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/lockdep.h> 50 #include <linux/netdevice.h> 51 #include <linux/skbuff.h> /* struct sk_buff */ 52 #include <linux/mm.h> 53 #include <linux/security.h> 54 #include <linux/slab.h> 55 #include <linux/uaccess.h> 56 #include <linux/memcontrol.h> 57 #include <linux/res_counter.h> 58 #include <linux/jump_label.h> 59 60 #include <linux/filter.h> 61 #include <linux/rculist_nulls.h> 62 #include <linux/poll.h> 63 64 #include <linux/atomic.h> 65 #include <net/dst.h> 66 #include <net/checksum.h> 67 68 struct cgroup; 69 struct cgroup_subsys; 70 #ifdef CONFIG_NET 71 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss); 72 void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss); 73 #else 74 static inline 75 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss) 76 { 77 return 0; 78 } 79 static inline 80 void mem_cgroup_sockets_destroy(struct cgroup *cgrp, struct cgroup_subsys *ss) 81 { 82 } 83 #endif 84 /* 85 * This structure really needs to be cleaned up. 86 * Most of it is for TCP, and not used by any of 87 * the other protocols. 88 */ 89 90 /* Define this to get the SOCK_DBG debugging facility. */ 91 #define SOCK_DEBUGGING 92 #ifdef SOCK_DEBUGGING 93 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 94 printk(KERN_DEBUG msg); } while (0) 95 #else 96 /* Validate arguments and do nothing */ 97 static inline __printf(2, 3) 98 void SOCK_DEBUG(struct sock *sk, const char *msg, ...) 99 { 100 } 101 #endif 102 103 /* This is the per-socket lock. The spinlock provides a synchronization 104 * between user contexts and software interrupt processing, whereas the 105 * mini-semaphore synchronizes multiple users amongst themselves. 106 */ 107 typedef struct { 108 spinlock_t slock; 109 int owned; 110 wait_queue_head_t wq; 111 /* 112 * We express the mutex-alike socket_lock semantics 113 * to the lock validator by explicitly managing 114 * the slock as a lock variant (in addition to 115 * the slock itself): 116 */ 117 #ifdef CONFIG_DEBUG_LOCK_ALLOC 118 struct lockdep_map dep_map; 119 #endif 120 } socket_lock_t; 121 122 struct sock; 123 struct proto; 124 struct net; 125 126 /** 127 * struct sock_common - minimal network layer representation of sockets 128 * @skc_daddr: Foreign IPv4 addr 129 * @skc_rcv_saddr: Bound local IPv4 addr 130 * @skc_hash: hash value used with various protocol lookup tables 131 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 132 * @skc_family: network address family 133 * @skc_state: Connection state 134 * @skc_reuse: %SO_REUSEADDR setting 135 * @skc_bound_dev_if: bound device index if != 0 136 * @skc_bind_node: bind hash linkage for various protocol lookup tables 137 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 138 * @skc_prot: protocol handlers inside a network family 139 * @skc_net: reference to the network namespace of this socket 140 * @skc_node: main hash linkage for various protocol lookup tables 141 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 142 * @skc_tx_queue_mapping: tx queue number for this connection 143 * @skc_refcnt: reference count 144 * 145 * This is the minimal network layer representation of sockets, the header 146 * for struct sock and struct inet_timewait_sock. 147 */ 148 struct sock_common { 149 /* skc_daddr and skc_rcv_saddr must be grouped : 150 * cf INET_MATCH() and INET_TW_MATCH() 151 */ 152 __be32 skc_daddr; 153 __be32 skc_rcv_saddr; 154 155 union { 156 unsigned int skc_hash; 157 __u16 skc_u16hashes[2]; 158 }; 159 unsigned short skc_family; 160 volatile unsigned char skc_state; 161 unsigned char skc_reuse; 162 int skc_bound_dev_if; 163 union { 164 struct hlist_node skc_bind_node; 165 struct hlist_nulls_node skc_portaddr_node; 166 }; 167 struct proto *skc_prot; 168 #ifdef CONFIG_NET_NS 169 struct net *skc_net; 170 #endif 171 /* 172 * fields between dontcopy_begin/dontcopy_end 173 * are not copied in sock_copy() 174 */ 175 /* private: */ 176 int skc_dontcopy_begin[0]; 177 /* public: */ 178 union { 179 struct hlist_node skc_node; 180 struct hlist_nulls_node skc_nulls_node; 181 }; 182 int skc_tx_queue_mapping; 183 atomic_t skc_refcnt; 184 /* private: */ 185 int skc_dontcopy_end[0]; 186 /* public: */ 187 }; 188 189 struct cg_proto; 190 /** 191 * struct sock - network layer representation of sockets 192 * @__sk_common: shared layout with inet_timewait_sock 193 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 194 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 195 * @sk_lock: synchronizer 196 * @sk_rcvbuf: size of receive buffer in bytes 197 * @sk_wq: sock wait queue and async head 198 * @sk_dst_cache: destination cache 199 * @sk_dst_lock: destination cache lock 200 * @sk_policy: flow policy 201 * @sk_receive_queue: incoming packets 202 * @sk_wmem_alloc: transmit queue bytes committed 203 * @sk_write_queue: Packet sending queue 204 * @sk_async_wait_queue: DMA copied packets 205 * @sk_omem_alloc: "o" is "option" or "other" 206 * @sk_wmem_queued: persistent queue size 207 * @sk_forward_alloc: space allocated forward 208 * @sk_allocation: allocation mode 209 * @sk_sndbuf: size of send buffer in bytes 210 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 211 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 212 * @sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets 213 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 214 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 215 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 216 * @sk_gso_max_size: Maximum GSO segment size to build 217 * @sk_lingertime: %SO_LINGER l_linger setting 218 * @sk_backlog: always used with the per-socket spinlock held 219 * @sk_callback_lock: used with the callbacks in the end of this struct 220 * @sk_error_queue: rarely used 221 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 222 * IPV6_ADDRFORM for instance) 223 * @sk_err: last error 224 * @sk_err_soft: errors that don't cause failure but are the cause of a 225 * persistent failure not just 'timed out' 226 * @sk_drops: raw/udp drops counter 227 * @sk_ack_backlog: current listen backlog 228 * @sk_max_ack_backlog: listen backlog set in listen() 229 * @sk_priority: %SO_PRIORITY setting 230 * @sk_cgrp_prioidx: socket group's priority map index 231 * @sk_type: socket type (%SOCK_STREAM, etc) 232 * @sk_protocol: which protocol this socket belongs in this network family 233 * @sk_peer_pid: &struct pid for this socket's peer 234 * @sk_peer_cred: %SO_PEERCRED setting 235 * @sk_rcvlowat: %SO_RCVLOWAT setting 236 * @sk_rcvtimeo: %SO_RCVTIMEO setting 237 * @sk_sndtimeo: %SO_SNDTIMEO setting 238 * @sk_rxhash: flow hash received from netif layer 239 * @sk_filter: socket filtering instructions 240 * @sk_protinfo: private area, net family specific, when not using slab 241 * @sk_timer: sock cleanup timer 242 * @sk_stamp: time stamp of last packet received 243 * @sk_socket: Identd and reporting IO signals 244 * @sk_user_data: RPC layer private data 245 * @sk_sndmsg_page: cached page for sendmsg 246 * @sk_sndmsg_off: cached offset for sendmsg 247 * @sk_send_head: front of stuff to transmit 248 * @sk_security: used by security modules 249 * @sk_mark: generic packet mark 250 * @sk_classid: this socket's cgroup classid 251 * @sk_cgrp: this socket's cgroup-specific proto data 252 * @sk_write_pending: a write to stream socket waits to start 253 * @sk_state_change: callback to indicate change in the state of the sock 254 * @sk_data_ready: callback to indicate there is data to be processed 255 * @sk_write_space: callback to indicate there is bf sending space available 256 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 257 * @sk_backlog_rcv: callback to process the backlog 258 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 259 */ 260 struct sock { 261 /* 262 * Now struct inet_timewait_sock also uses sock_common, so please just 263 * don't add nothing before this first member (__sk_common) --acme 264 */ 265 struct sock_common __sk_common; 266 #define sk_node __sk_common.skc_node 267 #define sk_nulls_node __sk_common.skc_nulls_node 268 #define sk_refcnt __sk_common.skc_refcnt 269 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 270 271 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 272 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 273 #define sk_hash __sk_common.skc_hash 274 #define sk_family __sk_common.skc_family 275 #define sk_state __sk_common.skc_state 276 #define sk_reuse __sk_common.skc_reuse 277 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 278 #define sk_bind_node __sk_common.skc_bind_node 279 #define sk_prot __sk_common.skc_prot 280 #define sk_net __sk_common.skc_net 281 socket_lock_t sk_lock; 282 struct sk_buff_head sk_receive_queue; 283 /* 284 * The backlog queue is special, it is always used with 285 * the per-socket spinlock held and requires low latency 286 * access. Therefore we special case it's implementation. 287 * Note : rmem_alloc is in this structure to fill a hole 288 * on 64bit arches, not because its logically part of 289 * backlog. 290 */ 291 struct { 292 atomic_t rmem_alloc; 293 int len; 294 struct sk_buff *head; 295 struct sk_buff *tail; 296 } sk_backlog; 297 #define sk_rmem_alloc sk_backlog.rmem_alloc 298 int sk_forward_alloc; 299 #ifdef CONFIG_RPS 300 __u32 sk_rxhash; 301 #endif 302 atomic_t sk_drops; 303 int sk_rcvbuf; 304 305 struct sk_filter __rcu *sk_filter; 306 struct socket_wq __rcu *sk_wq; 307 308 #ifdef CONFIG_NET_DMA 309 struct sk_buff_head sk_async_wait_queue; 310 #endif 311 312 #ifdef CONFIG_XFRM 313 struct xfrm_policy *sk_policy[2]; 314 #endif 315 unsigned long sk_flags; 316 struct dst_entry *sk_dst_cache; 317 spinlock_t sk_dst_lock; 318 atomic_t sk_wmem_alloc; 319 atomic_t sk_omem_alloc; 320 int sk_sndbuf; 321 struct sk_buff_head sk_write_queue; 322 kmemcheck_bitfield_begin(flags); 323 unsigned int sk_shutdown : 2, 324 sk_no_check : 2, 325 sk_userlocks : 4, 326 sk_protocol : 8, 327 sk_type : 16; 328 kmemcheck_bitfield_end(flags); 329 int sk_wmem_queued; 330 gfp_t sk_allocation; 331 netdev_features_t sk_route_caps; 332 netdev_features_t sk_route_nocaps; 333 int sk_gso_type; 334 unsigned int sk_gso_max_size; 335 int sk_rcvlowat; 336 unsigned long sk_lingertime; 337 struct sk_buff_head sk_error_queue; 338 struct proto *sk_prot_creator; 339 rwlock_t sk_callback_lock; 340 int sk_err, 341 sk_err_soft; 342 unsigned short sk_ack_backlog; 343 unsigned short sk_max_ack_backlog; 344 __u32 sk_priority; 345 #ifdef CONFIG_CGROUPS 346 __u32 sk_cgrp_prioidx; 347 #endif 348 struct pid *sk_peer_pid; 349 const struct cred *sk_peer_cred; 350 long sk_rcvtimeo; 351 long sk_sndtimeo; 352 void *sk_protinfo; 353 struct timer_list sk_timer; 354 ktime_t sk_stamp; 355 struct socket *sk_socket; 356 void *sk_user_data; 357 struct page *sk_sndmsg_page; 358 struct sk_buff *sk_send_head; 359 __u32 sk_sndmsg_off; 360 __s32 sk_peek_off; 361 int sk_write_pending; 362 #ifdef CONFIG_SECURITY 363 void *sk_security; 364 #endif 365 __u32 sk_mark; 366 u32 sk_classid; 367 struct cg_proto *sk_cgrp; 368 void (*sk_state_change)(struct sock *sk); 369 void (*sk_data_ready)(struct sock *sk, int bytes); 370 void (*sk_write_space)(struct sock *sk); 371 void (*sk_error_report)(struct sock *sk); 372 int (*sk_backlog_rcv)(struct sock *sk, 373 struct sk_buff *skb); 374 void (*sk_destruct)(struct sock *sk); 375 }; 376 377 static inline int sk_peek_offset(struct sock *sk, int flags) 378 { 379 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0)) 380 return sk->sk_peek_off; 381 else 382 return 0; 383 } 384 385 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 386 { 387 if (sk->sk_peek_off >= 0) { 388 if (sk->sk_peek_off >= val) 389 sk->sk_peek_off -= val; 390 else 391 sk->sk_peek_off = 0; 392 } 393 } 394 395 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 396 { 397 if (sk->sk_peek_off >= 0) 398 sk->sk_peek_off += val; 399 } 400 401 /* 402 * Hashed lists helper routines 403 */ 404 static inline struct sock *sk_entry(const struct hlist_node *node) 405 { 406 return hlist_entry(node, struct sock, sk_node); 407 } 408 409 static inline struct sock *__sk_head(const struct hlist_head *head) 410 { 411 return hlist_entry(head->first, struct sock, sk_node); 412 } 413 414 static inline struct sock *sk_head(const struct hlist_head *head) 415 { 416 return hlist_empty(head) ? NULL : __sk_head(head); 417 } 418 419 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 420 { 421 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 422 } 423 424 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 425 { 426 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 427 } 428 429 static inline struct sock *sk_next(const struct sock *sk) 430 { 431 return sk->sk_node.next ? 432 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL; 433 } 434 435 static inline struct sock *sk_nulls_next(const struct sock *sk) 436 { 437 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 438 hlist_nulls_entry(sk->sk_nulls_node.next, 439 struct sock, sk_nulls_node) : 440 NULL; 441 } 442 443 static inline int sk_unhashed(const struct sock *sk) 444 { 445 return hlist_unhashed(&sk->sk_node); 446 } 447 448 static inline int sk_hashed(const struct sock *sk) 449 { 450 return !sk_unhashed(sk); 451 } 452 453 static __inline__ void sk_node_init(struct hlist_node *node) 454 { 455 node->pprev = NULL; 456 } 457 458 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node) 459 { 460 node->pprev = NULL; 461 } 462 463 static __inline__ void __sk_del_node(struct sock *sk) 464 { 465 __hlist_del(&sk->sk_node); 466 } 467 468 /* NB: equivalent to hlist_del_init_rcu */ 469 static __inline__ int __sk_del_node_init(struct sock *sk) 470 { 471 if (sk_hashed(sk)) { 472 __sk_del_node(sk); 473 sk_node_init(&sk->sk_node); 474 return 1; 475 } 476 return 0; 477 } 478 479 /* Grab socket reference count. This operation is valid only 480 when sk is ALREADY grabbed f.e. it is found in hash table 481 or a list and the lookup is made under lock preventing hash table 482 modifications. 483 */ 484 485 static inline void sock_hold(struct sock *sk) 486 { 487 atomic_inc(&sk->sk_refcnt); 488 } 489 490 /* Ungrab socket in the context, which assumes that socket refcnt 491 cannot hit zero, f.e. it is true in context of any socketcall. 492 */ 493 static inline void __sock_put(struct sock *sk) 494 { 495 atomic_dec(&sk->sk_refcnt); 496 } 497 498 static __inline__ int sk_del_node_init(struct sock *sk) 499 { 500 int rc = __sk_del_node_init(sk); 501 502 if (rc) { 503 /* paranoid for a while -acme */ 504 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 505 __sock_put(sk); 506 } 507 return rc; 508 } 509 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 510 511 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk) 512 { 513 if (sk_hashed(sk)) { 514 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 515 return 1; 516 } 517 return 0; 518 } 519 520 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk) 521 { 522 int rc = __sk_nulls_del_node_init_rcu(sk); 523 524 if (rc) { 525 /* paranoid for a while -acme */ 526 WARN_ON(atomic_read(&sk->sk_refcnt) == 1); 527 __sock_put(sk); 528 } 529 return rc; 530 } 531 532 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list) 533 { 534 hlist_add_head(&sk->sk_node, list); 535 } 536 537 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list) 538 { 539 sock_hold(sk); 540 __sk_add_node(sk, list); 541 } 542 543 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 544 { 545 sock_hold(sk); 546 hlist_add_head_rcu(&sk->sk_node, list); 547 } 548 549 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 550 { 551 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 552 } 553 554 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 555 { 556 sock_hold(sk); 557 __sk_nulls_add_node_rcu(sk, list); 558 } 559 560 static __inline__ void __sk_del_bind_node(struct sock *sk) 561 { 562 __hlist_del(&sk->sk_bind_node); 563 } 564 565 static __inline__ void sk_add_bind_node(struct sock *sk, 566 struct hlist_head *list) 567 { 568 hlist_add_head(&sk->sk_bind_node, list); 569 } 570 571 #define sk_for_each(__sk, node, list) \ 572 hlist_for_each_entry(__sk, node, list, sk_node) 573 #define sk_for_each_rcu(__sk, node, list) \ 574 hlist_for_each_entry_rcu(__sk, node, list, sk_node) 575 #define sk_nulls_for_each(__sk, node, list) \ 576 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 577 #define sk_nulls_for_each_rcu(__sk, node, list) \ 578 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 579 #define sk_for_each_from(__sk, node) \ 580 if (__sk && ({ node = &(__sk)->sk_node; 1; })) \ 581 hlist_for_each_entry_from(__sk, node, sk_node) 582 #define sk_nulls_for_each_from(__sk, node) \ 583 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 584 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 585 #define sk_for_each_safe(__sk, node, tmp, list) \ 586 hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node) 587 #define sk_for_each_bound(__sk, node, list) \ 588 hlist_for_each_entry(__sk, node, list, sk_bind_node) 589 590 /* Sock flags */ 591 enum sock_flags { 592 SOCK_DEAD, 593 SOCK_DONE, 594 SOCK_URGINLINE, 595 SOCK_KEEPOPEN, 596 SOCK_LINGER, 597 SOCK_DESTROY, 598 SOCK_BROADCAST, 599 SOCK_TIMESTAMP, 600 SOCK_ZAPPED, 601 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 602 SOCK_DBG, /* %SO_DEBUG setting */ 603 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 604 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 605 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 606 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */ 607 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */ 608 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */ 609 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */ 610 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 611 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */ 612 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */ 613 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */ 614 SOCK_FASYNC, /* fasync() active */ 615 SOCK_RXQ_OVFL, 616 SOCK_ZEROCOPY, /* buffers from userspace */ 617 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 618 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 619 * Will use last 4 bytes of packet sent from 620 * user-space instead. 621 */ 622 }; 623 624 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 625 { 626 nsk->sk_flags = osk->sk_flags; 627 } 628 629 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 630 { 631 __set_bit(flag, &sk->sk_flags); 632 } 633 634 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 635 { 636 __clear_bit(flag, &sk->sk_flags); 637 } 638 639 static inline int sock_flag(struct sock *sk, enum sock_flags flag) 640 { 641 return test_bit(flag, &sk->sk_flags); 642 } 643 644 static inline void sk_acceptq_removed(struct sock *sk) 645 { 646 sk->sk_ack_backlog--; 647 } 648 649 static inline void sk_acceptq_added(struct sock *sk) 650 { 651 sk->sk_ack_backlog++; 652 } 653 654 static inline int sk_acceptq_is_full(struct sock *sk) 655 { 656 return sk->sk_ack_backlog > sk->sk_max_ack_backlog; 657 } 658 659 /* 660 * Compute minimal free write space needed to queue new packets. 661 */ 662 static inline int sk_stream_min_wspace(struct sock *sk) 663 { 664 return sk->sk_wmem_queued >> 1; 665 } 666 667 static inline int sk_stream_wspace(struct sock *sk) 668 { 669 return sk->sk_sndbuf - sk->sk_wmem_queued; 670 } 671 672 extern void sk_stream_write_space(struct sock *sk); 673 674 static inline int sk_stream_memory_free(struct sock *sk) 675 { 676 return sk->sk_wmem_queued < sk->sk_sndbuf; 677 } 678 679 /* OOB backlog add */ 680 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 681 { 682 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 683 skb_dst_force(skb); 684 685 if (!sk->sk_backlog.tail) 686 sk->sk_backlog.head = skb; 687 else 688 sk->sk_backlog.tail->next = skb; 689 690 sk->sk_backlog.tail = skb; 691 skb->next = NULL; 692 } 693 694 /* 695 * Take into account size of receive queue and backlog queue 696 * Do not take into account this skb truesize, 697 * to allow even a single big packet to come. 698 */ 699 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb) 700 { 701 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 702 703 return qsize > sk->sk_rcvbuf; 704 } 705 706 /* The per-socket spinlock must be held here. */ 707 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb) 708 { 709 if (sk_rcvqueues_full(sk, skb)) 710 return -ENOBUFS; 711 712 __sk_add_backlog(sk, skb); 713 sk->sk_backlog.len += skb->truesize; 714 return 0; 715 } 716 717 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 718 { 719 return sk->sk_backlog_rcv(sk, skb); 720 } 721 722 static inline void sock_rps_record_flow(const struct sock *sk) 723 { 724 #ifdef CONFIG_RPS 725 struct rps_sock_flow_table *sock_flow_table; 726 727 rcu_read_lock(); 728 sock_flow_table = rcu_dereference(rps_sock_flow_table); 729 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash); 730 rcu_read_unlock(); 731 #endif 732 } 733 734 static inline void sock_rps_reset_flow(const struct sock *sk) 735 { 736 #ifdef CONFIG_RPS 737 struct rps_sock_flow_table *sock_flow_table; 738 739 rcu_read_lock(); 740 sock_flow_table = rcu_dereference(rps_sock_flow_table); 741 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash); 742 rcu_read_unlock(); 743 #endif 744 } 745 746 static inline void sock_rps_save_rxhash(struct sock *sk, 747 const struct sk_buff *skb) 748 { 749 #ifdef CONFIG_RPS 750 if (unlikely(sk->sk_rxhash != skb->rxhash)) { 751 sock_rps_reset_flow(sk); 752 sk->sk_rxhash = skb->rxhash; 753 } 754 #endif 755 } 756 757 static inline void sock_rps_reset_rxhash(struct sock *sk) 758 { 759 #ifdef CONFIG_RPS 760 sock_rps_reset_flow(sk); 761 sk->sk_rxhash = 0; 762 #endif 763 } 764 765 #define sk_wait_event(__sk, __timeo, __condition) \ 766 ({ int __rc; \ 767 release_sock(__sk); \ 768 __rc = __condition; \ 769 if (!__rc) { \ 770 *(__timeo) = schedule_timeout(*(__timeo)); \ 771 } \ 772 lock_sock(__sk); \ 773 __rc = __condition; \ 774 __rc; \ 775 }) 776 777 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 778 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 779 extern void sk_stream_wait_close(struct sock *sk, long timeo_p); 780 extern int sk_stream_error(struct sock *sk, int flags, int err); 781 extern void sk_stream_kill_queues(struct sock *sk); 782 783 extern int sk_wait_data(struct sock *sk, long *timeo); 784 785 struct request_sock_ops; 786 struct timewait_sock_ops; 787 struct inet_hashinfo; 788 struct raw_hashinfo; 789 struct module; 790 791 /* Networking protocol blocks we attach to sockets. 792 * socket layer -> transport layer interface 793 * transport -> network interface is defined by struct inet_proto 794 */ 795 struct proto { 796 void (*close)(struct sock *sk, 797 long timeout); 798 int (*connect)(struct sock *sk, 799 struct sockaddr *uaddr, 800 int addr_len); 801 int (*disconnect)(struct sock *sk, int flags); 802 803 struct sock * (*accept) (struct sock *sk, int flags, int *err); 804 805 int (*ioctl)(struct sock *sk, int cmd, 806 unsigned long arg); 807 int (*init)(struct sock *sk); 808 void (*destroy)(struct sock *sk); 809 void (*shutdown)(struct sock *sk, int how); 810 int (*setsockopt)(struct sock *sk, int level, 811 int optname, char __user *optval, 812 unsigned int optlen); 813 int (*getsockopt)(struct sock *sk, int level, 814 int optname, char __user *optval, 815 int __user *option); 816 #ifdef CONFIG_COMPAT 817 int (*compat_setsockopt)(struct sock *sk, 818 int level, 819 int optname, char __user *optval, 820 unsigned int optlen); 821 int (*compat_getsockopt)(struct sock *sk, 822 int level, 823 int optname, char __user *optval, 824 int __user *option); 825 int (*compat_ioctl)(struct sock *sk, 826 unsigned int cmd, unsigned long arg); 827 #endif 828 int (*sendmsg)(struct kiocb *iocb, struct sock *sk, 829 struct msghdr *msg, size_t len); 830 int (*recvmsg)(struct kiocb *iocb, struct sock *sk, 831 struct msghdr *msg, 832 size_t len, int noblock, int flags, 833 int *addr_len); 834 int (*sendpage)(struct sock *sk, struct page *page, 835 int offset, size_t size, int flags); 836 int (*bind)(struct sock *sk, 837 struct sockaddr *uaddr, int addr_len); 838 839 int (*backlog_rcv) (struct sock *sk, 840 struct sk_buff *skb); 841 842 /* Keeping track of sk's, looking them up, and port selection methods. */ 843 void (*hash)(struct sock *sk); 844 void (*unhash)(struct sock *sk); 845 void (*rehash)(struct sock *sk); 846 int (*get_port)(struct sock *sk, unsigned short snum); 847 void (*clear_sk)(struct sock *sk, int size); 848 849 /* Keeping track of sockets in use */ 850 #ifdef CONFIG_PROC_FS 851 unsigned int inuse_idx; 852 #endif 853 854 /* Memory pressure */ 855 void (*enter_memory_pressure)(struct sock *sk); 856 atomic_long_t *memory_allocated; /* Current allocated memory. */ 857 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 858 /* 859 * Pressure flag: try to collapse. 860 * Technical note: it is used by multiple contexts non atomically. 861 * All the __sk_mem_schedule() is of this nature: accounting 862 * is strict, actions are advisory and have some latency. 863 */ 864 int *memory_pressure; 865 long *sysctl_mem; 866 int *sysctl_wmem; 867 int *sysctl_rmem; 868 int max_header; 869 bool no_autobind; 870 871 struct kmem_cache *slab; 872 unsigned int obj_size; 873 int slab_flags; 874 875 struct percpu_counter *orphan_count; 876 877 struct request_sock_ops *rsk_prot; 878 struct timewait_sock_ops *twsk_prot; 879 880 union { 881 struct inet_hashinfo *hashinfo; 882 struct udp_table *udp_table; 883 struct raw_hashinfo *raw_hash; 884 } h; 885 886 struct module *owner; 887 888 char name[32]; 889 890 struct list_head node; 891 #ifdef SOCK_REFCNT_DEBUG 892 atomic_t socks; 893 #endif 894 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM 895 /* 896 * cgroup specific init/deinit functions. Called once for all 897 * protocols that implement it, from cgroups populate function. 898 * This function has to setup any files the protocol want to 899 * appear in the kmem cgroup filesystem. 900 */ 901 int (*init_cgroup)(struct cgroup *cgrp, 902 struct cgroup_subsys *ss); 903 void (*destroy_cgroup)(struct cgroup *cgrp, 904 struct cgroup_subsys *ss); 905 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg); 906 #endif 907 }; 908 909 struct cg_proto { 910 void (*enter_memory_pressure)(struct sock *sk); 911 struct res_counter *memory_allocated; /* Current allocated memory. */ 912 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 913 int *memory_pressure; 914 long *sysctl_mem; 915 /* 916 * memcg field is used to find which memcg we belong directly 917 * Each memcg struct can hold more than one cg_proto, so container_of 918 * won't really cut. 919 * 920 * The elegant solution would be having an inverse function to 921 * proto_cgroup in struct proto, but that means polluting the structure 922 * for everybody, instead of just for memcg users. 923 */ 924 struct mem_cgroup *memcg; 925 }; 926 927 extern int proto_register(struct proto *prot, int alloc_slab); 928 extern void proto_unregister(struct proto *prot); 929 930 #ifdef SOCK_REFCNT_DEBUG 931 static inline void sk_refcnt_debug_inc(struct sock *sk) 932 { 933 atomic_inc(&sk->sk_prot->socks); 934 } 935 936 static inline void sk_refcnt_debug_dec(struct sock *sk) 937 { 938 atomic_dec(&sk->sk_prot->socks); 939 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 940 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 941 } 942 943 inline void sk_refcnt_debug_release(const struct sock *sk) 944 { 945 if (atomic_read(&sk->sk_refcnt) != 1) 946 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 947 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt)); 948 } 949 #else /* SOCK_REFCNT_DEBUG */ 950 #define sk_refcnt_debug_inc(sk) do { } while (0) 951 #define sk_refcnt_debug_dec(sk) do { } while (0) 952 #define sk_refcnt_debug_release(sk) do { } while (0) 953 #endif /* SOCK_REFCNT_DEBUG */ 954 955 #if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM) && defined(CONFIG_NET) 956 extern struct jump_label_key memcg_socket_limit_enabled; 957 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 958 struct cg_proto *cg_proto) 959 { 960 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg)); 961 } 962 #define mem_cgroup_sockets_enabled static_branch(&memcg_socket_limit_enabled) 963 #else 964 #define mem_cgroup_sockets_enabled 0 965 static inline struct cg_proto *parent_cg_proto(struct proto *proto, 966 struct cg_proto *cg_proto) 967 { 968 return NULL; 969 } 970 #endif 971 972 973 static inline bool sk_has_memory_pressure(const struct sock *sk) 974 { 975 return sk->sk_prot->memory_pressure != NULL; 976 } 977 978 static inline bool sk_under_memory_pressure(const struct sock *sk) 979 { 980 if (!sk->sk_prot->memory_pressure) 981 return false; 982 983 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 984 return !!*sk->sk_cgrp->memory_pressure; 985 986 return !!*sk->sk_prot->memory_pressure; 987 } 988 989 static inline void sk_leave_memory_pressure(struct sock *sk) 990 { 991 int *memory_pressure = sk->sk_prot->memory_pressure; 992 993 if (!memory_pressure) 994 return; 995 996 if (*memory_pressure) 997 *memory_pressure = 0; 998 999 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1000 struct cg_proto *cg_proto = sk->sk_cgrp; 1001 struct proto *prot = sk->sk_prot; 1002 1003 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1004 if (*cg_proto->memory_pressure) 1005 *cg_proto->memory_pressure = 0; 1006 } 1007 1008 } 1009 1010 static inline void sk_enter_memory_pressure(struct sock *sk) 1011 { 1012 if (!sk->sk_prot->enter_memory_pressure) 1013 return; 1014 1015 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1016 struct cg_proto *cg_proto = sk->sk_cgrp; 1017 struct proto *prot = sk->sk_prot; 1018 1019 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1020 cg_proto->enter_memory_pressure(sk); 1021 } 1022 1023 sk->sk_prot->enter_memory_pressure(sk); 1024 } 1025 1026 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1027 { 1028 long *prot = sk->sk_prot->sysctl_mem; 1029 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1030 prot = sk->sk_cgrp->sysctl_mem; 1031 return prot[index]; 1032 } 1033 1034 static inline void memcg_memory_allocated_add(struct cg_proto *prot, 1035 unsigned long amt, 1036 int *parent_status) 1037 { 1038 struct res_counter *fail; 1039 int ret; 1040 1041 ret = res_counter_charge_nofail(prot->memory_allocated, 1042 amt << PAGE_SHIFT, &fail); 1043 if (ret < 0) 1044 *parent_status = OVER_LIMIT; 1045 } 1046 1047 static inline void memcg_memory_allocated_sub(struct cg_proto *prot, 1048 unsigned long amt) 1049 { 1050 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT); 1051 } 1052 1053 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot) 1054 { 1055 u64 ret; 1056 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE); 1057 return ret >> PAGE_SHIFT; 1058 } 1059 1060 static inline long 1061 sk_memory_allocated(const struct sock *sk) 1062 { 1063 struct proto *prot = sk->sk_prot; 1064 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1065 return memcg_memory_allocated_read(sk->sk_cgrp); 1066 1067 return atomic_long_read(prot->memory_allocated); 1068 } 1069 1070 static inline long 1071 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status) 1072 { 1073 struct proto *prot = sk->sk_prot; 1074 1075 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1076 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status); 1077 /* update the root cgroup regardless */ 1078 atomic_long_add_return(amt, prot->memory_allocated); 1079 return memcg_memory_allocated_read(sk->sk_cgrp); 1080 } 1081 1082 return atomic_long_add_return(amt, prot->memory_allocated); 1083 } 1084 1085 static inline void 1086 sk_memory_allocated_sub(struct sock *sk, int amt) 1087 { 1088 struct proto *prot = sk->sk_prot; 1089 1090 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1091 memcg_memory_allocated_sub(sk->sk_cgrp, amt); 1092 1093 atomic_long_sub(amt, prot->memory_allocated); 1094 } 1095 1096 static inline void sk_sockets_allocated_dec(struct sock *sk) 1097 { 1098 struct proto *prot = sk->sk_prot; 1099 1100 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1101 struct cg_proto *cg_proto = sk->sk_cgrp; 1102 1103 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1104 percpu_counter_dec(cg_proto->sockets_allocated); 1105 } 1106 1107 percpu_counter_dec(prot->sockets_allocated); 1108 } 1109 1110 static inline void sk_sockets_allocated_inc(struct sock *sk) 1111 { 1112 struct proto *prot = sk->sk_prot; 1113 1114 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { 1115 struct cg_proto *cg_proto = sk->sk_cgrp; 1116 1117 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto)) 1118 percpu_counter_inc(cg_proto->sockets_allocated); 1119 } 1120 1121 percpu_counter_inc(prot->sockets_allocated); 1122 } 1123 1124 static inline int 1125 sk_sockets_allocated_read_positive(struct sock *sk) 1126 { 1127 struct proto *prot = sk->sk_prot; 1128 1129 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) 1130 return percpu_counter_sum_positive(sk->sk_cgrp->sockets_allocated); 1131 1132 return percpu_counter_sum_positive(prot->sockets_allocated); 1133 } 1134 1135 static inline int 1136 proto_sockets_allocated_sum_positive(struct proto *prot) 1137 { 1138 return percpu_counter_sum_positive(prot->sockets_allocated); 1139 } 1140 1141 static inline long 1142 proto_memory_allocated(struct proto *prot) 1143 { 1144 return atomic_long_read(prot->memory_allocated); 1145 } 1146 1147 static inline bool 1148 proto_memory_pressure(struct proto *prot) 1149 { 1150 if (!prot->memory_pressure) 1151 return false; 1152 return !!*prot->memory_pressure; 1153 } 1154 1155 1156 #ifdef CONFIG_PROC_FS 1157 /* Called with local bh disabled */ 1158 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1159 extern int sock_prot_inuse_get(struct net *net, struct proto *proto); 1160 #else 1161 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot, 1162 int inc) 1163 { 1164 } 1165 #endif 1166 1167 1168 /* With per-bucket locks this operation is not-atomic, so that 1169 * this version is not worse. 1170 */ 1171 static inline void __sk_prot_rehash(struct sock *sk) 1172 { 1173 sk->sk_prot->unhash(sk); 1174 sk->sk_prot->hash(sk); 1175 } 1176 1177 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size); 1178 1179 /* About 10 seconds */ 1180 #define SOCK_DESTROY_TIME (10*HZ) 1181 1182 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1183 #define PROT_SOCK 1024 1184 1185 #define SHUTDOWN_MASK 3 1186 #define RCV_SHUTDOWN 1 1187 #define SEND_SHUTDOWN 2 1188 1189 #define SOCK_SNDBUF_LOCK 1 1190 #define SOCK_RCVBUF_LOCK 2 1191 #define SOCK_BINDADDR_LOCK 4 1192 #define SOCK_BINDPORT_LOCK 8 1193 1194 /* sock_iocb: used to kick off async processing of socket ios */ 1195 struct sock_iocb { 1196 struct list_head list; 1197 1198 int flags; 1199 int size; 1200 struct socket *sock; 1201 struct sock *sk; 1202 struct scm_cookie *scm; 1203 struct msghdr *msg, async_msg; 1204 struct kiocb *kiocb; 1205 }; 1206 1207 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb) 1208 { 1209 return (struct sock_iocb *)iocb->private; 1210 } 1211 1212 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si) 1213 { 1214 return si->kiocb; 1215 } 1216 1217 struct socket_alloc { 1218 struct socket socket; 1219 struct inode vfs_inode; 1220 }; 1221 1222 static inline struct socket *SOCKET_I(struct inode *inode) 1223 { 1224 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1225 } 1226 1227 static inline struct inode *SOCK_INODE(struct socket *socket) 1228 { 1229 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1230 } 1231 1232 /* 1233 * Functions for memory accounting 1234 */ 1235 extern int __sk_mem_schedule(struct sock *sk, int size, int kind); 1236 extern void __sk_mem_reclaim(struct sock *sk); 1237 1238 #define SK_MEM_QUANTUM ((int)PAGE_SIZE) 1239 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1240 #define SK_MEM_SEND 0 1241 #define SK_MEM_RECV 1 1242 1243 static inline int sk_mem_pages(int amt) 1244 { 1245 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1246 } 1247 1248 static inline int sk_has_account(struct sock *sk) 1249 { 1250 /* return true if protocol supports memory accounting */ 1251 return !!sk->sk_prot->memory_allocated; 1252 } 1253 1254 static inline int sk_wmem_schedule(struct sock *sk, int size) 1255 { 1256 if (!sk_has_account(sk)) 1257 return 1; 1258 return size <= sk->sk_forward_alloc || 1259 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1260 } 1261 1262 static inline int sk_rmem_schedule(struct sock *sk, int size) 1263 { 1264 if (!sk_has_account(sk)) 1265 return 1; 1266 return size <= sk->sk_forward_alloc || 1267 __sk_mem_schedule(sk, size, SK_MEM_RECV); 1268 } 1269 1270 static inline void sk_mem_reclaim(struct sock *sk) 1271 { 1272 if (!sk_has_account(sk)) 1273 return; 1274 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1275 __sk_mem_reclaim(sk); 1276 } 1277 1278 static inline void sk_mem_reclaim_partial(struct sock *sk) 1279 { 1280 if (!sk_has_account(sk)) 1281 return; 1282 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1283 __sk_mem_reclaim(sk); 1284 } 1285 1286 static inline void sk_mem_charge(struct sock *sk, int size) 1287 { 1288 if (!sk_has_account(sk)) 1289 return; 1290 sk->sk_forward_alloc -= size; 1291 } 1292 1293 static inline void sk_mem_uncharge(struct sock *sk, int size) 1294 { 1295 if (!sk_has_account(sk)) 1296 return; 1297 sk->sk_forward_alloc += size; 1298 } 1299 1300 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1301 { 1302 sock_set_flag(sk, SOCK_QUEUE_SHRUNK); 1303 sk->sk_wmem_queued -= skb->truesize; 1304 sk_mem_uncharge(sk, skb->truesize); 1305 __kfree_skb(skb); 1306 } 1307 1308 /* Used by processes to "lock" a socket state, so that 1309 * interrupts and bottom half handlers won't change it 1310 * from under us. It essentially blocks any incoming 1311 * packets, so that we won't get any new data or any 1312 * packets that change the state of the socket. 1313 * 1314 * While locked, BH processing will add new packets to 1315 * the backlog queue. This queue is processed by the 1316 * owner of the socket lock right before it is released. 1317 * 1318 * Since ~2.3.5 it is also exclusive sleep lock serializing 1319 * accesses from user process context. 1320 */ 1321 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned) 1322 1323 /* 1324 * Macro so as to not evaluate some arguments when 1325 * lockdep is not enabled. 1326 * 1327 * Mark both the sk_lock and the sk_lock.slock as a 1328 * per-address-family lock class. 1329 */ 1330 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1331 do { \ 1332 sk->sk_lock.owned = 0; \ 1333 init_waitqueue_head(&sk->sk_lock.wq); \ 1334 spin_lock_init(&(sk)->sk_lock.slock); \ 1335 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1336 sizeof((sk)->sk_lock)); \ 1337 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1338 (skey), (sname)); \ 1339 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1340 } while (0) 1341 1342 extern void lock_sock_nested(struct sock *sk, int subclass); 1343 1344 static inline void lock_sock(struct sock *sk) 1345 { 1346 lock_sock_nested(sk, 0); 1347 } 1348 1349 extern void release_sock(struct sock *sk); 1350 1351 /* BH context may only use the following locking interface. */ 1352 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1353 #define bh_lock_sock_nested(__sk) \ 1354 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1355 SINGLE_DEPTH_NESTING) 1356 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1357 1358 extern bool lock_sock_fast(struct sock *sk); 1359 /** 1360 * unlock_sock_fast - complement of lock_sock_fast 1361 * @sk: socket 1362 * @slow: slow mode 1363 * 1364 * fast unlock socket for user context. 1365 * If slow mode is on, we call regular release_sock() 1366 */ 1367 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1368 { 1369 if (slow) 1370 release_sock(sk); 1371 else 1372 spin_unlock_bh(&sk->sk_lock.slock); 1373 } 1374 1375 1376 extern struct sock *sk_alloc(struct net *net, int family, 1377 gfp_t priority, 1378 struct proto *prot); 1379 extern void sk_free(struct sock *sk); 1380 extern void sk_release_kernel(struct sock *sk); 1381 extern struct sock *sk_clone_lock(const struct sock *sk, 1382 const gfp_t priority); 1383 1384 extern struct sk_buff *sock_wmalloc(struct sock *sk, 1385 unsigned long size, int force, 1386 gfp_t priority); 1387 extern struct sk_buff *sock_rmalloc(struct sock *sk, 1388 unsigned long size, int force, 1389 gfp_t priority); 1390 extern void sock_wfree(struct sk_buff *skb); 1391 extern void sock_rfree(struct sk_buff *skb); 1392 1393 extern int sock_setsockopt(struct socket *sock, int level, 1394 int op, char __user *optval, 1395 unsigned int optlen); 1396 1397 extern int sock_getsockopt(struct socket *sock, int level, 1398 int op, char __user *optval, 1399 int __user *optlen); 1400 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1401 unsigned long size, 1402 int noblock, 1403 int *errcode); 1404 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk, 1405 unsigned long header_len, 1406 unsigned long data_len, 1407 int noblock, 1408 int *errcode); 1409 extern void *sock_kmalloc(struct sock *sk, int size, 1410 gfp_t priority); 1411 extern void sock_kfree_s(struct sock *sk, void *mem, int size); 1412 extern void sk_send_sigurg(struct sock *sk); 1413 1414 #ifdef CONFIG_CGROUPS 1415 extern void sock_update_classid(struct sock *sk); 1416 #else 1417 static inline void sock_update_classid(struct sock *sk) 1418 { 1419 } 1420 #endif 1421 1422 /* 1423 * Functions to fill in entries in struct proto_ops when a protocol 1424 * does not implement a particular function. 1425 */ 1426 extern int sock_no_bind(struct socket *, 1427 struct sockaddr *, int); 1428 extern int sock_no_connect(struct socket *, 1429 struct sockaddr *, int, int); 1430 extern int sock_no_socketpair(struct socket *, 1431 struct socket *); 1432 extern int sock_no_accept(struct socket *, 1433 struct socket *, int); 1434 extern int sock_no_getname(struct socket *, 1435 struct sockaddr *, int *, int); 1436 extern unsigned int sock_no_poll(struct file *, struct socket *, 1437 struct poll_table_struct *); 1438 extern int sock_no_ioctl(struct socket *, unsigned int, 1439 unsigned long); 1440 extern int sock_no_listen(struct socket *, int); 1441 extern int sock_no_shutdown(struct socket *, int); 1442 extern int sock_no_getsockopt(struct socket *, int , int, 1443 char __user *, int __user *); 1444 extern int sock_no_setsockopt(struct socket *, int, int, 1445 char __user *, unsigned int); 1446 extern int sock_no_sendmsg(struct kiocb *, struct socket *, 1447 struct msghdr *, size_t); 1448 extern int sock_no_recvmsg(struct kiocb *, struct socket *, 1449 struct msghdr *, size_t, int); 1450 extern int sock_no_mmap(struct file *file, 1451 struct socket *sock, 1452 struct vm_area_struct *vma); 1453 extern ssize_t sock_no_sendpage(struct socket *sock, 1454 struct page *page, 1455 int offset, size_t size, 1456 int flags); 1457 1458 /* 1459 * Functions to fill in entries in struct proto_ops when a protocol 1460 * uses the inet style. 1461 */ 1462 extern int sock_common_getsockopt(struct socket *sock, int level, int optname, 1463 char __user *optval, int __user *optlen); 1464 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock, 1465 struct msghdr *msg, size_t size, int flags); 1466 extern int sock_common_setsockopt(struct socket *sock, int level, int optname, 1467 char __user *optval, unsigned int optlen); 1468 extern int compat_sock_common_getsockopt(struct socket *sock, int level, 1469 int optname, char __user *optval, int __user *optlen); 1470 extern int compat_sock_common_setsockopt(struct socket *sock, int level, 1471 int optname, char __user *optval, unsigned int optlen); 1472 1473 extern void sk_common_release(struct sock *sk); 1474 1475 /* 1476 * Default socket callbacks and setup code 1477 */ 1478 1479 /* Initialise core socket variables */ 1480 extern void sock_init_data(struct socket *sock, struct sock *sk); 1481 1482 extern void sk_filter_release_rcu(struct rcu_head *rcu); 1483 1484 /** 1485 * sk_filter_release - release a socket filter 1486 * @fp: filter to remove 1487 * 1488 * Remove a filter from a socket and release its resources. 1489 */ 1490 1491 static inline void sk_filter_release(struct sk_filter *fp) 1492 { 1493 if (atomic_dec_and_test(&fp->refcnt)) 1494 call_rcu(&fp->rcu, sk_filter_release_rcu); 1495 } 1496 1497 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1498 { 1499 unsigned int size = sk_filter_len(fp); 1500 1501 atomic_sub(size, &sk->sk_omem_alloc); 1502 sk_filter_release(fp); 1503 } 1504 1505 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1506 { 1507 atomic_inc(&fp->refcnt); 1508 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc); 1509 } 1510 1511 /* 1512 * Socket reference counting postulates. 1513 * 1514 * * Each user of socket SHOULD hold a reference count. 1515 * * Each access point to socket (an hash table bucket, reference from a list, 1516 * running timer, skb in flight MUST hold a reference count. 1517 * * When reference count hits 0, it means it will never increase back. 1518 * * When reference count hits 0, it means that no references from 1519 * outside exist to this socket and current process on current CPU 1520 * is last user and may/should destroy this socket. 1521 * * sk_free is called from any context: process, BH, IRQ. When 1522 * it is called, socket has no references from outside -> sk_free 1523 * may release descendant resources allocated by the socket, but 1524 * to the time when it is called, socket is NOT referenced by any 1525 * hash tables, lists etc. 1526 * * Packets, delivered from outside (from network or from another process) 1527 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1528 * when they sit in queue. Otherwise, packets will leak to hole, when 1529 * socket is looked up by one cpu and unhasing is made by another CPU. 1530 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1531 * (leak to backlog). Packet socket does all the processing inside 1532 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1533 * use separate SMP lock, so that they are prone too. 1534 */ 1535 1536 /* Ungrab socket and destroy it, if it was the last reference. */ 1537 static inline void sock_put(struct sock *sk) 1538 { 1539 if (atomic_dec_and_test(&sk->sk_refcnt)) 1540 sk_free(sk); 1541 } 1542 1543 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1544 const int nested); 1545 1546 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1547 { 1548 sk->sk_tx_queue_mapping = tx_queue; 1549 } 1550 1551 static inline void sk_tx_queue_clear(struct sock *sk) 1552 { 1553 sk->sk_tx_queue_mapping = -1; 1554 } 1555 1556 static inline int sk_tx_queue_get(const struct sock *sk) 1557 { 1558 return sk ? sk->sk_tx_queue_mapping : -1; 1559 } 1560 1561 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1562 { 1563 sk_tx_queue_clear(sk); 1564 sk->sk_socket = sock; 1565 } 1566 1567 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1568 { 1569 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1570 return &rcu_dereference_raw(sk->sk_wq)->wait; 1571 } 1572 /* Detach socket from process context. 1573 * Announce socket dead, detach it from wait queue and inode. 1574 * Note that parent inode held reference count on this struct sock, 1575 * we do not release it in this function, because protocol 1576 * probably wants some additional cleanups or even continuing 1577 * to work with this socket (TCP). 1578 */ 1579 static inline void sock_orphan(struct sock *sk) 1580 { 1581 write_lock_bh(&sk->sk_callback_lock); 1582 sock_set_flag(sk, SOCK_DEAD); 1583 sk_set_socket(sk, NULL); 1584 sk->sk_wq = NULL; 1585 write_unlock_bh(&sk->sk_callback_lock); 1586 } 1587 1588 static inline void sock_graft(struct sock *sk, struct socket *parent) 1589 { 1590 write_lock_bh(&sk->sk_callback_lock); 1591 sk->sk_wq = parent->wq; 1592 parent->sk = sk; 1593 sk_set_socket(sk, parent); 1594 security_sock_graft(sk, parent); 1595 write_unlock_bh(&sk->sk_callback_lock); 1596 } 1597 1598 extern int sock_i_uid(struct sock *sk); 1599 extern unsigned long sock_i_ino(struct sock *sk); 1600 1601 static inline struct dst_entry * 1602 __sk_dst_get(struct sock *sk) 1603 { 1604 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) || 1605 lockdep_is_held(&sk->sk_lock.slock)); 1606 } 1607 1608 static inline struct dst_entry * 1609 sk_dst_get(struct sock *sk) 1610 { 1611 struct dst_entry *dst; 1612 1613 rcu_read_lock(); 1614 dst = rcu_dereference(sk->sk_dst_cache); 1615 if (dst) 1616 dst_hold(dst); 1617 rcu_read_unlock(); 1618 return dst; 1619 } 1620 1621 extern void sk_reset_txq(struct sock *sk); 1622 1623 static inline void dst_negative_advice(struct sock *sk) 1624 { 1625 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 1626 1627 if (dst && dst->ops->negative_advice) { 1628 ndst = dst->ops->negative_advice(dst); 1629 1630 if (ndst != dst) { 1631 rcu_assign_pointer(sk->sk_dst_cache, ndst); 1632 sk_reset_txq(sk); 1633 } 1634 } 1635 } 1636 1637 static inline void 1638 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 1639 { 1640 struct dst_entry *old_dst; 1641 1642 sk_tx_queue_clear(sk); 1643 /* 1644 * This can be called while sk is owned by the caller only, 1645 * with no state that can be checked in a rcu_dereference_check() cond 1646 */ 1647 old_dst = rcu_dereference_raw(sk->sk_dst_cache); 1648 rcu_assign_pointer(sk->sk_dst_cache, dst); 1649 dst_release(old_dst); 1650 } 1651 1652 static inline void 1653 sk_dst_set(struct sock *sk, struct dst_entry *dst) 1654 { 1655 spin_lock(&sk->sk_dst_lock); 1656 __sk_dst_set(sk, dst); 1657 spin_unlock(&sk->sk_dst_lock); 1658 } 1659 1660 static inline void 1661 __sk_dst_reset(struct sock *sk) 1662 { 1663 __sk_dst_set(sk, NULL); 1664 } 1665 1666 static inline void 1667 sk_dst_reset(struct sock *sk) 1668 { 1669 spin_lock(&sk->sk_dst_lock); 1670 __sk_dst_reset(sk); 1671 spin_unlock(&sk->sk_dst_lock); 1672 } 1673 1674 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 1675 1676 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 1677 1678 static inline int sk_can_gso(const struct sock *sk) 1679 { 1680 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 1681 } 1682 1683 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 1684 1685 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 1686 { 1687 sk->sk_route_nocaps |= flags; 1688 sk->sk_route_caps &= ~flags; 1689 } 1690 1691 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 1692 char __user *from, char *to, 1693 int copy, int offset) 1694 { 1695 if (skb->ip_summed == CHECKSUM_NONE) { 1696 int err = 0; 1697 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err); 1698 if (err) 1699 return err; 1700 skb->csum = csum_block_add(skb->csum, csum, offset); 1701 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 1702 if (!access_ok(VERIFY_READ, from, copy) || 1703 __copy_from_user_nocache(to, from, copy)) 1704 return -EFAULT; 1705 } else if (copy_from_user(to, from, copy)) 1706 return -EFAULT; 1707 1708 return 0; 1709 } 1710 1711 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 1712 char __user *from, int copy) 1713 { 1714 int err, offset = skb->len; 1715 1716 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 1717 copy, offset); 1718 if (err) 1719 __skb_trim(skb, offset); 1720 1721 return err; 1722 } 1723 1724 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from, 1725 struct sk_buff *skb, 1726 struct page *page, 1727 int off, int copy) 1728 { 1729 int err; 1730 1731 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 1732 copy, skb->len); 1733 if (err) 1734 return err; 1735 1736 skb->len += copy; 1737 skb->data_len += copy; 1738 skb->truesize += copy; 1739 sk->sk_wmem_queued += copy; 1740 sk_mem_charge(sk, copy); 1741 return 0; 1742 } 1743 1744 static inline int skb_copy_to_page(struct sock *sk, char __user *from, 1745 struct sk_buff *skb, struct page *page, 1746 int off, int copy) 1747 { 1748 if (skb->ip_summed == CHECKSUM_NONE) { 1749 int err = 0; 1750 __wsum csum = csum_and_copy_from_user(from, 1751 page_address(page) + off, 1752 copy, 0, &err); 1753 if (err) 1754 return err; 1755 skb->csum = csum_block_add(skb->csum, csum, skb->len); 1756 } else if (copy_from_user(page_address(page) + off, from, copy)) 1757 return -EFAULT; 1758 1759 skb->len += copy; 1760 skb->data_len += copy; 1761 skb->truesize += copy; 1762 sk->sk_wmem_queued += copy; 1763 sk_mem_charge(sk, copy); 1764 return 0; 1765 } 1766 1767 /** 1768 * sk_wmem_alloc_get - returns write allocations 1769 * @sk: socket 1770 * 1771 * Returns sk_wmem_alloc minus initial offset of one 1772 */ 1773 static inline int sk_wmem_alloc_get(const struct sock *sk) 1774 { 1775 return atomic_read(&sk->sk_wmem_alloc) - 1; 1776 } 1777 1778 /** 1779 * sk_rmem_alloc_get - returns read allocations 1780 * @sk: socket 1781 * 1782 * Returns sk_rmem_alloc 1783 */ 1784 static inline int sk_rmem_alloc_get(const struct sock *sk) 1785 { 1786 return atomic_read(&sk->sk_rmem_alloc); 1787 } 1788 1789 /** 1790 * sk_has_allocations - check if allocations are outstanding 1791 * @sk: socket 1792 * 1793 * Returns true if socket has write or read allocations 1794 */ 1795 static inline int sk_has_allocations(const struct sock *sk) 1796 { 1797 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 1798 } 1799 1800 /** 1801 * wq_has_sleeper - check if there are any waiting processes 1802 * @wq: struct socket_wq 1803 * 1804 * Returns true if socket_wq has waiting processes 1805 * 1806 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory 1807 * barrier call. They were added due to the race found within the tcp code. 1808 * 1809 * Consider following tcp code paths: 1810 * 1811 * CPU1 CPU2 1812 * 1813 * sys_select receive packet 1814 * ... ... 1815 * __add_wait_queue update tp->rcv_nxt 1816 * ... ... 1817 * tp->rcv_nxt check sock_def_readable 1818 * ... { 1819 * schedule rcu_read_lock(); 1820 * wq = rcu_dereference(sk->sk_wq); 1821 * if (wq && waitqueue_active(&wq->wait)) 1822 * wake_up_interruptible(&wq->wait) 1823 * ... 1824 * } 1825 * 1826 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 1827 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 1828 * could then endup calling schedule and sleep forever if there are no more 1829 * data on the socket. 1830 * 1831 */ 1832 static inline bool wq_has_sleeper(struct socket_wq *wq) 1833 { 1834 1835 /* 1836 * We need to be sure we are in sync with the 1837 * add_wait_queue modifications to the wait queue. 1838 * 1839 * This memory barrier is paired in the sock_poll_wait. 1840 */ 1841 smp_mb(); 1842 return wq && waitqueue_active(&wq->wait); 1843 } 1844 1845 /** 1846 * sock_poll_wait - place memory barrier behind the poll_wait call. 1847 * @filp: file 1848 * @wait_address: socket wait queue 1849 * @p: poll_table 1850 * 1851 * See the comments in the wq_has_sleeper function. 1852 */ 1853 static inline void sock_poll_wait(struct file *filp, 1854 wait_queue_head_t *wait_address, poll_table *p) 1855 { 1856 if (p && wait_address) { 1857 poll_wait(filp, wait_address, p); 1858 /* 1859 * We need to be sure we are in sync with the 1860 * socket flags modification. 1861 * 1862 * This memory barrier is paired in the wq_has_sleeper. 1863 */ 1864 smp_mb(); 1865 } 1866 } 1867 1868 /* 1869 * Queue a received datagram if it will fit. Stream and sequenced 1870 * protocols can't normally use this as they need to fit buffers in 1871 * and play with them. 1872 * 1873 * Inlined as it's very short and called for pretty much every 1874 * packet ever received. 1875 */ 1876 1877 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk) 1878 { 1879 skb_orphan(skb); 1880 skb->sk = sk; 1881 skb->destructor = sock_wfree; 1882 /* 1883 * We used to take a refcount on sk, but following operation 1884 * is enough to guarantee sk_free() wont free this sock until 1885 * all in-flight packets are completed 1886 */ 1887 atomic_add(skb->truesize, &sk->sk_wmem_alloc); 1888 } 1889 1890 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 1891 { 1892 skb_orphan(skb); 1893 skb->sk = sk; 1894 skb->destructor = sock_rfree; 1895 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 1896 sk_mem_charge(sk, skb->truesize); 1897 } 1898 1899 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer, 1900 unsigned long expires); 1901 1902 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer); 1903 1904 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 1905 1906 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 1907 1908 /* 1909 * Recover an error report and clear atomically 1910 */ 1911 1912 static inline int sock_error(struct sock *sk) 1913 { 1914 int err; 1915 if (likely(!sk->sk_err)) 1916 return 0; 1917 err = xchg(&sk->sk_err, 0); 1918 return -err; 1919 } 1920 1921 static inline unsigned long sock_wspace(struct sock *sk) 1922 { 1923 int amt = 0; 1924 1925 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 1926 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc); 1927 if (amt < 0) 1928 amt = 0; 1929 } 1930 return amt; 1931 } 1932 1933 static inline void sk_wake_async(struct sock *sk, int how, int band) 1934 { 1935 if (sock_flag(sk, SOCK_FASYNC)) 1936 sock_wake_async(sk->sk_socket, how, band); 1937 } 1938 1939 #define SOCK_MIN_SNDBUF 2048 1940 /* 1941 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need 1942 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak 1943 */ 1944 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff)) 1945 1946 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 1947 { 1948 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) { 1949 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 1950 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF); 1951 } 1952 } 1953 1954 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp); 1955 1956 static inline struct page *sk_stream_alloc_page(struct sock *sk) 1957 { 1958 struct page *page = NULL; 1959 1960 page = alloc_pages(sk->sk_allocation, 0); 1961 if (!page) { 1962 sk_enter_memory_pressure(sk); 1963 sk_stream_moderate_sndbuf(sk); 1964 } 1965 return page; 1966 } 1967 1968 /* 1969 * Default write policy as shown to user space via poll/select/SIGIO 1970 */ 1971 static inline int sock_writeable(const struct sock *sk) 1972 { 1973 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1); 1974 } 1975 1976 static inline gfp_t gfp_any(void) 1977 { 1978 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 1979 } 1980 1981 static inline long sock_rcvtimeo(const struct sock *sk, int noblock) 1982 { 1983 return noblock ? 0 : sk->sk_rcvtimeo; 1984 } 1985 1986 static inline long sock_sndtimeo(const struct sock *sk, int noblock) 1987 { 1988 return noblock ? 0 : sk->sk_sndtimeo; 1989 } 1990 1991 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 1992 { 1993 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1; 1994 } 1995 1996 /* Alas, with timeout socket operations are not restartable. 1997 * Compare this to poll(). 1998 */ 1999 static inline int sock_intr_errno(long timeo) 2000 { 2001 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2002 } 2003 2004 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2005 struct sk_buff *skb); 2006 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2007 struct sk_buff *skb); 2008 2009 static __inline__ void 2010 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2011 { 2012 ktime_t kt = skb->tstamp; 2013 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2014 2015 /* 2016 * generate control messages if 2017 * - receive time stamping in software requested (SOCK_RCVTSTAMP 2018 * or SOCK_TIMESTAMPING_RX_SOFTWARE) 2019 * - software time stamp available and wanted 2020 * (SOCK_TIMESTAMPING_SOFTWARE) 2021 * - hardware time stamps available and wanted 2022 * (SOCK_TIMESTAMPING_SYS_HARDWARE or 2023 * SOCK_TIMESTAMPING_RAW_HARDWARE) 2024 */ 2025 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2026 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) || 2027 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) || 2028 (hwtstamps->hwtstamp.tv64 && 2029 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) || 2030 (hwtstamps->syststamp.tv64 && 2031 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))) 2032 __sock_recv_timestamp(msg, sk, skb); 2033 else 2034 sk->sk_stamp = kt; 2035 2036 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2037 __sock_recv_wifi_status(msg, sk, skb); 2038 } 2039 2040 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2041 struct sk_buff *skb); 2042 2043 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2044 struct sk_buff *skb) 2045 { 2046 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2047 (1UL << SOCK_RCVTSTAMP) | \ 2048 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 2049 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \ 2050 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \ 2051 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE)) 2052 2053 if (sk->sk_flags & FLAGS_TS_OR_DROPS) 2054 __sock_recv_ts_and_drops(msg, sk, skb); 2055 else 2056 sk->sk_stamp = skb->tstamp; 2057 } 2058 2059 /** 2060 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2061 * @sk: socket sending this packet 2062 * @tx_flags: filled with instructions for time stamping 2063 * 2064 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if 2065 * parameters are invalid. 2066 */ 2067 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags); 2068 2069 /** 2070 * sk_eat_skb - Release a skb if it is no longer needed 2071 * @sk: socket to eat this skb from 2072 * @skb: socket buffer to eat 2073 * @copied_early: flag indicating whether DMA operations copied this data early 2074 * 2075 * This routine must be called with interrupts disabled or with the socket 2076 * locked so that the sk_buff queue operation is ok. 2077 */ 2078 #ifdef CONFIG_NET_DMA 2079 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 2080 { 2081 __skb_unlink(skb, &sk->sk_receive_queue); 2082 if (!copied_early) 2083 __kfree_skb(skb); 2084 else 2085 __skb_queue_tail(&sk->sk_async_wait_queue, skb); 2086 } 2087 #else 2088 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early) 2089 { 2090 __skb_unlink(skb, &sk->sk_receive_queue); 2091 __kfree_skb(skb); 2092 } 2093 #endif 2094 2095 static inline 2096 struct net *sock_net(const struct sock *sk) 2097 { 2098 return read_pnet(&sk->sk_net); 2099 } 2100 2101 static inline 2102 void sock_net_set(struct sock *sk, struct net *net) 2103 { 2104 write_pnet(&sk->sk_net, net); 2105 } 2106 2107 /* 2108 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace. 2109 * They should not hold a reference to a namespace in order to allow 2110 * to stop it. 2111 * Sockets after sk_change_net should be released using sk_release_kernel 2112 */ 2113 static inline void sk_change_net(struct sock *sk, struct net *net) 2114 { 2115 put_net(sock_net(sk)); 2116 sock_net_set(sk, hold_net(net)); 2117 } 2118 2119 static inline struct sock *skb_steal_sock(struct sk_buff *skb) 2120 { 2121 if (unlikely(skb->sk)) { 2122 struct sock *sk = skb->sk; 2123 2124 skb->destructor = NULL; 2125 skb->sk = NULL; 2126 return sk; 2127 } 2128 return NULL; 2129 } 2130 2131 extern void sock_enable_timestamp(struct sock *sk, int flag); 2132 extern int sock_get_timestamp(struct sock *, struct timeval __user *); 2133 extern int sock_get_timestampns(struct sock *, struct timespec __user *); 2134 2135 /* 2136 * Enable debug/info messages 2137 */ 2138 extern int net_msg_warn; 2139 #define NETDEBUG(fmt, args...) \ 2140 do { if (net_msg_warn) printk(fmt,##args); } while (0) 2141 2142 #define LIMIT_NETDEBUG(fmt, args...) \ 2143 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0) 2144 2145 extern __u32 sysctl_wmem_max; 2146 extern __u32 sysctl_rmem_max; 2147 2148 extern void sk_init(void); 2149 2150 extern int sysctl_optmem_max; 2151 2152 extern __u32 sysctl_wmem_default; 2153 extern __u32 sysctl_rmem_default; 2154 2155 #endif /* _SOCK_H */ 2156