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 <linux/llist.h> 67 #include <net/dst.h> 68 #include <net/checksum.h> 69 #include <net/tcp_states.h> 70 #include <linux/net_tstamp.h> 71 #include <net/l3mdev.h> 72 #include <uapi/linux/socket.h> 73 74 /* 75 * This structure really needs to be cleaned up. 76 * Most of it is for TCP, and not used by any of 77 * the other protocols. 78 */ 79 80 /* Define this to get the SOCK_DBG debugging facility. */ 81 #define SOCK_DEBUGGING 82 #ifdef SOCK_DEBUGGING 83 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 84 printk(KERN_DEBUG msg); } while (0) 85 #else 86 /* Validate arguments and do nothing */ 87 static inline __printf(2, 3) 88 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 89 { 90 } 91 #endif 92 93 /* This is the per-socket lock. The spinlock provides a synchronization 94 * between user contexts and software interrupt processing, whereas the 95 * mini-semaphore synchronizes multiple users amongst themselves. 96 */ 97 typedef struct { 98 spinlock_t slock; 99 int owned; 100 wait_queue_head_t wq; 101 /* 102 * We express the mutex-alike socket_lock semantics 103 * to the lock validator by explicitly managing 104 * the slock as a lock variant (in addition to 105 * the slock itself): 106 */ 107 #ifdef CONFIG_DEBUG_LOCK_ALLOC 108 struct lockdep_map dep_map; 109 #endif 110 } socket_lock_t; 111 112 struct sock; 113 struct proto; 114 struct net; 115 116 typedef __u32 __bitwise __portpair; 117 typedef __u64 __bitwise __addrpair; 118 119 /** 120 * struct sock_common - minimal network layer representation of sockets 121 * @skc_daddr: Foreign IPv4 addr 122 * @skc_rcv_saddr: Bound local IPv4 addr 123 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 124 * @skc_hash: hash value used with various protocol lookup tables 125 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 126 * @skc_dport: placeholder for inet_dport/tw_dport 127 * @skc_num: placeholder for inet_num/tw_num 128 * @skc_portpair: __u32 union of @skc_dport & @skc_num 129 * @skc_family: network address family 130 * @skc_state: Connection state 131 * @skc_reuse: %SO_REUSEADDR setting 132 * @skc_reuseport: %SO_REUSEPORT setting 133 * @skc_ipv6only: socket is IPV6 only 134 * @skc_net_refcnt: socket is using net ref counting 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_v6_daddr: IPV6 destination address 141 * @skc_v6_rcv_saddr: IPV6 source address 142 * @skc_cookie: socket's cookie value 143 * @skc_node: main hash linkage for various protocol lookup tables 144 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 145 * @skc_tx_queue_mapping: tx queue number for this connection 146 * @skc_rx_queue_mapping: rx queue number for this connection 147 * @skc_flags: place holder for sk_flags 148 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 149 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 150 * @skc_listener: connection request listener socket (aka rsk_listener) 151 * [union with @skc_flags] 152 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 153 * [union with @skc_flags] 154 * @skc_incoming_cpu: record/match cpu processing incoming packets 155 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 156 * [union with @skc_incoming_cpu] 157 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 158 * [union with @skc_incoming_cpu] 159 * @skc_refcnt: reference count 160 * 161 * This is the minimal network layer representation of sockets, the header 162 * for struct sock and struct inet_timewait_sock. 163 */ 164 struct sock_common { 165 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 166 * address on 64bit arches : cf INET_MATCH() 167 */ 168 union { 169 __addrpair skc_addrpair; 170 struct { 171 __be32 skc_daddr; 172 __be32 skc_rcv_saddr; 173 }; 174 }; 175 union { 176 unsigned int skc_hash; 177 __u16 skc_u16hashes[2]; 178 }; 179 /* skc_dport && skc_num must be grouped as well */ 180 union { 181 __portpair skc_portpair; 182 struct { 183 __be16 skc_dport; 184 __u16 skc_num; 185 }; 186 }; 187 188 unsigned short skc_family; 189 volatile unsigned char skc_state; 190 unsigned char skc_reuse:4; 191 unsigned char skc_reuseport:1; 192 unsigned char skc_ipv6only:1; 193 unsigned char skc_net_refcnt:1; 194 int skc_bound_dev_if; 195 union { 196 struct hlist_node skc_bind_node; 197 struct hlist_node skc_portaddr_node; 198 }; 199 struct proto *skc_prot; 200 possible_net_t skc_net; 201 202 #if IS_ENABLED(CONFIG_IPV6) 203 struct in6_addr skc_v6_daddr; 204 struct in6_addr skc_v6_rcv_saddr; 205 #endif 206 207 atomic64_t skc_cookie; 208 209 /* following fields are padding to force 210 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 211 * assuming IPV6 is enabled. We use this padding differently 212 * for different kind of 'sockets' 213 */ 214 union { 215 unsigned long skc_flags; 216 struct sock *skc_listener; /* request_sock */ 217 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 218 }; 219 /* 220 * fields between dontcopy_begin/dontcopy_end 221 * are not copied in sock_copy() 222 */ 223 /* private: */ 224 int skc_dontcopy_begin[0]; 225 /* public: */ 226 union { 227 struct hlist_node skc_node; 228 struct hlist_nulls_node skc_nulls_node; 229 }; 230 unsigned short skc_tx_queue_mapping; 231 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 232 unsigned short skc_rx_queue_mapping; 233 #endif 234 union { 235 int skc_incoming_cpu; 236 u32 skc_rcv_wnd; 237 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 238 }; 239 240 refcount_t skc_refcnt; 241 /* private: */ 242 int skc_dontcopy_end[0]; 243 union { 244 u32 skc_rxhash; 245 u32 skc_window_clamp; 246 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 247 }; 248 /* public: */ 249 }; 250 251 struct bpf_local_storage; 252 253 /** 254 * struct sock - network layer representation of sockets 255 * @__sk_common: shared layout with inet_timewait_sock 256 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 257 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 258 * @sk_lock: synchronizer 259 * @sk_kern_sock: True if sock is using kernel lock classes 260 * @sk_rcvbuf: size of receive buffer in bytes 261 * @sk_wq: sock wait queue and async head 262 * @sk_rx_dst: receive input route used by early demux 263 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 264 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 265 * @sk_dst_cache: destination cache 266 * @sk_dst_pending_confirm: need to confirm neighbour 267 * @sk_policy: flow policy 268 * @sk_receive_queue: incoming packets 269 * @sk_wmem_alloc: transmit queue bytes committed 270 * @sk_tsq_flags: TCP Small Queues flags 271 * @sk_write_queue: Packet sending queue 272 * @sk_omem_alloc: "o" is "option" or "other" 273 * @sk_wmem_queued: persistent queue size 274 * @sk_forward_alloc: space allocated forward 275 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 276 * @sk_napi_id: id of the last napi context to receive data for sk 277 * @sk_ll_usec: usecs to busypoll when there is no data 278 * @sk_allocation: allocation mode 279 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 280 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 281 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 282 * @sk_sndbuf: size of send buffer in bytes 283 * @__sk_flags_offset: empty field used to determine location of bitfield 284 * @sk_padding: unused element for alignment 285 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 286 * @sk_no_check_rx: allow zero checksum in RX packets 287 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 288 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 289 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 290 * @sk_gso_max_size: Maximum GSO segment size to build 291 * @sk_gso_max_segs: Maximum number of GSO segments 292 * @sk_pacing_shift: scaling factor for TCP Small Queues 293 * @sk_lingertime: %SO_LINGER l_linger setting 294 * @sk_backlog: always used with the per-socket spinlock held 295 * @defer_list: head of llist storing skbs to be freed 296 * @sk_callback_lock: used with the callbacks in the end of this struct 297 * @sk_error_queue: rarely used 298 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 299 * IPV6_ADDRFORM for instance) 300 * @sk_err: last error 301 * @sk_err_soft: errors that don't cause failure but are the cause of a 302 * persistent failure not just 'timed out' 303 * @sk_drops: raw/udp drops counter 304 * @sk_ack_backlog: current listen backlog 305 * @sk_max_ack_backlog: listen backlog set in listen() 306 * @sk_uid: user id of owner 307 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 308 * @sk_busy_poll_budget: napi processing budget when busypolling 309 * @sk_priority: %SO_PRIORITY setting 310 * @sk_type: socket type (%SOCK_STREAM, etc) 311 * @sk_protocol: which protocol this socket belongs in this network family 312 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 313 * @sk_peer_pid: &struct pid for this socket's peer 314 * @sk_peer_cred: %SO_PEERCRED setting 315 * @sk_rcvlowat: %SO_RCVLOWAT setting 316 * @sk_rcvtimeo: %SO_RCVTIMEO setting 317 * @sk_sndtimeo: %SO_SNDTIMEO setting 318 * @sk_txhash: computed flow hash for use on transmit 319 * @sk_filter: socket filtering instructions 320 * @sk_timer: sock cleanup timer 321 * @sk_stamp: time stamp of last packet received 322 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 323 * @sk_tsflags: SO_TIMESTAMPING flags 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_state_change: callback to indicate change in the state of the sock 340 * @sk_data_ready: callback to indicate there is data to be processed 341 * @sk_write_space: callback to indicate there is bf sending space available 342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 343 * @sk_backlog_rcv: callback to process the backlog 344 * @sk_validate_xmit_skb: ptr to an optional validate function 345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 346 * @sk_reuseport_cb: reuseport group container 347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 348 * @sk_rcu: used during RCU grace period 349 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 350 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 351 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 352 * @sk_txtime_unused: unused txtime flags 353 */ 354 struct sock { 355 /* 356 * Now struct inet_timewait_sock also uses sock_common, so please just 357 * don't add nothing before this first member (__sk_common) --acme 358 */ 359 struct sock_common __sk_common; 360 #define sk_node __sk_common.skc_node 361 #define sk_nulls_node __sk_common.skc_nulls_node 362 #define sk_refcnt __sk_common.skc_refcnt 363 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 364 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 365 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 366 #endif 367 368 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 369 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 370 #define sk_hash __sk_common.skc_hash 371 #define sk_portpair __sk_common.skc_portpair 372 #define sk_num __sk_common.skc_num 373 #define sk_dport __sk_common.skc_dport 374 #define sk_addrpair __sk_common.skc_addrpair 375 #define sk_daddr __sk_common.skc_daddr 376 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 377 #define sk_family __sk_common.skc_family 378 #define sk_state __sk_common.skc_state 379 #define sk_reuse __sk_common.skc_reuse 380 #define sk_reuseport __sk_common.skc_reuseport 381 #define sk_ipv6only __sk_common.skc_ipv6only 382 #define sk_net_refcnt __sk_common.skc_net_refcnt 383 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 384 #define sk_bind_node __sk_common.skc_bind_node 385 #define sk_prot __sk_common.skc_prot 386 #define sk_net __sk_common.skc_net 387 #define sk_v6_daddr __sk_common.skc_v6_daddr 388 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 389 #define sk_cookie __sk_common.skc_cookie 390 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 391 #define sk_flags __sk_common.skc_flags 392 #define sk_rxhash __sk_common.skc_rxhash 393 394 /* early demux fields */ 395 struct dst_entry *sk_rx_dst; 396 int sk_rx_dst_ifindex; 397 u32 sk_rx_dst_cookie; 398 399 socket_lock_t sk_lock; 400 atomic_t sk_drops; 401 int sk_rcvlowat; 402 struct sk_buff_head sk_error_queue; 403 struct sk_buff_head sk_receive_queue; 404 /* 405 * The backlog queue is special, it is always used with 406 * the per-socket spinlock held and requires low latency 407 * access. Therefore we special case it's implementation. 408 * Note : rmem_alloc is in this structure to fill a hole 409 * on 64bit arches, not because its logically part of 410 * backlog. 411 */ 412 struct { 413 atomic_t rmem_alloc; 414 int len; 415 struct sk_buff *head; 416 struct sk_buff *tail; 417 } sk_backlog; 418 struct llist_head defer_list; 419 420 #define sk_rmem_alloc sk_backlog.rmem_alloc 421 422 int sk_forward_alloc; 423 u32 sk_reserved_mem; 424 #ifdef CONFIG_NET_RX_BUSY_POLL 425 unsigned int sk_ll_usec; 426 /* ===== mostly read cache line ===== */ 427 unsigned int sk_napi_id; 428 #endif 429 int sk_rcvbuf; 430 431 struct sk_filter __rcu *sk_filter; 432 union { 433 struct socket_wq __rcu *sk_wq; 434 /* private: */ 435 struct socket_wq *sk_wq_raw; 436 /* public: */ 437 }; 438 #ifdef CONFIG_XFRM 439 struct xfrm_policy __rcu *sk_policy[2]; 440 #endif 441 442 struct dst_entry __rcu *sk_dst_cache; 443 atomic_t sk_omem_alloc; 444 int sk_sndbuf; 445 446 /* ===== cache line for TX ===== */ 447 int sk_wmem_queued; 448 refcount_t sk_wmem_alloc; 449 unsigned long sk_tsq_flags; 450 union { 451 struct sk_buff *sk_send_head; 452 struct rb_root tcp_rtx_queue; 453 }; 454 struct sk_buff_head sk_write_queue; 455 __s32 sk_peek_off; 456 int sk_write_pending; 457 __u32 sk_dst_pending_confirm; 458 u32 sk_pacing_status; /* see enum sk_pacing */ 459 long sk_sndtimeo; 460 struct timer_list sk_timer; 461 __u32 sk_priority; 462 __u32 sk_mark; 463 unsigned long sk_pacing_rate; /* bytes per second */ 464 unsigned long sk_max_pacing_rate; 465 struct page_frag sk_frag; 466 netdev_features_t sk_route_caps; 467 int sk_gso_type; 468 unsigned int sk_gso_max_size; 469 gfp_t sk_allocation; 470 __u32 sk_txhash; 471 472 /* 473 * Because of non atomicity rules, all 474 * changes are protected by socket lock. 475 */ 476 u8 sk_gso_disabled : 1, 477 sk_kern_sock : 1, 478 sk_no_check_tx : 1, 479 sk_no_check_rx : 1, 480 sk_userlocks : 4; 481 u8 sk_pacing_shift; 482 u16 sk_type; 483 u16 sk_protocol; 484 u16 sk_gso_max_segs; 485 unsigned long sk_lingertime; 486 struct proto *sk_prot_creator; 487 rwlock_t sk_callback_lock; 488 int sk_err, 489 sk_err_soft; 490 u32 sk_ack_backlog; 491 u32 sk_max_ack_backlog; 492 kuid_t sk_uid; 493 #ifdef CONFIG_NET_RX_BUSY_POLL 494 u8 sk_prefer_busy_poll; 495 u16 sk_busy_poll_budget; 496 #endif 497 spinlock_t sk_peer_lock; 498 int sk_bind_phc; 499 struct pid *sk_peer_pid; 500 const struct cred *sk_peer_cred; 501 502 long sk_rcvtimeo; 503 ktime_t sk_stamp; 504 #if BITS_PER_LONG==32 505 seqlock_t sk_stamp_seq; 506 #endif 507 u16 sk_tsflags; 508 u8 sk_shutdown; 509 u32 sk_tskey; 510 atomic_t sk_zckey; 511 512 u8 sk_clockid; 513 u8 sk_txtime_deadline_mode : 1, 514 sk_txtime_report_errors : 1, 515 sk_txtime_unused : 6; 516 517 struct socket *sk_socket; 518 void *sk_user_data; 519 #ifdef CONFIG_SECURITY 520 void *sk_security; 521 #endif 522 struct sock_cgroup_data sk_cgrp_data; 523 struct mem_cgroup *sk_memcg; 524 void (*sk_state_change)(struct sock *sk); 525 void (*sk_data_ready)(struct sock *sk); 526 void (*sk_write_space)(struct sock *sk); 527 void (*sk_error_report)(struct sock *sk); 528 int (*sk_backlog_rcv)(struct sock *sk, 529 struct sk_buff *skb); 530 #ifdef CONFIG_SOCK_VALIDATE_XMIT 531 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 532 struct net_device *dev, 533 struct sk_buff *skb); 534 #endif 535 void (*sk_destruct)(struct sock *sk); 536 struct sock_reuseport __rcu *sk_reuseport_cb; 537 #ifdef CONFIG_BPF_SYSCALL 538 struct bpf_local_storage __rcu *sk_bpf_storage; 539 #endif 540 struct rcu_head sk_rcu; 541 }; 542 543 enum sk_pacing { 544 SK_PACING_NONE = 0, 545 SK_PACING_NEEDED = 1, 546 SK_PACING_FQ = 2, 547 }; 548 549 /* Pointer stored in sk_user_data might not be suitable for copying 550 * when cloning the socket. For instance, it can point to a reference 551 * counted object. sk_user_data bottom bit is set if pointer must not 552 * be copied. 553 */ 554 #define SK_USER_DATA_NOCOPY 1UL 555 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 556 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 557 558 /** 559 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 560 * @sk: socket 561 */ 562 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 563 { 564 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 565 } 566 567 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 568 569 #define rcu_dereference_sk_user_data(sk) \ 570 ({ \ 571 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 572 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 573 }) 574 #define rcu_assign_sk_user_data(sk, ptr) \ 575 ({ \ 576 uintptr_t __tmp = (uintptr_t)(ptr); \ 577 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 578 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 579 }) 580 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 581 ({ \ 582 uintptr_t __tmp = (uintptr_t)(ptr); \ 583 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 584 rcu_assign_pointer(__sk_user_data((sk)), \ 585 __tmp | SK_USER_DATA_NOCOPY); \ 586 }) 587 588 /* 589 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 590 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 591 * on a socket means that the socket will reuse everybody else's port 592 * without looking at the other's sk_reuse value. 593 */ 594 595 #define SK_NO_REUSE 0 596 #define SK_CAN_REUSE 1 597 #define SK_FORCE_REUSE 2 598 599 int sk_set_peek_off(struct sock *sk, int val); 600 601 static inline int sk_peek_offset(struct sock *sk, int flags) 602 { 603 if (unlikely(flags & MSG_PEEK)) { 604 return READ_ONCE(sk->sk_peek_off); 605 } 606 607 return 0; 608 } 609 610 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 611 { 612 s32 off = READ_ONCE(sk->sk_peek_off); 613 614 if (unlikely(off >= 0)) { 615 off = max_t(s32, off - val, 0); 616 WRITE_ONCE(sk->sk_peek_off, off); 617 } 618 } 619 620 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 621 { 622 sk_peek_offset_bwd(sk, -val); 623 } 624 625 /* 626 * Hashed lists helper routines 627 */ 628 static inline struct sock *sk_entry(const struct hlist_node *node) 629 { 630 return hlist_entry(node, struct sock, sk_node); 631 } 632 633 static inline struct sock *__sk_head(const struct hlist_head *head) 634 { 635 return hlist_entry(head->first, struct sock, sk_node); 636 } 637 638 static inline struct sock *sk_head(const struct hlist_head *head) 639 { 640 return hlist_empty(head) ? NULL : __sk_head(head); 641 } 642 643 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 644 { 645 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 646 } 647 648 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 649 { 650 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 651 } 652 653 static inline struct sock *sk_next(const struct sock *sk) 654 { 655 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 656 } 657 658 static inline struct sock *sk_nulls_next(const struct sock *sk) 659 { 660 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 661 hlist_nulls_entry(sk->sk_nulls_node.next, 662 struct sock, sk_nulls_node) : 663 NULL; 664 } 665 666 static inline bool sk_unhashed(const struct sock *sk) 667 { 668 return hlist_unhashed(&sk->sk_node); 669 } 670 671 static inline bool sk_hashed(const struct sock *sk) 672 { 673 return !sk_unhashed(sk); 674 } 675 676 static inline void sk_node_init(struct hlist_node *node) 677 { 678 node->pprev = NULL; 679 } 680 681 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 682 { 683 node->pprev = NULL; 684 } 685 686 static inline void __sk_del_node(struct sock *sk) 687 { 688 __hlist_del(&sk->sk_node); 689 } 690 691 /* NB: equivalent to hlist_del_init_rcu */ 692 static inline bool __sk_del_node_init(struct sock *sk) 693 { 694 if (sk_hashed(sk)) { 695 __sk_del_node(sk); 696 sk_node_init(&sk->sk_node); 697 return true; 698 } 699 return false; 700 } 701 702 /* Grab socket reference count. This operation is valid only 703 when sk is ALREADY grabbed f.e. it is found in hash table 704 or a list and the lookup is made under lock preventing hash table 705 modifications. 706 */ 707 708 static __always_inline void sock_hold(struct sock *sk) 709 { 710 refcount_inc(&sk->sk_refcnt); 711 } 712 713 /* Ungrab socket in the context, which assumes that socket refcnt 714 cannot hit zero, f.e. it is true in context of any socketcall. 715 */ 716 static __always_inline void __sock_put(struct sock *sk) 717 { 718 refcount_dec(&sk->sk_refcnt); 719 } 720 721 static inline bool sk_del_node_init(struct sock *sk) 722 { 723 bool rc = __sk_del_node_init(sk); 724 725 if (rc) { 726 /* paranoid for a while -acme */ 727 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 728 __sock_put(sk); 729 } 730 return rc; 731 } 732 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 733 734 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 735 { 736 if (sk_hashed(sk)) { 737 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 738 return true; 739 } 740 return false; 741 } 742 743 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 744 { 745 bool rc = __sk_nulls_del_node_init_rcu(sk); 746 747 if (rc) { 748 /* paranoid for a while -acme */ 749 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 750 __sock_put(sk); 751 } 752 return rc; 753 } 754 755 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 756 { 757 hlist_add_head(&sk->sk_node, list); 758 } 759 760 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 761 { 762 sock_hold(sk); 763 __sk_add_node(sk, list); 764 } 765 766 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 767 { 768 sock_hold(sk); 769 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 770 sk->sk_family == AF_INET6) 771 hlist_add_tail_rcu(&sk->sk_node, list); 772 else 773 hlist_add_head_rcu(&sk->sk_node, list); 774 } 775 776 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 777 { 778 sock_hold(sk); 779 hlist_add_tail_rcu(&sk->sk_node, list); 780 } 781 782 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 783 { 784 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 785 } 786 787 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 788 { 789 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 790 } 791 792 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 793 { 794 sock_hold(sk); 795 __sk_nulls_add_node_rcu(sk, list); 796 } 797 798 static inline void __sk_del_bind_node(struct sock *sk) 799 { 800 __hlist_del(&sk->sk_bind_node); 801 } 802 803 static inline void sk_add_bind_node(struct sock *sk, 804 struct hlist_head *list) 805 { 806 hlist_add_head(&sk->sk_bind_node, list); 807 } 808 809 #define sk_for_each(__sk, list) \ 810 hlist_for_each_entry(__sk, list, sk_node) 811 #define sk_for_each_rcu(__sk, list) \ 812 hlist_for_each_entry_rcu(__sk, list, sk_node) 813 #define sk_nulls_for_each(__sk, node, list) \ 814 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 815 #define sk_nulls_for_each_rcu(__sk, node, list) \ 816 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 817 #define sk_for_each_from(__sk) \ 818 hlist_for_each_entry_from(__sk, sk_node) 819 #define sk_nulls_for_each_from(__sk, node) \ 820 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 821 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 822 #define sk_for_each_safe(__sk, tmp, list) \ 823 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 824 #define sk_for_each_bound(__sk, list) \ 825 hlist_for_each_entry(__sk, list, sk_bind_node) 826 827 /** 828 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 829 * @tpos: the type * to use as a loop cursor. 830 * @pos: the &struct hlist_node to use as a loop cursor. 831 * @head: the head for your list. 832 * @offset: offset of hlist_node within the struct. 833 * 834 */ 835 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 836 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 837 pos != NULL && \ 838 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 839 pos = rcu_dereference(hlist_next_rcu(pos))) 840 841 static inline struct user_namespace *sk_user_ns(struct sock *sk) 842 { 843 /* Careful only use this in a context where these parameters 844 * can not change and must all be valid, such as recvmsg from 845 * userspace. 846 */ 847 return sk->sk_socket->file->f_cred->user_ns; 848 } 849 850 /* Sock flags */ 851 enum sock_flags { 852 SOCK_DEAD, 853 SOCK_DONE, 854 SOCK_URGINLINE, 855 SOCK_KEEPOPEN, 856 SOCK_LINGER, 857 SOCK_DESTROY, 858 SOCK_BROADCAST, 859 SOCK_TIMESTAMP, 860 SOCK_ZAPPED, 861 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 862 SOCK_DBG, /* %SO_DEBUG setting */ 863 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 864 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 865 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 866 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 867 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 868 SOCK_FASYNC, /* fasync() active */ 869 SOCK_RXQ_OVFL, 870 SOCK_ZEROCOPY, /* buffers from userspace */ 871 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 872 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 873 * Will use last 4 bytes of packet sent from 874 * user-space instead. 875 */ 876 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 877 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 878 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 879 SOCK_TXTIME, 880 SOCK_XDP, /* XDP is attached */ 881 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 882 }; 883 884 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 885 886 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 887 { 888 nsk->sk_flags = osk->sk_flags; 889 } 890 891 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 892 { 893 __set_bit(flag, &sk->sk_flags); 894 } 895 896 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 897 { 898 __clear_bit(flag, &sk->sk_flags); 899 } 900 901 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 902 int valbool) 903 { 904 if (valbool) 905 sock_set_flag(sk, bit); 906 else 907 sock_reset_flag(sk, bit); 908 } 909 910 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 911 { 912 return test_bit(flag, &sk->sk_flags); 913 } 914 915 #ifdef CONFIG_NET 916 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 917 static inline int sk_memalloc_socks(void) 918 { 919 return static_branch_unlikely(&memalloc_socks_key); 920 } 921 922 void __receive_sock(struct file *file); 923 #else 924 925 static inline int sk_memalloc_socks(void) 926 { 927 return 0; 928 } 929 930 static inline void __receive_sock(struct file *file) 931 { } 932 #endif 933 934 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 935 { 936 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 937 } 938 939 static inline void sk_acceptq_removed(struct sock *sk) 940 { 941 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 942 } 943 944 static inline void sk_acceptq_added(struct sock *sk) 945 { 946 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 947 } 948 949 /* Note: If you think the test should be: 950 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 951 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 952 */ 953 static inline bool sk_acceptq_is_full(const struct sock *sk) 954 { 955 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 956 } 957 958 /* 959 * Compute minimal free write space needed to queue new packets. 960 */ 961 static inline int sk_stream_min_wspace(const struct sock *sk) 962 { 963 return READ_ONCE(sk->sk_wmem_queued) >> 1; 964 } 965 966 static inline int sk_stream_wspace(const struct sock *sk) 967 { 968 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 969 } 970 971 static inline void sk_wmem_queued_add(struct sock *sk, int val) 972 { 973 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 974 } 975 976 void sk_stream_write_space(struct sock *sk); 977 978 /* OOB backlog add */ 979 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 980 { 981 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 982 skb_dst_force(skb); 983 984 if (!sk->sk_backlog.tail) 985 WRITE_ONCE(sk->sk_backlog.head, skb); 986 else 987 sk->sk_backlog.tail->next = skb; 988 989 WRITE_ONCE(sk->sk_backlog.tail, skb); 990 skb->next = NULL; 991 } 992 993 /* 994 * Take into account size of receive queue and backlog queue 995 * Do not take into account this skb truesize, 996 * to allow even a single big packet to come. 997 */ 998 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 999 { 1000 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1001 1002 return qsize > limit; 1003 } 1004 1005 /* The per-socket spinlock must be held here. */ 1006 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1007 unsigned int limit) 1008 { 1009 if (sk_rcvqueues_full(sk, limit)) 1010 return -ENOBUFS; 1011 1012 /* 1013 * If the skb was allocated from pfmemalloc reserves, only 1014 * allow SOCK_MEMALLOC sockets to use it as this socket is 1015 * helping free memory 1016 */ 1017 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1018 return -ENOMEM; 1019 1020 __sk_add_backlog(sk, skb); 1021 sk->sk_backlog.len += skb->truesize; 1022 return 0; 1023 } 1024 1025 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1026 1027 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1028 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1029 1030 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1031 { 1032 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1033 return __sk_backlog_rcv(sk, skb); 1034 1035 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1036 tcp_v6_do_rcv, 1037 tcp_v4_do_rcv, 1038 sk, skb); 1039 } 1040 1041 static inline void sk_incoming_cpu_update(struct sock *sk) 1042 { 1043 int cpu = raw_smp_processor_id(); 1044 1045 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1046 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1047 } 1048 1049 static inline void sock_rps_record_flow_hash(__u32 hash) 1050 { 1051 #ifdef CONFIG_RPS 1052 struct rps_sock_flow_table *sock_flow_table; 1053 1054 rcu_read_lock(); 1055 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1056 rps_record_sock_flow(sock_flow_table, hash); 1057 rcu_read_unlock(); 1058 #endif 1059 } 1060 1061 static inline void sock_rps_record_flow(const struct sock *sk) 1062 { 1063 #ifdef CONFIG_RPS 1064 if (static_branch_unlikely(&rfs_needed)) { 1065 /* Reading sk->sk_rxhash might incur an expensive cache line 1066 * miss. 1067 * 1068 * TCP_ESTABLISHED does cover almost all states where RFS 1069 * might be useful, and is cheaper [1] than testing : 1070 * IPv4: inet_sk(sk)->inet_daddr 1071 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1072 * OR an additional socket flag 1073 * [1] : sk_state and sk_prot are in the same cache line. 1074 */ 1075 if (sk->sk_state == TCP_ESTABLISHED) 1076 sock_rps_record_flow_hash(sk->sk_rxhash); 1077 } 1078 #endif 1079 } 1080 1081 static inline void sock_rps_save_rxhash(struct sock *sk, 1082 const struct sk_buff *skb) 1083 { 1084 #ifdef CONFIG_RPS 1085 if (unlikely(sk->sk_rxhash != skb->hash)) 1086 sk->sk_rxhash = skb->hash; 1087 #endif 1088 } 1089 1090 static inline void sock_rps_reset_rxhash(struct sock *sk) 1091 { 1092 #ifdef CONFIG_RPS 1093 sk->sk_rxhash = 0; 1094 #endif 1095 } 1096 1097 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1098 ({ int __rc; \ 1099 release_sock(__sk); \ 1100 __rc = __condition; \ 1101 if (!__rc) { \ 1102 *(__timeo) = wait_woken(__wait, \ 1103 TASK_INTERRUPTIBLE, \ 1104 *(__timeo)); \ 1105 } \ 1106 sched_annotate_sleep(); \ 1107 lock_sock(__sk); \ 1108 __rc = __condition; \ 1109 __rc; \ 1110 }) 1111 1112 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1113 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1114 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1115 int sk_stream_error(struct sock *sk, int flags, int err); 1116 void sk_stream_kill_queues(struct sock *sk); 1117 void sk_set_memalloc(struct sock *sk); 1118 void sk_clear_memalloc(struct sock *sk); 1119 1120 void __sk_flush_backlog(struct sock *sk); 1121 1122 static inline bool sk_flush_backlog(struct sock *sk) 1123 { 1124 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1125 __sk_flush_backlog(sk); 1126 return true; 1127 } 1128 return false; 1129 } 1130 1131 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1132 1133 struct request_sock_ops; 1134 struct timewait_sock_ops; 1135 struct inet_hashinfo; 1136 struct raw_hashinfo; 1137 struct smc_hashinfo; 1138 struct module; 1139 struct sk_psock; 1140 1141 /* 1142 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1143 * un-modified. Special care is taken when initializing object to zero. 1144 */ 1145 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1146 { 1147 if (offsetof(struct sock, sk_node.next) != 0) 1148 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1149 memset(&sk->sk_node.pprev, 0, 1150 size - offsetof(struct sock, sk_node.pprev)); 1151 } 1152 1153 /* Networking protocol blocks we attach to sockets. 1154 * socket layer -> transport layer interface 1155 */ 1156 struct proto { 1157 void (*close)(struct sock *sk, 1158 long timeout); 1159 int (*pre_connect)(struct sock *sk, 1160 struct sockaddr *uaddr, 1161 int addr_len); 1162 int (*connect)(struct sock *sk, 1163 struct sockaddr *uaddr, 1164 int addr_len); 1165 int (*disconnect)(struct sock *sk, int flags); 1166 1167 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1168 bool kern); 1169 1170 int (*ioctl)(struct sock *sk, int cmd, 1171 unsigned long arg); 1172 int (*init)(struct sock *sk); 1173 void (*destroy)(struct sock *sk); 1174 void (*shutdown)(struct sock *sk, int how); 1175 int (*setsockopt)(struct sock *sk, int level, 1176 int optname, sockptr_t optval, 1177 unsigned int optlen); 1178 int (*getsockopt)(struct sock *sk, int level, 1179 int optname, char __user *optval, 1180 int __user *option); 1181 void (*keepalive)(struct sock *sk, int valbool); 1182 #ifdef CONFIG_COMPAT 1183 int (*compat_ioctl)(struct sock *sk, 1184 unsigned int cmd, unsigned long arg); 1185 #endif 1186 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1187 size_t len); 1188 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1189 size_t len, int noblock, int flags, 1190 int *addr_len); 1191 int (*sendpage)(struct sock *sk, struct page *page, 1192 int offset, size_t size, int flags); 1193 int (*bind)(struct sock *sk, 1194 struct sockaddr *addr, int addr_len); 1195 int (*bind_add)(struct sock *sk, 1196 struct sockaddr *addr, int addr_len); 1197 1198 int (*backlog_rcv) (struct sock *sk, 1199 struct sk_buff *skb); 1200 bool (*bpf_bypass_getsockopt)(int level, 1201 int optname); 1202 1203 void (*release_cb)(struct sock *sk); 1204 1205 /* Keeping track of sk's, looking them up, and port selection methods. */ 1206 int (*hash)(struct sock *sk); 1207 void (*unhash)(struct sock *sk); 1208 void (*rehash)(struct sock *sk); 1209 int (*get_port)(struct sock *sk, unsigned short snum); 1210 #ifdef CONFIG_BPF_SYSCALL 1211 int (*psock_update_sk_prot)(struct sock *sk, 1212 struct sk_psock *psock, 1213 bool restore); 1214 #endif 1215 1216 /* Keeping track of sockets in use */ 1217 #ifdef CONFIG_PROC_FS 1218 unsigned int inuse_idx; 1219 #endif 1220 1221 #if IS_ENABLED(CONFIG_MPTCP) 1222 int (*forward_alloc_get)(const struct sock *sk); 1223 #endif 1224 1225 bool (*stream_memory_free)(const struct sock *sk, int wake); 1226 bool (*sock_is_readable)(struct sock *sk); 1227 /* Memory pressure */ 1228 void (*enter_memory_pressure)(struct sock *sk); 1229 void (*leave_memory_pressure)(struct sock *sk); 1230 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1231 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1232 1233 /* 1234 * Pressure flag: try to collapse. 1235 * Technical note: it is used by multiple contexts non atomically. 1236 * All the __sk_mem_schedule() is of this nature: accounting 1237 * is strict, actions are advisory and have some latency. 1238 */ 1239 unsigned long *memory_pressure; 1240 long *sysctl_mem; 1241 1242 int *sysctl_wmem; 1243 int *sysctl_rmem; 1244 u32 sysctl_wmem_offset; 1245 u32 sysctl_rmem_offset; 1246 1247 int max_header; 1248 bool no_autobind; 1249 1250 struct kmem_cache *slab; 1251 unsigned int obj_size; 1252 slab_flags_t slab_flags; 1253 unsigned int useroffset; /* Usercopy region offset */ 1254 unsigned int usersize; /* Usercopy region size */ 1255 1256 unsigned int __percpu *orphan_count; 1257 1258 struct request_sock_ops *rsk_prot; 1259 struct timewait_sock_ops *twsk_prot; 1260 1261 union { 1262 struct inet_hashinfo *hashinfo; 1263 struct udp_table *udp_table; 1264 struct raw_hashinfo *raw_hash; 1265 struct smc_hashinfo *smc_hash; 1266 } h; 1267 1268 struct module *owner; 1269 1270 char name[32]; 1271 1272 struct list_head node; 1273 #ifdef SOCK_REFCNT_DEBUG 1274 atomic_t socks; 1275 #endif 1276 int (*diag_destroy)(struct sock *sk, int err); 1277 } __randomize_layout; 1278 1279 int proto_register(struct proto *prot, int alloc_slab); 1280 void proto_unregister(struct proto *prot); 1281 int sock_load_diag_module(int family, int protocol); 1282 1283 #ifdef SOCK_REFCNT_DEBUG 1284 static inline void sk_refcnt_debug_inc(struct sock *sk) 1285 { 1286 atomic_inc(&sk->sk_prot->socks); 1287 } 1288 1289 static inline void sk_refcnt_debug_dec(struct sock *sk) 1290 { 1291 atomic_dec(&sk->sk_prot->socks); 1292 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1293 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1294 } 1295 1296 static inline void sk_refcnt_debug_release(const struct sock *sk) 1297 { 1298 if (refcount_read(&sk->sk_refcnt) != 1) 1299 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1300 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1301 } 1302 #else /* SOCK_REFCNT_DEBUG */ 1303 #define sk_refcnt_debug_inc(sk) do { } while (0) 1304 #define sk_refcnt_debug_dec(sk) do { } while (0) 1305 #define sk_refcnt_debug_release(sk) do { } while (0) 1306 #endif /* SOCK_REFCNT_DEBUG */ 1307 1308 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1309 1310 static inline int sk_forward_alloc_get(const struct sock *sk) 1311 { 1312 #if IS_ENABLED(CONFIG_MPTCP) 1313 if (sk->sk_prot->forward_alloc_get) 1314 return sk->sk_prot->forward_alloc_get(sk); 1315 #endif 1316 return sk->sk_forward_alloc; 1317 } 1318 1319 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1320 { 1321 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1322 return false; 1323 1324 return sk->sk_prot->stream_memory_free ? 1325 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1326 tcp_stream_memory_free, sk, wake) : true; 1327 } 1328 1329 static inline bool sk_stream_memory_free(const struct sock *sk) 1330 { 1331 return __sk_stream_memory_free(sk, 0); 1332 } 1333 1334 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1335 { 1336 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1337 __sk_stream_memory_free(sk, wake); 1338 } 1339 1340 static inline bool sk_stream_is_writeable(const struct sock *sk) 1341 { 1342 return __sk_stream_is_writeable(sk, 0); 1343 } 1344 1345 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1346 struct cgroup *ancestor) 1347 { 1348 #ifdef CONFIG_SOCK_CGROUP_DATA 1349 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1350 ancestor); 1351 #else 1352 return -ENOTSUPP; 1353 #endif 1354 } 1355 1356 static inline bool sk_has_memory_pressure(const struct sock *sk) 1357 { 1358 return sk->sk_prot->memory_pressure != NULL; 1359 } 1360 1361 static inline bool sk_under_memory_pressure(const struct sock *sk) 1362 { 1363 if (!sk->sk_prot->memory_pressure) 1364 return false; 1365 1366 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1367 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1368 return true; 1369 1370 return !!*sk->sk_prot->memory_pressure; 1371 } 1372 1373 static inline long 1374 sk_memory_allocated(const struct sock *sk) 1375 { 1376 return atomic_long_read(sk->sk_prot->memory_allocated); 1377 } 1378 1379 static inline long 1380 sk_memory_allocated_add(struct sock *sk, int amt) 1381 { 1382 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1383 } 1384 1385 static inline void 1386 sk_memory_allocated_sub(struct sock *sk, int amt) 1387 { 1388 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1389 } 1390 1391 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1392 1393 static inline void sk_sockets_allocated_dec(struct sock *sk) 1394 { 1395 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1396 SK_ALLOC_PERCPU_COUNTER_BATCH); 1397 } 1398 1399 static inline void sk_sockets_allocated_inc(struct sock *sk) 1400 { 1401 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1402 SK_ALLOC_PERCPU_COUNTER_BATCH); 1403 } 1404 1405 static inline u64 1406 sk_sockets_allocated_read_positive(struct sock *sk) 1407 { 1408 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1409 } 1410 1411 static inline int 1412 proto_sockets_allocated_sum_positive(struct proto *prot) 1413 { 1414 return percpu_counter_sum_positive(prot->sockets_allocated); 1415 } 1416 1417 static inline long 1418 proto_memory_allocated(struct proto *prot) 1419 { 1420 return atomic_long_read(prot->memory_allocated); 1421 } 1422 1423 static inline bool 1424 proto_memory_pressure(struct proto *prot) 1425 { 1426 if (!prot->memory_pressure) 1427 return false; 1428 return !!*prot->memory_pressure; 1429 } 1430 1431 1432 #ifdef CONFIG_PROC_FS 1433 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1434 struct prot_inuse { 1435 int all; 1436 int val[PROTO_INUSE_NR]; 1437 }; 1438 1439 static inline void sock_prot_inuse_add(const struct net *net, 1440 const struct proto *prot, int val) 1441 { 1442 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1443 } 1444 1445 static inline void sock_inuse_add(const struct net *net, int val) 1446 { 1447 this_cpu_add(net->core.prot_inuse->all, val); 1448 } 1449 1450 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1451 int sock_inuse_get(struct net *net); 1452 #else 1453 static inline void sock_prot_inuse_add(const struct net *net, 1454 const struct proto *prot, int val) 1455 { 1456 } 1457 1458 static inline void sock_inuse_add(const struct net *net, int val) 1459 { 1460 } 1461 #endif 1462 1463 1464 /* With per-bucket locks this operation is not-atomic, so that 1465 * this version is not worse. 1466 */ 1467 static inline int __sk_prot_rehash(struct sock *sk) 1468 { 1469 sk->sk_prot->unhash(sk); 1470 return sk->sk_prot->hash(sk); 1471 } 1472 1473 /* About 10 seconds */ 1474 #define SOCK_DESTROY_TIME (10*HZ) 1475 1476 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1477 #define PROT_SOCK 1024 1478 1479 #define SHUTDOWN_MASK 3 1480 #define RCV_SHUTDOWN 1 1481 #define SEND_SHUTDOWN 2 1482 1483 #define SOCK_BINDADDR_LOCK 4 1484 #define SOCK_BINDPORT_LOCK 8 1485 1486 struct socket_alloc { 1487 struct socket socket; 1488 struct inode vfs_inode; 1489 }; 1490 1491 static inline struct socket *SOCKET_I(struct inode *inode) 1492 { 1493 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1494 } 1495 1496 static inline struct inode *SOCK_INODE(struct socket *socket) 1497 { 1498 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1499 } 1500 1501 /* 1502 * Functions for memory accounting 1503 */ 1504 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1505 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1506 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1507 void __sk_mem_reclaim(struct sock *sk, int amount); 1508 1509 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1510 * do not necessarily have 16x time more memory than 4KB ones. 1511 */ 1512 #define SK_MEM_QUANTUM 4096 1513 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1514 #define SK_MEM_SEND 0 1515 #define SK_MEM_RECV 1 1516 1517 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1518 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1519 { 1520 long val = sk->sk_prot->sysctl_mem[index]; 1521 1522 #if PAGE_SIZE > SK_MEM_QUANTUM 1523 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1524 #elif PAGE_SIZE < SK_MEM_QUANTUM 1525 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1526 #endif 1527 return val; 1528 } 1529 1530 static inline int sk_mem_pages(int amt) 1531 { 1532 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1533 } 1534 1535 static inline bool sk_has_account(struct sock *sk) 1536 { 1537 /* return true if protocol supports memory accounting */ 1538 return !!sk->sk_prot->memory_allocated; 1539 } 1540 1541 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1542 { 1543 if (!sk_has_account(sk)) 1544 return true; 1545 return size <= sk->sk_forward_alloc || 1546 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1547 } 1548 1549 static inline bool 1550 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1551 { 1552 if (!sk_has_account(sk)) 1553 return true; 1554 return size <= sk->sk_forward_alloc || 1555 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1556 skb_pfmemalloc(skb); 1557 } 1558 1559 static inline int sk_unused_reserved_mem(const struct sock *sk) 1560 { 1561 int unused_mem; 1562 1563 if (likely(!sk->sk_reserved_mem)) 1564 return 0; 1565 1566 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1567 atomic_read(&sk->sk_rmem_alloc); 1568 1569 return unused_mem > 0 ? unused_mem : 0; 1570 } 1571 1572 static inline void sk_mem_reclaim(struct sock *sk) 1573 { 1574 int reclaimable; 1575 1576 if (!sk_has_account(sk)) 1577 return; 1578 1579 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1580 1581 if (reclaimable >= SK_MEM_QUANTUM) 1582 __sk_mem_reclaim(sk, reclaimable); 1583 } 1584 1585 static inline void sk_mem_reclaim_final(struct sock *sk) 1586 { 1587 sk->sk_reserved_mem = 0; 1588 sk_mem_reclaim(sk); 1589 } 1590 1591 static inline void sk_mem_reclaim_partial(struct sock *sk) 1592 { 1593 int reclaimable; 1594 1595 if (!sk_has_account(sk)) 1596 return; 1597 1598 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1599 1600 if (reclaimable > SK_MEM_QUANTUM) 1601 __sk_mem_reclaim(sk, reclaimable - 1); 1602 } 1603 1604 static inline void sk_mem_charge(struct sock *sk, int size) 1605 { 1606 if (!sk_has_account(sk)) 1607 return; 1608 sk->sk_forward_alloc -= size; 1609 } 1610 1611 /* the following macros control memory reclaiming in sk_mem_uncharge() 1612 */ 1613 #define SK_RECLAIM_THRESHOLD (1 << 21) 1614 #define SK_RECLAIM_CHUNK (1 << 20) 1615 1616 static inline void sk_mem_uncharge(struct sock *sk, int size) 1617 { 1618 int reclaimable; 1619 1620 if (!sk_has_account(sk)) 1621 return; 1622 sk->sk_forward_alloc += size; 1623 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1624 1625 /* Avoid a possible overflow. 1626 * TCP send queues can make this happen, if sk_mem_reclaim() 1627 * is not called and more than 2 GBytes are released at once. 1628 * 1629 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1630 * no need to hold that much forward allocation anyway. 1631 */ 1632 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD)) 1633 __sk_mem_reclaim(sk, SK_RECLAIM_CHUNK); 1634 } 1635 1636 static inline void sock_release_ownership(struct sock *sk) 1637 { 1638 if (sk->sk_lock.owned) { 1639 sk->sk_lock.owned = 0; 1640 1641 /* The sk_lock has mutex_unlock() semantics: */ 1642 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1643 } 1644 } 1645 1646 /* 1647 * Macro so as to not evaluate some arguments when 1648 * lockdep is not enabled. 1649 * 1650 * Mark both the sk_lock and the sk_lock.slock as a 1651 * per-address-family lock class. 1652 */ 1653 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1654 do { \ 1655 sk->sk_lock.owned = 0; \ 1656 init_waitqueue_head(&sk->sk_lock.wq); \ 1657 spin_lock_init(&(sk)->sk_lock.slock); \ 1658 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1659 sizeof((sk)->sk_lock)); \ 1660 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1661 (skey), (sname)); \ 1662 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1663 } while (0) 1664 1665 static inline bool lockdep_sock_is_held(const struct sock *sk) 1666 { 1667 return lockdep_is_held(&sk->sk_lock) || 1668 lockdep_is_held(&sk->sk_lock.slock); 1669 } 1670 1671 void lock_sock_nested(struct sock *sk, int subclass); 1672 1673 static inline void lock_sock(struct sock *sk) 1674 { 1675 lock_sock_nested(sk, 0); 1676 } 1677 1678 void __lock_sock(struct sock *sk); 1679 void __release_sock(struct sock *sk); 1680 void release_sock(struct sock *sk); 1681 1682 /* BH context may only use the following locking interface. */ 1683 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1684 #define bh_lock_sock_nested(__sk) \ 1685 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1686 SINGLE_DEPTH_NESTING) 1687 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1688 1689 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1690 1691 /** 1692 * lock_sock_fast - fast version of lock_sock 1693 * @sk: socket 1694 * 1695 * This version should be used for very small section, where process wont block 1696 * return false if fast path is taken: 1697 * 1698 * sk_lock.slock locked, owned = 0, BH disabled 1699 * 1700 * return true if slow path is taken: 1701 * 1702 * sk_lock.slock unlocked, owned = 1, BH enabled 1703 */ 1704 static inline bool lock_sock_fast(struct sock *sk) 1705 { 1706 /* The sk_lock has mutex_lock() semantics here. */ 1707 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1708 1709 return __lock_sock_fast(sk); 1710 } 1711 1712 /* fast socket lock variant for caller already holding a [different] socket lock */ 1713 static inline bool lock_sock_fast_nested(struct sock *sk) 1714 { 1715 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1716 1717 return __lock_sock_fast(sk); 1718 } 1719 1720 /** 1721 * unlock_sock_fast - complement of lock_sock_fast 1722 * @sk: socket 1723 * @slow: slow mode 1724 * 1725 * fast unlock socket for user context. 1726 * If slow mode is on, we call regular release_sock() 1727 */ 1728 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1729 __releases(&sk->sk_lock.slock) 1730 { 1731 if (slow) { 1732 release_sock(sk); 1733 __release(&sk->sk_lock.slock); 1734 } else { 1735 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1736 spin_unlock_bh(&sk->sk_lock.slock); 1737 } 1738 } 1739 1740 /* Used by processes to "lock" a socket state, so that 1741 * interrupts and bottom half handlers won't change it 1742 * from under us. It essentially blocks any incoming 1743 * packets, so that we won't get any new data or any 1744 * packets that change the state of the socket. 1745 * 1746 * While locked, BH processing will add new packets to 1747 * the backlog queue. This queue is processed by the 1748 * owner of the socket lock right before it is released. 1749 * 1750 * Since ~2.3.5 it is also exclusive sleep lock serializing 1751 * accesses from user process context. 1752 */ 1753 1754 static inline void sock_owned_by_me(const struct sock *sk) 1755 { 1756 #ifdef CONFIG_LOCKDEP 1757 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1758 #endif 1759 } 1760 1761 static inline bool sock_owned_by_user(const struct sock *sk) 1762 { 1763 sock_owned_by_me(sk); 1764 return sk->sk_lock.owned; 1765 } 1766 1767 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1768 { 1769 return sk->sk_lock.owned; 1770 } 1771 1772 /* no reclassification while locks are held */ 1773 static inline bool sock_allow_reclassification(const struct sock *csk) 1774 { 1775 struct sock *sk = (struct sock *)csk; 1776 1777 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1778 } 1779 1780 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1781 struct proto *prot, int kern); 1782 void sk_free(struct sock *sk); 1783 void sk_destruct(struct sock *sk); 1784 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1785 void sk_free_unlock_clone(struct sock *sk); 1786 1787 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1788 gfp_t priority); 1789 void __sock_wfree(struct sk_buff *skb); 1790 void sock_wfree(struct sk_buff *skb); 1791 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1792 gfp_t priority); 1793 void skb_orphan_partial(struct sk_buff *skb); 1794 void sock_rfree(struct sk_buff *skb); 1795 void sock_efree(struct sk_buff *skb); 1796 #ifdef CONFIG_INET 1797 void sock_edemux(struct sk_buff *skb); 1798 void sock_pfree(struct sk_buff *skb); 1799 #else 1800 #define sock_edemux sock_efree 1801 #endif 1802 1803 int sock_setsockopt(struct socket *sock, int level, int op, 1804 sockptr_t optval, unsigned int optlen); 1805 1806 int sock_getsockopt(struct socket *sock, int level, int op, 1807 char __user *optval, int __user *optlen); 1808 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1809 bool timeval, bool time32); 1810 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1811 int noblock, int *errcode); 1812 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1813 unsigned long data_len, int noblock, 1814 int *errcode, int max_page_order); 1815 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1816 void sock_kfree_s(struct sock *sk, void *mem, int size); 1817 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1818 void sk_send_sigurg(struct sock *sk); 1819 1820 struct sockcm_cookie { 1821 u64 transmit_time; 1822 u32 mark; 1823 u16 tsflags; 1824 }; 1825 1826 static inline void sockcm_init(struct sockcm_cookie *sockc, 1827 const struct sock *sk) 1828 { 1829 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1830 } 1831 1832 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1833 struct sockcm_cookie *sockc); 1834 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1835 struct sockcm_cookie *sockc); 1836 1837 /* 1838 * Functions to fill in entries in struct proto_ops when a protocol 1839 * does not implement a particular function. 1840 */ 1841 int sock_no_bind(struct socket *, struct sockaddr *, int); 1842 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1843 int sock_no_socketpair(struct socket *, struct socket *); 1844 int sock_no_accept(struct socket *, struct socket *, int, bool); 1845 int sock_no_getname(struct socket *, struct sockaddr *, int); 1846 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1847 int sock_no_listen(struct socket *, int); 1848 int sock_no_shutdown(struct socket *, int); 1849 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1850 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1851 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1852 int sock_no_mmap(struct file *file, struct socket *sock, 1853 struct vm_area_struct *vma); 1854 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1855 size_t size, int flags); 1856 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1857 int offset, size_t size, int flags); 1858 1859 /* 1860 * Functions to fill in entries in struct proto_ops when a protocol 1861 * uses the inet style. 1862 */ 1863 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1864 char __user *optval, int __user *optlen); 1865 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1866 int flags); 1867 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1868 sockptr_t optval, unsigned int optlen); 1869 1870 void sk_common_release(struct sock *sk); 1871 1872 /* 1873 * Default socket callbacks and setup code 1874 */ 1875 1876 /* Initialise core socket variables */ 1877 void sock_init_data(struct socket *sock, struct sock *sk); 1878 1879 /* 1880 * Socket reference counting postulates. 1881 * 1882 * * Each user of socket SHOULD hold a reference count. 1883 * * Each access point to socket (an hash table bucket, reference from a list, 1884 * running timer, skb in flight MUST hold a reference count. 1885 * * When reference count hits 0, it means it will never increase back. 1886 * * When reference count hits 0, it means that no references from 1887 * outside exist to this socket and current process on current CPU 1888 * is last user and may/should destroy this socket. 1889 * * sk_free is called from any context: process, BH, IRQ. When 1890 * it is called, socket has no references from outside -> sk_free 1891 * may release descendant resources allocated by the socket, but 1892 * to the time when it is called, socket is NOT referenced by any 1893 * hash tables, lists etc. 1894 * * Packets, delivered from outside (from network or from another process) 1895 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1896 * when they sit in queue. Otherwise, packets will leak to hole, when 1897 * socket is looked up by one cpu and unhasing is made by another CPU. 1898 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1899 * (leak to backlog). Packet socket does all the processing inside 1900 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1901 * use separate SMP lock, so that they are prone too. 1902 */ 1903 1904 /* Ungrab socket and destroy it, if it was the last reference. */ 1905 static inline void sock_put(struct sock *sk) 1906 { 1907 if (refcount_dec_and_test(&sk->sk_refcnt)) 1908 sk_free(sk); 1909 } 1910 /* Generic version of sock_put(), dealing with all sockets 1911 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1912 */ 1913 void sock_gen_put(struct sock *sk); 1914 1915 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1916 unsigned int trim_cap, bool refcounted); 1917 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1918 const int nested) 1919 { 1920 return __sk_receive_skb(sk, skb, nested, 1, true); 1921 } 1922 1923 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1924 { 1925 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1926 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1927 return; 1928 sk->sk_tx_queue_mapping = tx_queue; 1929 } 1930 1931 #define NO_QUEUE_MAPPING USHRT_MAX 1932 1933 static inline void sk_tx_queue_clear(struct sock *sk) 1934 { 1935 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1936 } 1937 1938 static inline int sk_tx_queue_get(const struct sock *sk) 1939 { 1940 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1941 return sk->sk_tx_queue_mapping; 1942 1943 return -1; 1944 } 1945 1946 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1947 { 1948 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1949 if (skb_rx_queue_recorded(skb)) { 1950 u16 rx_queue = skb_get_rx_queue(skb); 1951 1952 if (unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 1953 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 1954 } 1955 #endif 1956 } 1957 1958 static inline void sk_rx_queue_clear(struct sock *sk) 1959 { 1960 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1961 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 1962 #endif 1963 } 1964 1965 static inline int sk_rx_queue_get(const struct sock *sk) 1966 { 1967 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1968 if (sk) { 1969 int res = READ_ONCE(sk->sk_rx_queue_mapping); 1970 1971 if (res != NO_QUEUE_MAPPING) 1972 return res; 1973 } 1974 #endif 1975 1976 return -1; 1977 } 1978 1979 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1980 { 1981 sk->sk_socket = sock; 1982 } 1983 1984 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1985 { 1986 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1987 return &rcu_dereference_raw(sk->sk_wq)->wait; 1988 } 1989 /* Detach socket from process context. 1990 * Announce socket dead, detach it from wait queue and inode. 1991 * Note that parent inode held reference count on this struct sock, 1992 * we do not release it in this function, because protocol 1993 * probably wants some additional cleanups or even continuing 1994 * to work with this socket (TCP). 1995 */ 1996 static inline void sock_orphan(struct sock *sk) 1997 { 1998 write_lock_bh(&sk->sk_callback_lock); 1999 sock_set_flag(sk, SOCK_DEAD); 2000 sk_set_socket(sk, NULL); 2001 sk->sk_wq = NULL; 2002 write_unlock_bh(&sk->sk_callback_lock); 2003 } 2004 2005 static inline void sock_graft(struct sock *sk, struct socket *parent) 2006 { 2007 WARN_ON(parent->sk); 2008 write_lock_bh(&sk->sk_callback_lock); 2009 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2010 parent->sk = sk; 2011 sk_set_socket(sk, parent); 2012 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2013 security_sock_graft(sk, parent); 2014 write_unlock_bh(&sk->sk_callback_lock); 2015 } 2016 2017 kuid_t sock_i_uid(struct sock *sk); 2018 unsigned long sock_i_ino(struct sock *sk); 2019 2020 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2021 { 2022 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2023 } 2024 2025 static inline u32 net_tx_rndhash(void) 2026 { 2027 u32 v = prandom_u32(); 2028 2029 return v ?: 1; 2030 } 2031 2032 static inline void sk_set_txhash(struct sock *sk) 2033 { 2034 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2035 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2036 } 2037 2038 static inline bool sk_rethink_txhash(struct sock *sk) 2039 { 2040 if (sk->sk_txhash) { 2041 sk_set_txhash(sk); 2042 return true; 2043 } 2044 return false; 2045 } 2046 2047 static inline struct dst_entry * 2048 __sk_dst_get(struct sock *sk) 2049 { 2050 return rcu_dereference_check(sk->sk_dst_cache, 2051 lockdep_sock_is_held(sk)); 2052 } 2053 2054 static inline struct dst_entry * 2055 sk_dst_get(struct sock *sk) 2056 { 2057 struct dst_entry *dst; 2058 2059 rcu_read_lock(); 2060 dst = rcu_dereference(sk->sk_dst_cache); 2061 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2062 dst = NULL; 2063 rcu_read_unlock(); 2064 return dst; 2065 } 2066 2067 static inline void __dst_negative_advice(struct sock *sk) 2068 { 2069 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2070 2071 if (dst && dst->ops->negative_advice) { 2072 ndst = dst->ops->negative_advice(dst); 2073 2074 if (ndst != dst) { 2075 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2076 sk_tx_queue_clear(sk); 2077 sk->sk_dst_pending_confirm = 0; 2078 } 2079 } 2080 } 2081 2082 static inline void dst_negative_advice(struct sock *sk) 2083 { 2084 sk_rethink_txhash(sk); 2085 __dst_negative_advice(sk); 2086 } 2087 2088 static inline void 2089 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2090 { 2091 struct dst_entry *old_dst; 2092 2093 sk_tx_queue_clear(sk); 2094 sk->sk_dst_pending_confirm = 0; 2095 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2096 lockdep_sock_is_held(sk)); 2097 rcu_assign_pointer(sk->sk_dst_cache, dst); 2098 dst_release(old_dst); 2099 } 2100 2101 static inline void 2102 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2103 { 2104 struct dst_entry *old_dst; 2105 2106 sk_tx_queue_clear(sk); 2107 sk->sk_dst_pending_confirm = 0; 2108 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2109 dst_release(old_dst); 2110 } 2111 2112 static inline void 2113 __sk_dst_reset(struct sock *sk) 2114 { 2115 __sk_dst_set(sk, NULL); 2116 } 2117 2118 static inline void 2119 sk_dst_reset(struct sock *sk) 2120 { 2121 sk_dst_set(sk, NULL); 2122 } 2123 2124 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2125 2126 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2127 2128 static inline void sk_dst_confirm(struct sock *sk) 2129 { 2130 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2131 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2132 } 2133 2134 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2135 { 2136 if (skb_get_dst_pending_confirm(skb)) { 2137 struct sock *sk = skb->sk; 2138 unsigned long now = jiffies; 2139 2140 /* avoid dirtying neighbour */ 2141 if (READ_ONCE(n->confirmed) != now) 2142 WRITE_ONCE(n->confirmed, now); 2143 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2144 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2145 } 2146 } 2147 2148 bool sk_mc_loop(struct sock *sk); 2149 2150 static inline bool sk_can_gso(const struct sock *sk) 2151 { 2152 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2153 } 2154 2155 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2156 2157 static inline void sk_gso_disable(struct sock *sk) 2158 { 2159 sk->sk_gso_disabled = 1; 2160 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2161 } 2162 2163 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2164 struct iov_iter *from, char *to, 2165 int copy, int offset) 2166 { 2167 if (skb->ip_summed == CHECKSUM_NONE) { 2168 __wsum csum = 0; 2169 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2170 return -EFAULT; 2171 skb->csum = csum_block_add(skb->csum, csum, offset); 2172 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2173 if (!copy_from_iter_full_nocache(to, copy, from)) 2174 return -EFAULT; 2175 } else if (!copy_from_iter_full(to, copy, from)) 2176 return -EFAULT; 2177 2178 return 0; 2179 } 2180 2181 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2182 struct iov_iter *from, int copy) 2183 { 2184 int err, offset = skb->len; 2185 2186 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2187 copy, offset); 2188 if (err) 2189 __skb_trim(skb, offset); 2190 2191 return err; 2192 } 2193 2194 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2195 struct sk_buff *skb, 2196 struct page *page, 2197 int off, int copy) 2198 { 2199 int err; 2200 2201 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2202 copy, skb->len); 2203 if (err) 2204 return err; 2205 2206 skb->len += copy; 2207 skb->data_len += copy; 2208 skb->truesize += copy; 2209 sk_wmem_queued_add(sk, copy); 2210 sk_mem_charge(sk, copy); 2211 return 0; 2212 } 2213 2214 /** 2215 * sk_wmem_alloc_get - returns write allocations 2216 * @sk: socket 2217 * 2218 * Return: sk_wmem_alloc minus initial offset of one 2219 */ 2220 static inline int sk_wmem_alloc_get(const struct sock *sk) 2221 { 2222 return refcount_read(&sk->sk_wmem_alloc) - 1; 2223 } 2224 2225 /** 2226 * sk_rmem_alloc_get - returns read allocations 2227 * @sk: socket 2228 * 2229 * Return: sk_rmem_alloc 2230 */ 2231 static inline int sk_rmem_alloc_get(const struct sock *sk) 2232 { 2233 return atomic_read(&sk->sk_rmem_alloc); 2234 } 2235 2236 /** 2237 * sk_has_allocations - check if allocations are outstanding 2238 * @sk: socket 2239 * 2240 * Return: true if socket has write or read allocations 2241 */ 2242 static inline bool sk_has_allocations(const struct sock *sk) 2243 { 2244 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2245 } 2246 2247 /** 2248 * skwq_has_sleeper - check if there are any waiting processes 2249 * @wq: struct socket_wq 2250 * 2251 * Return: true if socket_wq has waiting processes 2252 * 2253 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2254 * barrier call. They were added due to the race found within the tcp code. 2255 * 2256 * Consider following tcp code paths:: 2257 * 2258 * CPU1 CPU2 2259 * sys_select receive packet 2260 * ... ... 2261 * __add_wait_queue update tp->rcv_nxt 2262 * ... ... 2263 * tp->rcv_nxt check sock_def_readable 2264 * ... { 2265 * schedule rcu_read_lock(); 2266 * wq = rcu_dereference(sk->sk_wq); 2267 * if (wq && waitqueue_active(&wq->wait)) 2268 * wake_up_interruptible(&wq->wait) 2269 * ... 2270 * } 2271 * 2272 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2273 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2274 * could then endup calling schedule and sleep forever if there are no more 2275 * data on the socket. 2276 * 2277 */ 2278 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2279 { 2280 return wq && wq_has_sleeper(&wq->wait); 2281 } 2282 2283 /** 2284 * sock_poll_wait - place memory barrier behind the poll_wait call. 2285 * @filp: file 2286 * @sock: socket to wait on 2287 * @p: poll_table 2288 * 2289 * See the comments in the wq_has_sleeper function. 2290 */ 2291 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2292 poll_table *p) 2293 { 2294 if (!poll_does_not_wait(p)) { 2295 poll_wait(filp, &sock->wq.wait, p); 2296 /* We need to be sure we are in sync with the 2297 * socket flags modification. 2298 * 2299 * This memory barrier is paired in the wq_has_sleeper. 2300 */ 2301 smp_mb(); 2302 } 2303 } 2304 2305 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2306 { 2307 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2308 u32 txhash = READ_ONCE(sk->sk_txhash); 2309 2310 if (txhash) { 2311 skb->l4_hash = 1; 2312 skb->hash = txhash; 2313 } 2314 } 2315 2316 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2317 2318 /* 2319 * Queue a received datagram if it will fit. Stream and sequenced 2320 * protocols can't normally use this as they need to fit buffers in 2321 * and play with them. 2322 * 2323 * Inlined as it's very short and called for pretty much every 2324 * packet ever received. 2325 */ 2326 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2327 { 2328 skb_orphan(skb); 2329 skb->sk = sk; 2330 skb->destructor = sock_rfree; 2331 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2332 sk_mem_charge(sk, skb->truesize); 2333 } 2334 2335 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2336 { 2337 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2338 skb_orphan(skb); 2339 skb->destructor = sock_efree; 2340 skb->sk = sk; 2341 return true; 2342 } 2343 return false; 2344 } 2345 2346 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2347 { 2348 if (skb->destructor != sock_wfree) { 2349 skb_orphan(skb); 2350 return; 2351 } 2352 skb->slow_gro = 1; 2353 } 2354 2355 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2356 unsigned long expires); 2357 2358 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2359 2360 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2361 2362 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2363 struct sk_buff *skb, unsigned int flags, 2364 void (*destructor)(struct sock *sk, 2365 struct sk_buff *skb)); 2366 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2367 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2368 2369 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2370 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2371 2372 /* 2373 * Recover an error report and clear atomically 2374 */ 2375 2376 static inline int sock_error(struct sock *sk) 2377 { 2378 int err; 2379 2380 /* Avoid an atomic operation for the common case. 2381 * This is racy since another cpu/thread can change sk_err under us. 2382 */ 2383 if (likely(data_race(!sk->sk_err))) 2384 return 0; 2385 2386 err = xchg(&sk->sk_err, 0); 2387 return -err; 2388 } 2389 2390 void sk_error_report(struct sock *sk); 2391 2392 static inline unsigned long sock_wspace(struct sock *sk) 2393 { 2394 int amt = 0; 2395 2396 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2397 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2398 if (amt < 0) 2399 amt = 0; 2400 } 2401 return amt; 2402 } 2403 2404 /* Note: 2405 * We use sk->sk_wq_raw, from contexts knowing this 2406 * pointer is not NULL and cannot disappear/change. 2407 */ 2408 static inline void sk_set_bit(int nr, struct sock *sk) 2409 { 2410 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2411 !sock_flag(sk, SOCK_FASYNC)) 2412 return; 2413 2414 set_bit(nr, &sk->sk_wq_raw->flags); 2415 } 2416 2417 static inline void sk_clear_bit(int nr, struct sock *sk) 2418 { 2419 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2420 !sock_flag(sk, SOCK_FASYNC)) 2421 return; 2422 2423 clear_bit(nr, &sk->sk_wq_raw->flags); 2424 } 2425 2426 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2427 { 2428 if (sock_flag(sk, SOCK_FASYNC)) { 2429 rcu_read_lock(); 2430 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2431 rcu_read_unlock(); 2432 } 2433 } 2434 2435 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2436 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2437 * Note: for send buffers, TCP works better if we can build two skbs at 2438 * minimum. 2439 */ 2440 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2441 2442 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2443 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2444 2445 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2446 { 2447 u32 val; 2448 2449 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2450 return; 2451 2452 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2453 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2454 2455 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2456 } 2457 2458 /** 2459 * sk_page_frag - return an appropriate page_frag 2460 * @sk: socket 2461 * 2462 * Use the per task page_frag instead of the per socket one for 2463 * optimization when we know that we're in the normal context and owns 2464 * everything that's associated with %current. 2465 * 2466 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest 2467 * inside other socket operations and end up recursing into sk_page_frag() 2468 * while it's already in use. 2469 * 2470 * Return: a per task page_frag if context allows that, 2471 * otherwise a per socket one. 2472 */ 2473 static inline struct page_frag *sk_page_frag(struct sock *sk) 2474 { 2475 if (gfpflags_normal_context(sk->sk_allocation)) 2476 return ¤t->task_frag; 2477 2478 return &sk->sk_frag; 2479 } 2480 2481 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2482 2483 /* 2484 * Default write policy as shown to user space via poll/select/SIGIO 2485 */ 2486 static inline bool sock_writeable(const struct sock *sk) 2487 { 2488 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2489 } 2490 2491 static inline gfp_t gfp_any(void) 2492 { 2493 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2494 } 2495 2496 static inline gfp_t gfp_memcg_charge(void) 2497 { 2498 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2499 } 2500 2501 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2502 { 2503 return noblock ? 0 : sk->sk_rcvtimeo; 2504 } 2505 2506 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2507 { 2508 return noblock ? 0 : sk->sk_sndtimeo; 2509 } 2510 2511 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2512 { 2513 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2514 2515 return v ?: 1; 2516 } 2517 2518 /* Alas, with timeout socket operations are not restartable. 2519 * Compare this to poll(). 2520 */ 2521 static inline int sock_intr_errno(long timeo) 2522 { 2523 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2524 } 2525 2526 struct sock_skb_cb { 2527 u32 dropcount; 2528 }; 2529 2530 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2531 * using skb->cb[] would keep using it directly and utilize its 2532 * alignement guarantee. 2533 */ 2534 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2535 sizeof(struct sock_skb_cb))) 2536 2537 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2538 SOCK_SKB_CB_OFFSET)) 2539 2540 #define sock_skb_cb_check_size(size) \ 2541 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2542 2543 static inline void 2544 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2545 { 2546 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2547 atomic_read(&sk->sk_drops) : 0; 2548 } 2549 2550 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2551 { 2552 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2553 2554 atomic_add(segs, &sk->sk_drops); 2555 } 2556 2557 static inline ktime_t sock_read_timestamp(struct sock *sk) 2558 { 2559 #if BITS_PER_LONG==32 2560 unsigned int seq; 2561 ktime_t kt; 2562 2563 do { 2564 seq = read_seqbegin(&sk->sk_stamp_seq); 2565 kt = sk->sk_stamp; 2566 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2567 2568 return kt; 2569 #else 2570 return READ_ONCE(sk->sk_stamp); 2571 #endif 2572 } 2573 2574 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2575 { 2576 #if BITS_PER_LONG==32 2577 write_seqlock(&sk->sk_stamp_seq); 2578 sk->sk_stamp = kt; 2579 write_sequnlock(&sk->sk_stamp_seq); 2580 #else 2581 WRITE_ONCE(sk->sk_stamp, kt); 2582 #endif 2583 } 2584 2585 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2586 struct sk_buff *skb); 2587 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2588 struct sk_buff *skb); 2589 2590 static inline void 2591 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2592 { 2593 ktime_t kt = skb->tstamp; 2594 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2595 2596 /* 2597 * generate control messages if 2598 * - receive time stamping in software requested 2599 * - software time stamp available and wanted 2600 * - hardware time stamps available and wanted 2601 */ 2602 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2603 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2604 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2605 (hwtstamps->hwtstamp && 2606 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2607 __sock_recv_timestamp(msg, sk, skb); 2608 else 2609 sock_write_timestamp(sk, kt); 2610 2611 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2612 __sock_recv_wifi_status(msg, sk, skb); 2613 } 2614 2615 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2616 struct sk_buff *skb); 2617 2618 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2619 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2620 struct sk_buff *skb) 2621 { 2622 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2623 (1UL << SOCK_RCVTSTAMP)) 2624 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2625 SOF_TIMESTAMPING_RAW_HARDWARE) 2626 2627 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2628 __sock_recv_ts_and_drops(msg, sk, skb); 2629 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2630 sock_write_timestamp(sk, skb->tstamp); 2631 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2632 sock_write_timestamp(sk, 0); 2633 } 2634 2635 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2636 2637 /** 2638 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2639 * @sk: socket sending this packet 2640 * @tsflags: timestamping flags to use 2641 * @tx_flags: completed with instructions for time stamping 2642 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2643 * 2644 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2645 */ 2646 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2647 __u8 *tx_flags, __u32 *tskey) 2648 { 2649 if (unlikely(tsflags)) { 2650 __sock_tx_timestamp(tsflags, tx_flags); 2651 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2652 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2653 *tskey = sk->sk_tskey++; 2654 } 2655 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2656 *tx_flags |= SKBTX_WIFI_STATUS; 2657 } 2658 2659 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2660 __u8 *tx_flags) 2661 { 2662 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2663 } 2664 2665 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2666 { 2667 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2668 &skb_shinfo(skb)->tskey); 2669 } 2670 2671 static inline bool sk_is_tcp(const struct sock *sk) 2672 { 2673 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2674 } 2675 2676 /** 2677 * sk_eat_skb - Release a skb if it is no longer needed 2678 * @sk: socket to eat this skb from 2679 * @skb: socket buffer to eat 2680 * 2681 * This routine must be called with interrupts disabled or with the socket 2682 * locked so that the sk_buff queue operation is ok. 2683 */ 2684 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2685 { 2686 __skb_unlink(skb, &sk->sk_receive_queue); 2687 __kfree_skb(skb); 2688 } 2689 2690 static inline 2691 struct net *sock_net(const struct sock *sk) 2692 { 2693 return read_pnet(&sk->sk_net); 2694 } 2695 2696 static inline 2697 void sock_net_set(struct sock *sk, struct net *net) 2698 { 2699 write_pnet(&sk->sk_net, net); 2700 } 2701 2702 static inline bool 2703 skb_sk_is_prefetched(struct sk_buff *skb) 2704 { 2705 #ifdef CONFIG_INET 2706 return skb->destructor == sock_pfree; 2707 #else 2708 return false; 2709 #endif /* CONFIG_INET */ 2710 } 2711 2712 /* This helper checks if a socket is a full socket, 2713 * ie _not_ a timewait or request socket. 2714 */ 2715 static inline bool sk_fullsock(const struct sock *sk) 2716 { 2717 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2718 } 2719 2720 static inline bool 2721 sk_is_refcounted(struct sock *sk) 2722 { 2723 /* Only full sockets have sk->sk_flags. */ 2724 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2725 } 2726 2727 /** 2728 * skb_steal_sock - steal a socket from an sk_buff 2729 * @skb: sk_buff to steal the socket from 2730 * @refcounted: is set to true if the socket is reference-counted 2731 */ 2732 static inline struct sock * 2733 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2734 { 2735 if (skb->sk) { 2736 struct sock *sk = skb->sk; 2737 2738 *refcounted = true; 2739 if (skb_sk_is_prefetched(skb)) 2740 *refcounted = sk_is_refcounted(sk); 2741 skb->destructor = NULL; 2742 skb->sk = NULL; 2743 return sk; 2744 } 2745 *refcounted = false; 2746 return NULL; 2747 } 2748 2749 /* Checks if this SKB belongs to an HW offloaded socket 2750 * and whether any SW fallbacks are required based on dev. 2751 * Check decrypted mark in case skb_orphan() cleared socket. 2752 */ 2753 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2754 struct net_device *dev) 2755 { 2756 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2757 struct sock *sk = skb->sk; 2758 2759 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2760 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2761 #ifdef CONFIG_TLS_DEVICE 2762 } else if (unlikely(skb->decrypted)) { 2763 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2764 kfree_skb(skb); 2765 skb = NULL; 2766 #endif 2767 } 2768 #endif 2769 2770 return skb; 2771 } 2772 2773 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2774 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2775 */ 2776 static inline bool sk_listener(const struct sock *sk) 2777 { 2778 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2779 } 2780 2781 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2782 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2783 int type); 2784 2785 bool sk_ns_capable(const struct sock *sk, 2786 struct user_namespace *user_ns, int cap); 2787 bool sk_capable(const struct sock *sk, int cap); 2788 bool sk_net_capable(const struct sock *sk, int cap); 2789 2790 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2791 2792 /* Take into consideration the size of the struct sk_buff overhead in the 2793 * determination of these values, since that is non-constant across 2794 * platforms. This makes socket queueing behavior and performance 2795 * not depend upon such differences. 2796 */ 2797 #define _SK_MEM_PACKETS 256 2798 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2799 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2800 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2801 2802 extern __u32 sysctl_wmem_max; 2803 extern __u32 sysctl_rmem_max; 2804 2805 extern int sysctl_tstamp_allow_data; 2806 extern int sysctl_optmem_max; 2807 2808 extern __u32 sysctl_wmem_default; 2809 extern __u32 sysctl_rmem_default; 2810 2811 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2812 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2813 2814 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2815 { 2816 /* Does this proto have per netns sysctl_wmem ? */ 2817 if (proto->sysctl_wmem_offset) 2818 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2819 2820 return *proto->sysctl_wmem; 2821 } 2822 2823 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2824 { 2825 /* Does this proto have per netns sysctl_rmem ? */ 2826 if (proto->sysctl_rmem_offset) 2827 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2828 2829 return *proto->sysctl_rmem; 2830 } 2831 2832 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2833 * Some wifi drivers need to tweak it to get more chunks. 2834 * They can use this helper from their ndo_start_xmit() 2835 */ 2836 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2837 { 2838 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2839 return; 2840 WRITE_ONCE(sk->sk_pacing_shift, val); 2841 } 2842 2843 /* if a socket is bound to a device, check that the given device 2844 * index is either the same or that the socket is bound to an L3 2845 * master device and the given device index is also enslaved to 2846 * that L3 master 2847 */ 2848 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2849 { 2850 int mdif; 2851 2852 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2853 return true; 2854 2855 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2856 if (mdif && mdif == sk->sk_bound_dev_if) 2857 return true; 2858 2859 return false; 2860 } 2861 2862 void sock_def_readable(struct sock *sk); 2863 2864 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2865 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2866 int sock_set_timestamping(struct sock *sk, int optname, 2867 struct so_timestamping timestamping); 2868 2869 void sock_enable_timestamps(struct sock *sk); 2870 void sock_no_linger(struct sock *sk); 2871 void sock_set_keepalive(struct sock *sk); 2872 void sock_set_priority(struct sock *sk, u32 priority); 2873 void sock_set_rcvbuf(struct sock *sk, int val); 2874 void sock_set_mark(struct sock *sk, u32 val); 2875 void sock_set_reuseaddr(struct sock *sk); 2876 void sock_set_reuseport(struct sock *sk); 2877 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2878 2879 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2880 2881 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2882 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2883 sockptr_t optval, int optlen, bool old_timeval); 2884 2885 static inline bool sk_is_readable(struct sock *sk) 2886 { 2887 if (sk->sk_prot->sock_is_readable) 2888 return sk->sk_prot->sock_is_readable(sk); 2889 return false; 2890 } 2891 #endif /* _SOCK_H */ 2892