1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/filter.h> 60 #include <linux/rculist_nulls.h> 61 #include <linux/poll.h> 62 #include <linux/sockptr.h> 63 #include <linux/indirect_call_wrapper.h> 64 #include <linux/atomic.h> 65 #include <linux/refcount.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 165 * address on 64bit arches : cf INET_MATCH() 166 */ 167 union { 168 __addrpair skc_addrpair; 169 struct { 170 __be32 skc_daddr; 171 __be32 skc_rcv_saddr; 172 }; 173 }; 174 union { 175 unsigned int skc_hash; 176 __u16 skc_u16hashes[2]; 177 }; 178 /* skc_dport && skc_num must be grouped as well */ 179 union { 180 __portpair skc_portpair; 181 struct { 182 __be16 skc_dport; 183 __u16 skc_num; 184 }; 185 }; 186 187 unsigned short skc_family; 188 volatile unsigned char skc_state; 189 unsigned char skc_reuse:4; 190 unsigned char skc_reuseport:1; 191 unsigned char skc_ipv6only:1; 192 unsigned char skc_net_refcnt:1; 193 int skc_bound_dev_if; 194 union { 195 struct hlist_node skc_bind_node; 196 struct hlist_node skc_portaddr_node; 197 }; 198 struct proto *skc_prot; 199 possible_net_t skc_net; 200 201 #if IS_ENABLED(CONFIG_IPV6) 202 struct in6_addr skc_v6_daddr; 203 struct in6_addr skc_v6_rcv_saddr; 204 #endif 205 206 atomic64_t skc_cookie; 207 208 /* following fields are padding to force 209 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 210 * assuming IPV6 is enabled. We use this padding differently 211 * for different kind of 'sockets' 212 */ 213 union { 214 unsigned long skc_flags; 215 struct sock *skc_listener; /* request_sock */ 216 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 217 }; 218 /* 219 * fields between dontcopy_begin/dontcopy_end 220 * are not copied in sock_copy() 221 */ 222 /* private: */ 223 int skc_dontcopy_begin[0]; 224 /* public: */ 225 union { 226 struct hlist_node skc_node; 227 struct hlist_nulls_node skc_nulls_node; 228 }; 229 unsigned short skc_tx_queue_mapping; 230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 231 unsigned short skc_rx_queue_mapping; 232 #endif 233 union { 234 int skc_incoming_cpu; 235 u32 skc_rcv_wnd; 236 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 237 }; 238 239 refcount_t skc_refcnt; 240 /* private: */ 241 int skc_dontcopy_end[0]; 242 union { 243 u32 skc_rxhash; 244 u32 skc_window_clamp; 245 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 246 }; 247 /* public: */ 248 }; 249 250 struct bpf_local_storage; 251 252 /** 253 * struct sock - network layer representation of sockets 254 * @__sk_common: shared layout with inet_timewait_sock 255 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 256 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 257 * @sk_lock: synchronizer 258 * @sk_kern_sock: True if sock is using kernel lock classes 259 * @sk_rcvbuf: size of receive buffer in bytes 260 * @sk_wq: sock wait queue and async head 261 * @sk_rx_dst: receive input route used by early demux 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 286 * @sk_route_forced_caps: static, forced route capabilities 287 * (set in tcp_init_sock()) 288 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 289 * @sk_gso_max_size: Maximum GSO segment size to build 290 * @sk_gso_max_segs: Maximum number of GSO segments 291 * @sk_pacing_shift: scaling factor for TCP Small Queues 292 * @sk_lingertime: %SO_LINGER l_linger setting 293 * @sk_backlog: always used with the per-socket spinlock held 294 * @sk_callback_lock: used with the callbacks in the end of this struct 295 * @sk_error_queue: rarely used 296 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 297 * IPV6_ADDRFORM for instance) 298 * @sk_err: last error 299 * @sk_err_soft: errors that don't cause failure but are the cause of a 300 * persistent failure not just 'timed out' 301 * @sk_drops: raw/udp drops counter 302 * @sk_ack_backlog: current listen backlog 303 * @sk_max_ack_backlog: listen backlog set in listen() 304 * @sk_uid: user id of owner 305 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 306 * @sk_busy_poll_budget: napi processing budget when busypolling 307 * @sk_priority: %SO_PRIORITY setting 308 * @sk_type: socket type (%SOCK_STREAM, etc) 309 * @sk_protocol: which protocol this socket belongs in this network family 310 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 311 * @sk_peer_pid: &struct pid for this socket's peer 312 * @sk_peer_cred: %SO_PEERCRED setting 313 * @sk_rcvlowat: %SO_RCVLOWAT setting 314 * @sk_rcvtimeo: %SO_RCVTIMEO setting 315 * @sk_sndtimeo: %SO_SNDTIMEO setting 316 * @sk_txhash: computed flow hash for use on transmit 317 * @sk_filter: socket filtering instructions 318 * @sk_timer: sock cleanup timer 319 * @sk_stamp: time stamp of last packet received 320 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 321 * @sk_tsflags: SO_TIMESTAMPING flags 322 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 323 * for timestamping 324 * @sk_tskey: counter to disambiguate concurrent tstamp requests 325 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 326 * @sk_socket: Identd and reporting IO signals 327 * @sk_user_data: RPC layer private data 328 * @sk_frag: cached page frag 329 * @sk_peek_off: current peek_offset value 330 * @sk_send_head: front of stuff to transmit 331 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 332 * @sk_security: used by security modules 333 * @sk_mark: generic packet mark 334 * @sk_cgrp_data: cgroup data for this cgroup 335 * @sk_memcg: this socket's memory cgroup association 336 * @sk_write_pending: a write to stream socket waits to start 337 * @sk_state_change: callback to indicate change in the state of the sock 338 * @sk_data_ready: callback to indicate there is data to be processed 339 * @sk_write_space: callback to indicate there is bf sending space available 340 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 341 * @sk_backlog_rcv: callback to process the backlog 342 * @sk_validate_xmit_skb: ptr to an optional validate function 343 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 344 * @sk_reuseport_cb: reuseport group container 345 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 346 * @sk_rcu: used during RCU grace period 347 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 348 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 349 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 350 * @sk_txtime_unused: unused txtime flags 351 */ 352 struct sock { 353 /* 354 * Now struct inet_timewait_sock also uses sock_common, so please just 355 * don't add nothing before this first member (__sk_common) --acme 356 */ 357 struct sock_common __sk_common; 358 #define sk_node __sk_common.skc_node 359 #define sk_nulls_node __sk_common.skc_nulls_node 360 #define sk_refcnt __sk_common.skc_refcnt 361 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 362 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 363 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 364 #endif 365 366 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 367 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 368 #define sk_hash __sk_common.skc_hash 369 #define sk_portpair __sk_common.skc_portpair 370 #define sk_num __sk_common.skc_num 371 #define sk_dport __sk_common.skc_dport 372 #define sk_addrpair __sk_common.skc_addrpair 373 #define sk_daddr __sk_common.skc_daddr 374 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 375 #define sk_family __sk_common.skc_family 376 #define sk_state __sk_common.skc_state 377 #define sk_reuse __sk_common.skc_reuse 378 #define sk_reuseport __sk_common.skc_reuseport 379 #define sk_ipv6only __sk_common.skc_ipv6only 380 #define sk_net_refcnt __sk_common.skc_net_refcnt 381 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 382 #define sk_bind_node __sk_common.skc_bind_node 383 #define sk_prot __sk_common.skc_prot 384 #define sk_net __sk_common.skc_net 385 #define sk_v6_daddr __sk_common.skc_v6_daddr 386 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 387 #define sk_cookie __sk_common.skc_cookie 388 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 389 #define sk_flags __sk_common.skc_flags 390 #define sk_rxhash __sk_common.skc_rxhash 391 392 socket_lock_t sk_lock; 393 atomic_t sk_drops; 394 int sk_rcvlowat; 395 struct sk_buff_head sk_error_queue; 396 struct sk_buff_head sk_receive_queue; 397 /* 398 * The backlog queue is special, it is always used with 399 * the per-socket spinlock held and requires low latency 400 * access. Therefore we special case it's implementation. 401 * Note : rmem_alloc is in this structure to fill a hole 402 * on 64bit arches, not because its logically part of 403 * backlog. 404 */ 405 struct { 406 atomic_t rmem_alloc; 407 int len; 408 struct sk_buff *head; 409 struct sk_buff *tail; 410 } sk_backlog; 411 #define sk_rmem_alloc sk_backlog.rmem_alloc 412 413 int sk_forward_alloc; 414 u32 sk_reserved_mem; 415 #ifdef CONFIG_NET_RX_BUSY_POLL 416 unsigned int sk_ll_usec; 417 /* ===== mostly read cache line ===== */ 418 unsigned int sk_napi_id; 419 #endif 420 int sk_rcvbuf; 421 422 struct sk_filter __rcu *sk_filter; 423 union { 424 struct socket_wq __rcu *sk_wq; 425 /* private: */ 426 struct socket_wq *sk_wq_raw; 427 /* public: */ 428 }; 429 #ifdef CONFIG_XFRM 430 struct xfrm_policy __rcu *sk_policy[2]; 431 #endif 432 struct dst_entry *sk_rx_dst; 433 struct dst_entry __rcu *sk_dst_cache; 434 atomic_t sk_omem_alloc; 435 int sk_sndbuf; 436 437 /* ===== cache line for TX ===== */ 438 int sk_wmem_queued; 439 refcount_t sk_wmem_alloc; 440 unsigned long sk_tsq_flags; 441 union { 442 struct sk_buff *sk_send_head; 443 struct rb_root tcp_rtx_queue; 444 }; 445 struct sk_buff_head sk_write_queue; 446 __s32 sk_peek_off; 447 int sk_write_pending; 448 __u32 sk_dst_pending_confirm; 449 u32 sk_pacing_status; /* see enum sk_pacing */ 450 long sk_sndtimeo; 451 struct timer_list sk_timer; 452 __u32 sk_priority; 453 __u32 sk_mark; 454 unsigned long sk_pacing_rate; /* bytes per second */ 455 unsigned long sk_max_pacing_rate; 456 struct page_frag sk_frag; 457 netdev_features_t sk_route_caps; 458 netdev_features_t sk_route_nocaps; 459 netdev_features_t sk_route_forced_caps; 460 int sk_gso_type; 461 unsigned int sk_gso_max_size; 462 gfp_t sk_allocation; 463 __u32 sk_txhash; 464 465 /* 466 * Because of non atomicity rules, all 467 * changes are protected by socket lock. 468 */ 469 u8 sk_padding : 1, 470 sk_kern_sock : 1, 471 sk_no_check_tx : 1, 472 sk_no_check_rx : 1, 473 sk_userlocks : 4; 474 u8 sk_pacing_shift; 475 u16 sk_type; 476 u16 sk_protocol; 477 u16 sk_gso_max_segs; 478 unsigned long sk_lingertime; 479 struct proto *sk_prot_creator; 480 rwlock_t sk_callback_lock; 481 int sk_err, 482 sk_err_soft; 483 u32 sk_ack_backlog; 484 u32 sk_max_ack_backlog; 485 kuid_t sk_uid; 486 #ifdef CONFIG_NET_RX_BUSY_POLL 487 u8 sk_prefer_busy_poll; 488 u16 sk_busy_poll_budget; 489 #endif 490 spinlock_t sk_peer_lock; 491 struct pid *sk_peer_pid; 492 const struct cred *sk_peer_cred; 493 494 long sk_rcvtimeo; 495 ktime_t sk_stamp; 496 #if BITS_PER_LONG==32 497 seqlock_t sk_stamp_seq; 498 #endif 499 u16 sk_tsflags; 500 int sk_bind_phc; 501 u8 sk_shutdown; 502 u32 sk_tskey; 503 atomic_t sk_zckey; 504 505 u8 sk_clockid; 506 u8 sk_txtime_deadline_mode : 1, 507 sk_txtime_report_errors : 1, 508 sk_txtime_unused : 6; 509 510 struct socket *sk_socket; 511 void *sk_user_data; 512 #ifdef CONFIG_SECURITY 513 void *sk_security; 514 #endif 515 struct sock_cgroup_data sk_cgrp_data; 516 struct mem_cgroup *sk_memcg; 517 void (*sk_state_change)(struct sock *sk); 518 void (*sk_data_ready)(struct sock *sk); 519 void (*sk_write_space)(struct sock *sk); 520 void (*sk_error_report)(struct sock *sk); 521 int (*sk_backlog_rcv)(struct sock *sk, 522 struct sk_buff *skb); 523 #ifdef CONFIG_SOCK_VALIDATE_XMIT 524 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 525 struct net_device *dev, 526 struct sk_buff *skb); 527 #endif 528 void (*sk_destruct)(struct sock *sk); 529 struct sock_reuseport __rcu *sk_reuseport_cb; 530 #ifdef CONFIG_BPF_SYSCALL 531 struct bpf_local_storage __rcu *sk_bpf_storage; 532 #endif 533 struct rcu_head sk_rcu; 534 }; 535 536 enum sk_pacing { 537 SK_PACING_NONE = 0, 538 SK_PACING_NEEDED = 1, 539 SK_PACING_FQ = 2, 540 }; 541 542 /* Pointer stored in sk_user_data might not be suitable for copying 543 * when cloning the socket. For instance, it can point to a reference 544 * counted object. sk_user_data bottom bit is set if pointer must not 545 * be copied. 546 */ 547 #define SK_USER_DATA_NOCOPY 1UL 548 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 549 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 550 551 /** 552 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 553 * @sk: socket 554 */ 555 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 556 { 557 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 558 } 559 560 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 561 562 #define rcu_dereference_sk_user_data(sk) \ 563 ({ \ 564 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 565 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 566 }) 567 #define rcu_assign_sk_user_data(sk, ptr) \ 568 ({ \ 569 uintptr_t __tmp = (uintptr_t)(ptr); \ 570 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 571 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 572 }) 573 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 574 ({ \ 575 uintptr_t __tmp = (uintptr_t)(ptr); \ 576 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 577 rcu_assign_pointer(__sk_user_data((sk)), \ 578 __tmp | SK_USER_DATA_NOCOPY); \ 579 }) 580 581 /* 582 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 583 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 584 * on a socket means that the socket will reuse everybody else's port 585 * without looking at the other's sk_reuse value. 586 */ 587 588 #define SK_NO_REUSE 0 589 #define SK_CAN_REUSE 1 590 #define SK_FORCE_REUSE 2 591 592 int sk_set_peek_off(struct sock *sk, int val); 593 594 static inline int sk_peek_offset(struct sock *sk, int flags) 595 { 596 if (unlikely(flags & MSG_PEEK)) { 597 return READ_ONCE(sk->sk_peek_off); 598 } 599 600 return 0; 601 } 602 603 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 604 { 605 s32 off = READ_ONCE(sk->sk_peek_off); 606 607 if (unlikely(off >= 0)) { 608 off = max_t(s32, off - val, 0); 609 WRITE_ONCE(sk->sk_peek_off, off); 610 } 611 } 612 613 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 614 { 615 sk_peek_offset_bwd(sk, -val); 616 } 617 618 /* 619 * Hashed lists helper routines 620 */ 621 static inline struct sock *sk_entry(const struct hlist_node *node) 622 { 623 return hlist_entry(node, struct sock, sk_node); 624 } 625 626 static inline struct sock *__sk_head(const struct hlist_head *head) 627 { 628 return hlist_entry(head->first, struct sock, sk_node); 629 } 630 631 static inline struct sock *sk_head(const struct hlist_head *head) 632 { 633 return hlist_empty(head) ? NULL : __sk_head(head); 634 } 635 636 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 637 { 638 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 639 } 640 641 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 642 { 643 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 644 } 645 646 static inline struct sock *sk_next(const struct sock *sk) 647 { 648 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 649 } 650 651 static inline struct sock *sk_nulls_next(const struct sock *sk) 652 { 653 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 654 hlist_nulls_entry(sk->sk_nulls_node.next, 655 struct sock, sk_nulls_node) : 656 NULL; 657 } 658 659 static inline bool sk_unhashed(const struct sock *sk) 660 { 661 return hlist_unhashed(&sk->sk_node); 662 } 663 664 static inline bool sk_hashed(const struct sock *sk) 665 { 666 return !sk_unhashed(sk); 667 } 668 669 static inline void sk_node_init(struct hlist_node *node) 670 { 671 node->pprev = NULL; 672 } 673 674 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 675 { 676 node->pprev = NULL; 677 } 678 679 static inline void __sk_del_node(struct sock *sk) 680 { 681 __hlist_del(&sk->sk_node); 682 } 683 684 /* NB: equivalent to hlist_del_init_rcu */ 685 static inline bool __sk_del_node_init(struct sock *sk) 686 { 687 if (sk_hashed(sk)) { 688 __sk_del_node(sk); 689 sk_node_init(&sk->sk_node); 690 return true; 691 } 692 return false; 693 } 694 695 /* Grab socket reference count. This operation is valid only 696 when sk is ALREADY grabbed f.e. it is found in hash table 697 or a list and the lookup is made under lock preventing hash table 698 modifications. 699 */ 700 701 static __always_inline void sock_hold(struct sock *sk) 702 { 703 refcount_inc(&sk->sk_refcnt); 704 } 705 706 /* Ungrab socket in the context, which assumes that socket refcnt 707 cannot hit zero, f.e. it is true in context of any socketcall. 708 */ 709 static __always_inline void __sock_put(struct sock *sk) 710 { 711 refcount_dec(&sk->sk_refcnt); 712 } 713 714 static inline bool sk_del_node_init(struct sock *sk) 715 { 716 bool rc = __sk_del_node_init(sk); 717 718 if (rc) { 719 /* paranoid for a while -acme */ 720 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 721 __sock_put(sk); 722 } 723 return rc; 724 } 725 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 726 727 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 728 { 729 if (sk_hashed(sk)) { 730 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 731 return true; 732 } 733 return false; 734 } 735 736 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 737 { 738 bool rc = __sk_nulls_del_node_init_rcu(sk); 739 740 if (rc) { 741 /* paranoid for a while -acme */ 742 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 743 __sock_put(sk); 744 } 745 return rc; 746 } 747 748 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 749 { 750 hlist_add_head(&sk->sk_node, list); 751 } 752 753 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 754 { 755 sock_hold(sk); 756 __sk_add_node(sk, list); 757 } 758 759 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 760 { 761 sock_hold(sk); 762 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 763 sk->sk_family == AF_INET6) 764 hlist_add_tail_rcu(&sk->sk_node, list); 765 else 766 hlist_add_head_rcu(&sk->sk_node, list); 767 } 768 769 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 770 { 771 sock_hold(sk); 772 hlist_add_tail_rcu(&sk->sk_node, list); 773 } 774 775 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 776 { 777 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 778 } 779 780 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 781 { 782 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 783 } 784 785 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 786 { 787 sock_hold(sk); 788 __sk_nulls_add_node_rcu(sk, list); 789 } 790 791 static inline void __sk_del_bind_node(struct sock *sk) 792 { 793 __hlist_del(&sk->sk_bind_node); 794 } 795 796 static inline void sk_add_bind_node(struct sock *sk, 797 struct hlist_head *list) 798 { 799 hlist_add_head(&sk->sk_bind_node, list); 800 } 801 802 #define sk_for_each(__sk, list) \ 803 hlist_for_each_entry(__sk, list, sk_node) 804 #define sk_for_each_rcu(__sk, list) \ 805 hlist_for_each_entry_rcu(__sk, list, sk_node) 806 #define sk_nulls_for_each(__sk, node, list) \ 807 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 808 #define sk_nulls_for_each_rcu(__sk, node, list) \ 809 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 810 #define sk_for_each_from(__sk) \ 811 hlist_for_each_entry_from(__sk, sk_node) 812 #define sk_nulls_for_each_from(__sk, node) \ 813 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 814 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 815 #define sk_for_each_safe(__sk, tmp, list) \ 816 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 817 #define sk_for_each_bound(__sk, list) \ 818 hlist_for_each_entry(__sk, list, sk_bind_node) 819 820 /** 821 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 822 * @tpos: the type * to use as a loop cursor. 823 * @pos: the &struct hlist_node to use as a loop cursor. 824 * @head: the head for your list. 825 * @offset: offset of hlist_node within the struct. 826 * 827 */ 828 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 829 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 830 pos != NULL && \ 831 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 832 pos = rcu_dereference(hlist_next_rcu(pos))) 833 834 static inline struct user_namespace *sk_user_ns(struct sock *sk) 835 { 836 /* Careful only use this in a context where these parameters 837 * can not change and must all be valid, such as recvmsg from 838 * userspace. 839 */ 840 return sk->sk_socket->file->f_cred->user_ns; 841 } 842 843 /* Sock flags */ 844 enum sock_flags { 845 SOCK_DEAD, 846 SOCK_DONE, 847 SOCK_URGINLINE, 848 SOCK_KEEPOPEN, 849 SOCK_LINGER, 850 SOCK_DESTROY, 851 SOCK_BROADCAST, 852 SOCK_TIMESTAMP, 853 SOCK_ZAPPED, 854 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 855 SOCK_DBG, /* %SO_DEBUG setting */ 856 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 857 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 858 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 859 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 860 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 861 SOCK_FASYNC, /* fasync() active */ 862 SOCK_RXQ_OVFL, 863 SOCK_ZEROCOPY, /* buffers from userspace */ 864 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 865 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 866 * Will use last 4 bytes of packet sent from 867 * user-space instead. 868 */ 869 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 870 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 871 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 872 SOCK_TXTIME, 873 SOCK_XDP, /* XDP is attached */ 874 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 875 }; 876 877 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 878 879 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 880 { 881 nsk->sk_flags = osk->sk_flags; 882 } 883 884 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 885 { 886 __set_bit(flag, &sk->sk_flags); 887 } 888 889 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 890 { 891 __clear_bit(flag, &sk->sk_flags); 892 } 893 894 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 895 int valbool) 896 { 897 if (valbool) 898 sock_set_flag(sk, bit); 899 else 900 sock_reset_flag(sk, bit); 901 } 902 903 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 904 { 905 return test_bit(flag, &sk->sk_flags); 906 } 907 908 #ifdef CONFIG_NET 909 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 910 static inline int sk_memalloc_socks(void) 911 { 912 return static_branch_unlikely(&memalloc_socks_key); 913 } 914 915 void __receive_sock(struct file *file); 916 #else 917 918 static inline int sk_memalloc_socks(void) 919 { 920 return 0; 921 } 922 923 static inline void __receive_sock(struct file *file) 924 { } 925 #endif 926 927 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 928 { 929 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 930 } 931 932 static inline void sk_acceptq_removed(struct sock *sk) 933 { 934 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 935 } 936 937 static inline void sk_acceptq_added(struct sock *sk) 938 { 939 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 940 } 941 942 /* Note: If you think the test should be: 943 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 944 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 945 */ 946 static inline bool sk_acceptq_is_full(const struct sock *sk) 947 { 948 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 949 } 950 951 /* 952 * Compute minimal free write space needed to queue new packets. 953 */ 954 static inline int sk_stream_min_wspace(const struct sock *sk) 955 { 956 return READ_ONCE(sk->sk_wmem_queued) >> 1; 957 } 958 959 static inline int sk_stream_wspace(const struct sock *sk) 960 { 961 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 962 } 963 964 static inline void sk_wmem_queued_add(struct sock *sk, int val) 965 { 966 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 967 } 968 969 void sk_stream_write_space(struct sock *sk); 970 971 /* OOB backlog add */ 972 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 973 { 974 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 975 skb_dst_force(skb); 976 977 if (!sk->sk_backlog.tail) 978 WRITE_ONCE(sk->sk_backlog.head, skb); 979 else 980 sk->sk_backlog.tail->next = skb; 981 982 WRITE_ONCE(sk->sk_backlog.tail, skb); 983 skb->next = NULL; 984 } 985 986 /* 987 * Take into account size of receive queue and backlog queue 988 * Do not take into account this skb truesize, 989 * to allow even a single big packet to come. 990 */ 991 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 992 { 993 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 994 995 return qsize > limit; 996 } 997 998 /* The per-socket spinlock must be held here. */ 999 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1000 unsigned int limit) 1001 { 1002 if (sk_rcvqueues_full(sk, limit)) 1003 return -ENOBUFS; 1004 1005 /* 1006 * If the skb was allocated from pfmemalloc reserves, only 1007 * allow SOCK_MEMALLOC sockets to use it as this socket is 1008 * helping free memory 1009 */ 1010 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1011 return -ENOMEM; 1012 1013 __sk_add_backlog(sk, skb); 1014 sk->sk_backlog.len += skb->truesize; 1015 return 0; 1016 } 1017 1018 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1019 1020 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1021 { 1022 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1023 return __sk_backlog_rcv(sk, skb); 1024 1025 return sk->sk_backlog_rcv(sk, skb); 1026 } 1027 1028 static inline void sk_incoming_cpu_update(struct sock *sk) 1029 { 1030 int cpu = raw_smp_processor_id(); 1031 1032 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1033 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1034 } 1035 1036 static inline void sock_rps_record_flow_hash(__u32 hash) 1037 { 1038 #ifdef CONFIG_RPS 1039 struct rps_sock_flow_table *sock_flow_table; 1040 1041 rcu_read_lock(); 1042 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1043 rps_record_sock_flow(sock_flow_table, hash); 1044 rcu_read_unlock(); 1045 #endif 1046 } 1047 1048 static inline void sock_rps_record_flow(const struct sock *sk) 1049 { 1050 #ifdef CONFIG_RPS 1051 if (static_branch_unlikely(&rfs_needed)) { 1052 /* Reading sk->sk_rxhash might incur an expensive cache line 1053 * miss. 1054 * 1055 * TCP_ESTABLISHED does cover almost all states where RFS 1056 * might be useful, and is cheaper [1] than testing : 1057 * IPv4: inet_sk(sk)->inet_daddr 1058 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1059 * OR an additional socket flag 1060 * [1] : sk_state and sk_prot are in the same cache line. 1061 */ 1062 if (sk->sk_state == TCP_ESTABLISHED) 1063 sock_rps_record_flow_hash(sk->sk_rxhash); 1064 } 1065 #endif 1066 } 1067 1068 static inline void sock_rps_save_rxhash(struct sock *sk, 1069 const struct sk_buff *skb) 1070 { 1071 #ifdef CONFIG_RPS 1072 if (unlikely(sk->sk_rxhash != skb->hash)) 1073 sk->sk_rxhash = skb->hash; 1074 #endif 1075 } 1076 1077 static inline void sock_rps_reset_rxhash(struct sock *sk) 1078 { 1079 #ifdef CONFIG_RPS 1080 sk->sk_rxhash = 0; 1081 #endif 1082 } 1083 1084 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1085 ({ int __rc; \ 1086 release_sock(__sk); \ 1087 __rc = __condition; \ 1088 if (!__rc) { \ 1089 *(__timeo) = wait_woken(__wait, \ 1090 TASK_INTERRUPTIBLE, \ 1091 *(__timeo)); \ 1092 } \ 1093 sched_annotate_sleep(); \ 1094 lock_sock(__sk); \ 1095 __rc = __condition; \ 1096 __rc; \ 1097 }) 1098 1099 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1100 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1101 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1102 int sk_stream_error(struct sock *sk, int flags, int err); 1103 void sk_stream_kill_queues(struct sock *sk); 1104 void sk_set_memalloc(struct sock *sk); 1105 void sk_clear_memalloc(struct sock *sk); 1106 1107 void __sk_flush_backlog(struct sock *sk); 1108 1109 static inline bool sk_flush_backlog(struct sock *sk) 1110 { 1111 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1112 __sk_flush_backlog(sk); 1113 return true; 1114 } 1115 return false; 1116 } 1117 1118 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1119 1120 struct request_sock_ops; 1121 struct timewait_sock_ops; 1122 struct inet_hashinfo; 1123 struct raw_hashinfo; 1124 struct smc_hashinfo; 1125 struct module; 1126 struct sk_psock; 1127 1128 /* 1129 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1130 * un-modified. Special care is taken when initializing object to zero. 1131 */ 1132 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1133 { 1134 if (offsetof(struct sock, sk_node.next) != 0) 1135 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1136 memset(&sk->sk_node.pprev, 0, 1137 size - offsetof(struct sock, sk_node.pprev)); 1138 } 1139 1140 /* Networking protocol blocks we attach to sockets. 1141 * socket layer -> transport layer interface 1142 */ 1143 struct proto { 1144 void (*close)(struct sock *sk, 1145 long timeout); 1146 int (*pre_connect)(struct sock *sk, 1147 struct sockaddr *uaddr, 1148 int addr_len); 1149 int (*connect)(struct sock *sk, 1150 struct sockaddr *uaddr, 1151 int addr_len); 1152 int (*disconnect)(struct sock *sk, int flags); 1153 1154 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1155 bool kern); 1156 1157 int (*ioctl)(struct sock *sk, int cmd, 1158 unsigned long arg); 1159 int (*init)(struct sock *sk); 1160 void (*destroy)(struct sock *sk); 1161 void (*shutdown)(struct sock *sk, int how); 1162 int (*setsockopt)(struct sock *sk, int level, 1163 int optname, sockptr_t optval, 1164 unsigned int optlen); 1165 int (*getsockopt)(struct sock *sk, int level, 1166 int optname, char __user *optval, 1167 int __user *option); 1168 void (*keepalive)(struct sock *sk, int valbool); 1169 #ifdef CONFIG_COMPAT 1170 int (*compat_ioctl)(struct sock *sk, 1171 unsigned int cmd, unsigned long arg); 1172 #endif 1173 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1174 size_t len); 1175 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1176 size_t len, int noblock, int flags, 1177 int *addr_len); 1178 int (*sendpage)(struct sock *sk, struct page *page, 1179 int offset, size_t size, int flags); 1180 int (*bind)(struct sock *sk, 1181 struct sockaddr *addr, int addr_len); 1182 int (*bind_add)(struct sock *sk, 1183 struct sockaddr *addr, int addr_len); 1184 1185 int (*backlog_rcv) (struct sock *sk, 1186 struct sk_buff *skb); 1187 bool (*bpf_bypass_getsockopt)(int level, 1188 int optname); 1189 1190 void (*release_cb)(struct sock *sk); 1191 1192 /* Keeping track of sk's, looking them up, and port selection methods. */ 1193 int (*hash)(struct sock *sk); 1194 void (*unhash)(struct sock *sk); 1195 void (*rehash)(struct sock *sk); 1196 int (*get_port)(struct sock *sk, unsigned short snum); 1197 #ifdef CONFIG_BPF_SYSCALL 1198 int (*psock_update_sk_prot)(struct sock *sk, 1199 struct sk_psock *psock, 1200 bool restore); 1201 #endif 1202 1203 /* Keeping track of sockets in use */ 1204 #ifdef CONFIG_PROC_FS 1205 unsigned int inuse_idx; 1206 #endif 1207 1208 bool (*stream_memory_free)(const struct sock *sk, int wake); 1209 bool (*stream_memory_read)(const struct sock *sk); 1210 /* Memory pressure */ 1211 void (*enter_memory_pressure)(struct sock *sk); 1212 void (*leave_memory_pressure)(struct sock *sk); 1213 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1214 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1215 /* 1216 * Pressure flag: try to collapse. 1217 * Technical note: it is used by multiple contexts non atomically. 1218 * All the __sk_mem_schedule() is of this nature: accounting 1219 * is strict, actions are advisory and have some latency. 1220 */ 1221 unsigned long *memory_pressure; 1222 long *sysctl_mem; 1223 1224 int *sysctl_wmem; 1225 int *sysctl_rmem; 1226 u32 sysctl_wmem_offset; 1227 u32 sysctl_rmem_offset; 1228 1229 int max_header; 1230 bool no_autobind; 1231 1232 struct kmem_cache *slab; 1233 unsigned int obj_size; 1234 slab_flags_t slab_flags; 1235 unsigned int useroffset; /* Usercopy region offset */ 1236 unsigned int usersize; /* Usercopy region size */ 1237 1238 unsigned int __percpu *orphan_count; 1239 1240 struct request_sock_ops *rsk_prot; 1241 struct timewait_sock_ops *twsk_prot; 1242 1243 union { 1244 struct inet_hashinfo *hashinfo; 1245 struct udp_table *udp_table; 1246 struct raw_hashinfo *raw_hash; 1247 struct smc_hashinfo *smc_hash; 1248 } h; 1249 1250 struct module *owner; 1251 1252 char name[32]; 1253 1254 struct list_head node; 1255 #ifdef SOCK_REFCNT_DEBUG 1256 atomic_t socks; 1257 #endif 1258 int (*diag_destroy)(struct sock *sk, int err); 1259 } __randomize_layout; 1260 1261 int proto_register(struct proto *prot, int alloc_slab); 1262 void proto_unregister(struct proto *prot); 1263 int sock_load_diag_module(int family, int protocol); 1264 1265 #ifdef SOCK_REFCNT_DEBUG 1266 static inline void sk_refcnt_debug_inc(struct sock *sk) 1267 { 1268 atomic_inc(&sk->sk_prot->socks); 1269 } 1270 1271 static inline void sk_refcnt_debug_dec(struct sock *sk) 1272 { 1273 atomic_dec(&sk->sk_prot->socks); 1274 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1275 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1276 } 1277 1278 static inline void sk_refcnt_debug_release(const struct sock *sk) 1279 { 1280 if (refcount_read(&sk->sk_refcnt) != 1) 1281 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1282 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1283 } 1284 #else /* SOCK_REFCNT_DEBUG */ 1285 #define sk_refcnt_debug_inc(sk) do { } while (0) 1286 #define sk_refcnt_debug_dec(sk) do { } while (0) 1287 #define sk_refcnt_debug_release(sk) do { } while (0) 1288 #endif /* SOCK_REFCNT_DEBUG */ 1289 1290 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1291 1292 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1293 { 1294 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1295 return false; 1296 1297 #ifdef CONFIG_INET 1298 return sk->sk_prot->stream_memory_free ? 1299 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free, 1300 tcp_stream_memory_free, 1301 sk, wake) : true; 1302 #else 1303 return sk->sk_prot->stream_memory_free ? 1304 sk->sk_prot->stream_memory_free(sk, wake) : true; 1305 #endif 1306 } 1307 1308 static inline bool sk_stream_memory_free(const struct sock *sk) 1309 { 1310 return __sk_stream_memory_free(sk, 0); 1311 } 1312 1313 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1314 { 1315 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1316 __sk_stream_memory_free(sk, wake); 1317 } 1318 1319 static inline bool sk_stream_is_writeable(const struct sock *sk) 1320 { 1321 return __sk_stream_is_writeable(sk, 0); 1322 } 1323 1324 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1325 struct cgroup *ancestor) 1326 { 1327 #ifdef CONFIG_SOCK_CGROUP_DATA 1328 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1329 ancestor); 1330 #else 1331 return -ENOTSUPP; 1332 #endif 1333 } 1334 1335 static inline bool sk_has_memory_pressure(const struct sock *sk) 1336 { 1337 return sk->sk_prot->memory_pressure != NULL; 1338 } 1339 1340 static inline bool sk_under_memory_pressure(const struct sock *sk) 1341 { 1342 if (!sk->sk_prot->memory_pressure) 1343 return false; 1344 1345 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1346 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1347 return true; 1348 1349 return !!*sk->sk_prot->memory_pressure; 1350 } 1351 1352 static inline long 1353 sk_memory_allocated(const struct sock *sk) 1354 { 1355 return atomic_long_read(sk->sk_prot->memory_allocated); 1356 } 1357 1358 static inline long 1359 sk_memory_allocated_add(struct sock *sk, int amt) 1360 { 1361 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1362 } 1363 1364 static inline void 1365 sk_memory_allocated_sub(struct sock *sk, int amt) 1366 { 1367 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1368 } 1369 1370 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1371 1372 static inline void sk_sockets_allocated_dec(struct sock *sk) 1373 { 1374 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1375 SK_ALLOC_PERCPU_COUNTER_BATCH); 1376 } 1377 1378 static inline void sk_sockets_allocated_inc(struct sock *sk) 1379 { 1380 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1381 SK_ALLOC_PERCPU_COUNTER_BATCH); 1382 } 1383 1384 static inline u64 1385 sk_sockets_allocated_read_positive(struct sock *sk) 1386 { 1387 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1388 } 1389 1390 static inline int 1391 proto_sockets_allocated_sum_positive(struct proto *prot) 1392 { 1393 return percpu_counter_sum_positive(prot->sockets_allocated); 1394 } 1395 1396 static inline long 1397 proto_memory_allocated(struct proto *prot) 1398 { 1399 return atomic_long_read(prot->memory_allocated); 1400 } 1401 1402 static inline bool 1403 proto_memory_pressure(struct proto *prot) 1404 { 1405 if (!prot->memory_pressure) 1406 return false; 1407 return !!*prot->memory_pressure; 1408 } 1409 1410 1411 #ifdef CONFIG_PROC_FS 1412 /* Called with local bh disabled */ 1413 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1414 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1415 int sock_inuse_get(struct net *net); 1416 #else 1417 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1418 int inc) 1419 { 1420 } 1421 #endif 1422 1423 1424 /* With per-bucket locks this operation is not-atomic, so that 1425 * this version is not worse. 1426 */ 1427 static inline int __sk_prot_rehash(struct sock *sk) 1428 { 1429 sk->sk_prot->unhash(sk); 1430 return sk->sk_prot->hash(sk); 1431 } 1432 1433 /* About 10 seconds */ 1434 #define SOCK_DESTROY_TIME (10*HZ) 1435 1436 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1437 #define PROT_SOCK 1024 1438 1439 #define SHUTDOWN_MASK 3 1440 #define RCV_SHUTDOWN 1 1441 #define SEND_SHUTDOWN 2 1442 1443 #define SOCK_BINDADDR_LOCK 4 1444 #define SOCK_BINDPORT_LOCK 8 1445 1446 struct socket_alloc { 1447 struct socket socket; 1448 struct inode vfs_inode; 1449 }; 1450 1451 static inline struct socket *SOCKET_I(struct inode *inode) 1452 { 1453 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1454 } 1455 1456 static inline struct inode *SOCK_INODE(struct socket *socket) 1457 { 1458 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1459 } 1460 1461 /* 1462 * Functions for memory accounting 1463 */ 1464 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1465 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1466 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1467 void __sk_mem_reclaim(struct sock *sk, int amount); 1468 1469 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1470 * do not necessarily have 16x time more memory than 4KB ones. 1471 */ 1472 #define SK_MEM_QUANTUM 4096 1473 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1474 #define SK_MEM_SEND 0 1475 #define SK_MEM_RECV 1 1476 1477 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1478 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1479 { 1480 long val = sk->sk_prot->sysctl_mem[index]; 1481 1482 #if PAGE_SIZE > SK_MEM_QUANTUM 1483 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1484 #elif PAGE_SIZE < SK_MEM_QUANTUM 1485 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1486 #endif 1487 return val; 1488 } 1489 1490 static inline int sk_mem_pages(int amt) 1491 { 1492 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1493 } 1494 1495 static inline bool sk_has_account(struct sock *sk) 1496 { 1497 /* return true if protocol supports memory accounting */ 1498 return !!sk->sk_prot->memory_allocated; 1499 } 1500 1501 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1502 { 1503 if (!sk_has_account(sk)) 1504 return true; 1505 return size <= sk->sk_forward_alloc || 1506 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1507 } 1508 1509 static inline bool 1510 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1511 { 1512 if (!sk_has_account(sk)) 1513 return true; 1514 return size <= sk->sk_forward_alloc || 1515 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1516 skb_pfmemalloc(skb); 1517 } 1518 1519 static inline int sk_unused_reserved_mem(const struct sock *sk) 1520 { 1521 int unused_mem; 1522 1523 if (likely(!sk->sk_reserved_mem)) 1524 return 0; 1525 1526 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1527 atomic_read(&sk->sk_rmem_alloc); 1528 1529 return unused_mem > 0 ? unused_mem : 0; 1530 } 1531 1532 static inline void sk_mem_reclaim(struct sock *sk) 1533 { 1534 int reclaimable; 1535 1536 if (!sk_has_account(sk)) 1537 return; 1538 1539 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1540 1541 if (reclaimable >= SK_MEM_QUANTUM) 1542 __sk_mem_reclaim(sk, reclaimable); 1543 } 1544 1545 static inline void sk_mem_reclaim_final(struct sock *sk) 1546 { 1547 sk->sk_reserved_mem = 0; 1548 sk_mem_reclaim(sk); 1549 } 1550 1551 static inline void sk_mem_reclaim_partial(struct sock *sk) 1552 { 1553 int reclaimable; 1554 1555 if (!sk_has_account(sk)) 1556 return; 1557 1558 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1559 1560 if (reclaimable > SK_MEM_QUANTUM) 1561 __sk_mem_reclaim(sk, reclaimable - 1); 1562 } 1563 1564 static inline void sk_mem_charge(struct sock *sk, int size) 1565 { 1566 if (!sk_has_account(sk)) 1567 return; 1568 sk->sk_forward_alloc -= size; 1569 } 1570 1571 static inline void sk_mem_uncharge(struct sock *sk, int size) 1572 { 1573 int reclaimable; 1574 1575 if (!sk_has_account(sk)) 1576 return; 1577 sk->sk_forward_alloc += size; 1578 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1579 1580 /* Avoid a possible overflow. 1581 * TCP send queues can make this happen, if sk_mem_reclaim() 1582 * is not called and more than 2 GBytes are released at once. 1583 * 1584 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1585 * no need to hold that much forward allocation anyway. 1586 */ 1587 if (unlikely(reclaimable >= 1 << 21)) 1588 __sk_mem_reclaim(sk, 1 << 20); 1589 } 1590 1591 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1592 { 1593 sk_wmem_queued_add(sk, -skb->truesize); 1594 sk_mem_uncharge(sk, skb->truesize); 1595 __kfree_skb(skb); 1596 } 1597 1598 static inline void sock_release_ownership(struct sock *sk) 1599 { 1600 if (sk->sk_lock.owned) { 1601 sk->sk_lock.owned = 0; 1602 1603 /* The sk_lock has mutex_unlock() semantics: */ 1604 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1605 } 1606 } 1607 1608 /* 1609 * Macro so as to not evaluate some arguments when 1610 * lockdep is not enabled. 1611 * 1612 * Mark both the sk_lock and the sk_lock.slock as a 1613 * per-address-family lock class. 1614 */ 1615 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1616 do { \ 1617 sk->sk_lock.owned = 0; \ 1618 init_waitqueue_head(&sk->sk_lock.wq); \ 1619 spin_lock_init(&(sk)->sk_lock.slock); \ 1620 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1621 sizeof((sk)->sk_lock)); \ 1622 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1623 (skey), (sname)); \ 1624 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1625 } while (0) 1626 1627 static inline bool lockdep_sock_is_held(const struct sock *sk) 1628 { 1629 return lockdep_is_held(&sk->sk_lock) || 1630 lockdep_is_held(&sk->sk_lock.slock); 1631 } 1632 1633 void lock_sock_nested(struct sock *sk, int subclass); 1634 1635 static inline void lock_sock(struct sock *sk) 1636 { 1637 lock_sock_nested(sk, 0); 1638 } 1639 1640 void __lock_sock(struct sock *sk); 1641 void __release_sock(struct sock *sk); 1642 void release_sock(struct sock *sk); 1643 1644 /* BH context may only use the following locking interface. */ 1645 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1646 #define bh_lock_sock_nested(__sk) \ 1647 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1648 SINGLE_DEPTH_NESTING) 1649 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1650 1651 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1652 1653 /** 1654 * lock_sock_fast - fast version of lock_sock 1655 * @sk: socket 1656 * 1657 * This version should be used for very small section, where process wont block 1658 * return false if fast path is taken: 1659 * 1660 * sk_lock.slock locked, owned = 0, BH disabled 1661 * 1662 * return true if slow path is taken: 1663 * 1664 * sk_lock.slock unlocked, owned = 1, BH enabled 1665 */ 1666 static inline bool lock_sock_fast(struct sock *sk) 1667 { 1668 /* The sk_lock has mutex_lock() semantics here. */ 1669 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1670 1671 return __lock_sock_fast(sk); 1672 } 1673 1674 /* fast socket lock variant for caller already holding a [different] socket lock */ 1675 static inline bool lock_sock_fast_nested(struct sock *sk) 1676 { 1677 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1678 1679 return __lock_sock_fast(sk); 1680 } 1681 1682 /** 1683 * unlock_sock_fast - complement of lock_sock_fast 1684 * @sk: socket 1685 * @slow: slow mode 1686 * 1687 * fast unlock socket for user context. 1688 * If slow mode is on, we call regular release_sock() 1689 */ 1690 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1691 __releases(&sk->sk_lock.slock) 1692 { 1693 if (slow) { 1694 release_sock(sk); 1695 __release(&sk->sk_lock.slock); 1696 } else { 1697 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1698 spin_unlock_bh(&sk->sk_lock.slock); 1699 } 1700 } 1701 1702 /* Used by processes to "lock" a socket state, so that 1703 * interrupts and bottom half handlers won't change it 1704 * from under us. It essentially blocks any incoming 1705 * packets, so that we won't get any new data or any 1706 * packets that change the state of the socket. 1707 * 1708 * While locked, BH processing will add new packets to 1709 * the backlog queue. This queue is processed by the 1710 * owner of the socket lock right before it is released. 1711 * 1712 * Since ~2.3.5 it is also exclusive sleep lock serializing 1713 * accesses from user process context. 1714 */ 1715 1716 static inline void sock_owned_by_me(const struct sock *sk) 1717 { 1718 #ifdef CONFIG_LOCKDEP 1719 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1720 #endif 1721 } 1722 1723 static inline bool sock_owned_by_user(const struct sock *sk) 1724 { 1725 sock_owned_by_me(sk); 1726 return sk->sk_lock.owned; 1727 } 1728 1729 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1730 { 1731 return sk->sk_lock.owned; 1732 } 1733 1734 /* no reclassification while locks are held */ 1735 static inline bool sock_allow_reclassification(const struct sock *csk) 1736 { 1737 struct sock *sk = (struct sock *)csk; 1738 1739 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1740 } 1741 1742 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1743 struct proto *prot, int kern); 1744 void sk_free(struct sock *sk); 1745 void sk_destruct(struct sock *sk); 1746 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1747 void sk_free_unlock_clone(struct sock *sk); 1748 1749 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1750 gfp_t priority); 1751 void __sock_wfree(struct sk_buff *skb); 1752 void sock_wfree(struct sk_buff *skb); 1753 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1754 gfp_t priority); 1755 void skb_orphan_partial(struct sk_buff *skb); 1756 void sock_rfree(struct sk_buff *skb); 1757 void sock_efree(struct sk_buff *skb); 1758 #ifdef CONFIG_INET 1759 void sock_edemux(struct sk_buff *skb); 1760 void sock_pfree(struct sk_buff *skb); 1761 #else 1762 #define sock_edemux sock_efree 1763 #endif 1764 1765 int sock_setsockopt(struct socket *sock, int level, int op, 1766 sockptr_t optval, unsigned int optlen); 1767 1768 int sock_getsockopt(struct socket *sock, int level, int op, 1769 char __user *optval, int __user *optlen); 1770 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1771 bool timeval, bool time32); 1772 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1773 int noblock, int *errcode); 1774 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1775 unsigned long data_len, int noblock, 1776 int *errcode, int max_page_order); 1777 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1778 void sock_kfree_s(struct sock *sk, void *mem, int size); 1779 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1780 void sk_send_sigurg(struct sock *sk); 1781 1782 struct sockcm_cookie { 1783 u64 transmit_time; 1784 u32 mark; 1785 u16 tsflags; 1786 }; 1787 1788 static inline void sockcm_init(struct sockcm_cookie *sockc, 1789 const struct sock *sk) 1790 { 1791 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1792 } 1793 1794 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1795 struct sockcm_cookie *sockc); 1796 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1797 struct sockcm_cookie *sockc); 1798 1799 /* 1800 * Functions to fill in entries in struct proto_ops when a protocol 1801 * does not implement a particular function. 1802 */ 1803 int sock_no_bind(struct socket *, struct sockaddr *, int); 1804 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1805 int sock_no_socketpair(struct socket *, struct socket *); 1806 int sock_no_accept(struct socket *, struct socket *, int, bool); 1807 int sock_no_getname(struct socket *, struct sockaddr *, int); 1808 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1809 int sock_no_listen(struct socket *, int); 1810 int sock_no_shutdown(struct socket *, int); 1811 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1812 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1813 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1814 int sock_no_mmap(struct file *file, struct socket *sock, 1815 struct vm_area_struct *vma); 1816 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1817 size_t size, int flags); 1818 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1819 int offset, size_t size, int flags); 1820 1821 /* 1822 * Functions to fill in entries in struct proto_ops when a protocol 1823 * uses the inet style. 1824 */ 1825 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1826 char __user *optval, int __user *optlen); 1827 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1828 int flags); 1829 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1830 sockptr_t optval, unsigned int optlen); 1831 1832 void sk_common_release(struct sock *sk); 1833 1834 /* 1835 * Default socket callbacks and setup code 1836 */ 1837 1838 /* Initialise core socket variables */ 1839 void sock_init_data(struct socket *sock, struct sock *sk); 1840 1841 /* 1842 * Socket reference counting postulates. 1843 * 1844 * * Each user of socket SHOULD hold a reference count. 1845 * * Each access point to socket (an hash table bucket, reference from a list, 1846 * running timer, skb in flight MUST hold a reference count. 1847 * * When reference count hits 0, it means it will never increase back. 1848 * * When reference count hits 0, it means that no references from 1849 * outside exist to this socket and current process on current CPU 1850 * is last user and may/should destroy this socket. 1851 * * sk_free is called from any context: process, BH, IRQ. When 1852 * it is called, socket has no references from outside -> sk_free 1853 * may release descendant resources allocated by the socket, but 1854 * to the time when it is called, socket is NOT referenced by any 1855 * hash tables, lists etc. 1856 * * Packets, delivered from outside (from network or from another process) 1857 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1858 * when they sit in queue. Otherwise, packets will leak to hole, when 1859 * socket is looked up by one cpu and unhasing is made by another CPU. 1860 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1861 * (leak to backlog). Packet socket does all the processing inside 1862 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1863 * use separate SMP lock, so that they are prone too. 1864 */ 1865 1866 /* Ungrab socket and destroy it, if it was the last reference. */ 1867 static inline void sock_put(struct sock *sk) 1868 { 1869 if (refcount_dec_and_test(&sk->sk_refcnt)) 1870 sk_free(sk); 1871 } 1872 /* Generic version of sock_put(), dealing with all sockets 1873 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1874 */ 1875 void sock_gen_put(struct sock *sk); 1876 1877 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1878 unsigned int trim_cap, bool refcounted); 1879 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1880 const int nested) 1881 { 1882 return __sk_receive_skb(sk, skb, nested, 1, true); 1883 } 1884 1885 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1886 { 1887 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1888 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1889 return; 1890 sk->sk_tx_queue_mapping = tx_queue; 1891 } 1892 1893 #define NO_QUEUE_MAPPING USHRT_MAX 1894 1895 static inline void sk_tx_queue_clear(struct sock *sk) 1896 { 1897 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1898 } 1899 1900 static inline int sk_tx_queue_get(const struct sock *sk) 1901 { 1902 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1903 return sk->sk_tx_queue_mapping; 1904 1905 return -1; 1906 } 1907 1908 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1909 { 1910 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1911 if (skb_rx_queue_recorded(skb)) { 1912 u16 rx_queue = skb_get_rx_queue(skb); 1913 1914 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING)) 1915 return; 1916 1917 sk->sk_rx_queue_mapping = rx_queue; 1918 } 1919 #endif 1920 } 1921 1922 static inline void sk_rx_queue_clear(struct sock *sk) 1923 { 1924 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1925 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING; 1926 #endif 1927 } 1928 1929 static inline int sk_rx_queue_get(const struct sock *sk) 1930 { 1931 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1932 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING) 1933 return sk->sk_rx_queue_mapping; 1934 #endif 1935 1936 return -1; 1937 } 1938 1939 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1940 { 1941 sk->sk_socket = sock; 1942 } 1943 1944 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1945 { 1946 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1947 return &rcu_dereference_raw(sk->sk_wq)->wait; 1948 } 1949 /* Detach socket from process context. 1950 * Announce socket dead, detach it from wait queue and inode. 1951 * Note that parent inode held reference count on this struct sock, 1952 * we do not release it in this function, because protocol 1953 * probably wants some additional cleanups or even continuing 1954 * to work with this socket (TCP). 1955 */ 1956 static inline void sock_orphan(struct sock *sk) 1957 { 1958 write_lock_bh(&sk->sk_callback_lock); 1959 sock_set_flag(sk, SOCK_DEAD); 1960 sk_set_socket(sk, NULL); 1961 sk->sk_wq = NULL; 1962 write_unlock_bh(&sk->sk_callback_lock); 1963 } 1964 1965 static inline void sock_graft(struct sock *sk, struct socket *parent) 1966 { 1967 WARN_ON(parent->sk); 1968 write_lock_bh(&sk->sk_callback_lock); 1969 rcu_assign_pointer(sk->sk_wq, &parent->wq); 1970 parent->sk = sk; 1971 sk_set_socket(sk, parent); 1972 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1973 security_sock_graft(sk, parent); 1974 write_unlock_bh(&sk->sk_callback_lock); 1975 } 1976 1977 kuid_t sock_i_uid(struct sock *sk); 1978 unsigned long sock_i_ino(struct sock *sk); 1979 1980 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1981 { 1982 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1983 } 1984 1985 static inline u32 net_tx_rndhash(void) 1986 { 1987 u32 v = prandom_u32(); 1988 1989 return v ?: 1; 1990 } 1991 1992 static inline void sk_set_txhash(struct sock *sk) 1993 { 1994 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 1995 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 1996 } 1997 1998 static inline bool sk_rethink_txhash(struct sock *sk) 1999 { 2000 if (sk->sk_txhash) { 2001 sk_set_txhash(sk); 2002 return true; 2003 } 2004 return false; 2005 } 2006 2007 static inline struct dst_entry * 2008 __sk_dst_get(struct sock *sk) 2009 { 2010 return rcu_dereference_check(sk->sk_dst_cache, 2011 lockdep_sock_is_held(sk)); 2012 } 2013 2014 static inline struct dst_entry * 2015 sk_dst_get(struct sock *sk) 2016 { 2017 struct dst_entry *dst; 2018 2019 rcu_read_lock(); 2020 dst = rcu_dereference(sk->sk_dst_cache); 2021 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2022 dst = NULL; 2023 rcu_read_unlock(); 2024 return dst; 2025 } 2026 2027 static inline void __dst_negative_advice(struct sock *sk) 2028 { 2029 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2030 2031 if (dst && dst->ops->negative_advice) { 2032 ndst = dst->ops->negative_advice(dst); 2033 2034 if (ndst != dst) { 2035 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2036 sk_tx_queue_clear(sk); 2037 sk->sk_dst_pending_confirm = 0; 2038 } 2039 } 2040 } 2041 2042 static inline void dst_negative_advice(struct sock *sk) 2043 { 2044 sk_rethink_txhash(sk); 2045 __dst_negative_advice(sk); 2046 } 2047 2048 static inline void 2049 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2050 { 2051 struct dst_entry *old_dst; 2052 2053 sk_tx_queue_clear(sk); 2054 sk->sk_dst_pending_confirm = 0; 2055 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2056 lockdep_sock_is_held(sk)); 2057 rcu_assign_pointer(sk->sk_dst_cache, dst); 2058 dst_release(old_dst); 2059 } 2060 2061 static inline void 2062 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2063 { 2064 struct dst_entry *old_dst; 2065 2066 sk_tx_queue_clear(sk); 2067 sk->sk_dst_pending_confirm = 0; 2068 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2069 dst_release(old_dst); 2070 } 2071 2072 static inline void 2073 __sk_dst_reset(struct sock *sk) 2074 { 2075 __sk_dst_set(sk, NULL); 2076 } 2077 2078 static inline void 2079 sk_dst_reset(struct sock *sk) 2080 { 2081 sk_dst_set(sk, NULL); 2082 } 2083 2084 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2085 2086 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2087 2088 static inline void sk_dst_confirm(struct sock *sk) 2089 { 2090 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2091 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2092 } 2093 2094 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2095 { 2096 if (skb_get_dst_pending_confirm(skb)) { 2097 struct sock *sk = skb->sk; 2098 unsigned long now = jiffies; 2099 2100 /* avoid dirtying neighbour */ 2101 if (READ_ONCE(n->confirmed) != now) 2102 WRITE_ONCE(n->confirmed, now); 2103 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2104 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2105 } 2106 } 2107 2108 bool sk_mc_loop(struct sock *sk); 2109 2110 static inline bool sk_can_gso(const struct sock *sk) 2111 { 2112 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2113 } 2114 2115 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2116 2117 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 2118 { 2119 sk->sk_route_nocaps |= flags; 2120 sk->sk_route_caps &= ~flags; 2121 } 2122 2123 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2124 struct iov_iter *from, char *to, 2125 int copy, int offset) 2126 { 2127 if (skb->ip_summed == CHECKSUM_NONE) { 2128 __wsum csum = 0; 2129 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2130 return -EFAULT; 2131 skb->csum = csum_block_add(skb->csum, csum, offset); 2132 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2133 if (!copy_from_iter_full_nocache(to, copy, from)) 2134 return -EFAULT; 2135 } else if (!copy_from_iter_full(to, copy, from)) 2136 return -EFAULT; 2137 2138 return 0; 2139 } 2140 2141 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2142 struct iov_iter *from, int copy) 2143 { 2144 int err, offset = skb->len; 2145 2146 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2147 copy, offset); 2148 if (err) 2149 __skb_trim(skb, offset); 2150 2151 return err; 2152 } 2153 2154 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2155 struct sk_buff *skb, 2156 struct page *page, 2157 int off, int copy) 2158 { 2159 int err; 2160 2161 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2162 copy, skb->len); 2163 if (err) 2164 return err; 2165 2166 skb->len += copy; 2167 skb->data_len += copy; 2168 skb->truesize += copy; 2169 sk_wmem_queued_add(sk, copy); 2170 sk_mem_charge(sk, copy); 2171 return 0; 2172 } 2173 2174 /** 2175 * sk_wmem_alloc_get - returns write allocations 2176 * @sk: socket 2177 * 2178 * Return: sk_wmem_alloc minus initial offset of one 2179 */ 2180 static inline int sk_wmem_alloc_get(const struct sock *sk) 2181 { 2182 return refcount_read(&sk->sk_wmem_alloc) - 1; 2183 } 2184 2185 /** 2186 * sk_rmem_alloc_get - returns read allocations 2187 * @sk: socket 2188 * 2189 * Return: sk_rmem_alloc 2190 */ 2191 static inline int sk_rmem_alloc_get(const struct sock *sk) 2192 { 2193 return atomic_read(&sk->sk_rmem_alloc); 2194 } 2195 2196 /** 2197 * sk_has_allocations - check if allocations are outstanding 2198 * @sk: socket 2199 * 2200 * Return: true if socket has write or read allocations 2201 */ 2202 static inline bool sk_has_allocations(const struct sock *sk) 2203 { 2204 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2205 } 2206 2207 /** 2208 * skwq_has_sleeper - check if there are any waiting processes 2209 * @wq: struct socket_wq 2210 * 2211 * Return: true if socket_wq has waiting processes 2212 * 2213 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2214 * barrier call. They were added due to the race found within the tcp code. 2215 * 2216 * Consider following tcp code paths:: 2217 * 2218 * CPU1 CPU2 2219 * sys_select receive packet 2220 * ... ... 2221 * __add_wait_queue update tp->rcv_nxt 2222 * ... ... 2223 * tp->rcv_nxt check sock_def_readable 2224 * ... { 2225 * schedule rcu_read_lock(); 2226 * wq = rcu_dereference(sk->sk_wq); 2227 * if (wq && waitqueue_active(&wq->wait)) 2228 * wake_up_interruptible(&wq->wait) 2229 * ... 2230 * } 2231 * 2232 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2233 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2234 * could then endup calling schedule and sleep forever if there are no more 2235 * data on the socket. 2236 * 2237 */ 2238 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2239 { 2240 return wq && wq_has_sleeper(&wq->wait); 2241 } 2242 2243 /** 2244 * sock_poll_wait - place memory barrier behind the poll_wait call. 2245 * @filp: file 2246 * @sock: socket to wait on 2247 * @p: poll_table 2248 * 2249 * See the comments in the wq_has_sleeper function. 2250 */ 2251 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2252 poll_table *p) 2253 { 2254 if (!poll_does_not_wait(p)) { 2255 poll_wait(filp, &sock->wq.wait, p); 2256 /* We need to be sure we are in sync with the 2257 * socket flags modification. 2258 * 2259 * This memory barrier is paired in the wq_has_sleeper. 2260 */ 2261 smp_mb(); 2262 } 2263 } 2264 2265 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2266 { 2267 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2268 u32 txhash = READ_ONCE(sk->sk_txhash); 2269 2270 if (txhash) { 2271 skb->l4_hash = 1; 2272 skb->hash = txhash; 2273 } 2274 } 2275 2276 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2277 2278 /* 2279 * Queue a received datagram if it will fit. Stream and sequenced 2280 * protocols can't normally use this as they need to fit buffers in 2281 * and play with them. 2282 * 2283 * Inlined as it's very short and called for pretty much every 2284 * packet ever received. 2285 */ 2286 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2287 { 2288 skb_orphan(skb); 2289 skb->sk = sk; 2290 skb->destructor = sock_rfree; 2291 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2292 sk_mem_charge(sk, skb->truesize); 2293 } 2294 2295 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2296 { 2297 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2298 skb_orphan(skb); 2299 skb->destructor = sock_efree; 2300 skb->sk = sk; 2301 return true; 2302 } 2303 return false; 2304 } 2305 2306 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2307 { 2308 if (skb->destructor != sock_wfree) { 2309 skb_orphan(skb); 2310 return; 2311 } 2312 skb->slow_gro = 1; 2313 } 2314 2315 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2316 unsigned long expires); 2317 2318 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2319 2320 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2321 2322 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2323 struct sk_buff *skb, unsigned int flags, 2324 void (*destructor)(struct sock *sk, 2325 struct sk_buff *skb)); 2326 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2327 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2328 2329 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2330 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2331 2332 /* 2333 * Recover an error report and clear atomically 2334 */ 2335 2336 static inline int sock_error(struct sock *sk) 2337 { 2338 int err; 2339 2340 /* Avoid an atomic operation for the common case. 2341 * This is racy since another cpu/thread can change sk_err under us. 2342 */ 2343 if (likely(data_race(!sk->sk_err))) 2344 return 0; 2345 2346 err = xchg(&sk->sk_err, 0); 2347 return -err; 2348 } 2349 2350 void sk_error_report(struct sock *sk); 2351 2352 static inline unsigned long sock_wspace(struct sock *sk) 2353 { 2354 int amt = 0; 2355 2356 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2357 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2358 if (amt < 0) 2359 amt = 0; 2360 } 2361 return amt; 2362 } 2363 2364 /* Note: 2365 * We use sk->sk_wq_raw, from contexts knowing this 2366 * pointer is not NULL and cannot disappear/change. 2367 */ 2368 static inline void sk_set_bit(int nr, struct sock *sk) 2369 { 2370 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2371 !sock_flag(sk, SOCK_FASYNC)) 2372 return; 2373 2374 set_bit(nr, &sk->sk_wq_raw->flags); 2375 } 2376 2377 static inline void sk_clear_bit(int nr, struct sock *sk) 2378 { 2379 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2380 !sock_flag(sk, SOCK_FASYNC)) 2381 return; 2382 2383 clear_bit(nr, &sk->sk_wq_raw->flags); 2384 } 2385 2386 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2387 { 2388 if (sock_flag(sk, SOCK_FASYNC)) { 2389 rcu_read_lock(); 2390 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2391 rcu_read_unlock(); 2392 } 2393 } 2394 2395 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2396 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2397 * Note: for send buffers, TCP works better if we can build two skbs at 2398 * minimum. 2399 */ 2400 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2401 2402 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2403 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2404 2405 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2406 { 2407 u32 val; 2408 2409 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2410 return; 2411 2412 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2413 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2414 2415 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2416 } 2417 2418 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2419 bool force_schedule); 2420 2421 /** 2422 * sk_page_frag - return an appropriate page_frag 2423 * @sk: socket 2424 * 2425 * Use the per task page_frag instead of the per socket one for 2426 * optimization when we know that we're in the normal context and owns 2427 * everything that's associated with %current. 2428 * 2429 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest 2430 * inside other socket operations and end up recursing into sk_page_frag() 2431 * while it's already in use. 2432 * 2433 * Return: a per task page_frag if context allows that, 2434 * otherwise a per socket one. 2435 */ 2436 static inline struct page_frag *sk_page_frag(struct sock *sk) 2437 { 2438 if (gfpflags_normal_context(sk->sk_allocation)) 2439 return ¤t->task_frag; 2440 2441 return &sk->sk_frag; 2442 } 2443 2444 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2445 2446 /* 2447 * Default write policy as shown to user space via poll/select/SIGIO 2448 */ 2449 static inline bool sock_writeable(const struct sock *sk) 2450 { 2451 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2452 } 2453 2454 static inline gfp_t gfp_any(void) 2455 { 2456 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2457 } 2458 2459 static inline gfp_t gfp_memcg_charge(void) 2460 { 2461 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2462 } 2463 2464 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2465 { 2466 return noblock ? 0 : sk->sk_rcvtimeo; 2467 } 2468 2469 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2470 { 2471 return noblock ? 0 : sk->sk_sndtimeo; 2472 } 2473 2474 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2475 { 2476 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2477 2478 return v ?: 1; 2479 } 2480 2481 /* Alas, with timeout socket operations are not restartable. 2482 * Compare this to poll(). 2483 */ 2484 static inline int sock_intr_errno(long timeo) 2485 { 2486 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2487 } 2488 2489 struct sock_skb_cb { 2490 u32 dropcount; 2491 }; 2492 2493 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2494 * using skb->cb[] would keep using it directly and utilize its 2495 * alignement guarantee. 2496 */ 2497 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2498 sizeof(struct sock_skb_cb))) 2499 2500 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2501 SOCK_SKB_CB_OFFSET)) 2502 2503 #define sock_skb_cb_check_size(size) \ 2504 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2505 2506 static inline void 2507 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2508 { 2509 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2510 atomic_read(&sk->sk_drops) : 0; 2511 } 2512 2513 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2514 { 2515 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2516 2517 atomic_add(segs, &sk->sk_drops); 2518 } 2519 2520 static inline ktime_t sock_read_timestamp(struct sock *sk) 2521 { 2522 #if BITS_PER_LONG==32 2523 unsigned int seq; 2524 ktime_t kt; 2525 2526 do { 2527 seq = read_seqbegin(&sk->sk_stamp_seq); 2528 kt = sk->sk_stamp; 2529 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2530 2531 return kt; 2532 #else 2533 return READ_ONCE(sk->sk_stamp); 2534 #endif 2535 } 2536 2537 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2538 { 2539 #if BITS_PER_LONG==32 2540 write_seqlock(&sk->sk_stamp_seq); 2541 sk->sk_stamp = kt; 2542 write_sequnlock(&sk->sk_stamp_seq); 2543 #else 2544 WRITE_ONCE(sk->sk_stamp, kt); 2545 #endif 2546 } 2547 2548 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2549 struct sk_buff *skb); 2550 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2551 struct sk_buff *skb); 2552 2553 static inline void 2554 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2555 { 2556 ktime_t kt = skb->tstamp; 2557 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2558 2559 /* 2560 * generate control messages if 2561 * - receive time stamping in software requested 2562 * - software time stamp available and wanted 2563 * - hardware time stamps available and wanted 2564 */ 2565 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2566 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2567 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2568 (hwtstamps->hwtstamp && 2569 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2570 __sock_recv_timestamp(msg, sk, skb); 2571 else 2572 sock_write_timestamp(sk, kt); 2573 2574 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2575 __sock_recv_wifi_status(msg, sk, skb); 2576 } 2577 2578 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2579 struct sk_buff *skb); 2580 2581 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2582 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2583 struct sk_buff *skb) 2584 { 2585 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2586 (1UL << SOCK_RCVTSTAMP)) 2587 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2588 SOF_TIMESTAMPING_RAW_HARDWARE) 2589 2590 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2591 __sock_recv_ts_and_drops(msg, sk, skb); 2592 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2593 sock_write_timestamp(sk, skb->tstamp); 2594 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2595 sock_write_timestamp(sk, 0); 2596 } 2597 2598 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2599 2600 /** 2601 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2602 * @sk: socket sending this packet 2603 * @tsflags: timestamping flags to use 2604 * @tx_flags: completed with instructions for time stamping 2605 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2606 * 2607 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2608 */ 2609 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2610 __u8 *tx_flags, __u32 *tskey) 2611 { 2612 if (unlikely(tsflags)) { 2613 __sock_tx_timestamp(tsflags, tx_flags); 2614 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2615 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2616 *tskey = sk->sk_tskey++; 2617 } 2618 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2619 *tx_flags |= SKBTX_WIFI_STATUS; 2620 } 2621 2622 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2623 __u8 *tx_flags) 2624 { 2625 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2626 } 2627 2628 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2629 { 2630 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2631 &skb_shinfo(skb)->tskey); 2632 } 2633 2634 /** 2635 * sk_eat_skb - Release a skb if it is no longer needed 2636 * @sk: socket to eat this skb from 2637 * @skb: socket buffer to eat 2638 * 2639 * This routine must be called with interrupts disabled or with the socket 2640 * locked so that the sk_buff queue operation is ok. 2641 */ 2642 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2643 { 2644 __skb_unlink(skb, &sk->sk_receive_queue); 2645 __kfree_skb(skb); 2646 } 2647 2648 static inline 2649 struct net *sock_net(const struct sock *sk) 2650 { 2651 return read_pnet(&sk->sk_net); 2652 } 2653 2654 static inline 2655 void sock_net_set(struct sock *sk, struct net *net) 2656 { 2657 write_pnet(&sk->sk_net, net); 2658 } 2659 2660 static inline bool 2661 skb_sk_is_prefetched(struct sk_buff *skb) 2662 { 2663 #ifdef CONFIG_INET 2664 return skb->destructor == sock_pfree; 2665 #else 2666 return false; 2667 #endif /* CONFIG_INET */ 2668 } 2669 2670 /* This helper checks if a socket is a full socket, 2671 * ie _not_ a timewait or request socket. 2672 */ 2673 static inline bool sk_fullsock(const struct sock *sk) 2674 { 2675 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2676 } 2677 2678 static inline bool 2679 sk_is_refcounted(struct sock *sk) 2680 { 2681 /* Only full sockets have sk->sk_flags. */ 2682 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2683 } 2684 2685 /** 2686 * skb_steal_sock - steal a socket from an sk_buff 2687 * @skb: sk_buff to steal the socket from 2688 * @refcounted: is set to true if the socket is reference-counted 2689 */ 2690 static inline struct sock * 2691 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2692 { 2693 if (skb->sk) { 2694 struct sock *sk = skb->sk; 2695 2696 *refcounted = true; 2697 if (skb_sk_is_prefetched(skb)) 2698 *refcounted = sk_is_refcounted(sk); 2699 skb->destructor = NULL; 2700 skb->sk = NULL; 2701 return sk; 2702 } 2703 *refcounted = false; 2704 return NULL; 2705 } 2706 2707 /* Checks if this SKB belongs to an HW offloaded socket 2708 * and whether any SW fallbacks are required based on dev. 2709 * Check decrypted mark in case skb_orphan() cleared socket. 2710 */ 2711 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2712 struct net_device *dev) 2713 { 2714 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2715 struct sock *sk = skb->sk; 2716 2717 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2718 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2719 #ifdef CONFIG_TLS_DEVICE 2720 } else if (unlikely(skb->decrypted)) { 2721 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2722 kfree_skb(skb); 2723 skb = NULL; 2724 #endif 2725 } 2726 #endif 2727 2728 return skb; 2729 } 2730 2731 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2732 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2733 */ 2734 static inline bool sk_listener(const struct sock *sk) 2735 { 2736 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2737 } 2738 2739 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2740 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2741 int type); 2742 2743 bool sk_ns_capable(const struct sock *sk, 2744 struct user_namespace *user_ns, int cap); 2745 bool sk_capable(const struct sock *sk, int cap); 2746 bool sk_net_capable(const struct sock *sk, int cap); 2747 2748 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2749 2750 /* Take into consideration the size of the struct sk_buff overhead in the 2751 * determination of these values, since that is non-constant across 2752 * platforms. This makes socket queueing behavior and performance 2753 * not depend upon such differences. 2754 */ 2755 #define _SK_MEM_PACKETS 256 2756 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2757 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2758 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2759 2760 extern __u32 sysctl_wmem_max; 2761 extern __u32 sysctl_rmem_max; 2762 2763 extern int sysctl_tstamp_allow_data; 2764 extern int sysctl_optmem_max; 2765 2766 extern __u32 sysctl_wmem_default; 2767 extern __u32 sysctl_rmem_default; 2768 2769 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2770 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2771 2772 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2773 { 2774 /* Does this proto have per netns sysctl_wmem ? */ 2775 if (proto->sysctl_wmem_offset) 2776 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2777 2778 return *proto->sysctl_wmem; 2779 } 2780 2781 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2782 { 2783 /* Does this proto have per netns sysctl_rmem ? */ 2784 if (proto->sysctl_rmem_offset) 2785 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2786 2787 return *proto->sysctl_rmem; 2788 } 2789 2790 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2791 * Some wifi drivers need to tweak it to get more chunks. 2792 * They can use this helper from their ndo_start_xmit() 2793 */ 2794 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2795 { 2796 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2797 return; 2798 WRITE_ONCE(sk->sk_pacing_shift, val); 2799 } 2800 2801 /* if a socket is bound to a device, check that the given device 2802 * index is either the same or that the socket is bound to an L3 2803 * master device and the given device index is also enslaved to 2804 * that L3 master 2805 */ 2806 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2807 { 2808 int mdif; 2809 2810 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2811 return true; 2812 2813 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2814 if (mdif && mdif == sk->sk_bound_dev_if) 2815 return true; 2816 2817 return false; 2818 } 2819 2820 void sock_def_readable(struct sock *sk); 2821 2822 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2823 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2824 int sock_set_timestamping(struct sock *sk, int optname, 2825 struct so_timestamping timestamping); 2826 2827 void sock_enable_timestamps(struct sock *sk); 2828 void sock_no_linger(struct sock *sk); 2829 void sock_set_keepalive(struct sock *sk); 2830 void sock_set_priority(struct sock *sk, u32 priority); 2831 void sock_set_rcvbuf(struct sock *sk, int val); 2832 void sock_set_mark(struct sock *sk, u32 val); 2833 void sock_set_reuseaddr(struct sock *sk); 2834 void sock_set_reuseport(struct sock *sk); 2835 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2836 2837 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2838 2839 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2840 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2841 sockptr_t optval, int optlen, bool old_timeval); 2842 2843 #endif /* _SOCK_H */ 2844