return static_key_false(&memalloc_socks); 704 } 705 #else 706 707 static inline int sk_memalloc_socks(void) 708 { 709 return 0; 710 } 711 712 #endif 713 714 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 715 { 716 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 717 } 718 719 static inline void sk_acceptq_removed(struct sock *sk) 720 { 721 sk->sk_ack_backlog--; 722 } 723 724 static inline void sk_acceptq_added(struct sock *sk) 725 { 726 sk->sk_ack_backlog++; 727 } 728 729 static inline bool sk_acceptq_is_full(const struct sock *sk) 730 { 731 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 732 } 733 734 /* 735 * Compute minimal free write space needed to queue new packets. 736 */ 737 static inline int sk_stream_min_wspace(const struct sock *sk) 738 { 739 return sk->sk_wmem_queued >> 1; 740 } 741 742 static inline int sk_stream_wspace(const struct sock *sk) 743 { 744 return sk->sk_sndbuf - sk->sk_wmem_queued; 745 } 746 747 extern void sk_stream_write_space(struct sock *sk); 748 749 static inline bool sk_stream_memory_free(const struct sock *sk) 750 { 751 return sk->sk_wmem_queued < sk->sk_sndbuf; 752 } 753 754 /* OOB backlog add */ 755 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 756 { 757 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 758 skb_dst_force(skb); 759 760 if (!sk->sk_backlog.tail) 761 sk->sk_backlog.head = skb; 762 else 763 sk->sk_backlog.tail->next = skb; 764 765 sk->sk_backlog.tail = skb; 766 skb->next = NULL; 767 } 768 769 /* 770 * Take into account size of receive queue and backlog queue 771 * Do not take into account this skb truesize, 772 * to allow even a single big packet to come. 773 */ 774 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb, 775 unsigned int limit) 776 { 777 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 778 779 return qsize > limit; 780 } 781 782 /* The per-socket spinlock must be held here. */ 783 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 784 unsigned int limit) 785 { 786 if (sk_rcvqueues_full(sk, skb, limit)) 787 return -ENOBUFS; 788 789 __sk_add_backlog(sk, skb); 790 sk->sk_backlog.len += skb->truesize; 791 return 0; 792 } 793 794 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 795 796 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 797 { 798 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 799 return __sk_backlog_rcv(sk, skb); 800 801 return sk->sk_backlog_rcv(sk, skb); 802 } 803 804 static inline void sock_rps_record_flow(const struct sock *sk) 805 { 806 #ifdef CONFIG_RPS 807 struct rps_sock_flow_table *sock_flow_table; 808 809 rcu_read_lock(); 810 sock_flow_table = rcu_dereference(rps_sock_flow_table); 811 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 812 rcu_read_unlock(); 813 #endif 814 } 815 816 static inline void sock_rps_reset_flow(const struct sock *sk) 817 { 818 #ifdef CONFIG_RPS 819 struct rps_sock_flow_table *sock_flow_table; 820 821 rcu_read_lock(); 822 sock_flow_table = rcu_dereference(rps_sock_flow_table); 823 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 824 rcu_read_unlock(); 825 #endif 826 } 827 828 static inline void sock_rps_save_rxhash(struct sock *sk, 829 const struct sk_buff *skb) 830 { 831 #ifdef CONFIG_RPS 832 if (unlikely(sk->sk_rxhash != skb->rxhash)) { 833 sock_rps_reset_flow(sk); 834 sk->sk_rxhash = skb->rxhash; 835 } 836 #endif 837 } 838 839 static inline void sock_rps_reset_rxhash(struct sock *sk) 840 { 841 #ifdef CONFIG_RPS 842 sock_rps_reset_flow(sk); 843 sk->sk_rxhash = 0; 844 #endif 845 } 846 847 #define sk_wait_event(__sk, __timeo, __condition) \ 848 ({ int __rc; \ 849 release_sock(__sk); \ 850 __rc = __condition; \ 851 if (!__rc) { \ 852 *(__timeo) = schedule_timeout(*(__timeo)); \ 853 } \ 854 lock_sock(__sk); \ 855 __rc = __condition; \ 856 __rc; \ 857 }) 858 859 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 860 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 861 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 862 extern int sk_stream_error(struct sock *sk, int flags, int err); 863 extern void sk_stream_kill_queues(struct sock *sk); 864 extern void sk_set_memalloc(struct sock *sk); 865 extern void sk_clear_memalloc(struct sock *sk); 866 867 extern int sk_wait_data(struct sock *sk, long *timeo); 868 869 struct request_sock_ops; 870 struct timewait_sock_ops; 871 struct inet_hashinfo; 872 struct raw_hashinfo; 873 struct module; 874 875 /* 876 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 877 * un-modified. Special care is taken when initializing object to zero. 878 */ 879 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 880 { 881 if (offsetof(struct sock, sk_node.next) != 0) 882 memset(sk, 0, offsetof(struct sock, sk_node.next)); 883 memset(&sk->sk_node.pprev, 0, 884 size - offsetof(struct sock, sk_node.pprev)); 885 } 886 887 /* Networking protocol blocks we attach to sockets. 888 * socket layer -> transport layer interface 889 * transport -> network interface is defined by struct inet_proto 890 */ 891 struct proto { 892 void (*close)(struct sock *sk, 893 long timeout); 894 int (*connect)(struct sock *sk, 895 struct sockaddr *uaddr, 896 int addr_len); 897 int (*disconnect)(struct sock *sk, int flags); 898 899 struct sock * (*accept)(struct sock *sk, int flags, int *err); 900 901 int (*ioctl)(struct sock *sk, int cmd, 902 unsigned long arg); 903 int (*init)(struct sock *sk); 904 void (*destroy)(struct sock *sk); 905 void (*shutdown)(struct sock *sk, int how); 906 int (*setsockopt)(struct sock *sk, int level, 907 int optname, char __user *optval, 908 unsigned int optlen); 909 int (*getsockopt)(struct sock *sk, int level, 910 int optname, char __user *optval, 911 int __user *option); 912 #ifdef CONFIG_COMPAT 913 int (*compat_setsockopt)(struct sock *sk, 914 int level, 915 int optname, char __user *optval, 916 unsigned int optlen); 917 int (*compat_getsockopt)(struct sock *sk, 918 int level, 919 int optname, char __user *optval, 920 int __user *option); 921 int (*compat_ioctl)(struct sock *sk, 922 unsigned int cmd, unsigned long arg); 923 #endif 924 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 925 struct msghdr *msg, size_t len); 926 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 927 struct msghdr *msg, 928 size_t len, int noblock, int flags, 929 int *addr_len); 930 int (*sendpage)(struct sock *sk, struct page *page, 931 int offset, size_t size, int flags); 932 int (*bind)(struct sock *sk, 933 struct sockaddr *uaddr, int addr_len); 934 935 int (*backlog_rcv) (struct sock *sk, 936 struct sk_buff *skb); 937 938 void (*release_cb)(struct sock *sk); 939 void (*mtu_reduced)(struct sock *sk); 940 941 /* Keeping track of sk's, looking them up, and port selection methods. */ 942 void (*hash)(struct sock *sk); 943 void (*unhash)(struct sock *sk); 944 void (*rehash)(struct sock *sk); 945 int (*get_port)(struct sock *sk, unsigned short snum); 946 void (*clear_sk)(struct sock *sk, int size); 947 948 /* Keeping track of sockets in use */ 949 #ifdef CONFIG_PROC_FS 950 unsigned int inuse_idx; 951 #endif 952 953 /* Memory pressure */ 954 void (*enter_memory_pressure)(struct sock *sk); 955 atomic_long_t *memory_allocated; /* Current allocated memory. */ 956 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 957 /* 958 * Pressure flag: try to collapse. 959 * Technical note: it is used by multiple contexts non atomically. 960 * All the __sk_mem_schedule() is of this nature: accounting 961 * is strict, actions are advisory and have some latency. 962 */ 963 int *memory_pressure; 964 long *sysctl_mem; 965 int *sysctl_wmem; 966 int *sysctl_rmem; 967 int max_header; 968 bool no_autobind; 969 970 struct kmem_cache *slab; 971 unsigned int obj_size; 972 int slab_flags; 973 974 struct percpu_counter *orphan_count; 975 976 struct request_sock_ops *rsk_prot; 977 struct timewait_sock_ops *twsk_prot; 978 979 union { 980 struct inet_hashinfo *hashinfo; 981 struct udp_table *udp_table; 982 struct raw_hashinfo *raw_hash; 983 } h; 984 985 struct module *owner; 986 987 char name[32]; 988 989 struct list_head node; 990 #ifdef SOCK_REFCNT_DEBUG 991 atomic_t socks; 992 #endif 993 #ifdef CONFIG_MEMCG_KMEM 994 /* 995 * cgroup specific init/deinit functions. Called once for all 996 * protocols that implement it, from cgroups populate function. 997 * This function has to setup any files the protocol want to 998 * appear in the kmem cgroup filesystem. 999 */ 1000 int (*init_cgroup)(struct mem_cgroup *memcg, 1001 struct cgroup_subsys *ss); 1002 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1003 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1004 #endif 1005 }; 1006 1007 /* 1008 * Bits in struct cg_proto.flags 1009 */ 1010 enum cg_proto_flags { 1011 /* Currently active and new sockets should be assigned to cgroups */ 1012 MEMCG_SOCK_ACTIVE, 1013 /* It was ever activated; we must disarm static keys on destruction */ 1014 MEMCG_SOCK_ACTIVATED, 1015 }; 1016 1017 struct cg_proto { 1018 void (*enter_memory_pressure)(struct sock *sk); 1019 struct res_counter *memory_allocated; /* Current allocated memory. */ 1020 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1021 int *memory_pressure; 1022 long *sysctl_mem; 1023 unsigned long flags; 1024 /* 1025 * memcg field is used to find which memcg we belong directly 1026 * Each memcg struct can hold more than one cg_proto, so container_of 1027 * won't really cut. 1028 * 1029 * The elegant solution would be having an inverse function to 1030 * proto_cgroup in struct proto, but that means polluting the structure 1031 * for everybody, instead of just for memcg users. 1032 */ 1033 struct mem_cgroup *memcg; 1034 }; 1035 1036 extern int proto_register(struct proto *prot, int alloc_slab); 1037 extern void proto_unregister(struct proto *prot); 1038 1039 static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1040 { 1041 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1042 } 1043 1044 static inline bool memcg_proto_activated(struct cg_proto *cg_proto) 1045 { 1046 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags); 1047 } 1048 1049 #ifdef SOCK_REFCNT_DEBUG 1050 static inline void sk_refcnt_debug_inc(struct sock *sk) 1051 { 1052 atomic_inc(&sk->sk_prot->socks); 1053 } 1054 1055 static inline void sk_refcnt_debug_dec(struct sock *sk) 1056 { 1057 atomic_dec(&sk->sk_prot->socks); 1058 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1059 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1060 } 1061 1062 static inline void sk_refcnt_debug_release(const struct sock *sk) 1063 { 1064 if (atomic_read(&sk->sk_refcnt) != 1) 1065 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1066 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1067 } 1068 #else /* SOCK_REFCNT_DEBUG */ 1069 #define sk_refcnt_debug_inc(sk) do { } while (0) 1070 #define sk_refcnt_debug_dec(sk) do { } while (0) 1071 #define sk_refcnt_debug_release(sk) do { } while (0) 1072 #endif /* SOCK_REFCNT_DEBUG */ 1073 1074 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1075 extern struct static_key memcg_socket_limit_enabled; 1076 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1077 struct cg_proto *cg_proto) 1078 { 1079 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1080 } 1081 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1082 #else 1083 #define mem_cgroup_sockets_enabled 0 1084 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1085 struct cg_proto *cg_proto) 1086 { 1087 return NULL; 1088 } 1089 #endif 1090 1091 1092 static inline bool sk_has_memory_pressure(const struct sock *sk) 1093 { 1094 return sk->sk_prot->memory_pressure != NULL; 1095 } 1096 1097 static inline bool sk_under_memory_pressure(const struct sock *sk) 1098 { 1099 if (!sk->sk_prot->memory_pressure) 1100 return false; 1101 1102 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1103 return !!*sk->sk_cgrp->memory_pressure; 1104 1105 return !!*sk->sk_prot->memory_pressure; 1106 } 1107 1108 static inline void sk_leave_memory_pressure(struct sock *sk) 1109 { 1110 int *memory_pressure = sk->sk_prot->memory_pressure; 1111 1112 if (!memory_pressure) 1113 return; 1114 1115 if (*memory_pressure) 1116 *memory_pressure = 0; 1117 1118 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1119 struct cg_proto *cg_proto = sk->sk_cgrp; 1120 struct proto *prot = sk->sk_prot; 1121 1122 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1123 if (*cg_proto->memory_pressure) 1124 *cg_proto->memory_pressure = 0; 1125 } 1126 1127 } 1128 1129 static inline void sk_enter_memory_pressure(struct sock *sk) 1130 { 1131 if (!sk->sk_prot->enter_memory_pressure) 1132 return; 1133 1134 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1135 struct cg_proto *cg_proto = sk->sk_cgrp; 1136 struct proto *prot = sk->sk_prot; 1137 1138 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1139 cg_proto->enter_memory_pressure(sk); 1140 } 1141 1142 sk->sk_prot->enter_memory_pressure(sk); 1143 } 1144 1145 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1146 { 1147 long *prot = sk->sk_prot->sysctl_mem; 1148 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1149 prot = sk->sk_cgrp->sysctl_mem; 1150 return prot[index]; 1151 } 1152 1153 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1154 unsigned long amt, 1155 int *parent_status) 1156 { 1157 struct res_counter *fail; 1158 int ret; 1159 1160 ret = res_counter_charge_nofail(prot->memory_allocated, 1161 amt << PAGE_SHIFT, &fail); 1162 if (ret < 0) 1163 *parent_status = OVER_LIMIT; 1164 } 1165 1166 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1167 unsigned long amt) 1168 { 1169 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT); 1170 } 1171 1172 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot) 1173 { 1174 u64 ret; 1175 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE); 1176 return ret >> PAGE_SHIFT; 1177 } 1178 1179 static inline long 1180 sk_memory_allocated(const struct sock *sk) 1181 { 1182 struct proto *prot = sk->sk_prot; 1183 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1184 return memcg_memory_allocated_read(sk->sk_cgrp); 1185 1186 return atomic_long_read(prot->memory_allocated); 1187 } 1188 1189 static inline long 1190 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1191 { 1192 struct proto *prot = sk->sk_prot; 1193 1194 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1195 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1196 /* update the root cgroup regardless */ 1197 atomic_long_add_return(amt, prot->memory_allocated); 1198 return memcg_memory_allocated_read(sk->sk_cgrp); 1199 } 1200 1201 return atomic_long_add_return(amt, prot->memory_allocated); 1202 } 1203 1204 static inline void 1205 sk_memory_allocated_sub(struct sock *sk, int amt) 1206 { 1207 struct proto *prot = sk->sk_prot; 1208 1209 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1210 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1211 1212 atomic_long_sub(amt, prot->memory_allocated); 1213 } 1214 1215 static inline void sk_sockets_allocated_dec(struct sock *sk) 1216 { 1217 struct proto *prot = sk->sk_prot; 1218 1219 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1220 struct cg_proto *cg_proto = sk->sk_cgrp; 1221 1222 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1223 percpu_counter_dec(cg_proto->sockets_allocated); 1224 } 1225 1226 percpu_counter_dec(prot->sockets_allocated); 1227 } 1228 1229 static inline void sk_sockets_allocated_inc(struct sock *sk) 1230 { 1231 struct proto *prot = sk->sk_prot; 1232 1233 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1234 struct cg_proto *cg_proto = sk->sk_cgrp; 1235 1236 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1237 percpu_counter_inc(cg_proto->sockets_allocated); 1238 } 1239 1240 percpu_counter_inc(prot->sockets_allocated); 1241 } 1242 1243 static inline int 1244 sk_sockets_allocated_read_positive(struct sock *sk) 1245 { 1246 struct proto *prot = sk->sk_prot; 1247 1248 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1249 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated); 1250 1251 return percpu_counter_read_positive(prot->sockets_allocated); 1252 } 1253 1254 static inline int 1255 proto_sockets_allocated_sum_positive(struct proto *prot) 1256 { 1257 return percpu_counter_sum_positive(prot->sockets_allocated); 1258 } 1259 1260 static inline long 1261 proto_memory_allocated(struct proto *prot) 1262 { 1263 return atomic_long_read(prot->memory_allocated); 1264 } 1265 1266 static inline bool 1267 proto_memory_pressure(struct proto *prot) 1268 { 1269 if (!prot->memory_pressure) 1270 return false; 1271 return !!*prot->memory_pressure; 1272 } 1273 1274 1275 #ifdef CONFIG_PROC_FS 1276 /* Called with local bh disabled */ 1277 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1278 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 1279 #else 1280 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1281 int inc) 1282 { 1283 } 1284 #endif 1285 1286 1287 /* With per-bucket locks this operation is not-atomic, so that 1288 * this version is not worse. 1289 */ 1290 static inline void __sk_prot_rehash(struct sock *sk) 1291 { 1292 sk->sk_prot->unhash(sk); 1293 sk->sk_prot->hash(sk); 1294 } 1295 1296 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1297 1298 /* About 10 seconds */ 1299 #define SOCK_DESTROY_TIME (10*HZ) 1300 1301 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1302 #define PROT_SOCK 1024 1303 1304 #define SHUTDOWN_MASK 3 1305 #define RCV_SHUTDOWN 1 1306 #define SEND_SHUTDOWN 2 1307 1308 #define SOCK_SNDBUF_LOCK 1 1309 #define SOCK_RCVBUF_LOCK 2 1310 #define SOCK_BINDADDR_LOCK 4 1311 #define SOCK_BINDPORT_LOCK 8 1312 1313 /* sock_iocb: used to kick off async processing of socket ios */ 1314 struct sock_iocb { 1315 struct list_head list; 1316 1317 int flags; 1318 int size; 1319 struct socket *sock; 1320 struct sock *sk; 1321 struct scm_cookie *scm; 1322 struct msghdr *msg, async_msg; 1323 struct kiocb *kiocb; 1324 }; 1325 1326 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 1327 { 1328 return (struct sock_iocb *)iocb->private; 1329 } 1330 1331 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 1332 { 1333 return si->kiocb; 1334 } 1335 1336 struct socket_alloc { 1337 struct socket socket; 1338 struct inode vfs_inode; 1339 }; 1340 1341 static inline struct socket *SOCKET_I(struct inode *inode) 1342 { 1343 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1344 } 1345 1346 static inline struct inode *SOCK_INODE(struct socket *socket) 1347 { 1348 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1349 } 1350 1351 /* 1352 * Functions for memory accounting 1353 */ 1354 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 1355 extern void __sk_mem_reclaim(struct sock *sk); 1356 1357 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1358 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1359 #define SK_MEM_SEND 0 1360 #define SK_MEM_RECV 1 1361 1362 static inline int sk_mem_pages(int amt) 1363 { 1364 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1365 } 1366 1367 static inline bool sk_has_account(struct sock *sk) 1368 { 1369 /* return true if protocol supports memory accounting */ 1370 return !!sk->sk_prot->memory_allocated; 1371 } 1372 1373 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1374 { 1375 if (!sk_has_account(sk)) 1376 return true; 1377 return size <= sk->sk_forward_alloc || 1378 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1379 } 1380 1381 static inline bool 1382 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1383 { 1384 if (!sk_has_account(sk)) 1385 return true; 1386 return size<= sk->sk_forward_alloc || 1387 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1388 skb_pfmemalloc(skb); 1389 } 1390 1391 static inline void sk_mem_reclaim(struct sock *sk) 1392 { 1393 if (!sk_has_account(sk)) 1394 return; 1395 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1396 __sk_mem_reclaim(sk); 1397 } 1398 1399 static inline void sk_mem_reclaim_partial(struct sock *sk) 1400 { 1401 if (!sk_has_account(sk)) 1402 return; 1403 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1404 __sk_mem_reclaim(sk); 1405 } 1406 1407 static inline void sk_mem_charge(struct sock *sk, int size) 1408 { 1409 if (!sk_has_account(sk)) 1410 return; 1411 sk->sk_forward_alloc -= size; 1412 } 1413 1414 static inline void sk_mem_uncharge(struct sock *sk, int size) 1415 { 1416 if (!sk_has_account(sk)) 1417 return; 1418 sk->sk_forward_alloc += size; 1419 } 1420 1421 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1422 { 1423 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1424 sk->sk_wmem_queued -= skb->truesize; 1425 sk_mem_uncharge(sk, skb->truesize); 1426 __kfree_skb(skb); 1427 } 1428 1429 /* Used by processes to "lock" a socket state, so that 1430 * interrupts and bottom half handlers won't change it 1431 * from under us. It essentially blocks any incoming 1432 * packets, so that we won't get any new data or any 1433 * packets that change the state of the socket. 1434 * 1435 * While locked, BH processing will add new packets to 1436 * the backlog queue. This queue is processed by the 1437 * owner of the socket lock right before it is released. 1438 * 1439 * Since ~2.3.5 it is also exclusive sleep lock serializing 1440 * accesses from user process context. 1441 */ 1442 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1443 1444 /* 1445 * Macro so as to not evaluate some arguments when 1446 * lockdep is not enabled. 1447 * 1448 * Mark both the sk_lock and the sk_lock.slock as a 1449 * per-address-family lock class. 1450 */ 1451 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1452 do { \ 1453 sk->sk_lock.owned = 0; \ 1454 init_waitqueue_head(&sk->sk_lock.wq); \ 1455 spin_lock_init(&(sk)->sk_lock.slock); \ 1456 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1457 sizeof((sk)->sk_lock)); \ 1458 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1459 (skey), (sname)); \ 1460 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1461 } while (0) 1462 1463 extern void lock_sock_nested(struct sock *sk, int subclass); 1464 1465 static inline void lock_sock(struct sock *sk) 1466 { 1467 lock_sock_nested(sk, 0); 1468 } 1469 1470 extern void release_sock(struct sock *sk); 1471 1472 /* BH context may only use the following locking interface. */ 1473 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1474 #define bh_lock_sock_nested(__sk) \ 1475 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1476 SINGLE_DEPTH_NESTING) 1477 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1478 1479 extern bool lock_sock_fast(struct sock *sk); 1480 /** 1481 * unlock_sock_fast - complement of lock_sock_fast 1482 * @sk: socket 1483 * @slow: slow mode 1484 * 1485 * fast unlock socket for user context. 1486 * If slow mode is on, we call regular release_sock() 1487 */ 1488 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1489 { 1490 if (slow) 1491 release_sock(sk); 1492 else 1493 spin_unlock_bh(&sk->sk_lock.slock); 1494 } 1495 1496 1497 extern struct sock *sk_alloc(struct net *net, int family, 1498 gfp_t priority, 1499 struct proto *prot); 1500 extern void sk_free(struct sock *sk); 1501 extern void sk_release_kernel(struct sock *sk); 1502 extern struct sock *sk_clone_lock(const struct sock *sk, 1503 const gfp_t priority); 1504 1505 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1506 unsigned long size, int force, 1507 gfp_t priority); 1508 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1509 unsigned long size, int force, 1510 gfp_t priority); 1511 extern void sock_wfree(struct sk_buff *skb); 1512 extern void sock_rfree(struct sk_buff *skb); 1513 extern void sock_edemux(struct sk_buff *skb); 1514 1515 extern int sock_setsockopt(struct socket *sock, int level, 1516 int op, char __user *optval, 1517 unsigned int optlen); 1518 1519 extern int sock_getsockopt(struct socket *sock, int level, 1520 int op, char __user *optval, 1521 int __user *optlen); 1522 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1523 unsigned long size, 1524 int noblock, 1525 int *errcode); 1526 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1527 unsigned long header_len, 1528 unsigned long data_len, 1529 int noblock, 1530 int *errcode); 1531 extern void *sock_kmalloc(struct sock *sk, int size, 1532 gfp_t priority); 1533 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1534 extern void sk_send_sigurg(struct sock *sk); 1535 1536 /* 1537 * Functions to fill in entries in struct proto_ops when a protocol 1538 * does not implement a particular function. 1539 */ 1540 extern int sock_no_bind(struct socket *, 1541 struct sockaddr *, int); 1542 extern int sock_no_connect(struct socket *, 1543 struct sockaddr *, int, int); 1544 extern int sock_no_socketpair(struct socket *, 1545 struct socket *); 1546 extern int sock_no_accept(struct socket *, 1547 struct socket *, int); 1548 extern int sock_no_getname(struct socket *, 1549 struct sockaddr *, int *, int); 1550 extern unsigned int sock_no_poll(struct file *, struct socket *, 1551 struct poll_table_struct *); 1552 extern int sock_no_ioctl(struct socket *, unsigned int, 1553 unsigned long); 1554 extern int sock_no_listen(struct socket *, int); 1555 extern int sock_no_shutdown(struct socket *, int); 1556 extern int sock_no_getsockopt(struct socket *, int , int, 1557 char __user *, int __user *); 1558 extern int sock_no_setsockopt(struct socket *, int, int, 1559 char __user *, unsigned int); 1560 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1561 struct msghdr *, size_t); 1562 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1563 struct msghdr *, size_t, int); 1564 extern int sock_no_mmap(struct file *file, 1565 struct socket *sock, 1566 struct vm_area_struct *vma); 1567 extern ssize_t sock_no_sendpage(struct socket *sock, 1568 struct page *page, 1569 int offset, size_t size, 1570 int flags); 1571 1572 /* 1573 * Functions to fill in entries in struct proto_ops when a protocol 1574 * uses the inet style. 1575 */ 1576 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1577 char __user *optval, int __user *optlen); 1578 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1579 struct msghdr *msg, size_t size, int flags); 1580 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1581 char __user *optval, unsigned int optlen); 1582 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1583 int optname, char __user *optval, int __user *optlen); 1584 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1585 int optname, char __user *optval, unsigned int optlen); 1586 1587 extern void sk_common_release(struct sock *sk); 1588 1589 /* 1590 * Default socket callbacks and setup code 1591 */ 1592 1593 /* Initialise core socket variables */ 1594 extern void sock_init_data(struct socket *sock, struct sock *sk); 1595 1596 extern void sk_filter_release_rcu(struct rcu_head *rcu); 1597 1598 /** 1599 * sk_filter_release - release a socket filter 1600 * @fp: filter to remove 1601 * 1602 * Remove a filter from a socket and release its resources. 1603 */ 1604 1605 static inline void sk_filter_release(struct sk_filter *fp) 1606 { 1607 if (atomic_dec_and_test(&fp->refcnt)) 1608 call_rcu(&fp->rcu, sk_filter_release_rcu); 1609 } 1610 1611 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1612 { 1613 unsigned int size = sk_filter_len(fp); 1614 1615 atomic_sub(size, &sk->sk_omem_alloc); 1616 sk_filter_release(fp); 1617 } 1618 1619 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1620 { 1621 atomic_inc(&fp->refcnt); 1622 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1623 } 1624 1625 /* 1626 * Socket reference counting postulates. 1627 * 1628 * * Each user of socket SHOULD hold a reference count. 1629 * * Each access point to socket (an hash table bucket, reference from a list, 1630 * running timer, skb in flight MUST hold a reference count. 1631 * * When reference count hits 0, it means it will never increase back. 1632 * * When reference count hits 0, it means that no references from 1633 * outside exist to this socket and current process on current CPU 1634 * is last user and may/should destroy this socket. 1635 * * sk_free is called from any context: process, BH, IRQ. When 1636 * it is called, socket has no references from outside -> sk_free 1637 * may release descendant resources allocated by the socket, but 1638 * to the time when it is called, socket is NOT referenced by any 1639 * hash tables, lists etc. 1640 * * Packets, delivered from outside (from network or from another process) 1641 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1642 * when they sit in queue. Otherwise, packets will leak to hole, when 1643 * socket is looked up by one cpu and unhasing is made by another CPU. 1644 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1645 * (leak to backlog). Packet socket does all the processing inside 1646 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1647 * use separate SMP lock, so that they are prone too. 1648 */ 1649 1650 /* Ungrab socket and destroy it, if it was the last reference. */ 1651 static inline void sock_put(struct sock *sk) 1652 { 1653 if (atomic_dec_and_test(&sk->sk_refcnt)) 1654 sk_free(sk); 1655 } 1656 1657 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1658 const int nested); 1659 1660 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1661 { 1662 sk->sk_tx_queue_mapping = tx_queue; 1663 } 1664 1665 static inline void sk_tx_queue_clear(struct sock *sk) 1666 { 1667 sk->sk_tx_queue_mapping = -1; 1668 } 1669 1670 static inline int sk_tx_queue_get(const struct sock *sk) 1671 { 1672 return sk ? sk->sk_tx_queue_mapping : -1; 1673 } 1674 1675 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1676 { 1677 sk_tx_queue_clear(sk); 1678 sk->sk_socket = sock; 1679 } 1680 1681 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1682 { 1683 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1684 return &rcu_dereference_raw(sk->sk_wq)->wait; 1685 } 1686 /* Detach socket from process context. 1687 * Announce socket dead, detach it from wait queue and inode. 1688 * Note that parent inode held reference count on this struct sock, 1689 * we do not release it in this function, because protocol 1690 * probably wants some additional cleanups or even continuing 1691 * to work with this socket (TCP). 1692 */ 1693 static inline void sock_orphan(struct sock *sk) 1694 { 1695 write_lock_bh(&sk->sk_callback_lock); 1696 sock_set_flag(sk, SOCK_DEAD); 1697 sk_set_socket(sk, NULL); 1698 sk->sk_wq = NULL; 1699 write_unlock_bh(&sk->sk_callback_lock); 1700 } 1701 1702 static inline void sock_graft(struct sock *sk, struct socket *parent) 1703 { 1704 write_lock_bh(&sk->sk_callback_lock); 1705 sk->sk_wq = parent->wq; 1706 parent->sk = sk; 1707 sk_set_socket(sk, parent); 1708 security_sock_graft(sk, parent); 1709 write_unlock_bh(&sk->sk_callback_lock); 1710 } 1711 1712 extern kuid_t sock_i_uid(struct sock *sk); 1713 extern unsigned long sock_i_ino(struct sock *sk); 1714 1715 static inline struct dst_entry * 1716 __sk_dst_get(struct sock *sk) 1717 { 1718 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1719 lockdep_is_held(&sk->sk_lock.slock)); 1720 } 1721 1722 static inline struct dst_entry * 1723 sk_dst_get(struct sock *sk) 1724 { 1725 struct dst_entry *dst; 1726 1727 rcu_read_lock(); 1728 dst = rcu_dereference(sk->sk_dst_cache); 1729 if (dst) 1730 dst_hold(dst); 1731 rcu_read_unlock(); 1732 return dst; 1733 } 1734 1735 extern void sk_reset_txq(struct sock *sk); 1736 1737 static inline void dst_negative_advice(struct sock *sk) 1738 { 1739 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1740 1741 if (dst && dst->ops->negative_advice) { 1742 ndst = dst->ops->negative_advice(dst); 1743 1744 if (ndst != dst) { 1745 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1746 sk_reset_txq(sk); 1747 } 1748 } 1749 } 1750 1751 static inline void 1752 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1753 { 1754 struct dst_entry *old_dst; 1755 1756 sk_tx_queue_clear(sk); 1757 /* 1758 * This can be called while sk is owned by the caller only, 1759 * with no state that can be checked in a rcu_dereference_check() cond 1760 */ 1761 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1762 rcu_assign_pointer(sk->sk_dst_cache, dst); 1763 dst_release(old_dst); 1764 } 1765 1766 static inline void 1767 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1768 { 1769 spin_lock(&sk->sk_dst_lock); 1770 __sk_dst_set(sk, dst); 1771 spin_unlock(&sk->sk_dst_lock); 1772 } 1773 1774 static inline void 1775 __sk_dst_reset(struct sock *sk) 1776 { 1777 __sk_dst_set(sk, NULL); 1778 } 1779 1780 static inline void 1781 sk_dst_reset(struct sock *sk) 1782 { 1783 spin_lock(&sk->sk_dst_lock); 1784 __sk_dst_reset(sk); 1785 spin_unlock(&sk->sk_dst_lock); 1786 } 1787 1788 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1789 1790 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1791 1792 static inline bool sk_can_gso(const struct sock *sk) 1793 { 1794 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1795 } 1796 1797 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1798 1799 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1800 { 1801 sk->sk_route_nocaps |= flags; 1802 sk->sk_route_caps &= ~flags; 1803 } 1804 1805 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1806 char __user *from, char *to, 1807 int copy, int offset) 1808 { 1809 if (skb->ip_summed == CHECKSUM_NONE) { 1810 int err = 0; 1811 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1812 if (err) 1813 return err; 1814 skb->csum = csum_block_add(skb->csum, csum, offset); 1815 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1816 if (!access_ok(VERIFY_READ, from, copy) || 1817 __copy_from_user_nocache(to, from, copy)) 1818 return -EFAULT; 1819 } else if (copy_from_user(to, from, copy)) 1820 return -EFAULT; 1821 1822 return 0; 1823 } 1824 1825 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1826 char __user *from, int copy) 1827 { 1828 int err, offset = skb->len; 1829 1830 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1831 copy, offset); 1832 if (err) 1833 __skb_trim(skb, offset); 1834 1835 return err; 1836 } 1837 1838 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from, 1839 struct sk_buff *skb, 1840 struct page *page, 1841 int off, int copy) 1842 { 1843 int err; 1844 1845 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1846 copy, skb->len); 1847 if (err) 1848 return err; 1849 1850 skb->len += copy; 1851 skb->data_len += copy; 1852 skb->truesize += copy; 1853 sk->sk_wmem_queued += copy; 1854 sk_mem_charge(sk, copy); 1855 return 0; 1856 } 1857 1858 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1859 struct sk_buff *skb, struct page *page, 1860 int off, int copy) 1861 { 1862 if (skb->ip_summed == CHECKSUM_NONE) { 1863 int err = 0; 1864 __wsum csum = csum_and_copy_from_user(from, 1865 page_address(page) + off, 1866 copy, 0, &err); 1867 if (err) 1868 return err; 1869 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1870 } else if (copy_from_user(page_address(page) + off, from, copy)) 1871 return -EFAULT; 1872 1873 skb->len += copy; 1874 skb->data_len += copy; 1875 skb->truesize += copy; 1876 sk->sk_wmem_queued += copy; 1877 sk_mem_charge(sk, copy); 1878 return 0; 1879 } 1880 1881 /** 1882 * sk_wmem_alloc_get - returns write allocations 1883 * @sk: socket 1884 * 1885 * Returns sk_wmem_alloc minus initial offset of one 1886 */ 1887 static inline int sk_wmem_alloc_get(const struct sock *sk) 1888 { 1889 return atomic_read(&sk->sk_wmem_alloc) - 1; 1890 } 1891 1892 /** 1893 * sk_rmem_alloc_get - returns read allocations 1894 * @sk: socket 1895 * 1896 * Returns sk_rmem_alloc 1897 */ 1898 static inline int sk_rmem_alloc_get(const struct sock *sk) 1899 { 1900 return atomic_read(&sk->sk_rmem_alloc); 1901 } 1902 1903 /** 1904 * sk_has_allocations - check if allocations are outstanding 1905 * @sk: socket 1906 * 1907 * Returns true if socket has write or read allocations 1908 */ 1909 static inline bool sk_has_allocations(const struct sock *sk) 1910 { 1911 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1912 } 1913 1914 /** 1915 * wq_has_sleeper - check if there are any waiting processes 1916 * @wq: struct socket_wq 1917 * 1918 * Returns true if socket_wq has waiting processes 1919 * 1920 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1921 * barrier call. They were added due to the race found within the tcp code. 1922 * 1923 * Consider following tcp code paths: 1924 * 1925 * CPU1 CPU2 1926 * 1927 * sys_select receive packet 1928 * ... ... 1929 * __add_wait_queue update tp->rcv_nxt 1930 * ... ... 1931 * tp->rcv_nxt check sock_def_readable 1932 * ... { 1933 * schedule rcu_read_lock(); 1934 * wq = rcu_dereference(sk->sk_wq); 1935 * if (wq && waitqueue_active(&wq->wait)) 1936 * wake_up_interruptible(&wq->wait) 1937 * ... 1938 * } 1939 * 1940 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1941 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1942 * could then endup calling schedule and sleep forever if there are no more 1943 * data on the socket. 1944 * 1945 */ 1946 static inline bool wq_has_sleeper(struct socket_wq *wq) 1947 { 1948 /* We need to be sure we are in sync with the 1949 * add_wait_queue modifications to the wait queue. 1950 * 1951 * This memory barrier is paired in the sock_poll_wait. 1952 */ 1953 smp_mb(); 1954 return wq && waitqueue_active(&wq->wait); 1955 } 1956 1957 /** 1958 * sock_poll_wait - place memory barrier behind the poll_wait call. 1959 * @filp: file 1960 * @wait_address: socket wait queue 1961 * @p: poll_table 1962 * 1963 * See the comments in the wq_has_sleeper function. 1964 */ 1965 static inline void sock_poll_wait(struct file *filp, 1966 wait_queue_head_t *wait_address, poll_table *p) 1967 { 1968 if (!poll_does_not_wait(p) && wait_address) { 1969 poll_wait(filp, wait_address, p); 1970 /* We need to be sure we are in sync with the 1971 * socket flags modification. 1972 * 1973 * This memory barrier is paired in the wq_has_sleeper. 1974 */ 1975 smp_mb(); 1976 } 1977 } 1978 1979 /* 1980 * Queue a received datagram if it will fit. Stream and sequenced 1981 * protocols can't normally use this as they need to fit buffers in 1982 * and play with them. 1983 * 1984 * Inlined as it's very short and called for pretty much every 1985 * packet ever received. 1986 */ 1987 1988 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1989 { 1990 skb_orphan(skb); 1991 skb->sk = sk; 1992 skb->destructor = sock_wfree; 1993 /* 1994 * We used to take a refcount on sk, but following operation 1995 * is enough to guarantee sk_free() wont free this sock until 1996 * all in-flight packets are completed 1997 */ 1998 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1999 } 2000 2001 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2002 { 2003 skb_orphan(skb); 2004 skb->sk = sk; 2005 skb->destructor = sock_rfree; 2006 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2007 sk_mem_charge(sk, skb->truesize); 2008 } 2009 2010 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2011 unsigned long expires); 2012 2013 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2014 2015 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2016 2017 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2018 2019 /* 2020 * Recover an error report and clear atomically 2021 */ 2022 2023 static inline int sock_error(struct sock *sk) 2024 { 2025 int err; 2026 if (likely(!sk->sk_err)) 2027 return 0; 2028 err = xchg(&sk->sk_err, 0); 2029 return -err; 2030 } 2031 2032 static inline unsigned long sock_wspace(struct sock *sk) 2033 { 2034 int amt = 0; 2035 2036 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2037 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2038 if (amt < 0) 2039 amt = 0; 2040 } 2041 return amt; 2042 } 2043 2044 static inline void sk_wake_async(struct sock *sk, int how, int band) 2045 { 2046 if (sock_flag(sk, SOCK_FASYNC)) 2047 sock_wake_async(sk->sk_socket, how, band); 2048 } 2049 2050 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2051 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2052 * Note: for send buffers, TCP works better if we can build two skbs at 2053 * minimum. 2054 */ 2055 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2056 2057 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2058 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2059 2060 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2061 { 2062 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2063 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2064 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2065 } 2066 } 2067 2068 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 2069 2070 /** 2071 * sk_page_frag - return an appropriate page_frag 2072 * @sk: socket 2073 * 2074 * If socket allocation mode allows current thread to sleep, it means its 2075 * safe to use the per task page_frag instead of the per socket one. 2076 */ 2077 static inline struct page_frag *sk_page_frag(struct sock *sk) 2078 { 2079 if (sk->sk_allocation & __GFP_WAIT) 2080 return ¤t->task_frag; 2081 2082 return &sk->sk_frag; 2083 } 2084 2085 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2086 2087 /* 2088 * Default write policy as shown to user space via poll/select/SIGIO 2089 */ 2090 static inline bool sock_writeable(const struct sock *sk) 2091 { 2092 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2093 } 2094 2095 static inline gfp_t gfp_any(void) 2096 { 2097 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2098 } 2099 2100 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2101 { 2102 return noblock ? 0 : sk->sk_rcvtimeo; 2103 } 2104 2105 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2106 { 2107 return noblock ? 0 : sk->sk_sndtimeo; 2108 } 2109 2110 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2111 { 2112 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2113 } 2114 2115 /* Alas, with timeout socket operations are not restartable. 2116 * Compare this to poll(). 2117 */ 2118 static inline int sock_intr_errno(long timeo) 2119 { 2120 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2121 } 2122 2123 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2124 struct sk_buff *skb); 2125 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2126 struct sk_buff *skb); 2127 2128 static inline void 2129 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2130 { 2131 ktime_t kt = skb->tstamp; 2132 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2133 2134 /* 2135 * generate control messages if 2136 * - receive time stamping in software requested (SOCK_RCVTSTAMP 2137 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 2138 * - software time stamp available and wanted 2139 * (SOCK_TIMESTAMPING_SOFTWARE) 2140 * - hardware time stamps available and wanted 2141 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 2142 * SOCK_TIMESTAMPING_RAW_HARDWARE) 2143 */ 2144 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2145 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 2146 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 2147 (hwtstamps->hwtstamp.tv64 && 2148 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 2149 (hwtstamps->syststamp.tv64 && 2150 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 2151 __sock_recv_timestamp(msg, sk, skb); 2152 else 2153 sk->sk_stamp = kt; 2154 2155 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2156 __sock_recv_wifi_status(msg, sk, skb); 2157 } 2158 2159 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2160 struct sk_buff *skb); 2161 2162 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2163 struct sk_buff *skb) 2164 { 2165 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2166 (1UL << SOCK_RCVTSTAMP) | \ 2167 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 2168 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 2169 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 2170 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 2171 2172 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 2173 __sock_recv_ts_and_drops(msg, sk, skb); 2174 else 2175 sk->sk_stamp = skb->tstamp; 2176 } 2177 2178 /** 2179 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2180 * @sk: socket sending this packet 2181 * @tx_flags: filled with instructions for time stamping 2182 * 2183 * Currently only depends on SOCK_TIMESTAMPING* flags. 2184 */ 2185 extern void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 2186 2187 /** 2188 * sk_eat_skb - Release a skb if it is no longer needed 2189 * @sk: socket to eat this skb from 2190 * @skb: socket buffer to eat 2191 * @copied_early: flag indicating whether DMA operations copied this data early 2192 * 2193 * This routine must be called with interrupts disabled or with the socket 2194 * locked so that the sk_buff queue operation is ok. 2195 */ 2196 #ifdef CONFIG_NET_DMA 2197 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2198 { 2199 __skb_unlink(skb, &sk->sk_receive_queue); 2200 if (!copied_early) 2201 __kfree_skb(skb); 2202 else 2203 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 2204 } 2205 #else 2206 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2207 { 2208 __skb_unlink(skb, &sk->sk_receive_queue); 2209 __kfree_skb(skb); 2210 } 2211 #endif 2212 2213 static inline 2214 struct net *sock_net(const struct sock *sk) 2215 { 2216 return read_pnet(&sk->sk_net); 2217 } 2218 2219 static inline 2220 void sock_net_set(struct sock *sk, struct net *net) 2221 { 2222 write_pnet(&sk->sk_net, net); 2223 } 2224 2225 /* 2226 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2227 * They should not hold a reference to a namespace in order to allow 2228 * to stop it. 2229 * Sockets after sk_change_net should be released using sk_release_kernel 2230 */ 2231 static inline void sk_change_net(struct sock *sk, struct net *net) 2232 { 2233 put_net(sock_net(sk)); 2234 sock_net_set(sk, hold_net(net)); 2235 } 2236 2237 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2238 { 2239 if (skb->sk) { 2240 struct sock *sk = skb->sk; 2241 2242 skb->destructor = NULL; 2243 skb->sk = NULL; 2244 return sk; 2245 } 2246 return NULL; 2247 } 2248 2249 extern void sock_enable_timestamp(struct sock *sk, int flag); 2250 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 2251 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 2252 2253 /* 2254 * Enable debug/info messages 2255 */ 2256 extern int net_msg_warn; 2257 #define NETDEBUG(fmt, args...) \ 2258 do { if (net_msg_warn) printk(fmt,##args); } while (0) 2259 2260 #define LIMIT_NETDEBUG(fmt, args...) \ 2261 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 2262 2263 extern __u32 sysctl_wmem_max; 2264 extern __u32 sysctl_rmem_max; 2265 2266 extern int sysctl_optmem_max; 2267 2268 extern __u32 sysctl_wmem_default; 2269 extern __u32 sysctl_rmem_default; 2270 2271 #endif /* _SOCK_H */ 2272