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/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 union { 165 __addrpair skc_addrpair; 166 struct { 167 __be32 skc_daddr; 168 __be32 skc_rcv_saddr; 169 }; 170 }; 171 union { 172 unsigned int skc_hash; 173 __u16 skc_u16hashes[2]; 174 }; 175 /* skc_dport && skc_num must be grouped as well */ 176 union { 177 __portpair skc_portpair; 178 struct { 179 __be16 skc_dport; 180 __u16 skc_num; 181 }; 182 }; 183 184 unsigned short skc_family; 185 volatile unsigned char skc_state; 186 unsigned char skc_reuse:4; 187 unsigned char skc_reuseport:1; 188 unsigned char skc_ipv6only:1; 189 unsigned char skc_net_refcnt:1; 190 int skc_bound_dev_if; 191 union { 192 struct hlist_node skc_bind_node; 193 struct hlist_node skc_portaddr_node; 194 }; 195 struct proto *skc_prot; 196 possible_net_t skc_net; 197 198 #if IS_ENABLED(CONFIG_IPV6) 199 struct in6_addr skc_v6_daddr; 200 struct in6_addr skc_v6_rcv_saddr; 201 #endif 202 203 atomic64_t skc_cookie; 204 205 /* following fields are padding to force 206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 207 * assuming IPV6 is enabled. We use this padding differently 208 * for different kind of 'sockets' 209 */ 210 union { 211 unsigned long skc_flags; 212 struct sock *skc_listener; /* request_sock */ 213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 214 }; 215 /* 216 * fields between dontcopy_begin/dontcopy_end 217 * are not copied in sock_copy() 218 */ 219 /* private: */ 220 int skc_dontcopy_begin[0]; 221 /* public: */ 222 union { 223 struct hlist_node skc_node; 224 struct hlist_nulls_node skc_nulls_node; 225 }; 226 unsigned short skc_tx_queue_mapping; 227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 228 unsigned short skc_rx_queue_mapping; 229 #endif 230 union { 231 int skc_incoming_cpu; 232 u32 skc_rcv_wnd; 233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 234 }; 235 236 refcount_t skc_refcnt; 237 /* private: */ 238 int skc_dontcopy_end[0]; 239 union { 240 u32 skc_rxhash; 241 u32 skc_window_clamp; 242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 243 }; 244 /* public: */ 245 }; 246 247 struct bpf_local_storage; 248 struct sk_filter; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 304 * @sk_busy_poll_budget: napi processing budget when busypolling 305 * @sk_priority: %SO_PRIORITY setting 306 * @sk_type: socket type (%SOCK_STREAM, etc) 307 * @sk_protocol: which protocol this socket belongs in this network family 308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_txrehash: enable TX hash rethink 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 322 * for timestamping 323 * @sk_tskey: counter to disambiguate concurrent tstamp requests 324 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 325 * @sk_socket: Identd and reporting IO signals 326 * @sk_user_data: RPC layer private data 327 * @sk_frag: cached page frag 328 * @sk_peek_off: current peek_offset value 329 * @sk_send_head: front of stuff to transmit 330 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 331 * @sk_security: used by security modules 332 * @sk_mark: generic packet mark 333 * @sk_cgrp_data: cgroup data for this cgroup 334 * @sk_memcg: this socket's memory cgroup association 335 * @sk_write_pending: a write to stream socket waits to start 336 * @sk_state_change: callback to indicate change in the state of the sock 337 * @sk_data_ready: callback to indicate there is data to be processed 338 * @sk_write_space: callback to indicate there is bf sending space available 339 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 340 * @sk_backlog_rcv: callback to process the backlog 341 * @sk_validate_xmit_skb: ptr to an optional validate function 342 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 343 * @sk_reuseport_cb: reuseport group container 344 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 345 * @sk_rcu: used during RCU grace period 346 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 347 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 348 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 349 * @sk_txtime_unused: unused txtime flags 350 * @ns_tracker: tracker for netns reference 351 * @sk_bind2_node: bind node in the bhash2 table 352 */ 353 struct sock { 354 /* 355 * Now struct inet_timewait_sock also uses sock_common, so please just 356 * don't add nothing before this first member (__sk_common) --acme 357 */ 358 struct sock_common __sk_common; 359 #define sk_node __sk_common.skc_node 360 #define sk_nulls_node __sk_common.skc_nulls_node 361 #define sk_refcnt __sk_common.skc_refcnt 362 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 363 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 364 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 365 #endif 366 367 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 368 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 369 #define sk_hash __sk_common.skc_hash 370 #define sk_portpair __sk_common.skc_portpair 371 #define sk_num __sk_common.skc_num 372 #define sk_dport __sk_common.skc_dport 373 #define sk_addrpair __sk_common.skc_addrpair 374 #define sk_daddr __sk_common.skc_daddr 375 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 376 #define sk_family __sk_common.skc_family 377 #define sk_state __sk_common.skc_state 378 #define sk_reuse __sk_common.skc_reuse 379 #define sk_reuseport __sk_common.skc_reuseport 380 #define sk_ipv6only __sk_common.skc_ipv6only 381 #define sk_net_refcnt __sk_common.skc_net_refcnt 382 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 383 #define sk_bind_node __sk_common.skc_bind_node 384 #define sk_prot __sk_common.skc_prot 385 #define sk_net __sk_common.skc_net 386 #define sk_v6_daddr __sk_common.skc_v6_daddr 387 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 388 #define sk_cookie __sk_common.skc_cookie 389 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 390 #define sk_flags __sk_common.skc_flags 391 #define sk_rxhash __sk_common.skc_rxhash 392 393 /* early demux fields */ 394 struct dst_entry __rcu *sk_rx_dst; 395 int sk_rx_dst_ifindex; 396 u32 sk_rx_dst_cookie; 397 398 socket_lock_t sk_lock; 399 atomic_t sk_drops; 400 int sk_rcvlowat; 401 struct sk_buff_head sk_error_queue; 402 struct sk_buff_head sk_receive_queue; 403 /* 404 * The backlog queue is special, it is always used with 405 * the per-socket spinlock held and requires low latency 406 * access. Therefore we special case it's implementation. 407 * Note : rmem_alloc is in this structure to fill a hole 408 * on 64bit arches, not because its logically part of 409 * backlog. 410 */ 411 struct { 412 atomic_t rmem_alloc; 413 int len; 414 struct sk_buff *head; 415 struct sk_buff *tail; 416 } sk_backlog; 417 418 #define sk_rmem_alloc sk_backlog.rmem_alloc 419 420 int sk_forward_alloc; 421 u32 sk_reserved_mem; 422 #ifdef CONFIG_NET_RX_BUSY_POLL 423 unsigned int sk_ll_usec; 424 /* ===== mostly read cache line ===== */ 425 unsigned int sk_napi_id; 426 #endif 427 int sk_rcvbuf; 428 429 struct sk_filter __rcu *sk_filter; 430 union { 431 struct socket_wq __rcu *sk_wq; 432 /* private: */ 433 struct socket_wq *sk_wq_raw; 434 /* public: */ 435 }; 436 #ifdef CONFIG_XFRM 437 struct xfrm_policy __rcu *sk_policy[2]; 438 #endif 439 440 struct dst_entry __rcu *sk_dst_cache; 441 atomic_t sk_omem_alloc; 442 int sk_sndbuf; 443 444 /* ===== cache line for TX ===== */ 445 int sk_wmem_queued; 446 refcount_t sk_wmem_alloc; 447 unsigned long sk_tsq_flags; 448 union { 449 struct sk_buff *sk_send_head; 450 struct rb_root tcp_rtx_queue; 451 }; 452 struct sk_buff_head sk_write_queue; 453 __s32 sk_peek_off; 454 int sk_write_pending; 455 __u32 sk_dst_pending_confirm; 456 u32 sk_pacing_status; /* see enum sk_pacing */ 457 long sk_sndtimeo; 458 struct timer_list sk_timer; 459 __u32 sk_priority; 460 __u32 sk_mark; 461 unsigned long sk_pacing_rate; /* bytes per second */ 462 unsigned long sk_max_pacing_rate; 463 struct page_frag sk_frag; 464 netdev_features_t sk_route_caps; 465 int sk_gso_type; 466 unsigned int sk_gso_max_size; 467 gfp_t sk_allocation; 468 __u32 sk_txhash; 469 470 /* 471 * Because of non atomicity rules, all 472 * changes are protected by socket lock. 473 */ 474 u8 sk_gso_disabled : 1, 475 sk_kern_sock : 1, 476 sk_no_check_tx : 1, 477 sk_no_check_rx : 1, 478 sk_userlocks : 4; 479 u8 sk_pacing_shift; 480 u16 sk_type; 481 u16 sk_protocol; 482 u16 sk_gso_max_segs; 483 unsigned long sk_lingertime; 484 struct proto *sk_prot_creator; 485 rwlock_t sk_callback_lock; 486 int sk_err, 487 sk_err_soft; 488 u32 sk_ack_backlog; 489 u32 sk_max_ack_backlog; 490 kuid_t sk_uid; 491 u8 sk_txrehash; 492 #ifdef CONFIG_NET_RX_BUSY_POLL 493 u8 sk_prefer_busy_poll; 494 u16 sk_busy_poll_budget; 495 #endif 496 spinlock_t sk_peer_lock; 497 int sk_bind_phc; 498 struct pid *sk_peer_pid; 499 const struct cred *sk_peer_cred; 500 501 long sk_rcvtimeo; 502 ktime_t sk_stamp; 503 #if BITS_PER_LONG==32 504 seqlock_t sk_stamp_seq; 505 #endif 506 u16 sk_tsflags; 507 u8 sk_shutdown; 508 atomic_t sk_tskey; 509 atomic_t sk_zckey; 510 511 u8 sk_clockid; 512 u8 sk_txtime_deadline_mode : 1, 513 sk_txtime_report_errors : 1, 514 sk_txtime_unused : 6; 515 516 struct socket *sk_socket; 517 void *sk_user_data; 518 #ifdef CONFIG_SECURITY 519 void *sk_security; 520 #endif 521 struct sock_cgroup_data sk_cgrp_data; 522 struct mem_cgroup *sk_memcg; 523 void (*sk_state_change)(struct sock *sk); 524 void (*sk_data_ready)(struct sock *sk); 525 void (*sk_write_space)(struct sock *sk); 526 void (*sk_error_report)(struct sock *sk); 527 int (*sk_backlog_rcv)(struct sock *sk, 528 struct sk_buff *skb); 529 #ifdef CONFIG_SOCK_VALIDATE_XMIT 530 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 531 struct net_device *dev, 532 struct sk_buff *skb); 533 #endif 534 void (*sk_destruct)(struct sock *sk); 535 struct sock_reuseport __rcu *sk_reuseport_cb; 536 #ifdef CONFIG_BPF_SYSCALL 537 struct bpf_local_storage __rcu *sk_bpf_storage; 538 #endif 539 struct rcu_head sk_rcu; 540 netns_tracker ns_tracker; 541 struct hlist_node sk_bind2_node; 542 }; 543 544 enum sk_pacing { 545 SK_PACING_NONE = 0, 546 SK_PACING_NEEDED = 1, 547 SK_PACING_FQ = 2, 548 }; 549 550 /* Pointer stored in sk_user_data might not be suitable for copying 551 * when cloning the socket. For instance, it can point to a reference 552 * counted object. sk_user_data bottom bit is set if pointer must not 553 * be copied. 554 */ 555 #define SK_USER_DATA_NOCOPY 1UL 556 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 557 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 558 559 /** 560 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 561 * @sk: socket 562 */ 563 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 564 { 565 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 566 } 567 568 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 569 570 #define rcu_dereference_sk_user_data(sk) \ 571 ({ \ 572 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 573 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 574 }) 575 #define rcu_assign_sk_user_data(sk, ptr) \ 576 ({ \ 577 uintptr_t __tmp = (uintptr_t)(ptr); \ 578 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 579 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 580 }) 581 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 582 ({ \ 583 uintptr_t __tmp = (uintptr_t)(ptr); \ 584 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 585 rcu_assign_pointer(__sk_user_data((sk)), \ 586 __tmp | SK_USER_DATA_NOCOPY); \ 587 }) 588 589 static inline 590 struct net *sock_net(const struct sock *sk) 591 { 592 return read_pnet(&sk->sk_net); 593 } 594 595 static inline 596 void sock_net_set(struct sock *sk, struct net *net) 597 { 598 write_pnet(&sk->sk_net, net); 599 } 600 601 /* 602 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 603 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 604 * on a socket means that the socket will reuse everybody else's port 605 * without looking at the other's sk_reuse value. 606 */ 607 608 #define SK_NO_REUSE 0 609 #define SK_CAN_REUSE 1 610 #define SK_FORCE_REUSE 2 611 612 int sk_set_peek_off(struct sock *sk, int val); 613 614 static inline int sk_peek_offset(struct sock *sk, int flags) 615 { 616 if (unlikely(flags & MSG_PEEK)) { 617 return READ_ONCE(sk->sk_peek_off); 618 } 619 620 return 0; 621 } 622 623 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 624 { 625 s32 off = READ_ONCE(sk->sk_peek_off); 626 627 if (unlikely(off >= 0)) { 628 off = max_t(s32, off - val, 0); 629 WRITE_ONCE(sk->sk_peek_off, off); 630 } 631 } 632 633 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 634 { 635 sk_peek_offset_bwd(sk, -val); 636 } 637 638 /* 639 * Hashed lists helper routines 640 */ 641 static inline struct sock *sk_entry(const struct hlist_node *node) 642 { 643 return hlist_entry(node, struct sock, sk_node); 644 } 645 646 static inline struct sock *__sk_head(const struct hlist_head *head) 647 { 648 return hlist_entry(head->first, struct sock, sk_node); 649 } 650 651 static inline struct sock *sk_head(const struct hlist_head *head) 652 { 653 return hlist_empty(head) ? NULL : __sk_head(head); 654 } 655 656 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 657 { 658 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 659 } 660 661 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 662 { 663 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 664 } 665 666 static inline struct sock *sk_next(const struct sock *sk) 667 { 668 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 669 } 670 671 static inline struct sock *sk_nulls_next(const struct sock *sk) 672 { 673 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 674 hlist_nulls_entry(sk->sk_nulls_node.next, 675 struct sock, sk_nulls_node) : 676 NULL; 677 } 678 679 static inline bool sk_unhashed(const struct sock *sk) 680 { 681 return hlist_unhashed(&sk->sk_node); 682 } 683 684 static inline bool sk_hashed(const struct sock *sk) 685 { 686 return !sk_unhashed(sk); 687 } 688 689 static inline void sk_node_init(struct hlist_node *node) 690 { 691 node->pprev = NULL; 692 } 693 694 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 695 { 696 node->pprev = NULL; 697 } 698 699 static inline void __sk_del_node(struct sock *sk) 700 { 701 __hlist_del(&sk->sk_node); 702 } 703 704 /* NB: equivalent to hlist_del_init_rcu */ 705 static inline bool __sk_del_node_init(struct sock *sk) 706 { 707 if (sk_hashed(sk)) { 708 __sk_del_node(sk); 709 sk_node_init(&sk->sk_node); 710 return true; 711 } 712 return false; 713 } 714 715 /* Grab socket reference count. This operation is valid only 716 when sk is ALREADY grabbed f.e. it is found in hash table 717 or a list and the lookup is made under lock preventing hash table 718 modifications. 719 */ 720 721 static __always_inline void sock_hold(struct sock *sk) 722 { 723 refcount_inc(&sk->sk_refcnt); 724 } 725 726 /* Ungrab socket in the context, which assumes that socket refcnt 727 cannot hit zero, f.e. it is true in context of any socketcall. 728 */ 729 static __always_inline void __sock_put(struct sock *sk) 730 { 731 refcount_dec(&sk->sk_refcnt); 732 } 733 734 static inline bool sk_del_node_init(struct sock *sk) 735 { 736 bool rc = __sk_del_node_init(sk); 737 738 if (rc) { 739 /* paranoid for a while -acme */ 740 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 741 __sock_put(sk); 742 } 743 return rc; 744 } 745 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 746 747 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 748 { 749 if (sk_hashed(sk)) { 750 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 751 return true; 752 } 753 return false; 754 } 755 756 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 757 { 758 bool rc = __sk_nulls_del_node_init_rcu(sk); 759 760 if (rc) { 761 /* paranoid for a while -acme */ 762 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 763 __sock_put(sk); 764 } 765 return rc; 766 } 767 768 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 769 { 770 hlist_add_head(&sk->sk_node, list); 771 } 772 773 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 774 { 775 sock_hold(sk); 776 __sk_add_node(sk, list); 777 } 778 779 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 780 { 781 sock_hold(sk); 782 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 783 sk->sk_family == AF_INET6) 784 hlist_add_tail_rcu(&sk->sk_node, list); 785 else 786 hlist_add_head_rcu(&sk->sk_node, list); 787 } 788 789 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 790 { 791 sock_hold(sk); 792 hlist_add_tail_rcu(&sk->sk_node, list); 793 } 794 795 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 796 { 797 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 798 } 799 800 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 801 { 802 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 803 } 804 805 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 806 { 807 sock_hold(sk); 808 __sk_nulls_add_node_rcu(sk, list); 809 } 810 811 static inline void __sk_del_bind_node(struct sock *sk) 812 { 813 __hlist_del(&sk->sk_bind_node); 814 } 815 816 static inline void sk_add_bind_node(struct sock *sk, 817 struct hlist_head *list) 818 { 819 hlist_add_head(&sk->sk_bind_node, list); 820 } 821 822 static inline void __sk_del_bind2_node(struct sock *sk) 823 { 824 __hlist_del(&sk->sk_bind2_node); 825 } 826 827 static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list) 828 { 829 hlist_add_head(&sk->sk_bind2_node, list); 830 } 831 832 #define sk_for_each(__sk, list) \ 833 hlist_for_each_entry(__sk, list, sk_node) 834 #define sk_for_each_rcu(__sk, list) \ 835 hlist_for_each_entry_rcu(__sk, list, sk_node) 836 #define sk_nulls_for_each(__sk, node, list) \ 837 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 838 #define sk_nulls_for_each_rcu(__sk, node, list) \ 839 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 840 #define sk_for_each_from(__sk) \ 841 hlist_for_each_entry_from(__sk, sk_node) 842 #define sk_nulls_for_each_from(__sk, node) \ 843 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 844 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 845 #define sk_for_each_safe(__sk, tmp, list) \ 846 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 847 #define sk_for_each_bound(__sk, list) \ 848 hlist_for_each_entry(__sk, list, sk_bind_node) 849 #define sk_for_each_bound_bhash2(__sk, list) \ 850 hlist_for_each_entry(__sk, list, sk_bind2_node) 851 852 /** 853 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 854 * @tpos: the type * to use as a loop cursor. 855 * @pos: the &struct hlist_node to use as a loop cursor. 856 * @head: the head for your list. 857 * @offset: offset of hlist_node within the struct. 858 * 859 */ 860 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 861 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 862 pos != NULL && \ 863 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 864 pos = rcu_dereference(hlist_next_rcu(pos))) 865 866 static inline struct user_namespace *sk_user_ns(struct sock *sk) 867 { 868 /* Careful only use this in a context where these parameters 869 * can not change and must all be valid, such as recvmsg from 870 * userspace. 871 */ 872 return sk->sk_socket->file->f_cred->user_ns; 873 } 874 875 /* Sock flags */ 876 enum sock_flags { 877 SOCK_DEAD, 878 SOCK_DONE, 879 SOCK_URGINLINE, 880 SOCK_KEEPOPEN, 881 SOCK_LINGER, 882 SOCK_DESTROY, 883 SOCK_BROADCAST, 884 SOCK_TIMESTAMP, 885 SOCK_ZAPPED, 886 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 887 SOCK_DBG, /* %SO_DEBUG setting */ 888 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 889 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 890 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 891 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 892 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 893 SOCK_FASYNC, /* fasync() active */ 894 SOCK_RXQ_OVFL, 895 SOCK_ZEROCOPY, /* buffers from userspace */ 896 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 897 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 898 * Will use last 4 bytes of packet sent from 899 * user-space instead. 900 */ 901 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 902 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 903 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 904 SOCK_TXTIME, 905 SOCK_XDP, /* XDP is attached */ 906 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 907 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 908 }; 909 910 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 911 912 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 913 { 914 nsk->sk_flags = osk->sk_flags; 915 } 916 917 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 918 { 919 __set_bit(flag, &sk->sk_flags); 920 } 921 922 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 923 { 924 __clear_bit(flag, &sk->sk_flags); 925 } 926 927 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 928 int valbool) 929 { 930 if (valbool) 931 sock_set_flag(sk, bit); 932 else 933 sock_reset_flag(sk, bit); 934 } 935 936 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 937 { 938 return test_bit(flag, &sk->sk_flags); 939 } 940 941 #ifdef CONFIG_NET 942 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 943 static inline int sk_memalloc_socks(void) 944 { 945 return static_branch_unlikely(&memalloc_socks_key); 946 } 947 948 void __receive_sock(struct file *file); 949 #else 950 951 static inline int sk_memalloc_socks(void) 952 { 953 return 0; 954 } 955 956 static inline void __receive_sock(struct file *file) 957 { } 958 #endif 959 960 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 961 { 962 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 963 } 964 965 static inline void sk_acceptq_removed(struct sock *sk) 966 { 967 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 968 } 969 970 static inline void sk_acceptq_added(struct sock *sk) 971 { 972 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 973 } 974 975 /* Note: If you think the test should be: 976 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 977 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 978 */ 979 static inline bool sk_acceptq_is_full(const struct sock *sk) 980 { 981 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 982 } 983 984 /* 985 * Compute minimal free write space needed to queue new packets. 986 */ 987 static inline int sk_stream_min_wspace(const struct sock *sk) 988 { 989 return READ_ONCE(sk->sk_wmem_queued) >> 1; 990 } 991 992 static inline int sk_stream_wspace(const struct sock *sk) 993 { 994 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 995 } 996 997 static inline void sk_wmem_queued_add(struct sock *sk, int val) 998 { 999 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1000 } 1001 1002 void sk_stream_write_space(struct sock *sk); 1003 1004 /* OOB backlog add */ 1005 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1006 { 1007 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1008 skb_dst_force(skb); 1009 1010 if (!sk->sk_backlog.tail) 1011 WRITE_ONCE(sk->sk_backlog.head, skb); 1012 else 1013 sk->sk_backlog.tail->next = skb; 1014 1015 WRITE_ONCE(sk->sk_backlog.tail, skb); 1016 skb->next = NULL; 1017 } 1018 1019 /* 1020 * Take into account size of receive queue and backlog queue 1021 * Do not take into account this skb truesize, 1022 * to allow even a single big packet to come. 1023 */ 1024 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1025 { 1026 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1027 1028 return qsize > limit; 1029 } 1030 1031 /* The per-socket spinlock must be held here. */ 1032 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1033 unsigned int limit) 1034 { 1035 if (sk_rcvqueues_full(sk, limit)) 1036 return -ENOBUFS; 1037 1038 /* 1039 * If the skb was allocated from pfmemalloc reserves, only 1040 * allow SOCK_MEMALLOC sockets to use it as this socket is 1041 * helping free memory 1042 */ 1043 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1044 return -ENOMEM; 1045 1046 __sk_add_backlog(sk, skb); 1047 sk->sk_backlog.len += skb->truesize; 1048 return 0; 1049 } 1050 1051 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1052 1053 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1054 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1055 1056 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1057 { 1058 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1059 return __sk_backlog_rcv(sk, skb); 1060 1061 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1062 tcp_v6_do_rcv, 1063 tcp_v4_do_rcv, 1064 sk, skb); 1065 } 1066 1067 static inline void sk_incoming_cpu_update(struct sock *sk) 1068 { 1069 int cpu = raw_smp_processor_id(); 1070 1071 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1072 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1073 } 1074 1075 static inline void sock_rps_record_flow_hash(__u32 hash) 1076 { 1077 #ifdef CONFIG_RPS 1078 struct rps_sock_flow_table *sock_flow_table; 1079 1080 rcu_read_lock(); 1081 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1082 rps_record_sock_flow(sock_flow_table, hash); 1083 rcu_read_unlock(); 1084 #endif 1085 } 1086 1087 static inline void sock_rps_record_flow(const struct sock *sk) 1088 { 1089 #ifdef CONFIG_RPS 1090 if (static_branch_unlikely(&rfs_needed)) { 1091 /* Reading sk->sk_rxhash might incur an expensive cache line 1092 * miss. 1093 * 1094 * TCP_ESTABLISHED does cover almost all states where RFS 1095 * might be useful, and is cheaper [1] than testing : 1096 * IPv4: inet_sk(sk)->inet_daddr 1097 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1098 * OR an additional socket flag 1099 * [1] : sk_state and sk_prot are in the same cache line. 1100 */ 1101 if (sk->sk_state == TCP_ESTABLISHED) 1102 sock_rps_record_flow_hash(sk->sk_rxhash); 1103 } 1104 #endif 1105 } 1106 1107 static inline void sock_rps_save_rxhash(struct sock *sk, 1108 const struct sk_buff *skb) 1109 { 1110 #ifdef CONFIG_RPS 1111 if (unlikely(sk->sk_rxhash != skb->hash)) 1112 sk->sk_rxhash = skb->hash; 1113 #endif 1114 } 1115 1116 static inline void sock_rps_reset_rxhash(struct sock *sk) 1117 { 1118 #ifdef CONFIG_RPS 1119 sk->sk_rxhash = 0; 1120 #endif 1121 } 1122 1123 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1124 ({ int __rc; \ 1125 release_sock(__sk); \ 1126 __rc = __condition; \ 1127 if (!__rc) { \ 1128 *(__timeo) = wait_woken(__wait, \ 1129 TASK_INTERRUPTIBLE, \ 1130 *(__timeo)); \ 1131 } \ 1132 sched_annotate_sleep(); \ 1133 lock_sock(__sk); \ 1134 __rc = __condition; \ 1135 __rc; \ 1136 }) 1137 1138 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1139 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1140 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1141 int sk_stream_error(struct sock *sk, int flags, int err); 1142 void sk_stream_kill_queues(struct sock *sk); 1143 void sk_set_memalloc(struct sock *sk); 1144 void sk_clear_memalloc(struct sock *sk); 1145 1146 void __sk_flush_backlog(struct sock *sk); 1147 1148 static inline bool sk_flush_backlog(struct sock *sk) 1149 { 1150 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1151 __sk_flush_backlog(sk); 1152 return true; 1153 } 1154 return false; 1155 } 1156 1157 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1158 1159 struct request_sock_ops; 1160 struct timewait_sock_ops; 1161 struct inet_hashinfo; 1162 struct raw_hashinfo; 1163 struct smc_hashinfo; 1164 struct module; 1165 struct sk_psock; 1166 1167 /* 1168 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1169 * un-modified. Special care is taken when initializing object to zero. 1170 */ 1171 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1172 { 1173 if (offsetof(struct sock, sk_node.next) != 0) 1174 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1175 memset(&sk->sk_node.pprev, 0, 1176 size - offsetof(struct sock, sk_node.pprev)); 1177 } 1178 1179 /* Networking protocol blocks we attach to sockets. 1180 * socket layer -> transport layer interface 1181 */ 1182 struct proto { 1183 void (*close)(struct sock *sk, 1184 long timeout); 1185 int (*pre_connect)(struct sock *sk, 1186 struct sockaddr *uaddr, 1187 int addr_len); 1188 int (*connect)(struct sock *sk, 1189 struct sockaddr *uaddr, 1190 int addr_len); 1191 int (*disconnect)(struct sock *sk, int flags); 1192 1193 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1194 bool kern); 1195 1196 int (*ioctl)(struct sock *sk, int cmd, 1197 unsigned long arg); 1198 int (*init)(struct sock *sk); 1199 void (*destroy)(struct sock *sk); 1200 void (*shutdown)(struct sock *sk, int how); 1201 int (*setsockopt)(struct sock *sk, int level, 1202 int optname, sockptr_t optval, 1203 unsigned int optlen); 1204 int (*getsockopt)(struct sock *sk, int level, 1205 int optname, char __user *optval, 1206 int __user *option); 1207 void (*keepalive)(struct sock *sk, int valbool); 1208 #ifdef CONFIG_COMPAT 1209 int (*compat_ioctl)(struct sock *sk, 1210 unsigned int cmd, unsigned long arg); 1211 #endif 1212 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1213 size_t len); 1214 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1215 size_t len, int flags, int *addr_len); 1216 int (*sendpage)(struct sock *sk, struct page *page, 1217 int offset, size_t size, int flags); 1218 int (*bind)(struct sock *sk, 1219 struct sockaddr *addr, int addr_len); 1220 int (*bind_add)(struct sock *sk, 1221 struct sockaddr *addr, int addr_len); 1222 1223 int (*backlog_rcv) (struct sock *sk, 1224 struct sk_buff *skb); 1225 bool (*bpf_bypass_getsockopt)(int level, 1226 int optname); 1227 1228 void (*release_cb)(struct sock *sk); 1229 1230 /* Keeping track of sk's, looking them up, and port selection methods. */ 1231 int (*hash)(struct sock *sk); 1232 void (*unhash)(struct sock *sk); 1233 void (*rehash)(struct sock *sk); 1234 int (*get_port)(struct sock *sk, unsigned short snum); 1235 void (*put_port)(struct sock *sk); 1236 #ifdef CONFIG_BPF_SYSCALL 1237 int (*psock_update_sk_prot)(struct sock *sk, 1238 struct sk_psock *psock, 1239 bool restore); 1240 #endif 1241 1242 /* Keeping track of sockets in use */ 1243 #ifdef CONFIG_PROC_FS 1244 unsigned int inuse_idx; 1245 #endif 1246 1247 #if IS_ENABLED(CONFIG_MPTCP) 1248 int (*forward_alloc_get)(const struct sock *sk); 1249 #endif 1250 1251 bool (*stream_memory_free)(const struct sock *sk, int wake); 1252 bool (*sock_is_readable)(struct sock *sk); 1253 /* Memory pressure */ 1254 void (*enter_memory_pressure)(struct sock *sk); 1255 void (*leave_memory_pressure)(struct sock *sk); 1256 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1257 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1258 1259 /* 1260 * Pressure flag: try to collapse. 1261 * Technical note: it is used by multiple contexts non atomically. 1262 * All the __sk_mem_schedule() is of this nature: accounting 1263 * is strict, actions are advisory and have some latency. 1264 */ 1265 unsigned long *memory_pressure; 1266 long *sysctl_mem; 1267 1268 int *sysctl_wmem; 1269 int *sysctl_rmem; 1270 u32 sysctl_wmem_offset; 1271 u32 sysctl_rmem_offset; 1272 1273 int max_header; 1274 bool no_autobind; 1275 1276 struct kmem_cache *slab; 1277 unsigned int obj_size; 1278 slab_flags_t slab_flags; 1279 unsigned int useroffset; /* Usercopy region offset */ 1280 unsigned int usersize; /* Usercopy region size */ 1281 1282 unsigned int __percpu *orphan_count; 1283 1284 struct request_sock_ops *rsk_prot; 1285 struct timewait_sock_ops *twsk_prot; 1286 1287 union { 1288 struct inet_hashinfo *hashinfo; 1289 struct udp_table *udp_table; 1290 struct raw_hashinfo *raw_hash; 1291 struct smc_hashinfo *smc_hash; 1292 } h; 1293 1294 struct module *owner; 1295 1296 char name[32]; 1297 1298 struct list_head node; 1299 #ifdef SOCK_REFCNT_DEBUG 1300 atomic_t socks; 1301 #endif 1302 int (*diag_destroy)(struct sock *sk, int err); 1303 } __randomize_layout; 1304 1305 int proto_register(struct proto *prot, int alloc_slab); 1306 void proto_unregister(struct proto *prot); 1307 int sock_load_diag_module(int family, int protocol); 1308 1309 #ifdef SOCK_REFCNT_DEBUG 1310 static inline void sk_refcnt_debug_inc(struct sock *sk) 1311 { 1312 atomic_inc(&sk->sk_prot->socks); 1313 } 1314 1315 static inline void sk_refcnt_debug_dec(struct sock *sk) 1316 { 1317 atomic_dec(&sk->sk_prot->socks); 1318 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1319 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1320 } 1321 1322 static inline void sk_refcnt_debug_release(const struct sock *sk) 1323 { 1324 if (refcount_read(&sk->sk_refcnt) != 1) 1325 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1326 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1327 } 1328 #else /* SOCK_REFCNT_DEBUG */ 1329 #define sk_refcnt_debug_inc(sk) do { } while (0) 1330 #define sk_refcnt_debug_dec(sk) do { } while (0) 1331 #define sk_refcnt_debug_release(sk) do { } while (0) 1332 #endif /* SOCK_REFCNT_DEBUG */ 1333 1334 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1335 1336 static inline int sk_forward_alloc_get(const struct sock *sk) 1337 { 1338 #if IS_ENABLED(CONFIG_MPTCP) 1339 if (sk->sk_prot->forward_alloc_get) 1340 return sk->sk_prot->forward_alloc_get(sk); 1341 #endif 1342 return sk->sk_forward_alloc; 1343 } 1344 1345 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1346 { 1347 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1348 return false; 1349 1350 return sk->sk_prot->stream_memory_free ? 1351 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1352 tcp_stream_memory_free, sk, wake) : true; 1353 } 1354 1355 static inline bool sk_stream_memory_free(const struct sock *sk) 1356 { 1357 return __sk_stream_memory_free(sk, 0); 1358 } 1359 1360 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1361 { 1362 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1363 __sk_stream_memory_free(sk, wake); 1364 } 1365 1366 static inline bool sk_stream_is_writeable(const struct sock *sk) 1367 { 1368 return __sk_stream_is_writeable(sk, 0); 1369 } 1370 1371 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1372 struct cgroup *ancestor) 1373 { 1374 #ifdef CONFIG_SOCK_CGROUP_DATA 1375 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1376 ancestor); 1377 #else 1378 return -ENOTSUPP; 1379 #endif 1380 } 1381 1382 static inline bool sk_has_memory_pressure(const struct sock *sk) 1383 { 1384 return sk->sk_prot->memory_pressure != NULL; 1385 } 1386 1387 static inline bool sk_under_memory_pressure(const struct sock *sk) 1388 { 1389 if (!sk->sk_prot->memory_pressure) 1390 return false; 1391 1392 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1393 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1394 return true; 1395 1396 return !!*sk->sk_prot->memory_pressure; 1397 } 1398 1399 static inline long 1400 sk_memory_allocated(const struct sock *sk) 1401 { 1402 return atomic_long_read(sk->sk_prot->memory_allocated); 1403 } 1404 1405 static inline long 1406 sk_memory_allocated_add(struct sock *sk, int amt) 1407 { 1408 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1409 } 1410 1411 static inline void 1412 sk_memory_allocated_sub(struct sock *sk, int amt) 1413 { 1414 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1415 } 1416 1417 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1418 1419 static inline void sk_sockets_allocated_dec(struct sock *sk) 1420 { 1421 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1422 SK_ALLOC_PERCPU_COUNTER_BATCH); 1423 } 1424 1425 static inline void sk_sockets_allocated_inc(struct sock *sk) 1426 { 1427 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1428 SK_ALLOC_PERCPU_COUNTER_BATCH); 1429 } 1430 1431 static inline u64 1432 sk_sockets_allocated_read_positive(struct sock *sk) 1433 { 1434 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1435 } 1436 1437 static inline int 1438 proto_sockets_allocated_sum_positive(struct proto *prot) 1439 { 1440 return percpu_counter_sum_positive(prot->sockets_allocated); 1441 } 1442 1443 static inline long 1444 proto_memory_allocated(struct proto *prot) 1445 { 1446 return atomic_long_read(prot->memory_allocated); 1447 } 1448 1449 static inline bool 1450 proto_memory_pressure(struct proto *prot) 1451 { 1452 if (!prot->memory_pressure) 1453 return false; 1454 return !!*prot->memory_pressure; 1455 } 1456 1457 1458 #ifdef CONFIG_PROC_FS 1459 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1460 struct prot_inuse { 1461 int all; 1462 int val[PROTO_INUSE_NR]; 1463 }; 1464 1465 static inline void sock_prot_inuse_add(const struct net *net, 1466 const struct proto *prot, int val) 1467 { 1468 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1469 } 1470 1471 static inline void sock_inuse_add(const struct net *net, int val) 1472 { 1473 this_cpu_add(net->core.prot_inuse->all, val); 1474 } 1475 1476 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1477 int sock_inuse_get(struct net *net); 1478 #else 1479 static inline void sock_prot_inuse_add(const struct net *net, 1480 const struct proto *prot, int val) 1481 { 1482 } 1483 1484 static inline void sock_inuse_add(const struct net *net, int val) 1485 { 1486 } 1487 #endif 1488 1489 1490 /* With per-bucket locks this operation is not-atomic, so that 1491 * this version is not worse. 1492 */ 1493 static inline int __sk_prot_rehash(struct sock *sk) 1494 { 1495 sk->sk_prot->unhash(sk); 1496 return sk->sk_prot->hash(sk); 1497 } 1498 1499 /* About 10 seconds */ 1500 #define SOCK_DESTROY_TIME (10*HZ) 1501 1502 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1503 #define PROT_SOCK 1024 1504 1505 #define SHUTDOWN_MASK 3 1506 #define RCV_SHUTDOWN 1 1507 #define SEND_SHUTDOWN 2 1508 1509 #define SOCK_BINDADDR_LOCK 4 1510 #define SOCK_BINDPORT_LOCK 8 1511 1512 struct socket_alloc { 1513 struct socket socket; 1514 struct inode vfs_inode; 1515 }; 1516 1517 static inline struct socket *SOCKET_I(struct inode *inode) 1518 { 1519 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1520 } 1521 1522 static inline struct inode *SOCK_INODE(struct socket *socket) 1523 { 1524 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1525 } 1526 1527 /* 1528 * Functions for memory accounting 1529 */ 1530 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1531 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1532 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1533 void __sk_mem_reclaim(struct sock *sk, int amount); 1534 1535 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1536 * do not necessarily have 16x time more memory than 4KB ones. 1537 */ 1538 #define SK_MEM_QUANTUM 4096 1539 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1540 #define SK_MEM_SEND 0 1541 #define SK_MEM_RECV 1 1542 1543 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1544 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1545 { 1546 long val = sk->sk_prot->sysctl_mem[index]; 1547 1548 #if PAGE_SIZE > SK_MEM_QUANTUM 1549 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1550 #elif PAGE_SIZE < SK_MEM_QUANTUM 1551 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1552 #endif 1553 return val; 1554 } 1555 1556 static inline int sk_mem_pages(int amt) 1557 { 1558 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1559 } 1560 1561 static inline bool sk_has_account(struct sock *sk) 1562 { 1563 /* return true if protocol supports memory accounting */ 1564 return !!sk->sk_prot->memory_allocated; 1565 } 1566 1567 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1568 { 1569 if (!sk_has_account(sk)) 1570 return true; 1571 return size <= sk->sk_forward_alloc || 1572 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1573 } 1574 1575 static inline bool 1576 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1577 { 1578 if (!sk_has_account(sk)) 1579 return true; 1580 return size <= sk->sk_forward_alloc || 1581 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1582 skb_pfmemalloc(skb); 1583 } 1584 1585 static inline int sk_unused_reserved_mem(const struct sock *sk) 1586 { 1587 int unused_mem; 1588 1589 if (likely(!sk->sk_reserved_mem)) 1590 return 0; 1591 1592 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1593 atomic_read(&sk->sk_rmem_alloc); 1594 1595 return unused_mem > 0 ? unused_mem : 0; 1596 } 1597 1598 static inline void sk_mem_reclaim(struct sock *sk) 1599 { 1600 int reclaimable; 1601 1602 if (!sk_has_account(sk)) 1603 return; 1604 1605 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1606 1607 if (reclaimable >= SK_MEM_QUANTUM) 1608 __sk_mem_reclaim(sk, reclaimable); 1609 } 1610 1611 static inline void sk_mem_reclaim_final(struct sock *sk) 1612 { 1613 sk->sk_reserved_mem = 0; 1614 sk_mem_reclaim(sk); 1615 } 1616 1617 static inline void sk_mem_reclaim_partial(struct sock *sk) 1618 { 1619 int reclaimable; 1620 1621 if (!sk_has_account(sk)) 1622 return; 1623 1624 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1625 1626 if (reclaimable > SK_MEM_QUANTUM) 1627 __sk_mem_reclaim(sk, reclaimable - 1); 1628 } 1629 1630 static inline void sk_mem_charge(struct sock *sk, int size) 1631 { 1632 if (!sk_has_account(sk)) 1633 return; 1634 sk->sk_forward_alloc -= size; 1635 } 1636 1637 /* the following macros control memory reclaiming in sk_mem_uncharge() 1638 */ 1639 #define SK_RECLAIM_THRESHOLD (1 << 21) 1640 #define SK_RECLAIM_CHUNK (1 << 20) 1641 1642 static inline void sk_mem_uncharge(struct sock *sk, int size) 1643 { 1644 int reclaimable; 1645 1646 if (!sk_has_account(sk)) 1647 return; 1648 sk->sk_forward_alloc += size; 1649 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1650 1651 /* Avoid a possible overflow. 1652 * TCP send queues can make this happen, if sk_mem_reclaim() 1653 * is not called and more than 2 GBytes are released at once. 1654 * 1655 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1656 * no need to hold that much forward allocation anyway. 1657 */ 1658 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD)) 1659 __sk_mem_reclaim(sk, SK_RECLAIM_CHUNK); 1660 } 1661 1662 /* 1663 * Macro so as to not evaluate some arguments when 1664 * lockdep is not enabled. 1665 * 1666 * Mark both the sk_lock and the sk_lock.slock as a 1667 * per-address-family lock class. 1668 */ 1669 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1670 do { \ 1671 sk->sk_lock.owned = 0; \ 1672 init_waitqueue_head(&sk->sk_lock.wq); \ 1673 spin_lock_init(&(sk)->sk_lock.slock); \ 1674 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1675 sizeof((sk)->sk_lock)); \ 1676 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1677 (skey), (sname)); \ 1678 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1679 } while (0) 1680 1681 static inline bool lockdep_sock_is_held(const struct sock *sk) 1682 { 1683 return lockdep_is_held(&sk->sk_lock) || 1684 lockdep_is_held(&sk->sk_lock.slock); 1685 } 1686 1687 void lock_sock_nested(struct sock *sk, int subclass); 1688 1689 static inline void lock_sock(struct sock *sk) 1690 { 1691 lock_sock_nested(sk, 0); 1692 } 1693 1694 void __lock_sock(struct sock *sk); 1695 void __release_sock(struct sock *sk); 1696 void release_sock(struct sock *sk); 1697 1698 /* BH context may only use the following locking interface. */ 1699 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1700 #define bh_lock_sock_nested(__sk) \ 1701 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1702 SINGLE_DEPTH_NESTING) 1703 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1704 1705 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1706 1707 /** 1708 * lock_sock_fast - fast version of lock_sock 1709 * @sk: socket 1710 * 1711 * This version should be used for very small section, where process wont block 1712 * return false if fast path is taken: 1713 * 1714 * sk_lock.slock locked, owned = 0, BH disabled 1715 * 1716 * return true if slow path is taken: 1717 * 1718 * sk_lock.slock unlocked, owned = 1, BH enabled 1719 */ 1720 static inline bool lock_sock_fast(struct sock *sk) 1721 { 1722 /* The sk_lock has mutex_lock() semantics here. */ 1723 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1724 1725 return __lock_sock_fast(sk); 1726 } 1727 1728 /* fast socket lock variant for caller already holding a [different] socket lock */ 1729 static inline bool lock_sock_fast_nested(struct sock *sk) 1730 { 1731 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1732 1733 return __lock_sock_fast(sk); 1734 } 1735 1736 /** 1737 * unlock_sock_fast - complement of lock_sock_fast 1738 * @sk: socket 1739 * @slow: slow mode 1740 * 1741 * fast unlock socket for user context. 1742 * If slow mode is on, we call regular release_sock() 1743 */ 1744 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1745 __releases(&sk->sk_lock.slock) 1746 { 1747 if (slow) { 1748 release_sock(sk); 1749 __release(&sk->sk_lock.slock); 1750 } else { 1751 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1752 spin_unlock_bh(&sk->sk_lock.slock); 1753 } 1754 } 1755 1756 /* Used by processes to "lock" a socket state, so that 1757 * interrupts and bottom half handlers won't change it 1758 * from under us. It essentially blocks any incoming 1759 * packets, so that we won't get any new data or any 1760 * packets that change the state of the socket. 1761 * 1762 * While locked, BH processing will add new packets to 1763 * the backlog queue. This queue is processed by the 1764 * owner of the socket lock right before it is released. 1765 * 1766 * Since ~2.3.5 it is also exclusive sleep lock serializing 1767 * accesses from user process context. 1768 */ 1769 1770 static inline void sock_owned_by_me(const struct sock *sk) 1771 { 1772 #ifdef CONFIG_LOCKDEP 1773 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1774 #endif 1775 } 1776 1777 static inline bool sock_owned_by_user(const struct sock *sk) 1778 { 1779 sock_owned_by_me(sk); 1780 return sk->sk_lock.owned; 1781 } 1782 1783 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1784 { 1785 return sk->sk_lock.owned; 1786 } 1787 1788 static inline void sock_release_ownership(struct sock *sk) 1789 { 1790 if (sock_owned_by_user_nocheck(sk)) { 1791 sk->sk_lock.owned = 0; 1792 1793 /* The sk_lock has mutex_unlock() semantics: */ 1794 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1795 } 1796 } 1797 1798 /* no reclassification while locks are held */ 1799 static inline bool sock_allow_reclassification(const struct sock *csk) 1800 { 1801 struct sock *sk = (struct sock *)csk; 1802 1803 return !sock_owned_by_user_nocheck(sk) && 1804 !spin_is_locked(&sk->sk_lock.slock); 1805 } 1806 1807 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1808 struct proto *prot, int kern); 1809 void sk_free(struct sock *sk); 1810 void sk_destruct(struct sock *sk); 1811 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1812 void sk_free_unlock_clone(struct sock *sk); 1813 1814 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1815 gfp_t priority); 1816 void __sock_wfree(struct sk_buff *skb); 1817 void sock_wfree(struct sk_buff *skb); 1818 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1819 gfp_t priority); 1820 void skb_orphan_partial(struct sk_buff *skb); 1821 void sock_rfree(struct sk_buff *skb); 1822 void sock_efree(struct sk_buff *skb); 1823 #ifdef CONFIG_INET 1824 void sock_edemux(struct sk_buff *skb); 1825 void sock_pfree(struct sk_buff *skb); 1826 #else 1827 #define sock_edemux sock_efree 1828 #endif 1829 1830 int sock_setsockopt(struct socket *sock, int level, int op, 1831 sockptr_t optval, unsigned int optlen); 1832 1833 int sock_getsockopt(struct socket *sock, int level, int op, 1834 char __user *optval, int __user *optlen); 1835 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1836 bool timeval, bool time32); 1837 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1838 unsigned long data_len, int noblock, 1839 int *errcode, int max_page_order); 1840 1841 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1842 unsigned long size, 1843 int noblock, int *errcode) 1844 { 1845 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1846 } 1847 1848 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1849 void sock_kfree_s(struct sock *sk, void *mem, int size); 1850 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1851 void sk_send_sigurg(struct sock *sk); 1852 1853 struct sockcm_cookie { 1854 u64 transmit_time; 1855 u32 mark; 1856 u16 tsflags; 1857 }; 1858 1859 static inline void sockcm_init(struct sockcm_cookie *sockc, 1860 const struct sock *sk) 1861 { 1862 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1863 } 1864 1865 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1866 struct sockcm_cookie *sockc); 1867 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1868 struct sockcm_cookie *sockc); 1869 1870 /* 1871 * Functions to fill in entries in struct proto_ops when a protocol 1872 * does not implement a particular function. 1873 */ 1874 int sock_no_bind(struct socket *, struct sockaddr *, int); 1875 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1876 int sock_no_socketpair(struct socket *, struct socket *); 1877 int sock_no_accept(struct socket *, struct socket *, int, bool); 1878 int sock_no_getname(struct socket *, struct sockaddr *, int); 1879 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1880 int sock_no_listen(struct socket *, int); 1881 int sock_no_shutdown(struct socket *, int); 1882 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1883 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1884 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1885 int sock_no_mmap(struct file *file, struct socket *sock, 1886 struct vm_area_struct *vma); 1887 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1888 size_t size, int flags); 1889 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1890 int offset, size_t size, int flags); 1891 1892 /* 1893 * Functions to fill in entries in struct proto_ops when a protocol 1894 * uses the inet style. 1895 */ 1896 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1897 char __user *optval, int __user *optlen); 1898 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1899 int flags); 1900 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1901 sockptr_t optval, unsigned int optlen); 1902 1903 void sk_common_release(struct sock *sk); 1904 1905 /* 1906 * Default socket callbacks and setup code 1907 */ 1908 1909 /* Initialise core socket variables */ 1910 void sock_init_data(struct socket *sock, struct sock *sk); 1911 1912 /* 1913 * Socket reference counting postulates. 1914 * 1915 * * Each user of socket SHOULD hold a reference count. 1916 * * Each access point to socket (an hash table bucket, reference from a list, 1917 * running timer, skb in flight MUST hold a reference count. 1918 * * When reference count hits 0, it means it will never increase back. 1919 * * When reference count hits 0, it means that no references from 1920 * outside exist to this socket and current process on current CPU 1921 * is last user and may/should destroy this socket. 1922 * * sk_free is called from any context: process, BH, IRQ. When 1923 * it is called, socket has no references from outside -> sk_free 1924 * may release descendant resources allocated by the socket, but 1925 * to the time when it is called, socket is NOT referenced by any 1926 * hash tables, lists etc. 1927 * * Packets, delivered from outside (from network or from another process) 1928 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1929 * when they sit in queue. Otherwise, packets will leak to hole, when 1930 * socket is looked up by one cpu and unhasing is made by another CPU. 1931 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1932 * (leak to backlog). Packet socket does all the processing inside 1933 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1934 * use separate SMP lock, so that they are prone too. 1935 */ 1936 1937 /* Ungrab socket and destroy it, if it was the last reference. */ 1938 static inline void sock_put(struct sock *sk) 1939 { 1940 if (refcount_dec_and_test(&sk->sk_refcnt)) 1941 sk_free(sk); 1942 } 1943 /* Generic version of sock_put(), dealing with all sockets 1944 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1945 */ 1946 void sock_gen_put(struct sock *sk); 1947 1948 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1949 unsigned int trim_cap, bool refcounted); 1950 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1951 const int nested) 1952 { 1953 return __sk_receive_skb(sk, skb, nested, 1, true); 1954 } 1955 1956 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1957 { 1958 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1959 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1960 return; 1961 sk->sk_tx_queue_mapping = tx_queue; 1962 } 1963 1964 #define NO_QUEUE_MAPPING USHRT_MAX 1965 1966 static inline void sk_tx_queue_clear(struct sock *sk) 1967 { 1968 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1969 } 1970 1971 static inline int sk_tx_queue_get(const struct sock *sk) 1972 { 1973 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1974 return sk->sk_tx_queue_mapping; 1975 1976 return -1; 1977 } 1978 1979 static inline void __sk_rx_queue_set(struct sock *sk, 1980 const struct sk_buff *skb, 1981 bool force_set) 1982 { 1983 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1984 if (skb_rx_queue_recorded(skb)) { 1985 u16 rx_queue = skb_get_rx_queue(skb); 1986 1987 if (force_set || 1988 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 1989 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 1990 } 1991 #endif 1992 } 1993 1994 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1995 { 1996 __sk_rx_queue_set(sk, skb, true); 1997 } 1998 1999 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2000 { 2001 __sk_rx_queue_set(sk, skb, false); 2002 } 2003 2004 static inline void sk_rx_queue_clear(struct sock *sk) 2005 { 2006 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2007 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2008 #endif 2009 } 2010 2011 static inline int sk_rx_queue_get(const struct sock *sk) 2012 { 2013 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2014 if (sk) { 2015 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2016 2017 if (res != NO_QUEUE_MAPPING) 2018 return res; 2019 } 2020 #endif 2021 2022 return -1; 2023 } 2024 2025 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2026 { 2027 sk->sk_socket = sock; 2028 } 2029 2030 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2031 { 2032 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2033 return &rcu_dereference_raw(sk->sk_wq)->wait; 2034 } 2035 /* Detach socket from process context. 2036 * Announce socket dead, detach it from wait queue and inode. 2037 * Note that parent inode held reference count on this struct sock, 2038 * we do not release it in this function, because protocol 2039 * probably wants some additional cleanups or even continuing 2040 * to work with this socket (TCP). 2041 */ 2042 static inline void sock_orphan(struct sock *sk) 2043 { 2044 write_lock_bh(&sk->sk_callback_lock); 2045 sock_set_flag(sk, SOCK_DEAD); 2046 sk_set_socket(sk, NULL); 2047 sk->sk_wq = NULL; 2048 write_unlock_bh(&sk->sk_callback_lock); 2049 } 2050 2051 static inline void sock_graft(struct sock *sk, struct socket *parent) 2052 { 2053 WARN_ON(parent->sk); 2054 write_lock_bh(&sk->sk_callback_lock); 2055 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2056 parent->sk = sk; 2057 sk_set_socket(sk, parent); 2058 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2059 security_sock_graft(sk, parent); 2060 write_unlock_bh(&sk->sk_callback_lock); 2061 } 2062 2063 kuid_t sock_i_uid(struct sock *sk); 2064 unsigned long sock_i_ino(struct sock *sk); 2065 2066 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2067 { 2068 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2069 } 2070 2071 static inline u32 net_tx_rndhash(void) 2072 { 2073 u32 v = prandom_u32(); 2074 2075 return v ?: 1; 2076 } 2077 2078 static inline void sk_set_txhash(struct sock *sk) 2079 { 2080 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2081 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2082 } 2083 2084 static inline bool sk_rethink_txhash(struct sock *sk) 2085 { 2086 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2087 sk_set_txhash(sk); 2088 return true; 2089 } 2090 return false; 2091 } 2092 2093 static inline struct dst_entry * 2094 __sk_dst_get(struct sock *sk) 2095 { 2096 return rcu_dereference_check(sk->sk_dst_cache, 2097 lockdep_sock_is_held(sk)); 2098 } 2099 2100 static inline struct dst_entry * 2101 sk_dst_get(struct sock *sk) 2102 { 2103 struct dst_entry *dst; 2104 2105 rcu_read_lock(); 2106 dst = rcu_dereference(sk->sk_dst_cache); 2107 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2108 dst = NULL; 2109 rcu_read_unlock(); 2110 return dst; 2111 } 2112 2113 static inline void __dst_negative_advice(struct sock *sk) 2114 { 2115 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2116 2117 if (dst && dst->ops->negative_advice) { 2118 ndst = dst->ops->negative_advice(dst); 2119 2120 if (ndst != dst) { 2121 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2122 sk_tx_queue_clear(sk); 2123 sk->sk_dst_pending_confirm = 0; 2124 } 2125 } 2126 } 2127 2128 static inline void dst_negative_advice(struct sock *sk) 2129 { 2130 sk_rethink_txhash(sk); 2131 __dst_negative_advice(sk); 2132 } 2133 2134 static inline void 2135 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2136 { 2137 struct dst_entry *old_dst; 2138 2139 sk_tx_queue_clear(sk); 2140 sk->sk_dst_pending_confirm = 0; 2141 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2142 lockdep_sock_is_held(sk)); 2143 rcu_assign_pointer(sk->sk_dst_cache, dst); 2144 dst_release(old_dst); 2145 } 2146 2147 static inline void 2148 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2149 { 2150 struct dst_entry *old_dst; 2151 2152 sk_tx_queue_clear(sk); 2153 sk->sk_dst_pending_confirm = 0; 2154 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2155 dst_release(old_dst); 2156 } 2157 2158 static inline void 2159 __sk_dst_reset(struct sock *sk) 2160 { 2161 __sk_dst_set(sk, NULL); 2162 } 2163 2164 static inline void 2165 sk_dst_reset(struct sock *sk) 2166 { 2167 sk_dst_set(sk, NULL); 2168 } 2169 2170 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2171 2172 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2173 2174 static inline void sk_dst_confirm(struct sock *sk) 2175 { 2176 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2177 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2178 } 2179 2180 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2181 { 2182 if (skb_get_dst_pending_confirm(skb)) { 2183 struct sock *sk = skb->sk; 2184 2185 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2186 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2187 neigh_confirm(n); 2188 } 2189 } 2190 2191 bool sk_mc_loop(struct sock *sk); 2192 2193 static inline bool sk_can_gso(const struct sock *sk) 2194 { 2195 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2196 } 2197 2198 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2199 2200 static inline void sk_gso_disable(struct sock *sk) 2201 { 2202 sk->sk_gso_disabled = 1; 2203 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2204 } 2205 2206 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2207 struct iov_iter *from, char *to, 2208 int copy, int offset) 2209 { 2210 if (skb->ip_summed == CHECKSUM_NONE) { 2211 __wsum csum = 0; 2212 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2213 return -EFAULT; 2214 skb->csum = csum_block_add(skb->csum, csum, offset); 2215 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2216 if (!copy_from_iter_full_nocache(to, copy, from)) 2217 return -EFAULT; 2218 } else if (!copy_from_iter_full(to, copy, from)) 2219 return -EFAULT; 2220 2221 return 0; 2222 } 2223 2224 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2225 struct iov_iter *from, int copy) 2226 { 2227 int err, offset = skb->len; 2228 2229 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2230 copy, offset); 2231 if (err) 2232 __skb_trim(skb, offset); 2233 2234 return err; 2235 } 2236 2237 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2238 struct sk_buff *skb, 2239 struct page *page, 2240 int off, int copy) 2241 { 2242 int err; 2243 2244 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2245 copy, skb->len); 2246 if (err) 2247 return err; 2248 2249 skb->len += copy; 2250 skb->data_len += copy; 2251 skb->truesize += copy; 2252 sk_wmem_queued_add(sk, copy); 2253 sk_mem_charge(sk, copy); 2254 return 0; 2255 } 2256 2257 /** 2258 * sk_wmem_alloc_get - returns write allocations 2259 * @sk: socket 2260 * 2261 * Return: sk_wmem_alloc minus initial offset of one 2262 */ 2263 static inline int sk_wmem_alloc_get(const struct sock *sk) 2264 { 2265 return refcount_read(&sk->sk_wmem_alloc) - 1; 2266 } 2267 2268 /** 2269 * sk_rmem_alloc_get - returns read allocations 2270 * @sk: socket 2271 * 2272 * Return: sk_rmem_alloc 2273 */ 2274 static inline int sk_rmem_alloc_get(const struct sock *sk) 2275 { 2276 return atomic_read(&sk->sk_rmem_alloc); 2277 } 2278 2279 /** 2280 * sk_has_allocations - check if allocations are outstanding 2281 * @sk: socket 2282 * 2283 * Return: true if socket has write or read allocations 2284 */ 2285 static inline bool sk_has_allocations(const struct sock *sk) 2286 { 2287 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2288 } 2289 2290 /** 2291 * skwq_has_sleeper - check if there are any waiting processes 2292 * @wq: struct socket_wq 2293 * 2294 * Return: true if socket_wq has waiting processes 2295 * 2296 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2297 * barrier call. They were added due to the race found within the tcp code. 2298 * 2299 * Consider following tcp code paths:: 2300 * 2301 * CPU1 CPU2 2302 * sys_select receive packet 2303 * ... ... 2304 * __add_wait_queue update tp->rcv_nxt 2305 * ... ... 2306 * tp->rcv_nxt check sock_def_readable 2307 * ... { 2308 * schedule rcu_read_lock(); 2309 * wq = rcu_dereference(sk->sk_wq); 2310 * if (wq && waitqueue_active(&wq->wait)) 2311 * wake_up_interruptible(&wq->wait) 2312 * ... 2313 * } 2314 * 2315 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2316 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2317 * could then endup calling schedule and sleep forever if there are no more 2318 * data on the socket. 2319 * 2320 */ 2321 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2322 { 2323 return wq && wq_has_sleeper(&wq->wait); 2324 } 2325 2326 /** 2327 * sock_poll_wait - place memory barrier behind the poll_wait call. 2328 * @filp: file 2329 * @sock: socket to wait on 2330 * @p: poll_table 2331 * 2332 * See the comments in the wq_has_sleeper function. 2333 */ 2334 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2335 poll_table *p) 2336 { 2337 if (!poll_does_not_wait(p)) { 2338 poll_wait(filp, &sock->wq.wait, p); 2339 /* We need to be sure we are in sync with the 2340 * socket flags modification. 2341 * 2342 * This memory barrier is paired in the wq_has_sleeper. 2343 */ 2344 smp_mb(); 2345 } 2346 } 2347 2348 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2349 { 2350 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2351 u32 txhash = READ_ONCE(sk->sk_txhash); 2352 2353 if (txhash) { 2354 skb->l4_hash = 1; 2355 skb->hash = txhash; 2356 } 2357 } 2358 2359 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2360 2361 /* 2362 * Queue a received datagram if it will fit. Stream and sequenced 2363 * protocols can't normally use this as they need to fit buffers in 2364 * and play with them. 2365 * 2366 * Inlined as it's very short and called for pretty much every 2367 * packet ever received. 2368 */ 2369 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2370 { 2371 skb_orphan(skb); 2372 skb->sk = sk; 2373 skb->destructor = sock_rfree; 2374 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2375 sk_mem_charge(sk, skb->truesize); 2376 } 2377 2378 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2379 { 2380 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2381 skb_orphan(skb); 2382 skb->destructor = sock_efree; 2383 skb->sk = sk; 2384 return true; 2385 } 2386 return false; 2387 } 2388 2389 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2390 { 2391 if (skb->destructor != sock_wfree) { 2392 skb_orphan(skb); 2393 return; 2394 } 2395 skb->slow_gro = 1; 2396 } 2397 2398 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2399 unsigned long expires); 2400 2401 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2402 2403 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2404 2405 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2406 struct sk_buff *skb, unsigned int flags, 2407 void (*destructor)(struct sock *sk, 2408 struct sk_buff *skb)); 2409 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2410 2411 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2412 enum skb_drop_reason *reason); 2413 2414 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2415 { 2416 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2417 } 2418 2419 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2420 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2421 2422 /* 2423 * Recover an error report and clear atomically 2424 */ 2425 2426 static inline int sock_error(struct sock *sk) 2427 { 2428 int err; 2429 2430 /* Avoid an atomic operation for the common case. 2431 * This is racy since another cpu/thread can change sk_err under us. 2432 */ 2433 if (likely(data_race(!sk->sk_err))) 2434 return 0; 2435 2436 err = xchg(&sk->sk_err, 0); 2437 return -err; 2438 } 2439 2440 void sk_error_report(struct sock *sk); 2441 2442 static inline unsigned long sock_wspace(struct sock *sk) 2443 { 2444 int amt = 0; 2445 2446 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2447 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2448 if (amt < 0) 2449 amt = 0; 2450 } 2451 return amt; 2452 } 2453 2454 /* Note: 2455 * We use sk->sk_wq_raw, from contexts knowing this 2456 * pointer is not NULL and cannot disappear/change. 2457 */ 2458 static inline void sk_set_bit(int nr, struct sock *sk) 2459 { 2460 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2461 !sock_flag(sk, SOCK_FASYNC)) 2462 return; 2463 2464 set_bit(nr, &sk->sk_wq_raw->flags); 2465 } 2466 2467 static inline void sk_clear_bit(int nr, struct sock *sk) 2468 { 2469 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2470 !sock_flag(sk, SOCK_FASYNC)) 2471 return; 2472 2473 clear_bit(nr, &sk->sk_wq_raw->flags); 2474 } 2475 2476 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2477 { 2478 if (sock_flag(sk, SOCK_FASYNC)) { 2479 rcu_read_lock(); 2480 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2481 rcu_read_unlock(); 2482 } 2483 } 2484 2485 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2486 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2487 * Note: for send buffers, TCP works better if we can build two skbs at 2488 * minimum. 2489 */ 2490 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2491 2492 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2493 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2494 2495 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2496 { 2497 u32 val; 2498 2499 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2500 return; 2501 2502 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2503 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2504 2505 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2506 } 2507 2508 /** 2509 * sk_page_frag - return an appropriate page_frag 2510 * @sk: socket 2511 * 2512 * Use the per task page_frag instead of the per socket one for 2513 * optimization when we know that we're in process context and own 2514 * everything that's associated with %current. 2515 * 2516 * Both direct reclaim and page faults can nest inside other 2517 * socket operations and end up recursing into sk_page_frag() 2518 * while it's already in use: explicitly avoid task page_frag 2519 * usage if the caller is potentially doing any of them. 2520 * This assumes that page fault handlers use the GFP_NOFS flags. 2521 * 2522 * Return: a per task page_frag if context allows that, 2523 * otherwise a per socket one. 2524 */ 2525 static inline struct page_frag *sk_page_frag(struct sock *sk) 2526 { 2527 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) == 2528 (__GFP_DIRECT_RECLAIM | __GFP_FS)) 2529 return ¤t->task_frag; 2530 2531 return &sk->sk_frag; 2532 } 2533 2534 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2535 2536 /* 2537 * Default write policy as shown to user space via poll/select/SIGIO 2538 */ 2539 static inline bool sock_writeable(const struct sock *sk) 2540 { 2541 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2542 } 2543 2544 static inline gfp_t gfp_any(void) 2545 { 2546 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2547 } 2548 2549 static inline gfp_t gfp_memcg_charge(void) 2550 { 2551 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2552 } 2553 2554 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2555 { 2556 return noblock ? 0 : sk->sk_rcvtimeo; 2557 } 2558 2559 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2560 { 2561 return noblock ? 0 : sk->sk_sndtimeo; 2562 } 2563 2564 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2565 { 2566 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2567 2568 return v ?: 1; 2569 } 2570 2571 /* Alas, with timeout socket operations are not restartable. 2572 * Compare this to poll(). 2573 */ 2574 static inline int sock_intr_errno(long timeo) 2575 { 2576 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2577 } 2578 2579 struct sock_skb_cb { 2580 u32 dropcount; 2581 }; 2582 2583 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2584 * using skb->cb[] would keep using it directly and utilize its 2585 * alignement guarantee. 2586 */ 2587 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2588 sizeof(struct sock_skb_cb))) 2589 2590 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2591 SOCK_SKB_CB_OFFSET)) 2592 2593 #define sock_skb_cb_check_size(size) \ 2594 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2595 2596 static inline void 2597 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2598 { 2599 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2600 atomic_read(&sk->sk_drops) : 0; 2601 } 2602 2603 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2604 { 2605 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2606 2607 atomic_add(segs, &sk->sk_drops); 2608 } 2609 2610 static inline ktime_t sock_read_timestamp(struct sock *sk) 2611 { 2612 #if BITS_PER_LONG==32 2613 unsigned int seq; 2614 ktime_t kt; 2615 2616 do { 2617 seq = read_seqbegin(&sk->sk_stamp_seq); 2618 kt = sk->sk_stamp; 2619 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2620 2621 return kt; 2622 #else 2623 return READ_ONCE(sk->sk_stamp); 2624 #endif 2625 } 2626 2627 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2628 { 2629 #if BITS_PER_LONG==32 2630 write_seqlock(&sk->sk_stamp_seq); 2631 sk->sk_stamp = kt; 2632 write_sequnlock(&sk->sk_stamp_seq); 2633 #else 2634 WRITE_ONCE(sk->sk_stamp, kt); 2635 #endif 2636 } 2637 2638 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2639 struct sk_buff *skb); 2640 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2641 struct sk_buff *skb); 2642 2643 static inline void 2644 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2645 { 2646 ktime_t kt = skb->tstamp; 2647 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2648 2649 /* 2650 * generate control messages if 2651 * - receive time stamping in software requested 2652 * - software time stamp available and wanted 2653 * - hardware time stamps available and wanted 2654 */ 2655 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2656 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2657 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2658 (hwtstamps->hwtstamp && 2659 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2660 __sock_recv_timestamp(msg, sk, skb); 2661 else 2662 sock_write_timestamp(sk, kt); 2663 2664 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2665 __sock_recv_wifi_status(msg, sk, skb); 2666 } 2667 2668 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2669 struct sk_buff *skb); 2670 2671 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2672 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2673 struct sk_buff *skb) 2674 { 2675 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2676 (1UL << SOCK_RCVTSTAMP) | \ 2677 (1UL << SOCK_RCVMARK)) 2678 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2679 SOF_TIMESTAMPING_RAW_HARDWARE) 2680 2681 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2682 __sock_recv_cmsgs(msg, sk, skb); 2683 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2684 sock_write_timestamp(sk, skb->tstamp); 2685 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2686 sock_write_timestamp(sk, 0); 2687 } 2688 2689 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2690 2691 /** 2692 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2693 * @sk: socket sending this packet 2694 * @tsflags: timestamping flags to use 2695 * @tx_flags: completed with instructions for time stamping 2696 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2697 * 2698 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2699 */ 2700 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2701 __u8 *tx_flags, __u32 *tskey) 2702 { 2703 if (unlikely(tsflags)) { 2704 __sock_tx_timestamp(tsflags, tx_flags); 2705 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2706 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2707 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2708 } 2709 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2710 *tx_flags |= SKBTX_WIFI_STATUS; 2711 } 2712 2713 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2714 __u8 *tx_flags) 2715 { 2716 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2717 } 2718 2719 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2720 { 2721 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2722 &skb_shinfo(skb)->tskey); 2723 } 2724 2725 static inline bool sk_is_tcp(const struct sock *sk) 2726 { 2727 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2728 } 2729 2730 /** 2731 * sk_eat_skb - Release a skb if it is no longer needed 2732 * @sk: socket to eat this skb from 2733 * @skb: socket buffer to eat 2734 * 2735 * This routine must be called with interrupts disabled or with the socket 2736 * locked so that the sk_buff queue operation is ok. 2737 */ 2738 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2739 { 2740 __skb_unlink(skb, &sk->sk_receive_queue); 2741 __kfree_skb(skb); 2742 } 2743 2744 static inline bool 2745 skb_sk_is_prefetched(struct sk_buff *skb) 2746 { 2747 #ifdef CONFIG_INET 2748 return skb->destructor == sock_pfree; 2749 #else 2750 return false; 2751 #endif /* CONFIG_INET */ 2752 } 2753 2754 /* This helper checks if a socket is a full socket, 2755 * ie _not_ a timewait or request socket. 2756 */ 2757 static inline bool sk_fullsock(const struct sock *sk) 2758 { 2759 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2760 } 2761 2762 static inline bool 2763 sk_is_refcounted(struct sock *sk) 2764 { 2765 /* Only full sockets have sk->sk_flags. */ 2766 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2767 } 2768 2769 /** 2770 * skb_steal_sock - steal a socket from an sk_buff 2771 * @skb: sk_buff to steal the socket from 2772 * @refcounted: is set to true if the socket is reference-counted 2773 */ 2774 static inline struct sock * 2775 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2776 { 2777 if (skb->sk) { 2778 struct sock *sk = skb->sk; 2779 2780 *refcounted = true; 2781 if (skb_sk_is_prefetched(skb)) 2782 *refcounted = sk_is_refcounted(sk); 2783 skb->destructor = NULL; 2784 skb->sk = NULL; 2785 return sk; 2786 } 2787 *refcounted = false; 2788 return NULL; 2789 } 2790 2791 /* Checks if this SKB belongs to an HW offloaded socket 2792 * and whether any SW fallbacks are required based on dev. 2793 * Check decrypted mark in case skb_orphan() cleared socket. 2794 */ 2795 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2796 struct net_device *dev) 2797 { 2798 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2799 struct sock *sk = skb->sk; 2800 2801 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2802 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2803 #ifdef CONFIG_TLS_DEVICE 2804 } else if (unlikely(skb->decrypted)) { 2805 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2806 kfree_skb(skb); 2807 skb = NULL; 2808 #endif 2809 } 2810 #endif 2811 2812 return skb; 2813 } 2814 2815 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2816 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2817 */ 2818 static inline bool sk_listener(const struct sock *sk) 2819 { 2820 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2821 } 2822 2823 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2824 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2825 int type); 2826 2827 bool sk_ns_capable(const struct sock *sk, 2828 struct user_namespace *user_ns, int cap); 2829 bool sk_capable(const struct sock *sk, int cap); 2830 bool sk_net_capable(const struct sock *sk, int cap); 2831 2832 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2833 2834 /* Take into consideration the size of the struct sk_buff overhead in the 2835 * determination of these values, since that is non-constant across 2836 * platforms. This makes socket queueing behavior and performance 2837 * not depend upon such differences. 2838 */ 2839 #define _SK_MEM_PACKETS 256 2840 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2841 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2842 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2843 2844 extern __u32 sysctl_wmem_max; 2845 extern __u32 sysctl_rmem_max; 2846 2847 extern int sysctl_tstamp_allow_data; 2848 extern int sysctl_optmem_max; 2849 2850 extern __u32 sysctl_wmem_default; 2851 extern __u32 sysctl_rmem_default; 2852 2853 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2854 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2855 2856 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2857 { 2858 /* Does this proto have per netns sysctl_wmem ? */ 2859 if (proto->sysctl_wmem_offset) 2860 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2861 2862 return *proto->sysctl_wmem; 2863 } 2864 2865 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2866 { 2867 /* Does this proto have per netns sysctl_rmem ? */ 2868 if (proto->sysctl_rmem_offset) 2869 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2870 2871 return *proto->sysctl_rmem; 2872 } 2873 2874 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2875 * Some wifi drivers need to tweak it to get more chunks. 2876 * They can use this helper from their ndo_start_xmit() 2877 */ 2878 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2879 { 2880 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2881 return; 2882 WRITE_ONCE(sk->sk_pacing_shift, val); 2883 } 2884 2885 /* if a socket is bound to a device, check that the given device 2886 * index is either the same or that the socket is bound to an L3 2887 * master device and the given device index is also enslaved to 2888 * that L3 master 2889 */ 2890 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2891 { 2892 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2893 int mdif; 2894 2895 if (!bound_dev_if || bound_dev_if == dif) 2896 return true; 2897 2898 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2899 if (mdif && mdif == bound_dev_if) 2900 return true; 2901 2902 return false; 2903 } 2904 2905 void sock_def_readable(struct sock *sk); 2906 2907 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2908 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2909 int sock_set_timestamping(struct sock *sk, int optname, 2910 struct so_timestamping timestamping); 2911 2912 void sock_enable_timestamps(struct sock *sk); 2913 void sock_no_linger(struct sock *sk); 2914 void sock_set_keepalive(struct sock *sk); 2915 void sock_set_priority(struct sock *sk, u32 priority); 2916 void sock_set_rcvbuf(struct sock *sk, int val); 2917 void sock_set_mark(struct sock *sk, u32 val); 2918 void sock_set_reuseaddr(struct sock *sk); 2919 void sock_set_reuseport(struct sock *sk); 2920 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2921 2922 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2923 2924 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2925 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2926 sockptr_t optval, int optlen, bool old_timeval); 2927 2928 static inline bool sk_is_readable(struct sock *sk) 2929 { 2930 if (sk->sk_prot->sock_is_readable) 2931 return sk->sk_prot->sock_is_readable(sk); 2932 return false; 2933 } 2934 #endif /* _SOCK_H */ 2935