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_sndbuf: size of send buffer in bytes 237 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 238 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 239 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets 240 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 241 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 242 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 243 * @sk_gso_max_size: Maximum GSO segment size to build 244 * @sk_gso_max_segs: Maximum number of GSO segments 245 * @sk_lingertime: %SO_LINGER l_linger setting 246 * @sk_backlog: always used with the per-socket spinlock held 247 * @sk_callback_lock: used with the callbacks in the end of this struct 248 * @sk_error_queue: rarely used 249 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 250 * IPV6_ADDRFORM for instance) 251 * @sk_err: last error 252 * @sk_err_soft: errors that don't cause failure but are the cause of a 253 * persistent failure not just 'timed out' 254 * @sk_drops: raw/udp drops counter 255 * @sk_ack_backlog: current listen backlog 256 * @sk_max_ack_backlog: listen backlog set in listen() 257 * @sk_priority: %SO_PRIORITY setting 258 * @sk_cgrp_prioidx: socket group's priority map index 259 * @sk_type: socket type (%SOCK_STREAM, etc) 260 * @sk_protocol: which protocol this socket belongs in this network family 261 * @sk_peer_pid: &struct pid for this socket's peer 262 * @sk_peer_cred: %SO_PEERCRED setting 263 * @sk_rcvlowat: %SO_RCVLOWAT setting 264 * @sk_rcvtimeo: %SO_RCVTIMEO setting 265 * @sk_sndtimeo: %SO_SNDTIMEO setting 266 * @sk_rxhash: flow hash received from netif layer 267 * @sk_filter: socket filtering instructions 268 * @sk_protinfo: private area, net family specific, when not using slab 269 * @sk_timer: sock cleanup timer 270 * @sk_stamp: time stamp of last packet received 271 * @sk_socket: Identd and reporting IO signals 272 * @sk_user_data: RPC layer private data 273 * @sk_frag: cached page frag 274 * @sk_peek_off: current peek_offset value 275 * @sk_send_head: front of stuff to transmit 276 * @sk_security: used by security modules 277 * @sk_mark: generic packet mark 278 * @sk_classid: this socket's cgroup classid 279 * @sk_cgrp: this socket's cgroup-specific proto data 280 * @sk_write_pending: a write to stream socket waits to start 281 * @sk_state_change: callback to indicate change in the state of the sock 282 * @sk_data_ready: callback to indicate there is data to be processed 283 * @sk_write_space: callback to indicate there is bf sending space available 284 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 285 * @sk_backlog_rcv: callback to process the backlog 286 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 287 */ 288 struct sock { 289 /* 290 * Now struct inet_timewait_sock also uses sock_common, so please just 291 * don't add nothing before this first member (__sk_common) --acme 292 */ 293 struct sock_common __sk_common; 294 #define sk_node __sk_common.skc_node 295 #define sk_nulls_node __sk_common.skc_nulls_node 296 #define sk_refcnt __sk_common.skc_refcnt 297 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 298 299 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 300 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 301 #define sk_hash __sk_common.skc_hash 302 #define sk_family __sk_common.skc_family 303 #define sk_state __sk_common.skc_state 304 #define sk_reuse __sk_common.skc_reuse 305 #define sk_reuseport __sk_common.skc_reuseport 306 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 307 #define sk_bind_node __sk_common.skc_bind_node 308 #define sk_prot __sk_common.skc_prot 309 #define sk_net __sk_common.skc_net 310 socket_lock_t sk_lock; 311 struct sk_buff_head sk_receive_queue; 312 /* 313 * The backlog queue is special, it is always used with 314 * the per-socket spinlock held and requires low latency 315 * access. Therefore we special case it's implementation. 316 * Note : rmem_alloc is in this structure to fill a hole 317 * on 64bit arches, not because its logically part of 318 * backlog. 319 */ 320 struct { 321 atomic_t rmem_alloc; 322 int len; 323 struct sk_buff *head; 324 struct sk_buff *tail; 325 } sk_backlog; 326 #define sk_rmem_alloc sk_backlog.rmem_alloc 327 int sk_forward_alloc; 328 #ifdef CONFIG_RPS 329 __u32 sk_rxhash; 330 #endif 331 #ifdef CONFIG_NET_RX_BUSY_POLL 332 unsigned int sk_napi_id; 333 unsigned int sk_ll_usec; 334 #endif 335 atomic_t sk_drops; 336 int sk_rcvbuf; 337 338 struct sk_filter __rcu *sk_filter; 339 struct socket_wq __rcu *sk_wq; 340 341 #ifdef CONFIG_NET_DMA 342 struct sk_buff_head sk_async_wait_queue; 343 #endif 344 345 #ifdef CONFIG_XFRM 346 struct xfrm_policy *sk_policy[2]; 347 #endif 348 unsigned long sk_flags; 349 struct dst_entry *sk_rx_dst; 350 struct dst_entry __rcu *sk_dst_cache; 351 spinlock_t sk_dst_lock; 352 atomic_t sk_wmem_alloc; 353 atomic_t sk_omem_alloc; 354 int sk_sndbuf; 355 struct sk_buff_head sk_write_queue; 356 kmemcheck_bitfield_begin(flags); 357 unsigned int sk_shutdown : 2, 358 sk_no_check : 2, 359 sk_userlocks : 4, 360 sk_protocol : 8, 361 sk_type : 16; 362 kmemcheck_bitfield_end(flags); 363 int sk_wmem_queued; 364 gfp_t sk_allocation; 365 u32 sk_pacing_rate; /* bytes per second */ 366 netdev_features_t sk_route_caps; 367 netdev_features_t sk_route_nocaps; 368 int sk_gso_type; 369 unsigned int sk_gso_max_size; 370 u16 sk_gso_max_segs; 371 int sk_rcvlowat; 372 unsigned long sk_lingertime; 373 struct sk_buff_head sk_error_queue; 374 struct proto *sk_prot_creator; 375 rwlock_t sk_callback_lock; 376 int sk_err, 377 sk_err_soft; 378 unsigned short sk_ack_backlog; 379 unsigned short sk_max_ack_backlog; 380 __u32 sk_priority; 381 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP) 382 __u32 sk_cgrp_prioidx; 383 #endif 384 struct pid *sk_peer_pid; 385 const struct cred *sk_peer_cred; 386 long sk_rcvtimeo; 387 long sk_sndtimeo; 388 void *sk_protinfo; 389 struct timer_list sk_timer; 390 ktime_t sk_stamp; 391 struct socket *sk_socket; 392 void *sk_user_data; 393 struct page_frag sk_frag; 394 struct sk_buff *sk_send_head; 395 __s32 sk_peek_off; 396 int sk_write_pending; 397 #ifdef CONFIG_SECURITY 398 void *sk_security; 399 #endif 400 __u32 sk_mark; 401 u32 sk_classid; 402 struct cg_proto *sk_cgrp; 403 void (*sk_state_change)(struct sock *sk); 404 void (*sk_data_ready)(struct sock *sk, int bytes); 405 void (*sk_write_space)(struct sock *sk); 406 void (*sk_error_report)(struct sock *sk); 407 int (*sk_backlog_rcv)(struct sock *sk, 408 struct sk_buff *skb); 409 void (*sk_destruct)(struct sock *sk); 410 }; 411 412 /* 413 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 414 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 415 * on a socket means that the socket will reuse everybody else's port 416 * without looking at the other's sk_reuse value. 417 */ 418 419 #define SK_NO_REUSE 0 420 #define SK_CAN_REUSE 1 421 #define SK_FORCE_REUSE 2 422 423 static inline int sk_peek_offset(struct sock *sk, int flags) 424 { 425 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 426 return sk->sk_peek_off; 427 else 428 return 0; 429 } 430 431 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 432 { 433 if (sk->sk_peek_off >= 0) { 434 if (sk->sk_peek_off >= val) 435 sk->sk_peek_off -= val; 436 else 437 sk->sk_peek_off = 0; 438 } 439 } 440 441 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 442 { 443 if (sk->sk_peek_off >= 0) 444 sk->sk_peek_off += val; 445 } 446 447 /* 448 * Hashed lists helper routines 449 */ 450 static inline struct sock *sk_entry(const struct hlist_node *node) 451 { 452 return hlist_entry(node, struct sock, sk_node); 453 } 454 455 static inline struct sock *__sk_head(const struct hlist_head *head) 456 { 457 return hlist_entry(head->first, struct sock, sk_node); 458 } 459 460 static inline struct sock *sk_head(const struct hlist_head *head) 461 { 462 return hlist_empty(head) ? NULL : __sk_head(head); 463 } 464 465 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 466 { 467 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 468 } 469 470 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 471 { 472 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 473 } 474 475 static inline struct sock *sk_next(const struct sock *sk) 476 { 477 return sk->sk_node.next ? 478 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 479 } 480 481 static inline struct sock *sk_nulls_next(const struct sock *sk) 482 { 483 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 484 hlist_nulls_entry(sk->sk_nulls_node.next, 485 struct sock, sk_nulls_node) : 486 NULL; 487 } 488 489 static inline bool sk_unhashed(const struct sock *sk) 490 { 491 return hlist_unhashed(&sk->sk_node); 492 } 493 494 static inline bool sk_hashed(const struct sock *sk) 495 { 496 return !sk_unhashed(sk); 497 } 498 499 static inline void sk_node_init(struct hlist_node *node) 500 { 501 node->pprev = NULL; 502 } 503 504 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 505 { 506 node->pprev = NULL; 507 } 508 509 static inline void __sk_del_node(struct sock *sk) 510 { 511 __hlist_del(&sk->sk_node); 512 } 513 514 /* NB: equivalent to hlist_del_init_rcu */ 515 static inline bool __sk_del_node_init(struct sock *sk) 516 { 517 if (sk_hashed(sk)) { 518 __sk_del_node(sk); 519 sk_node_init(&sk->sk_node); 520 return true; 521 } 522 return false; 523 } 524 525 /* Grab socket reference count. This operation is valid only 526 when sk is ALREADY grabbed f.e. it is found in hash table 527 or a list and the lookup is made under lock preventing hash table 528 modifications. 529 */ 530 531 static inline void sock_hold(struct sock *sk) 532 { 533 atomic_inc(&sk->sk_refcnt); 534 } 535 536 /* Ungrab socket in the context, which assumes that socket refcnt 537 cannot hit zero, f.e. it is true in context of any socketcall. 538 */ 539 static inline void __sock_put(struct sock *sk) 540 { 541 atomic_dec(&sk->sk_refcnt); 542 } 543 544 static inline bool sk_del_node_init(struct sock *sk) 545 { 546 bool rc = __sk_del_node_init(sk); 547 548 if (rc) { 549 /* paranoid for a while -acme */ 550 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 551 __sock_put(sk); 552 } 553 return rc; 554 } 555 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 556 557 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 558 { 559 if (sk_hashed(sk)) { 560 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 561 return true; 562 } 563 return false; 564 } 565 566 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 567 { 568 bool rc = __sk_nulls_del_node_init_rcu(sk); 569 570 if (rc) { 571 /* paranoid for a while -acme */ 572 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 573 __sock_put(sk); 574 } 575 return rc; 576 } 577 578 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 579 { 580 hlist_add_head(&sk->sk_node, list); 581 } 582 583 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 584 { 585 sock_hold(sk); 586 __sk_add_node(sk, list); 587 } 588 589 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 590 { 591 sock_hold(sk); 592 hlist_add_head_rcu(&sk->sk_node, list); 593 } 594 595 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 596 { 597 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 598 } 599 600 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 601 { 602 sock_hold(sk); 603 __sk_nulls_add_node_rcu(sk, list); 604 } 605 606 static inline void __sk_del_bind_node(struct sock *sk) 607 { 608 __hlist_del(&sk->sk_bind_node); 609 } 610 611 static inline void sk_add_bind_node(struct sock *sk, 612 struct hlist_head *list) 613 { 614 hlist_add_head(&sk->sk_bind_node, list); 615 } 616 617 #define sk_for_each(__sk, list) \ 618 hlist_for_each_entry(__sk, list, sk_node) 619 #define sk_for_each_rcu(__sk, list) \ 620 hlist_for_each_entry_rcu(__sk, list, sk_node) 621 #define sk_nulls_for_each(__sk, node, list) \ 622 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 623 #define sk_nulls_for_each_rcu(__sk, node, list) \ 624 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 625 #define sk_for_each_from(__sk) \ 626 hlist_for_each_entry_from(__sk, sk_node) 627 #define sk_nulls_for_each_from(__sk, node) \ 628 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 629 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 630 #define sk_for_each_safe(__sk, tmp, list) \ 631 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 632 #define sk_for_each_bound(__sk, list) \ 633 hlist_for_each_entry(__sk, list, sk_bind_node) 634 635 static inline struct user_namespace *sk_user_ns(struct sock *sk) 636 { 637 /* Careful only use this in a context where these parameters 638 * can not change and must all be valid, such as recvmsg from 639 * userspace. 640 */ 641 return sk->sk_socket->file->f_cred->user_ns; 642 } 643 644 /* Sock flags */ 645 enum sock_flags { 646 SOCK_DEAD, 647 SOCK_DONE, 648 SOCK_URGINLINE, 649 SOCK_KEEPOPEN, 650 SOCK_LINGER, 651 SOCK_DESTROY, 652 SOCK_BROADCAST, 653 SOCK_TIMESTAMP, 654 SOCK_ZAPPED, 655 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 656 SOCK_DBG, /* %SO_DEBUG setting */ 657 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 658 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 659 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 660 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 661 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 662 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 663 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 664 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 665 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 666 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 667 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 668 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 669 SOCK_FASYNC, /* fasync() active */ 670 SOCK_RXQ_OVFL, 671 SOCK_ZEROCOPY, /* buffers from userspace */ 672 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 673 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 674 * Will use last 4 bytes of packet sent from 675 * user-space instead. 676 */ 677 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 678 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 679 }; 680 681 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 682 { 683 nsk->sk_flags = osk->sk_flags; 684 } 685 686 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 687 { 688 __set_bit(flag, &sk->sk_flags); 689 } 690 691 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 692 { 693 __clear_bit(flag, &sk->sk_flags); 694 } 695 696 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 697 { 698 return test_bit(flag, &sk->sk_flags); 699 } 700 701 #ifdef CONFIG_NET 702 extern struct static_key memalloc_socks; 703 static inline int sk_memalloc_socks(void) 704 { 705 return static_key_false(&memalloc_socks); 706 } 707 #else 708 709 static inline int sk_memalloc_socks(void) 710 { 711 return 0; 712 } 713 714 #endif 715 716 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask) 717 { 718 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC); 719 } 720 721 static inline void sk_acceptq_removed(struct sock *sk) 722 { 723 sk->sk_ack_backlog--; 724 } 725 726 static inline void sk_acceptq_added(struct sock *sk) 727 { 728 sk->sk_ack_backlog++; 729 } 730 731 static inline bool sk_acceptq_is_full(const struct sock *sk) 732 { 733 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 734 } 735 736 /* 737 * Compute minimal free write space needed to queue new packets. 738 */ 739 static inline int sk_stream_min_wspace(const struct sock *sk) 740 { 741 return sk->sk_wmem_queued >> 1; 742 } 743 744 static inline int sk_stream_wspace(const struct sock *sk) 745 { 746 return sk->sk_sndbuf - sk->sk_wmem_queued; 747 } 748 749 extern void sk_stream_write_space(struct sock *sk); 750 751 /* OOB backlog add */ 752 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 753 { 754 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 755 skb_dst_force(skb); 756 757 if (!sk->sk_backlog.tail) 758 sk->sk_backlog.head = skb; 759 else 760 sk->sk_backlog.tail->next = skb; 761 762 sk->sk_backlog.tail = skb; 763 skb->next = NULL; 764 } 765 766 /* 767 * Take into account size of receive queue and backlog queue 768 * Do not take into account this skb truesize, 769 * to allow even a single big packet to come. 770 */ 771 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb, 772 unsigned int limit) 773 { 774 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 775 776 return qsize > limit; 777 } 778 779 /* The per-socket spinlock must be held here. */ 780 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 781 unsigned int limit) 782 { 783 if (sk_rcvqueues_full(sk, skb, limit)) 784 return -ENOBUFS; 785 786 __sk_add_backlog(sk, skb); 787 sk->sk_backlog.len += skb->truesize; 788 return 0; 789 } 790 791 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 792 793 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 794 { 795 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 796 return __sk_backlog_rcv(sk, skb); 797 798 return sk->sk_backlog_rcv(sk, skb); 799 } 800 801 static inline void sock_rps_record_flow(const struct sock *sk) 802 { 803 #ifdef CONFIG_RPS 804 struct rps_sock_flow_table *sock_flow_table; 805 806 rcu_read_lock(); 807 sock_flow_table = rcu_dereference(rps_sock_flow_table); 808 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 809 rcu_read_unlock(); 810 #endif 811 } 812 813 static inline void sock_rps_reset_flow(const struct sock *sk) 814 { 815 #ifdef CONFIG_RPS 816 struct rps_sock_flow_table *sock_flow_table; 817 818 rcu_read_lock(); 819 sock_flow_table = rcu_dereference(rps_sock_flow_table); 820 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 821 rcu_read_unlock(); 822 #endif 823 } 824 825 static inline void sock_rps_save_rxhash(struct sock *sk, 826 const struct sk_buff *skb) 827 { 828 #ifdef CONFIG_RPS 829 if (unlikely(sk->sk_rxhash != skb->rxhash)) { 830 sock_rps_reset_flow(sk); 831 sk->sk_rxhash = skb->rxhash; 832 } 833 #endif 834 } 835 836 static inline void sock_rps_reset_rxhash(struct sock *sk) 837 { 838 #ifdef CONFIG_RPS 839 sock_rps_reset_flow(sk); 840 sk->sk_rxhash = 0; 841 #endif 842 } 843 844 #define sk_wait_event(__sk, __timeo, __condition) \ 845 ({ int __rc; \ 846 release_sock(__sk); \ 847 __rc = __condition; \ 848 if (!__rc) { \ 849 *(__timeo) = schedule_timeout(*(__timeo)); \ 850 } \ 851 lock_sock(__sk); \ 852 __rc = __condition; \ 853 __rc; \ 854 }) 855 856 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 857 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 858 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 859 extern int sk_stream_error(struct sock *sk, int flags, int err); 860 extern void sk_stream_kill_queues(struct sock *sk); 861 extern void sk_set_memalloc(struct sock *sk); 862 extern void sk_clear_memalloc(struct sock *sk); 863 864 extern int sk_wait_data(struct sock *sk, long *timeo); 865 866 struct request_sock_ops; 867 struct timewait_sock_ops; 868 struct inet_hashinfo; 869 struct raw_hashinfo; 870 struct module; 871 872 /* 873 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes 874 * un-modified. Special care is taken when initializing object to zero. 875 */ 876 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 877 { 878 if (offsetof(struct sock, sk_node.next) != 0) 879 memset(sk, 0, offsetof(struct sock, sk_node.next)); 880 memset(&sk->sk_node.pprev, 0, 881 size - offsetof(struct sock, sk_node.pprev)); 882 } 883 884 /* Networking protocol blocks we attach to sockets. 885 * socket layer -> transport layer interface 886 * transport -> network interface is defined by struct inet_proto 887 */ 888 struct proto { 889 void (*close)(struct sock *sk, 890 long timeout); 891 int (*connect)(struct sock *sk, 892 struct sockaddr *uaddr, 893 int addr_len); 894 int (*disconnect)(struct sock *sk, int flags); 895 896 struct sock * (*accept)(struct sock *sk, int flags, int *err); 897 898 int (*ioctl)(struct sock *sk, int cmd, 899 unsigned long arg); 900 int (*init)(struct sock *sk); 901 void (*destroy)(struct sock *sk); 902 void (*shutdown)(struct sock *sk, int how); 903 int (*setsockopt)(struct sock *sk, int level, 904 int optname, char __user *optval, 905 unsigned int optlen); 906 int (*getsockopt)(struct sock *sk, int level, 907 int optname, char __user *optval, 908 int __user *option); 909 #ifdef CONFIG_COMPAT 910 int (*compat_setsockopt)(struct sock *sk, 911 int level, 912 int optname, char __user *optval, 913 unsigned int optlen); 914 int (*compat_getsockopt)(struct sock *sk, 915 int level, 916 int optname, char __user *optval, 917 int __user *option); 918 int (*compat_ioctl)(struct sock *sk, 919 unsigned int cmd, unsigned long arg); 920 #endif 921 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 922 struct msghdr *msg, size_t len); 923 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 924 struct msghdr *msg, 925 size_t len, int noblock, int flags, 926 int *addr_len); 927 int (*sendpage)(struct sock *sk, struct page *page, 928 int offset, size_t size, int flags); 929 int (*bind)(struct sock *sk, 930 struct sockaddr *uaddr, int addr_len); 931 932 int (*backlog_rcv) (struct sock *sk, 933 struct sk_buff *skb); 934 935 void (*release_cb)(struct sock *sk); 936 void (*mtu_reduced)(struct sock *sk); 937 938 /* Keeping track of sk's, looking them up, and port selection methods. */ 939 void (*hash)(struct sock *sk); 940 void (*unhash)(struct sock *sk); 941 void (*rehash)(struct sock *sk); 942 int (*get_port)(struct sock *sk, unsigned short snum); 943 void (*clear_sk)(struct sock *sk, int size); 944 945 /* Keeping track of sockets in use */ 946 #ifdef CONFIG_PROC_FS 947 unsigned int inuse_idx; 948 #endif 949 950 bool (*stream_memory_free)(const struct sock *sk); 951 /* Memory pressure */ 952 void (*enter_memory_pressure)(struct sock *sk); 953 atomic_long_t *memory_allocated; /* Current allocated memory. */ 954 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 955 /* 956 * Pressure flag: try to collapse. 957 * Technical note: it is used by multiple contexts non atomically. 958 * All the __sk_mem_schedule() is of this nature: accounting 959 * is strict, actions are advisory and have some latency. 960 */ 961 int *memory_pressure; 962 long *sysctl_mem; 963 int *sysctl_wmem; 964 int *sysctl_rmem; 965 int max_header; 966 bool no_autobind; 967 968 struct kmem_cache *slab; 969 unsigned int obj_size; 970 int slab_flags; 971 972 struct percpu_counter *orphan_count; 973 974 struct request_sock_ops *rsk_prot; 975 struct timewait_sock_ops *twsk_prot; 976 977 union { 978 struct inet_hashinfo *hashinfo; 979 struct udp_table *udp_table; 980 struct raw_hashinfo *raw_hash; 981 } h; 982 983 struct module *owner; 984 985 char name[32]; 986 987 struct list_head node; 988 #ifdef SOCK_REFCNT_DEBUG 989 atomic_t socks; 990 #endif 991 #ifdef CONFIG_MEMCG_KMEM 992 /* 993 * cgroup specific init/deinit functions. Called once for all 994 * protocols that implement it, from cgroups populate function. 995 * This function has to setup any files the protocol want to 996 * appear in the kmem cgroup filesystem. 997 */ 998 int (*init_cgroup)(struct mem_cgroup *memcg, 999 struct cgroup_subsys *ss); 1000 void (*destroy_cgroup)(struct mem_cgroup *memcg); 1001 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 1002 #endif 1003 }; 1004 1005 /* 1006 * Bits in struct cg_proto.flags 1007 */ 1008 enum cg_proto_flags { 1009 /* Currently active and new sockets should be assigned to cgroups */ 1010 MEMCG_SOCK_ACTIVE, 1011 /* It was ever activated; we must disarm static keys on destruction */ 1012 MEMCG_SOCK_ACTIVATED, 1013 }; 1014 1015 struct cg_proto { 1016 void (*enter_memory_pressure)(struct sock *sk); 1017 struct res_counter *memory_allocated; /* Current allocated memory. */ 1018 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1019 int *memory_pressure; 1020 long *sysctl_mem; 1021 unsigned long flags; 1022 /* 1023 * memcg field is used to find which memcg we belong directly 1024 * Each memcg struct can hold more than one cg_proto, so container_of 1025 * won't really cut. 1026 * 1027 * The elegant solution would be having an inverse function to 1028 * proto_cgroup in struct proto, but that means polluting the structure 1029 * for everybody, instead of just for memcg users. 1030 */ 1031 struct mem_cgroup *memcg; 1032 }; 1033 1034 extern int proto_register(struct proto *prot, int alloc_slab); 1035 extern void proto_unregister(struct proto *prot); 1036 1037 static inline bool memcg_proto_active(struct cg_proto *cg_proto) 1038 { 1039 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags); 1040 } 1041 1042 static inline bool memcg_proto_activated(struct cg_proto *cg_proto) 1043 { 1044 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags); 1045 } 1046 1047 #ifdef SOCK_REFCNT_DEBUG 1048 static inline void sk_refcnt_debug_inc(struct sock *sk) 1049 { 1050 atomic_inc(&sk->sk_prot->socks); 1051 } 1052 1053 static inline void sk_refcnt_debug_dec(struct sock *sk) 1054 { 1055 atomic_dec(&sk->sk_prot->socks); 1056 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1057 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1058 } 1059 1060 static inline void sk_refcnt_debug_release(const struct sock *sk) 1061 { 1062 if (atomic_read(&sk->sk_refcnt) != 1) 1063 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1064 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 1065 } 1066 #else /* SOCK_REFCNT_DEBUG */ 1067 #define sk_refcnt_debug_inc(sk) do { } while (0) 1068 #define sk_refcnt_debug_dec(sk) do { } while (0) 1069 #define sk_refcnt_debug_release(sk) do { } while (0) 1070 #endif /* SOCK_REFCNT_DEBUG */ 1071 1072 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET) 1073 extern struct static_key memcg_socket_limit_enabled; 1074 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1075 struct cg_proto *cg_proto) 1076 { 1077 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 1078 } 1079 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled) 1080 #else 1081 #define mem_cgroup_sockets_enabled 0 1082 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 1083 struct cg_proto *cg_proto) 1084 { 1085 return NULL; 1086 } 1087 #endif 1088 1089 static inline bool sk_stream_memory_free(const struct sock *sk) 1090 { 1091 if (sk->sk_wmem_queued >= sk->sk_sndbuf) 1092 return false; 1093 1094 return sk->sk_prot->stream_memory_free ? 1095 sk->sk_prot->stream_memory_free(sk) : true; 1096 } 1097 1098 static inline bool sk_stream_is_writeable(const struct sock *sk) 1099 { 1100 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1101 sk_stream_memory_free(sk); 1102 } 1103 1104 1105 static inline bool sk_has_memory_pressure(const struct sock *sk) 1106 { 1107 return sk->sk_prot->memory_pressure != NULL; 1108 } 1109 1110 static inline bool sk_under_memory_pressure(const struct sock *sk) 1111 { 1112 if (!sk->sk_prot->memory_pressure) 1113 return false; 1114 1115 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1116 return !!*sk->sk_cgrp->memory_pressure; 1117 1118 return !!*sk->sk_prot->memory_pressure; 1119 } 1120 1121 static inline void sk_leave_memory_pressure(struct sock *sk) 1122 { 1123 int *memory_pressure = sk->sk_prot->memory_pressure; 1124 1125 if (!memory_pressure) 1126 return; 1127 1128 if (*memory_pressure) 1129 *memory_pressure = 0; 1130 1131 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1132 struct cg_proto *cg_proto = sk->sk_cgrp; 1133 struct proto *prot = sk->sk_prot; 1134 1135 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1136 if (*cg_proto->memory_pressure) 1137 *cg_proto->memory_pressure = 0; 1138 } 1139 1140 } 1141 1142 static inline void sk_enter_memory_pressure(struct sock *sk) 1143 { 1144 if (!sk->sk_prot->enter_memory_pressure) 1145 return; 1146 1147 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1148 struct cg_proto *cg_proto = sk->sk_cgrp; 1149 struct proto *prot = sk->sk_prot; 1150 1151 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1152 cg_proto->enter_memory_pressure(sk); 1153 } 1154 1155 sk->sk_prot->enter_memory_pressure(sk); 1156 } 1157 1158 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1159 { 1160 long *prot = sk->sk_prot->sysctl_mem; 1161 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1162 prot = sk->sk_cgrp->sysctl_mem; 1163 return prot[index]; 1164 } 1165 1166 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1167 unsigned long amt, 1168 int *parent_status) 1169 { 1170 struct res_counter *fail; 1171 int ret; 1172 1173 ret = res_counter_charge_nofail(prot->memory_allocated, 1174 amt << PAGE_SHIFT, &fail); 1175 if (ret < 0) 1176 *parent_status = OVER_LIMIT; 1177 } 1178 1179 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1180 unsigned long amt) 1181 { 1182 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT); 1183 } 1184 1185 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot) 1186 { 1187 u64 ret; 1188 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE); 1189 return ret >> PAGE_SHIFT; 1190 } 1191 1192 static inline long 1193 sk_memory_allocated(const struct sock *sk) 1194 { 1195 struct proto *prot = sk->sk_prot; 1196 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1197 return memcg_memory_allocated_read(sk->sk_cgrp); 1198 1199 return atomic_long_read(prot->memory_allocated); 1200 } 1201 1202 static inline long 1203 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1204 { 1205 struct proto *prot = sk->sk_prot; 1206 1207 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1208 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1209 /* update the root cgroup regardless */ 1210 atomic_long_add_return(amt, prot->memory_allocated); 1211 return memcg_memory_allocated_read(sk->sk_cgrp); 1212 } 1213 1214 return atomic_long_add_return(amt, prot->memory_allocated); 1215 } 1216 1217 static inline void 1218 sk_memory_allocated_sub(struct sock *sk, int amt) 1219 { 1220 struct proto *prot = sk->sk_prot; 1221 1222 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1223 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1224 1225 atomic_long_sub(amt, prot->memory_allocated); 1226 } 1227 1228 static inline void sk_sockets_allocated_dec(struct sock *sk) 1229 { 1230 struct proto *prot = sk->sk_prot; 1231 1232 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1233 struct cg_proto *cg_proto = sk->sk_cgrp; 1234 1235 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1236 percpu_counter_dec(cg_proto->sockets_allocated); 1237 } 1238 1239 percpu_counter_dec(prot->sockets_allocated); 1240 } 1241 1242 static inline void sk_sockets_allocated_inc(struct sock *sk) 1243 { 1244 struct proto *prot = sk->sk_prot; 1245 1246 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1247 struct cg_proto *cg_proto = sk->sk_cgrp; 1248 1249 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1250 percpu_counter_inc(cg_proto->sockets_allocated); 1251 } 1252 1253 percpu_counter_inc(prot->sockets_allocated); 1254 } 1255 1256 static inline int 1257 sk_sockets_allocated_read_positive(struct sock *sk) 1258 { 1259 struct proto *prot = sk->sk_prot; 1260 1261 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1262 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated); 1263 1264 return percpu_counter_read_positive(prot->sockets_allocated); 1265 } 1266 1267 static inline int 1268 proto_sockets_allocated_sum_positive(struct proto *prot) 1269 { 1270 return percpu_counter_sum_positive(prot->sockets_allocated); 1271 } 1272 1273 static inline long 1274 proto_memory_allocated(struct proto *prot) 1275 { 1276 return atomic_long_read(prot->memory_allocated); 1277 } 1278 1279 static inline bool 1280 proto_memory_pressure(struct proto *prot) 1281 { 1282 if (!prot->memory_pressure) 1283 return false; 1284 return !!*prot->memory_pressure; 1285 } 1286 1287 1288 #ifdef CONFIG_PROC_FS 1289 /* Called with local bh disabled */ 1290 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1291 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 1292 #else 1293 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1294 int inc) 1295 { 1296 } 1297 #endif 1298 1299 1300 /* With per-bucket locks this operation is not-atomic, so that 1301 * this version is not worse. 1302 */ 1303 static inline void __sk_prot_rehash(struct sock *sk) 1304 { 1305 sk->sk_prot->unhash(sk); 1306 sk->sk_prot->hash(sk); 1307 } 1308 1309 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1310 1311 /* About 10 seconds */ 1312 #define SOCK_DESTROY_TIME (10*HZ) 1313 1314 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1315 #define PROT_SOCK 1024 1316 1317 #define SHUTDOWN_MASK 3 1318 #define RCV_SHUTDOWN 1 1319 #define SEND_SHUTDOWN 2 1320 1321 #define SOCK_SNDBUF_LOCK 1 1322 #define SOCK_RCVBUF_LOCK 2 1323 #define SOCK_BINDADDR_LOCK 4 1324 #define SOCK_BINDPORT_LOCK 8 1325 1326 /* sock_iocb: used to kick off async processing of socket ios */ 1327 struct sock_iocb { 1328 struct list_head list; 1329 1330 int flags; 1331 int size; 1332 struct socket *sock; 1333 struct sock *sk; 1334 struct scm_cookie *scm; 1335 struct msghdr *msg, async_msg; 1336 struct kiocb *kiocb; 1337 }; 1338 1339 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 1340 { 1341 return (struct sock_iocb *)iocb->private; 1342 } 1343 1344 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 1345 { 1346 return si->kiocb; 1347 } 1348 1349 struct socket_alloc { 1350 struct socket socket; 1351 struct inode vfs_inode; 1352 }; 1353 1354 static inline struct socket *SOCKET_I(struct inode *inode) 1355 { 1356 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1357 } 1358 1359 static inline struct inode *SOCK_INODE(struct socket *socket) 1360 { 1361 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1362 } 1363 1364 /* 1365 * Functions for memory accounting 1366 */ 1367 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 1368 extern void __sk_mem_reclaim(struct sock *sk); 1369 1370 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1371 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1372 #define SK_MEM_SEND 0 1373 #define SK_MEM_RECV 1 1374 1375 static inline int sk_mem_pages(int amt) 1376 { 1377 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1378 } 1379 1380 static inline bool sk_has_account(struct sock *sk) 1381 { 1382 /* return true if protocol supports memory accounting */ 1383 return !!sk->sk_prot->memory_allocated; 1384 } 1385 1386 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1387 { 1388 if (!sk_has_account(sk)) 1389 return true; 1390 return size <= sk->sk_forward_alloc || 1391 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1392 } 1393 1394 static inline bool 1395 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1396 { 1397 if (!sk_has_account(sk)) 1398 return true; 1399 return size<= sk->sk_forward_alloc || 1400 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1401 skb_pfmemalloc(skb); 1402 } 1403 1404 static inline void sk_mem_reclaim(struct sock *sk) 1405 { 1406 if (!sk_has_account(sk)) 1407 return; 1408 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1409 __sk_mem_reclaim(sk); 1410 } 1411 1412 static inline void sk_mem_reclaim_partial(struct sock *sk) 1413 { 1414 if (!sk_has_account(sk)) 1415 return; 1416 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1417 __sk_mem_reclaim(sk); 1418 } 1419 1420 static inline void sk_mem_charge(struct sock *sk, int size) 1421 { 1422 if (!sk_has_account(sk)) 1423 return; 1424 sk->sk_forward_alloc -= size; 1425 } 1426 1427 static inline void sk_mem_uncharge(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_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1435 { 1436 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1437 sk->sk_wmem_queued -= skb->truesize; 1438 sk_mem_uncharge(sk, skb->truesize); 1439 __kfree_skb(skb); 1440 } 1441 1442 /* Used by processes to "lock" a socket state, so that 1443 * interrupts and bottom half handlers won't change it 1444 * from under us. It essentially blocks any incoming 1445 * packets, so that we won't get any new data or any 1446 * packets that change the state of the socket. 1447 * 1448 * While locked, BH processing will add new packets to 1449 * the backlog queue. This queue is processed by the 1450 * owner of the socket lock right before it is released. 1451 * 1452 * Since ~2.3.5 it is also exclusive sleep lock serializing 1453 * accesses from user process context. 1454 */ 1455 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1456 1457 /* 1458 * Macro so as to not evaluate some arguments when 1459 * lockdep is not enabled. 1460 * 1461 * Mark both the sk_lock and the sk_lock.slock as a 1462 * per-address-family lock class. 1463 */ 1464 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1465 do { \ 1466 sk->sk_lock.owned = 0; \ 1467 init_waitqueue_head(&sk->sk_lock.wq); \ 1468 spin_lock_init(&(sk)->sk_lock.slock); \ 1469 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1470 sizeof((sk)->sk_lock)); \ 1471 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1472 (skey), (sname)); \ 1473 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1474 } while (0) 1475 1476 extern void lock_sock_nested(struct sock *sk, int subclass); 1477 1478 static inline void lock_sock(struct sock *sk) 1479 { 1480 lock_sock_nested(sk, 0); 1481 } 1482 1483 extern void release_sock(struct sock *sk); 1484 1485 /* BH context may only use the following locking interface. */ 1486 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1487 #define bh_lock_sock_nested(__sk) \ 1488 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1489 SINGLE_DEPTH_NESTING) 1490 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1491 1492 extern bool lock_sock_fast(struct sock *sk); 1493 /** 1494 * unlock_sock_fast - complement of lock_sock_fast 1495 * @sk: socket 1496 * @slow: slow mode 1497 * 1498 * fast unlock socket for user context. 1499 * If slow mode is on, we call regular release_sock() 1500 */ 1501 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1502 { 1503 if (slow) 1504 release_sock(sk); 1505 else 1506 spin_unlock_bh(&sk->sk_lock.slock); 1507 } 1508 1509 1510 extern struct sock *sk_alloc(struct net *net, int family, 1511 gfp_t priority, 1512 struct proto *prot); 1513 extern void sk_free(struct sock *sk); 1514 extern void sk_release_kernel(struct sock *sk); 1515 extern struct sock *sk_clone_lock(const struct sock *sk, 1516 const gfp_t priority); 1517 1518 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1519 unsigned long size, int force, 1520 gfp_t priority); 1521 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1522 unsigned long size, int force, 1523 gfp_t priority); 1524 extern void sock_wfree(struct sk_buff *skb); 1525 extern void skb_orphan_partial(struct sk_buff *skb); 1526 extern void sock_rfree(struct sk_buff *skb); 1527 extern void sock_edemux(struct sk_buff *skb); 1528 1529 extern int sock_setsockopt(struct socket *sock, int level, 1530 int op, char __user *optval, 1531 unsigned int optlen); 1532 1533 extern int sock_getsockopt(struct socket *sock, int level, 1534 int op, char __user *optval, 1535 int __user *optlen); 1536 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1537 unsigned long size, 1538 int noblock, 1539 int *errcode); 1540 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1541 unsigned long header_len, 1542 unsigned long data_len, 1543 int noblock, 1544 int *errcode, 1545 int max_page_order); 1546 extern void *sock_kmalloc(struct sock *sk, int size, 1547 gfp_t priority); 1548 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1549 extern void sk_send_sigurg(struct sock *sk); 1550 1551 /* 1552 * Functions to fill in entries in struct proto_ops when a protocol 1553 * does not implement a particular function. 1554 */ 1555 extern int sock_no_bind(struct socket *, 1556 struct sockaddr *, int); 1557 extern int sock_no_connect(struct socket *, 1558 struct sockaddr *, int, int); 1559 extern int sock_no_socketpair(struct socket *, 1560 struct socket *); 1561 extern int sock_no_accept(struct socket *, 1562 struct socket *, int); 1563 extern int sock_no_getname(struct socket *, 1564 struct sockaddr *, int *, int); 1565 extern unsigned int sock_no_poll(struct file *, struct socket *, 1566 struct poll_table_struct *); 1567 extern int sock_no_ioctl(struct socket *, unsigned int, 1568 unsigned long); 1569 extern int sock_no_listen(struct socket *, int); 1570 extern int sock_no_shutdown(struct socket *, int); 1571 extern int sock_no_getsockopt(struct socket *, int , int, 1572 char __user *, int __user *); 1573 extern int sock_no_setsockopt(struct socket *, int, int, 1574 char __user *, unsigned int); 1575 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1576 struct msghdr *, size_t); 1577 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1578 struct msghdr *, size_t, int); 1579 extern int sock_no_mmap(struct file *file, 1580 struct socket *sock, 1581 struct vm_area_struct *vma); 1582 extern ssize_t sock_no_sendpage(struct socket *sock, 1583 struct page *page, 1584 int offset, size_t size, 1585 int flags); 1586 1587 /* 1588 * Functions to fill in entries in struct proto_ops when a protocol 1589 * uses the inet style. 1590 */ 1591 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1592 char __user *optval, int __user *optlen); 1593 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1594 struct msghdr *msg, size_t size, int flags); 1595 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1596 char __user *optval, unsigned int optlen); 1597 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1598 int optname, char __user *optval, int __user *optlen); 1599 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1600 int optname, char __user *optval, unsigned int optlen); 1601 1602 extern void sk_common_release(struct sock *sk); 1603 1604 /* 1605 * Default socket callbacks and setup code 1606 */ 1607 1608 /* Initialise core socket variables */ 1609 extern void sock_init_data(struct socket *sock, struct sock *sk); 1610 1611 extern void sk_filter_release_rcu(struct rcu_head *rcu); 1612 1613 /** 1614 * sk_filter_release - release a socket filter 1615 * @fp: filter to remove 1616 * 1617 * Remove a filter from a socket and release its resources. 1618 */ 1619 1620 static inline void sk_filter_release(struct sk_filter *fp) 1621 { 1622 if (atomic_dec_and_test(&fp->refcnt)) 1623 call_rcu(&fp->rcu, sk_filter_release_rcu); 1624 } 1625 1626 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1627 { 1628 unsigned int size = sk_filter_len(fp); 1629 1630 atomic_sub(size, &sk->sk_omem_alloc); 1631 sk_filter_release(fp); 1632 } 1633 1634 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1635 { 1636 atomic_inc(&fp->refcnt); 1637 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1638 } 1639 1640 /* 1641 * Socket reference counting postulates. 1642 * 1643 * * Each user of socket SHOULD hold a reference count. 1644 * * Each access point to socket (an hash table bucket, reference from a list, 1645 * running timer, skb in flight MUST hold a reference count. 1646 * * When reference count hits 0, it means it will never increase back. 1647 * * When reference count hits 0, it means that no references from 1648 * outside exist to this socket and current process on current CPU 1649 * is last user and may/should destroy this socket. 1650 * * sk_free is called from any context: process, BH, IRQ. When 1651 * it is called, socket has no references from outside -> sk_free 1652 * may release descendant resources allocated by the socket, but 1653 * to the time when it is called, socket is NOT referenced by any 1654 * hash tables, lists etc. 1655 * * Packets, delivered from outside (from network or from another process) 1656 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1657 * when they sit in queue. Otherwise, packets will leak to hole, when 1658 * socket is looked up by one cpu and unhasing is made by another CPU. 1659 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1660 * (leak to backlog). Packet socket does all the processing inside 1661 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1662 * use separate SMP lock, so that they are prone too. 1663 */ 1664 1665 /* Ungrab socket and destroy it, if it was the last reference. */ 1666 static inline void sock_put(struct sock *sk) 1667 { 1668 if (atomic_dec_and_test(&sk->sk_refcnt)) 1669 sk_free(sk); 1670 } 1671 1672 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1673 const int nested); 1674 1675 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1676 { 1677 sk->sk_tx_queue_mapping = tx_queue; 1678 } 1679 1680 static inline void sk_tx_queue_clear(struct sock *sk) 1681 { 1682 sk->sk_tx_queue_mapping = -1; 1683 } 1684 1685 static inline int sk_tx_queue_get(const struct sock *sk) 1686 { 1687 return sk ? sk->sk_tx_queue_mapping : -1; 1688 } 1689 1690 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1691 { 1692 sk_tx_queue_clear(sk); 1693 sk->sk_socket = sock; 1694 } 1695 1696 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1697 { 1698 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1699 return &rcu_dereference_raw(sk->sk_wq)->wait; 1700 } 1701 /* Detach socket from process context. 1702 * Announce socket dead, detach it from wait queue and inode. 1703 * Note that parent inode held reference count on this struct sock, 1704 * we do not release it in this function, because protocol 1705 * probably wants some additional cleanups or even continuing 1706 * to work with this socket (TCP). 1707 */ 1708 static inline void sock_orphan(struct sock *sk) 1709 { 1710 write_lock_bh(&sk->sk_callback_lock); 1711 sock_set_flag(sk, SOCK_DEAD); 1712 sk_set_socket(sk, NULL); 1713 sk->sk_wq = NULL; 1714 write_unlock_bh(&sk->sk_callback_lock); 1715 } 1716 1717 static inline void sock_graft(struct sock *sk, struct socket *parent) 1718 { 1719 write_lock_bh(&sk->sk_callback_lock); 1720 sk->sk_wq = parent->wq; 1721 parent->sk = sk; 1722 sk_set_socket(sk, parent); 1723 security_sock_graft(sk, parent); 1724 write_unlock_bh(&sk->sk_callback_lock); 1725 } 1726 1727 extern kuid_t sock_i_uid(struct sock *sk); 1728 extern unsigned long sock_i_ino(struct sock *sk); 1729 1730 static inline struct dst_entry * 1731 __sk_dst_get(struct sock *sk) 1732 { 1733 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1734 lockdep_is_held(&sk->sk_lock.slock)); 1735 } 1736 1737 static inline struct dst_entry * 1738 sk_dst_get(struct sock *sk) 1739 { 1740 struct dst_entry *dst; 1741 1742 rcu_read_lock(); 1743 dst = rcu_dereference(sk->sk_dst_cache); 1744 if (dst) 1745 dst_hold(dst); 1746 rcu_read_unlock(); 1747 return dst; 1748 } 1749 1750 extern void sk_reset_txq(struct sock *sk); 1751 1752 static inline void dst_negative_advice(struct sock *sk) 1753 { 1754 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1755 1756 if (dst && dst->ops->negative_advice) { 1757 ndst = dst->ops->negative_advice(dst); 1758 1759 if (ndst != dst) { 1760 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1761 sk_reset_txq(sk); 1762 } 1763 } 1764 } 1765 1766 static inline void 1767 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1768 { 1769 struct dst_entry *old_dst; 1770 1771 sk_tx_queue_clear(sk); 1772 /* 1773 * This can be called while sk is owned by the caller only, 1774 * with no state that can be checked in a rcu_dereference_check() cond 1775 */ 1776 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1777 rcu_assign_pointer(sk->sk_dst_cache, dst); 1778 dst_release(old_dst); 1779 } 1780 1781 static inline void 1782 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1783 { 1784 spin_lock(&sk->sk_dst_lock); 1785 __sk_dst_set(sk, dst); 1786 spin_unlock(&sk->sk_dst_lock); 1787 } 1788 1789 static inline void 1790 __sk_dst_reset(struct sock *sk) 1791 { 1792 __sk_dst_set(sk, NULL); 1793 } 1794 1795 static inline void 1796 sk_dst_reset(struct sock *sk) 1797 { 1798 spin_lock(&sk->sk_dst_lock); 1799 __sk_dst_reset(sk); 1800 spin_unlock(&sk->sk_dst_lock); 1801 } 1802 1803 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1804 1805 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1806 1807 static inline bool sk_can_gso(const struct sock *sk) 1808 { 1809 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1810 } 1811 1812 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1813 1814 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1815 { 1816 sk->sk_route_nocaps |= flags; 1817 sk->sk_route_caps &= ~flags; 1818 } 1819 1820 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1821 char __user *from, char *to, 1822 int copy, int offset) 1823 { 1824 if (skb->ip_summed == CHECKSUM_NONE) { 1825 int err = 0; 1826 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1827 if (err) 1828 return err; 1829 skb->csum = csum_block_add(skb->csum, csum, offset); 1830 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1831 if (!access_ok(VERIFY_READ, from, copy) || 1832 __copy_from_user_nocache(to, from, copy)) 1833 return -EFAULT; 1834 } else if (copy_from_user(to, from, copy)) 1835 return -EFAULT; 1836 1837 return 0; 1838 } 1839 1840 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1841 char __user *from, int copy) 1842 { 1843 int err, offset = skb->len; 1844 1845 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1846 copy, offset); 1847 if (err) 1848 __skb_trim(skb, offset); 1849 1850 return err; 1851 } 1852 1853 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from, 1854 struct sk_buff *skb, 1855 struct page *page, 1856 int off, int copy) 1857 { 1858 int err; 1859 1860 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1861 copy, skb->len); 1862 if (err) 1863 return err; 1864 1865 skb->len += copy; 1866 skb->data_len += copy; 1867 skb->truesize += copy; 1868 sk->sk_wmem_queued += copy; 1869 sk_mem_charge(sk, copy); 1870 return 0; 1871 } 1872 1873 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1874 struct sk_buff *skb, struct page *page, 1875 int off, int copy) 1876 { 1877 if (skb->ip_summed == CHECKSUM_NONE) { 1878 int err = 0; 1879 __wsum csum = csum_and_copy_from_user(from, 1880 page_address(page) + off, 1881 copy, 0, &err); 1882 if (err) 1883 return err; 1884 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1885 } else if (copy_from_user(page_address(page) + off, from, copy)) 1886 return -EFAULT; 1887 1888 skb->len += copy; 1889 skb->data_len += copy; 1890 skb->truesize += copy; 1891 sk->sk_wmem_queued += copy; 1892 sk_mem_charge(sk, copy); 1893 return 0; 1894 } 1895 1896 /** 1897 * sk_wmem_alloc_get - returns write allocations 1898 * @sk: socket 1899 * 1900 * Returns sk_wmem_alloc minus initial offset of one 1901 */ 1902 static inline int sk_wmem_alloc_get(const struct sock *sk) 1903 { 1904 return atomic_read(&sk->sk_wmem_alloc) - 1; 1905 } 1906 1907 /** 1908 * sk_rmem_alloc_get - returns read allocations 1909 * @sk: socket 1910 * 1911 * Returns sk_rmem_alloc 1912 */ 1913 static inline int sk_rmem_alloc_get(const struct sock *sk) 1914 { 1915 return atomic_read(&sk->sk_rmem_alloc); 1916 } 1917 1918 /** 1919 * sk_has_allocations - check if allocations are outstanding 1920 * @sk: socket 1921 * 1922 * Returns true if socket has write or read allocations 1923 */ 1924 static inline bool sk_has_allocations(const struct sock *sk) 1925 { 1926 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1927 } 1928 1929 /** 1930 * wq_has_sleeper - check if there are any waiting processes 1931 * @wq: struct socket_wq 1932 * 1933 * Returns true if socket_wq has waiting processes 1934 * 1935 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1936 * barrier call. They were added due to the race found within the tcp code. 1937 * 1938 * Consider following tcp code paths: 1939 * 1940 * CPU1 CPU2 1941 * 1942 * sys_select receive packet 1943 * ... ... 1944 * __add_wait_queue update tp->rcv_nxt 1945 * ... ... 1946 * tp->rcv_nxt check sock_def_readable 1947 * ... { 1948 * schedule rcu_read_lock(); 1949 * wq = rcu_dereference(sk->sk_wq); 1950 * if (wq && waitqueue_active(&wq->wait)) 1951 * wake_up_interruptible(&wq->wait) 1952 * ... 1953 * } 1954 * 1955 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1956 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1957 * could then endup calling schedule and sleep forever if there are no more 1958 * data on the socket. 1959 * 1960 */ 1961 static inline bool wq_has_sleeper(struct socket_wq *wq) 1962 { 1963 /* We need to be sure we are in sync with the 1964 * add_wait_queue modifications to the wait queue. 1965 * 1966 * This memory barrier is paired in the sock_poll_wait. 1967 */ 1968 smp_mb(); 1969 return wq && waitqueue_active(&wq->wait); 1970 } 1971 1972 /** 1973 * sock_poll_wait - place memory barrier behind the poll_wait call. 1974 * @filp: file 1975 * @wait_address: socket wait queue 1976 * @p: poll_table 1977 * 1978 * See the comments in the wq_has_sleeper function. 1979 */ 1980 static inline void sock_poll_wait(struct file *filp, 1981 wait_queue_head_t *wait_address, poll_table *p) 1982 { 1983 if (!poll_does_not_wait(p) && wait_address) { 1984 poll_wait(filp, wait_address, p); 1985 /* We need to be sure we are in sync with the 1986 * socket flags modification. 1987 * 1988 * This memory barrier is paired in the wq_has_sleeper. 1989 */ 1990 smp_mb(); 1991 } 1992 } 1993 1994 /* 1995 * Queue a received datagram if it will fit. Stream and sequenced 1996 * protocols can't normally use this as they need to fit buffers in 1997 * and play with them. 1998 * 1999 * Inlined as it's very short and called for pretty much every 2000 * packet ever received. 2001 */ 2002 2003 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 2004 { 2005 skb_orphan(skb); 2006 skb->sk = sk; 2007 skb->destructor = sock_wfree; 2008 /* 2009 * We used to take a refcount on sk, but following operation 2010 * is enough to guarantee sk_free() wont free this sock until 2011 * all in-flight packets are completed 2012 */ 2013 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 2014 } 2015 2016 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2017 { 2018 skb_orphan(skb); 2019 skb->sk = sk; 2020 skb->destructor = sock_rfree; 2021 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2022 sk_mem_charge(sk, skb->truesize); 2023 } 2024 2025 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2026 unsigned long expires); 2027 2028 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2029 2030 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2031 2032 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2033 2034 /* 2035 * Recover an error report and clear atomically 2036 */ 2037 2038 static inline int sock_error(struct sock *sk) 2039 { 2040 int err; 2041 if (likely(!sk->sk_err)) 2042 return 0; 2043 err = xchg(&sk->sk_err, 0); 2044 return -err; 2045 } 2046 2047 static inline unsigned long sock_wspace(struct sock *sk) 2048 { 2049 int amt = 0; 2050 2051 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2052 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 2053 if (amt < 0) 2054 amt = 0; 2055 } 2056 return amt; 2057 } 2058 2059 static inline void sk_wake_async(struct sock *sk, int how, int band) 2060 { 2061 if (sock_flag(sk, SOCK_FASYNC)) 2062 sock_wake_async(sk->sk_socket, how, band); 2063 } 2064 2065 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2066 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2067 * Note: for send buffers, TCP works better if we can build two skbs at 2068 * minimum. 2069 */ 2070 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2071 2072 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2073 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2074 2075 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2076 { 2077 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 2078 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2079 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF); 2080 } 2081 } 2082 2083 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 2084 2085 /** 2086 * sk_page_frag - return an appropriate page_frag 2087 * @sk: socket 2088 * 2089 * If socket allocation mode allows current thread to sleep, it means its 2090 * safe to use the per task page_frag instead of the per socket one. 2091 */ 2092 static inline struct page_frag *sk_page_frag(struct sock *sk) 2093 { 2094 if (sk->sk_allocation & __GFP_WAIT) 2095 return ¤t->task_frag; 2096 2097 return &sk->sk_frag; 2098 } 2099 2100 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2101 2102 /* 2103 * Default write policy as shown to user space via poll/select/SIGIO 2104 */ 2105 static inline bool sock_writeable(const struct sock *sk) 2106 { 2107 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 2108 } 2109 2110 static inline gfp_t gfp_any(void) 2111 { 2112 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2113 } 2114 2115 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2116 { 2117 return noblock ? 0 : sk->sk_rcvtimeo; 2118 } 2119 2120 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2121 { 2122 return noblock ? 0 : sk->sk_sndtimeo; 2123 } 2124 2125 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2126 { 2127 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 2128 } 2129 2130 /* Alas, with timeout socket operations are not restartable. 2131 * Compare this to poll(). 2132 */ 2133 static inline int sock_intr_errno(long timeo) 2134 { 2135 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2136 } 2137 2138 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2139 struct sk_buff *skb); 2140 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2141 struct sk_buff *skb); 2142 2143 static inline void 2144 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2145 { 2146 ktime_t kt = skb->tstamp; 2147 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2148 2149 /* 2150 * generate control messages if 2151 * - receive time stamping in software requested (SOCK_RCVTSTAMP 2152 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 2153 * - software time stamp available and wanted 2154 * (SOCK_TIMESTAMPING_SOFTWARE) 2155 * - hardware time stamps available and wanted 2156 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 2157 * SOCK_TIMESTAMPING_RAW_HARDWARE) 2158 */ 2159 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2160 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 2161 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 2162 (hwtstamps->hwtstamp.tv64 && 2163 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 2164 (hwtstamps->syststamp.tv64 && 2165 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 2166 __sock_recv_timestamp(msg, sk, skb); 2167 else 2168 sk->sk_stamp = kt; 2169 2170 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2171 __sock_recv_wifi_status(msg, sk, skb); 2172 } 2173 2174 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2175 struct sk_buff *skb); 2176 2177 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2178 struct sk_buff *skb) 2179 { 2180 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2181 (1UL << SOCK_RCVTSTAMP) | \ 2182 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 2183 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 2184 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 2185 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 2186 2187 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 2188 __sock_recv_ts_and_drops(msg, sk, skb); 2189 else 2190 sk->sk_stamp = skb->tstamp; 2191 } 2192 2193 /** 2194 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2195 * @sk: socket sending this packet 2196 * @tx_flags: filled with instructions for time stamping 2197 * 2198 * Currently only depends on SOCK_TIMESTAMPING* flags. 2199 */ 2200 extern void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 2201 2202 /** 2203 * sk_eat_skb - Release a skb if it is no longer needed 2204 * @sk: socket to eat this skb from 2205 * @skb: socket buffer to eat 2206 * @copied_early: flag indicating whether DMA operations copied this data early 2207 * 2208 * This routine must be called with interrupts disabled or with the socket 2209 * locked so that the sk_buff queue operation is ok. 2210 */ 2211 #ifdef CONFIG_NET_DMA 2212 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2213 { 2214 __skb_unlink(skb, &sk->sk_receive_queue); 2215 if (!copied_early) 2216 __kfree_skb(skb); 2217 else 2218 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 2219 } 2220 #else 2221 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early) 2222 { 2223 __skb_unlink(skb, &sk->sk_receive_queue); 2224 __kfree_skb(skb); 2225 } 2226 #endif 2227 2228 static inline 2229 struct net *sock_net(const struct sock *sk) 2230 { 2231 return read_pnet(&sk->sk_net); 2232 } 2233 2234 static inline 2235 void sock_net_set(struct sock *sk, struct net *net) 2236 { 2237 write_pnet(&sk->sk_net, net); 2238 } 2239 2240 /* 2241 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2242 * They should not hold a reference to a namespace in order to allow 2243 * to stop it. 2244 * Sockets after sk_change_net should be released using sk_release_kernel 2245 */ 2246 static inline void sk_change_net(struct sock *sk, struct net *net) 2247 { 2248 put_net(sock_net(sk)); 2249 sock_net_set(sk, hold_net(net)); 2250 } 2251 2252 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2253 { 2254 if (skb->sk) { 2255 struct sock *sk = skb->sk; 2256 2257 skb->destructor = NULL; 2258 skb->sk = NULL; 2259 return sk; 2260 } 2261 return NULL; 2262 } 2263 2264 extern void sock_enable_timestamp(struct sock *sk, int flag); 2265 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 2266 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 2267 extern int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, 2268 int level, int type); 2269 2270 /* 2271 * Enable debug/info messages 2272 */ 2273 extern int net_msg_warn; 2274 #define NETDEBUG(fmt, args...) \ 2275 do { if (net_msg_warn) printk(fmt,##args); } while (0) 2276 2277 #define LIMIT_NETDEBUG(fmt, args...) \ 2278 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 2279 2280 extern __u32 sysctl_wmem_max; 2281 extern __u32 sysctl_rmem_max; 2282 2283 extern int sysctl_optmem_max; 2284 2285 extern __u32 sysctl_wmem_default; 2286 extern __u32 sysctl_rmem_default; 2287 2288 #endif /* _SOCK_H */ 2289