1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/filter.h> 60 #include <linux/rculist_nulls.h> 61 #include <linux/poll.h> 62 #include <linux/sockptr.h> 63 #include <linux/indirect_call_wrapper.h> 64 #include <linux/atomic.h> 65 #include <linux/refcount.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 165 * address on 64bit arches : cf INET_MATCH() 166 */ 167 union { 168 __addrpair skc_addrpair; 169 struct { 170 __be32 skc_daddr; 171 __be32 skc_rcv_saddr; 172 }; 173 }; 174 union { 175 unsigned int skc_hash; 176 __u16 skc_u16hashes[2]; 177 }; 178 /* skc_dport && skc_num must be grouped as well */ 179 union { 180 __portpair skc_portpair; 181 struct { 182 __be16 skc_dport; 183 __u16 skc_num; 184 }; 185 }; 186 187 unsigned short skc_family; 188 volatile unsigned char skc_state; 189 unsigned char skc_reuse:4; 190 unsigned char skc_reuseport:1; 191 unsigned char skc_ipv6only:1; 192 unsigned char skc_net_refcnt:1; 193 int skc_bound_dev_if; 194 union { 195 struct hlist_node skc_bind_node; 196 struct hlist_node skc_portaddr_node; 197 }; 198 struct proto *skc_prot; 199 possible_net_t skc_net; 200 201 #if IS_ENABLED(CONFIG_IPV6) 202 struct in6_addr skc_v6_daddr; 203 struct in6_addr skc_v6_rcv_saddr; 204 #endif 205 206 atomic64_t skc_cookie; 207 208 /* following fields are padding to force 209 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 210 * assuming IPV6 is enabled. We use this padding differently 211 * for different kind of 'sockets' 212 */ 213 union { 214 unsigned long skc_flags; 215 struct sock *skc_listener; /* request_sock */ 216 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 217 }; 218 /* 219 * fields between dontcopy_begin/dontcopy_end 220 * are not copied in sock_copy() 221 */ 222 /* private: */ 223 int skc_dontcopy_begin[0]; 224 /* public: */ 225 union { 226 struct hlist_node skc_node; 227 struct hlist_nulls_node skc_nulls_node; 228 }; 229 unsigned short skc_tx_queue_mapping; 230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 231 unsigned short skc_rx_queue_mapping; 232 #endif 233 union { 234 int skc_incoming_cpu; 235 u32 skc_rcv_wnd; 236 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 237 }; 238 239 refcount_t skc_refcnt; 240 /* private: */ 241 int skc_dontcopy_end[0]; 242 union { 243 u32 skc_rxhash; 244 u32 skc_window_clamp; 245 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 246 }; 247 /* public: */ 248 }; 249 250 struct bpf_local_storage; 251 252 /** 253 * struct sock - network layer representation of sockets 254 * @__sk_common: shared layout with inet_timewait_sock 255 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 256 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 257 * @sk_lock: synchronizer 258 * @sk_kern_sock: True if sock is using kernel lock classes 259 * @sk_rcvbuf: size of receive buffer in bytes 260 * @sk_wq: sock wait queue and async head 261 * @sk_rx_dst: receive input route used by early demux 262 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 263 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 264 * @sk_dst_cache: destination cache 265 * @sk_dst_pending_confirm: need to confirm neighbour 266 * @sk_policy: flow policy 267 * @sk_receive_queue: incoming packets 268 * @sk_wmem_alloc: transmit queue bytes committed 269 * @sk_tsq_flags: TCP Small Queues flags 270 * @sk_write_queue: Packet sending queue 271 * @sk_omem_alloc: "o" is "option" or "other" 272 * @sk_wmem_queued: persistent queue size 273 * @sk_forward_alloc: space allocated forward 274 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 275 * @sk_napi_id: id of the last napi context to receive data for sk 276 * @sk_ll_usec: usecs to busypoll when there is no data 277 * @sk_allocation: allocation mode 278 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 279 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 280 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 281 * @sk_sndbuf: size of send buffer in bytes 282 * @__sk_flags_offset: empty field used to determine location of bitfield 283 * @sk_padding: unused element for alignment 284 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 285 * @sk_no_check_rx: allow zero checksum in RX packets 286 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 287 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 288 * @sk_route_forced_caps: static, forced route capabilities 289 * (set in tcp_init_sock()) 290 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 291 * @sk_gso_max_size: Maximum GSO segment size to build 292 * @sk_gso_max_segs: Maximum number of GSO segments 293 * @sk_pacing_shift: scaling factor for TCP Small Queues 294 * @sk_lingertime: %SO_LINGER l_linger setting 295 * @sk_backlog: always used with the per-socket spinlock held 296 * @sk_callback_lock: used with the callbacks in the end of this struct 297 * @sk_error_queue: rarely used 298 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 299 * IPV6_ADDRFORM for instance) 300 * @sk_err: last error 301 * @sk_err_soft: errors that don't cause failure but are the cause of a 302 * persistent failure not just 'timed out' 303 * @sk_drops: raw/udp drops counter 304 * @sk_ack_backlog: current listen backlog 305 * @sk_max_ack_backlog: listen backlog set in listen() 306 * @sk_uid: user id of owner 307 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 308 * @sk_busy_poll_budget: napi processing budget when busypolling 309 * @sk_priority: %SO_PRIORITY setting 310 * @sk_type: socket type (%SOCK_STREAM, etc) 311 * @sk_protocol: which protocol this socket belongs in this network family 312 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 313 * @sk_peer_pid: &struct pid for this socket's peer 314 * @sk_peer_cred: %SO_PEERCRED setting 315 * @sk_rcvlowat: %SO_RCVLOWAT setting 316 * @sk_rcvtimeo: %SO_RCVTIMEO setting 317 * @sk_sndtimeo: %SO_SNDTIMEO setting 318 * @sk_txhash: computed flow hash for use on transmit 319 * @sk_filter: socket filtering instructions 320 * @sk_timer: sock cleanup timer 321 * @sk_stamp: time stamp of last packet received 322 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 323 * @sk_tsflags: SO_TIMESTAMPING flags 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_state_change: callback to indicate change in the state of the sock 340 * @sk_data_ready: callback to indicate there is data to be processed 341 * @sk_write_space: callback to indicate there is bf sending space available 342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 343 * @sk_backlog_rcv: callback to process the backlog 344 * @sk_validate_xmit_skb: ptr to an optional validate function 345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 346 * @sk_reuseport_cb: reuseport group container 347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 348 * @sk_rcu: used during RCU grace period 349 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 350 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 351 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 352 * @sk_txtime_unused: unused txtime flags 353 */ 354 struct sock { 355 /* 356 * Now struct inet_timewait_sock also uses sock_common, so please just 357 * don't add nothing before this first member (__sk_common) --acme 358 */ 359 struct sock_common __sk_common; 360 #define sk_node __sk_common.skc_node 361 #define sk_nulls_node __sk_common.skc_nulls_node 362 #define sk_refcnt __sk_common.skc_refcnt 363 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 364 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 365 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 366 #endif 367 368 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 369 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 370 #define sk_hash __sk_common.skc_hash 371 #define sk_portpair __sk_common.skc_portpair 372 #define sk_num __sk_common.skc_num 373 #define sk_dport __sk_common.skc_dport 374 #define sk_addrpair __sk_common.skc_addrpair 375 #define sk_daddr __sk_common.skc_daddr 376 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 377 #define sk_family __sk_common.skc_family 378 #define sk_state __sk_common.skc_state 379 #define sk_reuse __sk_common.skc_reuse 380 #define sk_reuseport __sk_common.skc_reuseport 381 #define sk_ipv6only __sk_common.skc_ipv6only 382 #define sk_net_refcnt __sk_common.skc_net_refcnt 383 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 384 #define sk_bind_node __sk_common.skc_bind_node 385 #define sk_prot __sk_common.skc_prot 386 #define sk_net __sk_common.skc_net 387 #define sk_v6_daddr __sk_common.skc_v6_daddr 388 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 389 #define sk_cookie __sk_common.skc_cookie 390 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 391 #define sk_flags __sk_common.skc_flags 392 #define sk_rxhash __sk_common.skc_rxhash 393 394 socket_lock_t sk_lock; 395 atomic_t sk_drops; 396 int sk_rcvlowat; 397 struct sk_buff_head sk_error_queue; 398 struct sk_buff_head sk_receive_queue; 399 /* 400 * The backlog queue is special, it is always used with 401 * the per-socket spinlock held and requires low latency 402 * access. Therefore we special case it's implementation. 403 * Note : rmem_alloc is in this structure to fill a hole 404 * on 64bit arches, not because its logically part of 405 * backlog. 406 */ 407 struct { 408 atomic_t rmem_alloc; 409 int len; 410 struct sk_buff *head; 411 struct sk_buff *tail; 412 } sk_backlog; 413 #define sk_rmem_alloc sk_backlog.rmem_alloc 414 415 int sk_forward_alloc; 416 u32 sk_reserved_mem; 417 #ifdef CONFIG_NET_RX_BUSY_POLL 418 unsigned int sk_ll_usec; 419 /* ===== mostly read cache line ===== */ 420 unsigned int sk_napi_id; 421 #endif 422 int sk_rcvbuf; 423 424 struct sk_filter __rcu *sk_filter; 425 union { 426 struct socket_wq __rcu *sk_wq; 427 /* private: */ 428 struct socket_wq *sk_wq_raw; 429 /* public: */ 430 }; 431 #ifdef CONFIG_XFRM 432 struct xfrm_policy __rcu *sk_policy[2]; 433 #endif 434 struct dst_entry *sk_rx_dst; 435 int sk_rx_dst_ifindex; 436 u32 sk_rx_dst_cookie; 437 438 struct dst_entry __rcu *sk_dst_cache; 439 atomic_t sk_omem_alloc; 440 int sk_sndbuf; 441 442 /* ===== cache line for TX ===== */ 443 int sk_wmem_queued; 444 refcount_t sk_wmem_alloc; 445 unsigned long sk_tsq_flags; 446 union { 447 struct sk_buff *sk_send_head; 448 struct rb_root tcp_rtx_queue; 449 }; 450 struct sk_buff_head sk_write_queue; 451 __s32 sk_peek_off; 452 int sk_write_pending; 453 __u32 sk_dst_pending_confirm; 454 u32 sk_pacing_status; /* see enum sk_pacing */ 455 long sk_sndtimeo; 456 struct timer_list sk_timer; 457 __u32 sk_priority; 458 __u32 sk_mark; 459 unsigned long sk_pacing_rate; /* bytes per second */ 460 unsigned long sk_max_pacing_rate; 461 struct page_frag sk_frag; 462 netdev_features_t sk_route_caps; 463 netdev_features_t sk_route_nocaps; 464 netdev_features_t sk_route_forced_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_padding : 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 #ifdef CONFIG_NET_RX_BUSY_POLL 492 u8 sk_prefer_busy_poll; 493 u16 sk_busy_poll_budget; 494 #endif 495 spinlock_t sk_peer_lock; 496 struct pid *sk_peer_pid; 497 const struct cred *sk_peer_cred; 498 499 long sk_rcvtimeo; 500 ktime_t sk_stamp; 501 #if BITS_PER_LONG==32 502 seqlock_t sk_stamp_seq; 503 #endif 504 u16 sk_tsflags; 505 int sk_bind_phc; 506 u8 sk_shutdown; 507 u32 sk_tskey; 508 atomic_t sk_zckey; 509 510 u8 sk_clockid; 511 u8 sk_txtime_deadline_mode : 1, 512 sk_txtime_report_errors : 1, 513 sk_txtime_unused : 6; 514 515 struct socket *sk_socket; 516 void *sk_user_data; 517 #ifdef CONFIG_SECURITY 518 void *sk_security; 519 #endif 520 struct sock_cgroup_data sk_cgrp_data; 521 struct mem_cgroup *sk_memcg; 522 void (*sk_state_change)(struct sock *sk); 523 void (*sk_data_ready)(struct sock *sk); 524 void (*sk_write_space)(struct sock *sk); 525 void (*sk_error_report)(struct sock *sk); 526 int (*sk_backlog_rcv)(struct sock *sk, 527 struct sk_buff *skb); 528 #ifdef CONFIG_SOCK_VALIDATE_XMIT 529 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 530 struct net_device *dev, 531 struct sk_buff *skb); 532 #endif 533 void (*sk_destruct)(struct sock *sk); 534 struct sock_reuseport __rcu *sk_reuseport_cb; 535 #ifdef CONFIG_BPF_SYSCALL 536 struct bpf_local_storage __rcu *sk_bpf_storage; 537 #endif 538 struct rcu_head sk_rcu; 539 }; 540 541 enum sk_pacing { 542 SK_PACING_NONE = 0, 543 SK_PACING_NEEDED = 1, 544 SK_PACING_FQ = 2, 545 }; 546 547 /* Pointer stored in sk_user_data might not be suitable for copying 548 * when cloning the socket. For instance, it can point to a reference 549 * counted object. sk_user_data bottom bit is set if pointer must not 550 * be copied. 551 */ 552 #define SK_USER_DATA_NOCOPY 1UL 553 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 554 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 555 556 /** 557 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 558 * @sk: socket 559 */ 560 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 561 { 562 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 563 } 564 565 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 566 567 #define rcu_dereference_sk_user_data(sk) \ 568 ({ \ 569 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 570 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 571 }) 572 #define rcu_assign_sk_user_data(sk, ptr) \ 573 ({ \ 574 uintptr_t __tmp = (uintptr_t)(ptr); \ 575 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 576 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 577 }) 578 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 579 ({ \ 580 uintptr_t __tmp = (uintptr_t)(ptr); \ 581 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 582 rcu_assign_pointer(__sk_user_data((sk)), \ 583 __tmp | SK_USER_DATA_NOCOPY); \ 584 }) 585 586 /* 587 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 588 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 589 * on a socket means that the socket will reuse everybody else's port 590 * without looking at the other's sk_reuse value. 591 */ 592 593 #define SK_NO_REUSE 0 594 #define SK_CAN_REUSE 1 595 #define SK_FORCE_REUSE 2 596 597 int sk_set_peek_off(struct sock *sk, int val); 598 599 static inline int sk_peek_offset(struct sock *sk, int flags) 600 { 601 if (unlikely(flags & MSG_PEEK)) { 602 return READ_ONCE(sk->sk_peek_off); 603 } 604 605 return 0; 606 } 607 608 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 609 { 610 s32 off = READ_ONCE(sk->sk_peek_off); 611 612 if (unlikely(off >= 0)) { 613 off = max_t(s32, off - val, 0); 614 WRITE_ONCE(sk->sk_peek_off, off); 615 } 616 } 617 618 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 619 { 620 sk_peek_offset_bwd(sk, -val); 621 } 622 623 /* 624 * Hashed lists helper routines 625 */ 626 static inline struct sock *sk_entry(const struct hlist_node *node) 627 { 628 return hlist_entry(node, struct sock, sk_node); 629 } 630 631 static inline struct sock *__sk_head(const struct hlist_head *head) 632 { 633 return hlist_entry(head->first, struct sock, sk_node); 634 } 635 636 static inline struct sock *sk_head(const struct hlist_head *head) 637 { 638 return hlist_empty(head) ? NULL : __sk_head(head); 639 } 640 641 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 642 { 643 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 644 } 645 646 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 647 { 648 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 649 } 650 651 static inline struct sock *sk_next(const struct sock *sk) 652 { 653 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 654 } 655 656 static inline struct sock *sk_nulls_next(const struct sock *sk) 657 { 658 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 659 hlist_nulls_entry(sk->sk_nulls_node.next, 660 struct sock, sk_nulls_node) : 661 NULL; 662 } 663 664 static inline bool sk_unhashed(const struct sock *sk) 665 { 666 return hlist_unhashed(&sk->sk_node); 667 } 668 669 static inline bool sk_hashed(const struct sock *sk) 670 { 671 return !sk_unhashed(sk); 672 } 673 674 static inline void sk_node_init(struct hlist_node *node) 675 { 676 node->pprev = NULL; 677 } 678 679 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 680 { 681 node->pprev = NULL; 682 } 683 684 static inline void __sk_del_node(struct sock *sk) 685 { 686 __hlist_del(&sk->sk_node); 687 } 688 689 /* NB: equivalent to hlist_del_init_rcu */ 690 static inline bool __sk_del_node_init(struct sock *sk) 691 { 692 if (sk_hashed(sk)) { 693 __sk_del_node(sk); 694 sk_node_init(&sk->sk_node); 695 return true; 696 } 697 return false; 698 } 699 700 /* Grab socket reference count. This operation is valid only 701 when sk is ALREADY grabbed f.e. it is found in hash table 702 or a list and the lookup is made under lock preventing hash table 703 modifications. 704 */ 705 706 static __always_inline void sock_hold(struct sock *sk) 707 { 708 refcount_inc(&sk->sk_refcnt); 709 } 710 711 /* Ungrab socket in the context, which assumes that socket refcnt 712 cannot hit zero, f.e. it is true in context of any socketcall. 713 */ 714 static __always_inline void __sock_put(struct sock *sk) 715 { 716 refcount_dec(&sk->sk_refcnt); 717 } 718 719 static inline bool sk_del_node_init(struct sock *sk) 720 { 721 bool rc = __sk_del_node_init(sk); 722 723 if (rc) { 724 /* paranoid for a while -acme */ 725 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 726 __sock_put(sk); 727 } 728 return rc; 729 } 730 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 731 732 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 733 { 734 if (sk_hashed(sk)) { 735 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 736 return true; 737 } 738 return false; 739 } 740 741 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 742 { 743 bool rc = __sk_nulls_del_node_init_rcu(sk); 744 745 if (rc) { 746 /* paranoid for a while -acme */ 747 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 748 __sock_put(sk); 749 } 750 return rc; 751 } 752 753 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 754 { 755 hlist_add_head(&sk->sk_node, list); 756 } 757 758 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 759 { 760 sock_hold(sk); 761 __sk_add_node(sk, list); 762 } 763 764 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 765 { 766 sock_hold(sk); 767 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 768 sk->sk_family == AF_INET6) 769 hlist_add_tail_rcu(&sk->sk_node, list); 770 else 771 hlist_add_head_rcu(&sk->sk_node, list); 772 } 773 774 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 775 { 776 sock_hold(sk); 777 hlist_add_tail_rcu(&sk->sk_node, list); 778 } 779 780 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 781 { 782 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 783 } 784 785 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 786 { 787 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 788 } 789 790 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 791 { 792 sock_hold(sk); 793 __sk_nulls_add_node_rcu(sk, list); 794 } 795 796 static inline void __sk_del_bind_node(struct sock *sk) 797 { 798 __hlist_del(&sk->sk_bind_node); 799 } 800 801 static inline void sk_add_bind_node(struct sock *sk, 802 struct hlist_head *list) 803 { 804 hlist_add_head(&sk->sk_bind_node, list); 805 } 806 807 #define sk_for_each(__sk, list) \ 808 hlist_for_each_entry(__sk, list, sk_node) 809 #define sk_for_each_rcu(__sk, list) \ 810 hlist_for_each_entry_rcu(__sk, list, sk_node) 811 #define sk_nulls_for_each(__sk, node, list) \ 812 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 813 #define sk_nulls_for_each_rcu(__sk, node, list) \ 814 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 815 #define sk_for_each_from(__sk) \ 816 hlist_for_each_entry_from(__sk, sk_node) 817 #define sk_nulls_for_each_from(__sk, node) \ 818 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 819 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 820 #define sk_for_each_safe(__sk, tmp, list) \ 821 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 822 #define sk_for_each_bound(__sk, list) \ 823 hlist_for_each_entry(__sk, list, sk_bind_node) 824 825 /** 826 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 827 * @tpos: the type * to use as a loop cursor. 828 * @pos: the &struct hlist_node to use as a loop cursor. 829 * @head: the head for your list. 830 * @offset: offset of hlist_node within the struct. 831 * 832 */ 833 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 834 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 835 pos != NULL && \ 836 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 837 pos = rcu_dereference(hlist_next_rcu(pos))) 838 839 static inline struct user_namespace *sk_user_ns(struct sock *sk) 840 { 841 /* Careful only use this in a context where these parameters 842 * can not change and must all be valid, such as recvmsg from 843 * userspace. 844 */ 845 return sk->sk_socket->file->f_cred->user_ns; 846 } 847 848 /* Sock flags */ 849 enum sock_flags { 850 SOCK_DEAD, 851 SOCK_DONE, 852 SOCK_URGINLINE, 853 SOCK_KEEPOPEN, 854 SOCK_LINGER, 855 SOCK_DESTROY, 856 SOCK_BROADCAST, 857 SOCK_TIMESTAMP, 858 SOCK_ZAPPED, 859 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 860 SOCK_DBG, /* %SO_DEBUG setting */ 861 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 862 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 863 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 864 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 865 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 866 SOCK_FASYNC, /* fasync() active */ 867 SOCK_RXQ_OVFL, 868 SOCK_ZEROCOPY, /* buffers from userspace */ 869 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 870 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 871 * Will use last 4 bytes of packet sent from 872 * user-space instead. 873 */ 874 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 875 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 876 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 877 SOCK_TXTIME, 878 SOCK_XDP, /* XDP is attached */ 879 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 880 }; 881 882 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 883 884 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 885 { 886 nsk->sk_flags = osk->sk_flags; 887 } 888 889 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 890 { 891 __set_bit(flag, &sk->sk_flags); 892 } 893 894 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 895 { 896 __clear_bit(flag, &sk->sk_flags); 897 } 898 899 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 900 int valbool) 901 { 902 if (valbool) 903 sock_set_flag(sk, bit); 904 else 905 sock_reset_flag(sk, bit); 906 } 907 908 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 909 { 910 return test_bit(flag, &sk->sk_flags); 911 } 912 913 #ifdef CONFIG_NET 914 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 915 static inline int sk_memalloc_socks(void) 916 { 917 return static_branch_unlikely(&memalloc_socks_key); 918 } 919 920 void __receive_sock(struct file *file); 921 #else 922 923 static inline int sk_memalloc_socks(void) 924 { 925 return 0; 926 } 927 928 static inline void __receive_sock(struct file *file) 929 { } 930 #endif 931 932 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 933 { 934 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 935 } 936 937 static inline void sk_acceptq_removed(struct sock *sk) 938 { 939 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 940 } 941 942 static inline void sk_acceptq_added(struct sock *sk) 943 { 944 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 945 } 946 947 /* Note: If you think the test should be: 948 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 949 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 950 */ 951 static inline bool sk_acceptq_is_full(const struct sock *sk) 952 { 953 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 954 } 955 956 /* 957 * Compute minimal free write space needed to queue new packets. 958 */ 959 static inline int sk_stream_min_wspace(const struct sock *sk) 960 { 961 return READ_ONCE(sk->sk_wmem_queued) >> 1; 962 } 963 964 static inline int sk_stream_wspace(const struct sock *sk) 965 { 966 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 967 } 968 969 static inline void sk_wmem_queued_add(struct sock *sk, int val) 970 { 971 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 972 } 973 974 void sk_stream_write_space(struct sock *sk); 975 976 /* OOB backlog add */ 977 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 978 { 979 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 980 skb_dst_force(skb); 981 982 if (!sk->sk_backlog.tail) 983 WRITE_ONCE(sk->sk_backlog.head, skb); 984 else 985 sk->sk_backlog.tail->next = skb; 986 987 WRITE_ONCE(sk->sk_backlog.tail, skb); 988 skb->next = NULL; 989 } 990 991 /* 992 * Take into account size of receive queue and backlog queue 993 * Do not take into account this skb truesize, 994 * to allow even a single big packet to come. 995 */ 996 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 997 { 998 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 999 1000 return qsize > limit; 1001 } 1002 1003 /* The per-socket spinlock must be held here. */ 1004 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1005 unsigned int limit) 1006 { 1007 if (sk_rcvqueues_full(sk, limit)) 1008 return -ENOBUFS; 1009 1010 /* 1011 * If the skb was allocated from pfmemalloc reserves, only 1012 * allow SOCK_MEMALLOC sockets to use it as this socket is 1013 * helping free memory 1014 */ 1015 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1016 return -ENOMEM; 1017 1018 __sk_add_backlog(sk, skb); 1019 sk->sk_backlog.len += skb->truesize; 1020 return 0; 1021 } 1022 1023 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1024 1025 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1026 { 1027 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1028 return __sk_backlog_rcv(sk, skb); 1029 1030 return sk->sk_backlog_rcv(sk, skb); 1031 } 1032 1033 static inline void sk_incoming_cpu_update(struct sock *sk) 1034 { 1035 int cpu = raw_smp_processor_id(); 1036 1037 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1038 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1039 } 1040 1041 static inline void sock_rps_record_flow_hash(__u32 hash) 1042 { 1043 #ifdef CONFIG_RPS 1044 struct rps_sock_flow_table *sock_flow_table; 1045 1046 rcu_read_lock(); 1047 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1048 rps_record_sock_flow(sock_flow_table, hash); 1049 rcu_read_unlock(); 1050 #endif 1051 } 1052 1053 static inline void sock_rps_record_flow(const struct sock *sk) 1054 { 1055 #ifdef CONFIG_RPS 1056 if (static_branch_unlikely(&rfs_needed)) { 1057 /* Reading sk->sk_rxhash might incur an expensive cache line 1058 * miss. 1059 * 1060 * TCP_ESTABLISHED does cover almost all states where RFS 1061 * might be useful, and is cheaper [1] than testing : 1062 * IPv4: inet_sk(sk)->inet_daddr 1063 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1064 * OR an additional socket flag 1065 * [1] : sk_state and sk_prot are in the same cache line. 1066 */ 1067 if (sk->sk_state == TCP_ESTABLISHED) 1068 sock_rps_record_flow_hash(sk->sk_rxhash); 1069 } 1070 #endif 1071 } 1072 1073 static inline void sock_rps_save_rxhash(struct sock *sk, 1074 const struct sk_buff *skb) 1075 { 1076 #ifdef CONFIG_RPS 1077 if (unlikely(sk->sk_rxhash != skb->hash)) 1078 sk->sk_rxhash = skb->hash; 1079 #endif 1080 } 1081 1082 static inline void sock_rps_reset_rxhash(struct sock *sk) 1083 { 1084 #ifdef CONFIG_RPS 1085 sk->sk_rxhash = 0; 1086 #endif 1087 } 1088 1089 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1090 ({ int __rc; \ 1091 release_sock(__sk); \ 1092 __rc = __condition; \ 1093 if (!__rc) { \ 1094 *(__timeo) = wait_woken(__wait, \ 1095 TASK_INTERRUPTIBLE, \ 1096 *(__timeo)); \ 1097 } \ 1098 sched_annotate_sleep(); \ 1099 lock_sock(__sk); \ 1100 __rc = __condition; \ 1101 __rc; \ 1102 }) 1103 1104 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1105 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1106 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1107 int sk_stream_error(struct sock *sk, int flags, int err); 1108 void sk_stream_kill_queues(struct sock *sk); 1109 void sk_set_memalloc(struct sock *sk); 1110 void sk_clear_memalloc(struct sock *sk); 1111 1112 void __sk_flush_backlog(struct sock *sk); 1113 1114 static inline bool sk_flush_backlog(struct sock *sk) 1115 { 1116 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1117 __sk_flush_backlog(sk); 1118 return true; 1119 } 1120 return false; 1121 } 1122 1123 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1124 1125 struct request_sock_ops; 1126 struct timewait_sock_ops; 1127 struct inet_hashinfo; 1128 struct raw_hashinfo; 1129 struct smc_hashinfo; 1130 struct module; 1131 struct sk_psock; 1132 1133 /* 1134 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1135 * un-modified. Special care is taken when initializing object to zero. 1136 */ 1137 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1138 { 1139 if (offsetof(struct sock, sk_node.next) != 0) 1140 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1141 memset(&sk->sk_node.pprev, 0, 1142 size - offsetof(struct sock, sk_node.pprev)); 1143 } 1144 1145 /* Networking protocol blocks we attach to sockets. 1146 * socket layer -> transport layer interface 1147 */ 1148 struct proto { 1149 void (*close)(struct sock *sk, 1150 long timeout); 1151 int (*pre_connect)(struct sock *sk, 1152 struct sockaddr *uaddr, 1153 int addr_len); 1154 int (*connect)(struct sock *sk, 1155 struct sockaddr *uaddr, 1156 int addr_len); 1157 int (*disconnect)(struct sock *sk, int flags); 1158 1159 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1160 bool kern); 1161 1162 int (*ioctl)(struct sock *sk, int cmd, 1163 unsigned long arg); 1164 int (*init)(struct sock *sk); 1165 void (*destroy)(struct sock *sk); 1166 void (*shutdown)(struct sock *sk, int how); 1167 int (*setsockopt)(struct sock *sk, int level, 1168 int optname, sockptr_t optval, 1169 unsigned int optlen); 1170 int (*getsockopt)(struct sock *sk, int level, 1171 int optname, char __user *optval, 1172 int __user *option); 1173 void (*keepalive)(struct sock *sk, int valbool); 1174 #ifdef CONFIG_COMPAT 1175 int (*compat_ioctl)(struct sock *sk, 1176 unsigned int cmd, unsigned long arg); 1177 #endif 1178 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1179 size_t len); 1180 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1181 size_t len, int noblock, int flags, 1182 int *addr_len); 1183 int (*sendpage)(struct sock *sk, struct page *page, 1184 int offset, size_t size, int flags); 1185 int (*bind)(struct sock *sk, 1186 struct sockaddr *addr, int addr_len); 1187 int (*bind_add)(struct sock *sk, 1188 struct sockaddr *addr, int addr_len); 1189 1190 int (*backlog_rcv) (struct sock *sk, 1191 struct sk_buff *skb); 1192 bool (*bpf_bypass_getsockopt)(int level, 1193 int optname); 1194 1195 void (*release_cb)(struct sock *sk); 1196 1197 /* Keeping track of sk's, looking them up, and port selection methods. */ 1198 int (*hash)(struct sock *sk); 1199 void (*unhash)(struct sock *sk); 1200 void (*rehash)(struct sock *sk); 1201 int (*get_port)(struct sock *sk, unsigned short snum); 1202 #ifdef CONFIG_BPF_SYSCALL 1203 int (*psock_update_sk_prot)(struct sock *sk, 1204 struct sk_psock *psock, 1205 bool restore); 1206 #endif 1207 1208 /* Keeping track of sockets in use */ 1209 #ifdef CONFIG_PROC_FS 1210 unsigned int inuse_idx; 1211 #endif 1212 1213 int (*forward_alloc_get)(const struct sock *sk); 1214 1215 bool (*stream_memory_free)(const struct sock *sk, int wake); 1216 bool (*sock_is_readable)(struct sock *sk); 1217 /* Memory pressure */ 1218 void (*enter_memory_pressure)(struct sock *sk); 1219 void (*leave_memory_pressure)(struct sock *sk); 1220 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1221 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1222 1223 /* 1224 * Pressure flag: try to collapse. 1225 * Technical note: it is used by multiple contexts non atomically. 1226 * All the __sk_mem_schedule() is of this nature: accounting 1227 * is strict, actions are advisory and have some latency. 1228 */ 1229 unsigned long *memory_pressure; 1230 long *sysctl_mem; 1231 1232 int *sysctl_wmem; 1233 int *sysctl_rmem; 1234 u32 sysctl_wmem_offset; 1235 u32 sysctl_rmem_offset; 1236 1237 int max_header; 1238 bool no_autobind; 1239 1240 struct kmem_cache *slab; 1241 unsigned int obj_size; 1242 slab_flags_t slab_flags; 1243 unsigned int useroffset; /* Usercopy region offset */ 1244 unsigned int usersize; /* Usercopy region size */ 1245 1246 unsigned int __percpu *orphan_count; 1247 1248 struct request_sock_ops *rsk_prot; 1249 struct timewait_sock_ops *twsk_prot; 1250 1251 union { 1252 struct inet_hashinfo *hashinfo; 1253 struct udp_table *udp_table; 1254 struct raw_hashinfo *raw_hash; 1255 struct smc_hashinfo *smc_hash; 1256 } h; 1257 1258 struct module *owner; 1259 1260 char name[32]; 1261 1262 struct list_head node; 1263 #ifdef SOCK_REFCNT_DEBUG 1264 atomic_t socks; 1265 #endif 1266 int (*diag_destroy)(struct sock *sk, int err); 1267 } __randomize_layout; 1268 1269 int proto_register(struct proto *prot, int alloc_slab); 1270 void proto_unregister(struct proto *prot); 1271 int sock_load_diag_module(int family, int protocol); 1272 1273 #ifdef SOCK_REFCNT_DEBUG 1274 static inline void sk_refcnt_debug_inc(struct sock *sk) 1275 { 1276 atomic_inc(&sk->sk_prot->socks); 1277 } 1278 1279 static inline void sk_refcnt_debug_dec(struct sock *sk) 1280 { 1281 atomic_dec(&sk->sk_prot->socks); 1282 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1283 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1284 } 1285 1286 static inline void sk_refcnt_debug_release(const struct sock *sk) 1287 { 1288 if (refcount_read(&sk->sk_refcnt) != 1) 1289 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1290 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1291 } 1292 #else /* SOCK_REFCNT_DEBUG */ 1293 #define sk_refcnt_debug_inc(sk) do { } while (0) 1294 #define sk_refcnt_debug_dec(sk) do { } while (0) 1295 #define sk_refcnt_debug_release(sk) do { } while (0) 1296 #endif /* SOCK_REFCNT_DEBUG */ 1297 1298 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1299 1300 static inline int sk_forward_alloc_get(const struct sock *sk) 1301 { 1302 if (!sk->sk_prot->forward_alloc_get) 1303 return sk->sk_forward_alloc; 1304 1305 return sk->sk_prot->forward_alloc_get(sk); 1306 } 1307 1308 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1309 { 1310 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1311 return false; 1312 1313 return sk->sk_prot->stream_memory_free ? 1314 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1315 tcp_stream_memory_free, sk, wake) : true; 1316 } 1317 1318 static inline bool sk_stream_memory_free(const struct sock *sk) 1319 { 1320 return __sk_stream_memory_free(sk, 0); 1321 } 1322 1323 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1324 { 1325 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1326 __sk_stream_memory_free(sk, wake); 1327 } 1328 1329 static inline bool sk_stream_is_writeable(const struct sock *sk) 1330 { 1331 return __sk_stream_is_writeable(sk, 0); 1332 } 1333 1334 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1335 struct cgroup *ancestor) 1336 { 1337 #ifdef CONFIG_SOCK_CGROUP_DATA 1338 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1339 ancestor); 1340 #else 1341 return -ENOTSUPP; 1342 #endif 1343 } 1344 1345 static inline bool sk_has_memory_pressure(const struct sock *sk) 1346 { 1347 return sk->sk_prot->memory_pressure != NULL; 1348 } 1349 1350 static inline bool sk_under_memory_pressure(const struct sock *sk) 1351 { 1352 if (!sk->sk_prot->memory_pressure) 1353 return false; 1354 1355 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1356 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1357 return true; 1358 1359 return !!*sk->sk_prot->memory_pressure; 1360 } 1361 1362 static inline long 1363 sk_memory_allocated(const struct sock *sk) 1364 { 1365 return atomic_long_read(sk->sk_prot->memory_allocated); 1366 } 1367 1368 static inline long 1369 sk_memory_allocated_add(struct sock *sk, int amt) 1370 { 1371 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1372 } 1373 1374 static inline void 1375 sk_memory_allocated_sub(struct sock *sk, int amt) 1376 { 1377 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1378 } 1379 1380 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1381 1382 static inline void sk_sockets_allocated_dec(struct sock *sk) 1383 { 1384 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1385 SK_ALLOC_PERCPU_COUNTER_BATCH); 1386 } 1387 1388 static inline void sk_sockets_allocated_inc(struct sock *sk) 1389 { 1390 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1391 SK_ALLOC_PERCPU_COUNTER_BATCH); 1392 } 1393 1394 static inline u64 1395 sk_sockets_allocated_read_positive(struct sock *sk) 1396 { 1397 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1398 } 1399 1400 static inline int 1401 proto_sockets_allocated_sum_positive(struct proto *prot) 1402 { 1403 return percpu_counter_sum_positive(prot->sockets_allocated); 1404 } 1405 1406 static inline long 1407 proto_memory_allocated(struct proto *prot) 1408 { 1409 return atomic_long_read(prot->memory_allocated); 1410 } 1411 1412 static inline bool 1413 proto_memory_pressure(struct proto *prot) 1414 { 1415 if (!prot->memory_pressure) 1416 return false; 1417 return !!*prot->memory_pressure; 1418 } 1419 1420 1421 #ifdef CONFIG_PROC_FS 1422 /* Called with local bh disabled */ 1423 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1424 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1425 int sock_inuse_get(struct net *net); 1426 #else 1427 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1428 int inc) 1429 { 1430 } 1431 #endif 1432 1433 1434 /* With per-bucket locks this operation is not-atomic, so that 1435 * this version is not worse. 1436 */ 1437 static inline int __sk_prot_rehash(struct sock *sk) 1438 { 1439 sk->sk_prot->unhash(sk); 1440 return sk->sk_prot->hash(sk); 1441 } 1442 1443 /* About 10 seconds */ 1444 #define SOCK_DESTROY_TIME (10*HZ) 1445 1446 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1447 #define PROT_SOCK 1024 1448 1449 #define SHUTDOWN_MASK 3 1450 #define RCV_SHUTDOWN 1 1451 #define SEND_SHUTDOWN 2 1452 1453 #define SOCK_BINDADDR_LOCK 4 1454 #define SOCK_BINDPORT_LOCK 8 1455 1456 struct socket_alloc { 1457 struct socket socket; 1458 struct inode vfs_inode; 1459 }; 1460 1461 static inline struct socket *SOCKET_I(struct inode *inode) 1462 { 1463 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1464 } 1465 1466 static inline struct inode *SOCK_INODE(struct socket *socket) 1467 { 1468 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1469 } 1470 1471 /* 1472 * Functions for memory accounting 1473 */ 1474 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1475 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1476 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1477 void __sk_mem_reclaim(struct sock *sk, int amount); 1478 1479 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1480 * do not necessarily have 16x time more memory than 4KB ones. 1481 */ 1482 #define SK_MEM_QUANTUM 4096 1483 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1484 #define SK_MEM_SEND 0 1485 #define SK_MEM_RECV 1 1486 1487 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1488 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1489 { 1490 long val = sk->sk_prot->sysctl_mem[index]; 1491 1492 #if PAGE_SIZE > SK_MEM_QUANTUM 1493 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1494 #elif PAGE_SIZE < SK_MEM_QUANTUM 1495 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1496 #endif 1497 return val; 1498 } 1499 1500 static inline int sk_mem_pages(int amt) 1501 { 1502 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1503 } 1504 1505 static inline bool sk_has_account(struct sock *sk) 1506 { 1507 /* return true if protocol supports memory accounting */ 1508 return !!sk->sk_prot->memory_allocated; 1509 } 1510 1511 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1512 { 1513 if (!sk_has_account(sk)) 1514 return true; 1515 return size <= sk->sk_forward_alloc || 1516 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1517 } 1518 1519 static inline bool 1520 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1521 { 1522 if (!sk_has_account(sk)) 1523 return true; 1524 return size <= sk->sk_forward_alloc || 1525 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1526 skb_pfmemalloc(skb); 1527 } 1528 1529 static inline int sk_unused_reserved_mem(const struct sock *sk) 1530 { 1531 int unused_mem; 1532 1533 if (likely(!sk->sk_reserved_mem)) 1534 return 0; 1535 1536 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1537 atomic_read(&sk->sk_rmem_alloc); 1538 1539 return unused_mem > 0 ? unused_mem : 0; 1540 } 1541 1542 static inline void sk_mem_reclaim(struct sock *sk) 1543 { 1544 int reclaimable; 1545 1546 if (!sk_has_account(sk)) 1547 return; 1548 1549 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1550 1551 if (reclaimable >= SK_MEM_QUANTUM) 1552 __sk_mem_reclaim(sk, reclaimable); 1553 } 1554 1555 static inline void sk_mem_reclaim_final(struct sock *sk) 1556 { 1557 sk->sk_reserved_mem = 0; 1558 sk_mem_reclaim(sk); 1559 } 1560 1561 static inline void sk_mem_reclaim_partial(struct sock *sk) 1562 { 1563 int reclaimable; 1564 1565 if (!sk_has_account(sk)) 1566 return; 1567 1568 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1569 1570 if (reclaimable > SK_MEM_QUANTUM) 1571 __sk_mem_reclaim(sk, reclaimable - 1); 1572 } 1573 1574 static inline void sk_mem_charge(struct sock *sk, int size) 1575 { 1576 if (!sk_has_account(sk)) 1577 return; 1578 sk->sk_forward_alloc -= size; 1579 } 1580 1581 /* the following macros control memory reclaiming in sk_mem_uncharge() 1582 */ 1583 #define SK_RECLAIM_THRESHOLD (1 << 21) 1584 #define SK_RECLAIM_CHUNK (1 << 20) 1585 1586 static inline void sk_mem_uncharge(struct sock *sk, int size) 1587 { 1588 int reclaimable; 1589 1590 if (!sk_has_account(sk)) 1591 return; 1592 sk->sk_forward_alloc += size; 1593 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1594 1595 /* Avoid a possible overflow. 1596 * TCP send queues can make this happen, if sk_mem_reclaim() 1597 * is not called and more than 2 GBytes are released at once. 1598 * 1599 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1600 * no need to hold that much forward allocation anyway. 1601 */ 1602 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD)) 1603 __sk_mem_reclaim(sk, SK_RECLAIM_CHUNK); 1604 } 1605 1606 static inline void sock_release_ownership(struct sock *sk) 1607 { 1608 if (sk->sk_lock.owned) { 1609 sk->sk_lock.owned = 0; 1610 1611 /* The sk_lock has mutex_unlock() semantics: */ 1612 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1613 } 1614 } 1615 1616 /* 1617 * Macro so as to not evaluate some arguments when 1618 * lockdep is not enabled. 1619 * 1620 * Mark both the sk_lock and the sk_lock.slock as a 1621 * per-address-family lock class. 1622 */ 1623 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1624 do { \ 1625 sk->sk_lock.owned = 0; \ 1626 init_waitqueue_head(&sk->sk_lock.wq); \ 1627 spin_lock_init(&(sk)->sk_lock.slock); \ 1628 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1629 sizeof((sk)->sk_lock)); \ 1630 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1631 (skey), (sname)); \ 1632 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1633 } while (0) 1634 1635 static inline bool lockdep_sock_is_held(const struct sock *sk) 1636 { 1637 return lockdep_is_held(&sk->sk_lock) || 1638 lockdep_is_held(&sk->sk_lock.slock); 1639 } 1640 1641 void lock_sock_nested(struct sock *sk, int subclass); 1642 1643 static inline void lock_sock(struct sock *sk) 1644 { 1645 lock_sock_nested(sk, 0); 1646 } 1647 1648 void __lock_sock(struct sock *sk); 1649 void __release_sock(struct sock *sk); 1650 void release_sock(struct sock *sk); 1651 1652 /* BH context may only use the following locking interface. */ 1653 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1654 #define bh_lock_sock_nested(__sk) \ 1655 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1656 SINGLE_DEPTH_NESTING) 1657 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1658 1659 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1660 1661 /** 1662 * lock_sock_fast - fast version of lock_sock 1663 * @sk: socket 1664 * 1665 * This version should be used for very small section, where process wont block 1666 * return false if fast path is taken: 1667 * 1668 * sk_lock.slock locked, owned = 0, BH disabled 1669 * 1670 * return true if slow path is taken: 1671 * 1672 * sk_lock.slock unlocked, owned = 1, BH enabled 1673 */ 1674 static inline bool lock_sock_fast(struct sock *sk) 1675 { 1676 /* The sk_lock has mutex_lock() semantics here. */ 1677 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1678 1679 return __lock_sock_fast(sk); 1680 } 1681 1682 /* fast socket lock variant for caller already holding a [different] socket lock */ 1683 static inline bool lock_sock_fast_nested(struct sock *sk) 1684 { 1685 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1686 1687 return __lock_sock_fast(sk); 1688 } 1689 1690 /** 1691 * unlock_sock_fast - complement of lock_sock_fast 1692 * @sk: socket 1693 * @slow: slow mode 1694 * 1695 * fast unlock socket for user context. 1696 * If slow mode is on, we call regular release_sock() 1697 */ 1698 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1699 __releases(&sk->sk_lock.slock) 1700 { 1701 if (slow) { 1702 release_sock(sk); 1703 __release(&sk->sk_lock.slock); 1704 } else { 1705 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1706 spin_unlock_bh(&sk->sk_lock.slock); 1707 } 1708 } 1709 1710 /* Used by processes to "lock" a socket state, so that 1711 * interrupts and bottom half handlers won't change it 1712 * from under us. It essentially blocks any incoming 1713 * packets, so that we won't get any new data or any 1714 * packets that change the state of the socket. 1715 * 1716 * While locked, BH processing will add new packets to 1717 * the backlog queue. This queue is processed by the 1718 * owner of the socket lock right before it is released. 1719 * 1720 * Since ~2.3.5 it is also exclusive sleep lock serializing 1721 * accesses from user process context. 1722 */ 1723 1724 static inline void sock_owned_by_me(const struct sock *sk) 1725 { 1726 #ifdef CONFIG_LOCKDEP 1727 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1728 #endif 1729 } 1730 1731 static inline bool sock_owned_by_user(const struct sock *sk) 1732 { 1733 sock_owned_by_me(sk); 1734 return sk->sk_lock.owned; 1735 } 1736 1737 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1738 { 1739 return sk->sk_lock.owned; 1740 } 1741 1742 /* no reclassification while locks are held */ 1743 static inline bool sock_allow_reclassification(const struct sock *csk) 1744 { 1745 struct sock *sk = (struct sock *)csk; 1746 1747 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1748 } 1749 1750 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1751 struct proto *prot, int kern); 1752 void sk_free(struct sock *sk); 1753 void sk_destruct(struct sock *sk); 1754 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1755 void sk_free_unlock_clone(struct sock *sk); 1756 1757 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1758 gfp_t priority); 1759 void __sock_wfree(struct sk_buff *skb); 1760 void sock_wfree(struct sk_buff *skb); 1761 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1762 gfp_t priority); 1763 void skb_orphan_partial(struct sk_buff *skb); 1764 void sock_rfree(struct sk_buff *skb); 1765 void sock_efree(struct sk_buff *skb); 1766 #ifdef CONFIG_INET 1767 void sock_edemux(struct sk_buff *skb); 1768 void sock_pfree(struct sk_buff *skb); 1769 #else 1770 #define sock_edemux sock_efree 1771 #endif 1772 1773 int sock_setsockopt(struct socket *sock, int level, int op, 1774 sockptr_t optval, unsigned int optlen); 1775 1776 int sock_getsockopt(struct socket *sock, int level, int op, 1777 char __user *optval, int __user *optlen); 1778 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1779 bool timeval, bool time32); 1780 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1781 int noblock, int *errcode); 1782 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1783 unsigned long data_len, int noblock, 1784 int *errcode, int max_page_order); 1785 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1786 void sock_kfree_s(struct sock *sk, void *mem, int size); 1787 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1788 void sk_send_sigurg(struct sock *sk); 1789 1790 struct sockcm_cookie { 1791 u64 transmit_time; 1792 u32 mark; 1793 u16 tsflags; 1794 }; 1795 1796 static inline void sockcm_init(struct sockcm_cookie *sockc, 1797 const struct sock *sk) 1798 { 1799 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1800 } 1801 1802 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1803 struct sockcm_cookie *sockc); 1804 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1805 struct sockcm_cookie *sockc); 1806 1807 /* 1808 * Functions to fill in entries in struct proto_ops when a protocol 1809 * does not implement a particular function. 1810 */ 1811 int sock_no_bind(struct socket *, struct sockaddr *, int); 1812 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1813 int sock_no_socketpair(struct socket *, struct socket *); 1814 int sock_no_accept(struct socket *, struct socket *, int, bool); 1815 int sock_no_getname(struct socket *, struct sockaddr *, int); 1816 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1817 int sock_no_listen(struct socket *, int); 1818 int sock_no_shutdown(struct socket *, int); 1819 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1820 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1821 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1822 int sock_no_mmap(struct file *file, struct socket *sock, 1823 struct vm_area_struct *vma); 1824 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1825 size_t size, int flags); 1826 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1827 int offset, size_t size, int flags); 1828 1829 /* 1830 * Functions to fill in entries in struct proto_ops when a protocol 1831 * uses the inet style. 1832 */ 1833 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1834 char __user *optval, int __user *optlen); 1835 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1836 int flags); 1837 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1838 sockptr_t optval, unsigned int optlen); 1839 1840 void sk_common_release(struct sock *sk); 1841 1842 /* 1843 * Default socket callbacks and setup code 1844 */ 1845 1846 /* Initialise core socket variables */ 1847 void sock_init_data(struct socket *sock, struct sock *sk); 1848 1849 /* 1850 * Socket reference counting postulates. 1851 * 1852 * * Each user of socket SHOULD hold a reference count. 1853 * * Each access point to socket (an hash table bucket, reference from a list, 1854 * running timer, skb in flight MUST hold a reference count. 1855 * * When reference count hits 0, it means it will never increase back. 1856 * * When reference count hits 0, it means that no references from 1857 * outside exist to this socket and current process on current CPU 1858 * is last user and may/should destroy this socket. 1859 * * sk_free is called from any context: process, BH, IRQ. When 1860 * it is called, socket has no references from outside -> sk_free 1861 * may release descendant resources allocated by the socket, but 1862 * to the time when it is called, socket is NOT referenced by any 1863 * hash tables, lists etc. 1864 * * Packets, delivered from outside (from network or from another process) 1865 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1866 * when they sit in queue. Otherwise, packets will leak to hole, when 1867 * socket is looked up by one cpu and unhasing is made by another CPU. 1868 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1869 * (leak to backlog). Packet socket does all the processing inside 1870 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1871 * use separate SMP lock, so that they are prone too. 1872 */ 1873 1874 /* Ungrab socket and destroy it, if it was the last reference. */ 1875 static inline void sock_put(struct sock *sk) 1876 { 1877 if (refcount_dec_and_test(&sk->sk_refcnt)) 1878 sk_free(sk); 1879 } 1880 /* Generic version of sock_put(), dealing with all sockets 1881 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1882 */ 1883 void sock_gen_put(struct sock *sk); 1884 1885 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1886 unsigned int trim_cap, bool refcounted); 1887 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1888 const int nested) 1889 { 1890 return __sk_receive_skb(sk, skb, nested, 1, true); 1891 } 1892 1893 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1894 { 1895 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1896 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1897 return; 1898 sk->sk_tx_queue_mapping = tx_queue; 1899 } 1900 1901 #define NO_QUEUE_MAPPING USHRT_MAX 1902 1903 static inline void sk_tx_queue_clear(struct sock *sk) 1904 { 1905 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1906 } 1907 1908 static inline int sk_tx_queue_get(const struct sock *sk) 1909 { 1910 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1911 return sk->sk_tx_queue_mapping; 1912 1913 return -1; 1914 } 1915 1916 static inline void __sk_rx_queue_set(struct sock *sk, 1917 const struct sk_buff *skb, 1918 bool force_set) 1919 { 1920 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1921 if (skb_rx_queue_recorded(skb)) { 1922 u16 rx_queue = skb_get_rx_queue(skb); 1923 1924 if (force_set || 1925 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 1926 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 1927 } 1928 #endif 1929 } 1930 1931 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1932 { 1933 __sk_rx_queue_set(sk, skb, true); 1934 } 1935 1936 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 1937 { 1938 __sk_rx_queue_set(sk, skb, false); 1939 } 1940 1941 static inline void sk_rx_queue_clear(struct sock *sk) 1942 { 1943 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1944 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 1945 #endif 1946 } 1947 1948 static inline int sk_rx_queue_get(const struct sock *sk) 1949 { 1950 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1951 if (sk) { 1952 int res = READ_ONCE(sk->sk_rx_queue_mapping); 1953 1954 if (res != NO_QUEUE_MAPPING) 1955 return res; 1956 } 1957 #endif 1958 1959 return -1; 1960 } 1961 1962 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1963 { 1964 sk->sk_socket = sock; 1965 } 1966 1967 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1968 { 1969 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1970 return &rcu_dereference_raw(sk->sk_wq)->wait; 1971 } 1972 /* Detach socket from process context. 1973 * Announce socket dead, detach it from wait queue and inode. 1974 * Note that parent inode held reference count on this struct sock, 1975 * we do not release it in this function, because protocol 1976 * probably wants some additional cleanups or even continuing 1977 * to work with this socket (TCP). 1978 */ 1979 static inline void sock_orphan(struct sock *sk) 1980 { 1981 write_lock_bh(&sk->sk_callback_lock); 1982 sock_set_flag(sk, SOCK_DEAD); 1983 sk_set_socket(sk, NULL); 1984 sk->sk_wq = NULL; 1985 write_unlock_bh(&sk->sk_callback_lock); 1986 } 1987 1988 static inline void sock_graft(struct sock *sk, struct socket *parent) 1989 { 1990 WARN_ON(parent->sk); 1991 write_lock_bh(&sk->sk_callback_lock); 1992 rcu_assign_pointer(sk->sk_wq, &parent->wq); 1993 parent->sk = sk; 1994 sk_set_socket(sk, parent); 1995 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1996 security_sock_graft(sk, parent); 1997 write_unlock_bh(&sk->sk_callback_lock); 1998 } 1999 2000 kuid_t sock_i_uid(struct sock *sk); 2001 unsigned long sock_i_ino(struct sock *sk); 2002 2003 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2004 { 2005 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2006 } 2007 2008 static inline u32 net_tx_rndhash(void) 2009 { 2010 u32 v = prandom_u32(); 2011 2012 return v ?: 1; 2013 } 2014 2015 static inline void sk_set_txhash(struct sock *sk) 2016 { 2017 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2018 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2019 } 2020 2021 static inline bool sk_rethink_txhash(struct sock *sk) 2022 { 2023 if (sk->sk_txhash) { 2024 sk_set_txhash(sk); 2025 return true; 2026 } 2027 return false; 2028 } 2029 2030 static inline struct dst_entry * 2031 __sk_dst_get(struct sock *sk) 2032 { 2033 return rcu_dereference_check(sk->sk_dst_cache, 2034 lockdep_sock_is_held(sk)); 2035 } 2036 2037 static inline struct dst_entry * 2038 sk_dst_get(struct sock *sk) 2039 { 2040 struct dst_entry *dst; 2041 2042 rcu_read_lock(); 2043 dst = rcu_dereference(sk->sk_dst_cache); 2044 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2045 dst = NULL; 2046 rcu_read_unlock(); 2047 return dst; 2048 } 2049 2050 static inline void __dst_negative_advice(struct sock *sk) 2051 { 2052 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2053 2054 if (dst && dst->ops->negative_advice) { 2055 ndst = dst->ops->negative_advice(dst); 2056 2057 if (ndst != dst) { 2058 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2059 sk_tx_queue_clear(sk); 2060 sk->sk_dst_pending_confirm = 0; 2061 } 2062 } 2063 } 2064 2065 static inline void dst_negative_advice(struct sock *sk) 2066 { 2067 sk_rethink_txhash(sk); 2068 __dst_negative_advice(sk); 2069 } 2070 2071 static inline void 2072 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2073 { 2074 struct dst_entry *old_dst; 2075 2076 sk_tx_queue_clear(sk); 2077 sk->sk_dst_pending_confirm = 0; 2078 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2079 lockdep_sock_is_held(sk)); 2080 rcu_assign_pointer(sk->sk_dst_cache, dst); 2081 dst_release(old_dst); 2082 } 2083 2084 static inline void 2085 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2086 { 2087 struct dst_entry *old_dst; 2088 2089 sk_tx_queue_clear(sk); 2090 sk->sk_dst_pending_confirm = 0; 2091 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2092 dst_release(old_dst); 2093 } 2094 2095 static inline void 2096 __sk_dst_reset(struct sock *sk) 2097 { 2098 __sk_dst_set(sk, NULL); 2099 } 2100 2101 static inline void 2102 sk_dst_reset(struct sock *sk) 2103 { 2104 sk_dst_set(sk, NULL); 2105 } 2106 2107 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2108 2109 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2110 2111 static inline void sk_dst_confirm(struct sock *sk) 2112 { 2113 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2114 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2115 } 2116 2117 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2118 { 2119 if (skb_get_dst_pending_confirm(skb)) { 2120 struct sock *sk = skb->sk; 2121 unsigned long now = jiffies; 2122 2123 /* avoid dirtying neighbour */ 2124 if (READ_ONCE(n->confirmed) != now) 2125 WRITE_ONCE(n->confirmed, now); 2126 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2127 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2128 } 2129 } 2130 2131 bool sk_mc_loop(struct sock *sk); 2132 2133 static inline bool sk_can_gso(const struct sock *sk) 2134 { 2135 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2136 } 2137 2138 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2139 2140 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 2141 { 2142 sk->sk_route_nocaps |= flags; 2143 sk->sk_route_caps &= ~flags; 2144 } 2145 2146 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2147 struct iov_iter *from, char *to, 2148 int copy, int offset) 2149 { 2150 if (skb->ip_summed == CHECKSUM_NONE) { 2151 __wsum csum = 0; 2152 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2153 return -EFAULT; 2154 skb->csum = csum_block_add(skb->csum, csum, offset); 2155 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2156 if (!copy_from_iter_full_nocache(to, copy, from)) 2157 return -EFAULT; 2158 } else if (!copy_from_iter_full(to, copy, from)) 2159 return -EFAULT; 2160 2161 return 0; 2162 } 2163 2164 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2165 struct iov_iter *from, int copy) 2166 { 2167 int err, offset = skb->len; 2168 2169 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2170 copy, offset); 2171 if (err) 2172 __skb_trim(skb, offset); 2173 2174 return err; 2175 } 2176 2177 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2178 struct sk_buff *skb, 2179 struct page *page, 2180 int off, int copy) 2181 { 2182 int err; 2183 2184 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2185 copy, skb->len); 2186 if (err) 2187 return err; 2188 2189 skb->len += copy; 2190 skb->data_len += copy; 2191 skb->truesize += copy; 2192 sk_wmem_queued_add(sk, copy); 2193 sk_mem_charge(sk, copy); 2194 return 0; 2195 } 2196 2197 /** 2198 * sk_wmem_alloc_get - returns write allocations 2199 * @sk: socket 2200 * 2201 * Return: sk_wmem_alloc minus initial offset of one 2202 */ 2203 static inline int sk_wmem_alloc_get(const struct sock *sk) 2204 { 2205 return refcount_read(&sk->sk_wmem_alloc) - 1; 2206 } 2207 2208 /** 2209 * sk_rmem_alloc_get - returns read allocations 2210 * @sk: socket 2211 * 2212 * Return: sk_rmem_alloc 2213 */ 2214 static inline int sk_rmem_alloc_get(const struct sock *sk) 2215 { 2216 return atomic_read(&sk->sk_rmem_alloc); 2217 } 2218 2219 /** 2220 * sk_has_allocations - check if allocations are outstanding 2221 * @sk: socket 2222 * 2223 * Return: true if socket has write or read allocations 2224 */ 2225 static inline bool sk_has_allocations(const struct sock *sk) 2226 { 2227 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2228 } 2229 2230 /** 2231 * skwq_has_sleeper - check if there are any waiting processes 2232 * @wq: struct socket_wq 2233 * 2234 * Return: true if socket_wq has waiting processes 2235 * 2236 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2237 * barrier call. They were added due to the race found within the tcp code. 2238 * 2239 * Consider following tcp code paths:: 2240 * 2241 * CPU1 CPU2 2242 * sys_select receive packet 2243 * ... ... 2244 * __add_wait_queue update tp->rcv_nxt 2245 * ... ... 2246 * tp->rcv_nxt check sock_def_readable 2247 * ... { 2248 * schedule rcu_read_lock(); 2249 * wq = rcu_dereference(sk->sk_wq); 2250 * if (wq && waitqueue_active(&wq->wait)) 2251 * wake_up_interruptible(&wq->wait) 2252 * ... 2253 * } 2254 * 2255 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2256 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2257 * could then endup calling schedule and sleep forever if there are no more 2258 * data on the socket. 2259 * 2260 */ 2261 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2262 { 2263 return wq && wq_has_sleeper(&wq->wait); 2264 } 2265 2266 /** 2267 * sock_poll_wait - place memory barrier behind the poll_wait call. 2268 * @filp: file 2269 * @sock: socket to wait on 2270 * @p: poll_table 2271 * 2272 * See the comments in the wq_has_sleeper function. 2273 */ 2274 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2275 poll_table *p) 2276 { 2277 if (!poll_does_not_wait(p)) { 2278 poll_wait(filp, &sock->wq.wait, p); 2279 /* We need to be sure we are in sync with the 2280 * socket flags modification. 2281 * 2282 * This memory barrier is paired in the wq_has_sleeper. 2283 */ 2284 smp_mb(); 2285 } 2286 } 2287 2288 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2289 { 2290 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2291 u32 txhash = READ_ONCE(sk->sk_txhash); 2292 2293 if (txhash) { 2294 skb->l4_hash = 1; 2295 skb->hash = txhash; 2296 } 2297 } 2298 2299 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2300 2301 /* 2302 * Queue a received datagram if it will fit. Stream and sequenced 2303 * protocols can't normally use this as they need to fit buffers in 2304 * and play with them. 2305 * 2306 * Inlined as it's very short and called for pretty much every 2307 * packet ever received. 2308 */ 2309 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2310 { 2311 skb_orphan(skb); 2312 skb->sk = sk; 2313 skb->destructor = sock_rfree; 2314 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2315 sk_mem_charge(sk, skb->truesize); 2316 } 2317 2318 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2319 { 2320 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2321 skb_orphan(skb); 2322 skb->destructor = sock_efree; 2323 skb->sk = sk; 2324 return true; 2325 } 2326 return false; 2327 } 2328 2329 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2330 { 2331 if (skb->destructor != sock_wfree) { 2332 skb_orphan(skb); 2333 return; 2334 } 2335 skb->slow_gro = 1; 2336 } 2337 2338 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2339 unsigned long expires); 2340 2341 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2342 2343 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2344 2345 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2346 struct sk_buff *skb, unsigned int flags, 2347 void (*destructor)(struct sock *sk, 2348 struct sk_buff *skb)); 2349 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2350 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2351 2352 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2353 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2354 2355 /* 2356 * Recover an error report and clear atomically 2357 */ 2358 2359 static inline int sock_error(struct sock *sk) 2360 { 2361 int err; 2362 2363 /* Avoid an atomic operation for the common case. 2364 * This is racy since another cpu/thread can change sk_err under us. 2365 */ 2366 if (likely(data_race(!sk->sk_err))) 2367 return 0; 2368 2369 err = xchg(&sk->sk_err, 0); 2370 return -err; 2371 } 2372 2373 void sk_error_report(struct sock *sk); 2374 2375 static inline unsigned long sock_wspace(struct sock *sk) 2376 { 2377 int amt = 0; 2378 2379 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2380 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2381 if (amt < 0) 2382 amt = 0; 2383 } 2384 return amt; 2385 } 2386 2387 /* Note: 2388 * We use sk->sk_wq_raw, from contexts knowing this 2389 * pointer is not NULL and cannot disappear/change. 2390 */ 2391 static inline void sk_set_bit(int nr, struct sock *sk) 2392 { 2393 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2394 !sock_flag(sk, SOCK_FASYNC)) 2395 return; 2396 2397 set_bit(nr, &sk->sk_wq_raw->flags); 2398 } 2399 2400 static inline void sk_clear_bit(int nr, struct sock *sk) 2401 { 2402 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2403 !sock_flag(sk, SOCK_FASYNC)) 2404 return; 2405 2406 clear_bit(nr, &sk->sk_wq_raw->flags); 2407 } 2408 2409 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2410 { 2411 if (sock_flag(sk, SOCK_FASYNC)) { 2412 rcu_read_lock(); 2413 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2414 rcu_read_unlock(); 2415 } 2416 } 2417 2418 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2419 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2420 * Note: for send buffers, TCP works better if we can build two skbs at 2421 * minimum. 2422 */ 2423 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2424 2425 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2426 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2427 2428 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2429 { 2430 u32 val; 2431 2432 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2433 return; 2434 2435 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2436 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2437 2438 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2439 } 2440 2441 /** 2442 * sk_page_frag - return an appropriate page_frag 2443 * @sk: socket 2444 * 2445 * Use the per task page_frag instead of the per socket one for 2446 * optimization when we know that we're in process context and own 2447 * everything that's associated with %current. 2448 * 2449 * Both direct reclaim and page faults can nest inside other 2450 * socket operations and end up recursing into sk_page_frag() 2451 * while it's already in use: explicitly avoid task page_frag 2452 * usage if the caller is potentially doing any of them. 2453 * This assumes that page fault handlers use the GFP_NOFS flags. 2454 * 2455 * Return: a per task page_frag if context allows that, 2456 * otherwise a per socket one. 2457 */ 2458 static inline struct page_frag *sk_page_frag(struct sock *sk) 2459 { 2460 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) == 2461 (__GFP_DIRECT_RECLAIM | __GFP_FS)) 2462 return ¤t->task_frag; 2463 2464 return &sk->sk_frag; 2465 } 2466 2467 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2468 2469 /* 2470 * Default write policy as shown to user space via poll/select/SIGIO 2471 */ 2472 static inline bool sock_writeable(const struct sock *sk) 2473 { 2474 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2475 } 2476 2477 static inline gfp_t gfp_any(void) 2478 { 2479 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2480 } 2481 2482 static inline gfp_t gfp_memcg_charge(void) 2483 { 2484 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2485 } 2486 2487 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2488 { 2489 return noblock ? 0 : sk->sk_rcvtimeo; 2490 } 2491 2492 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2493 { 2494 return noblock ? 0 : sk->sk_sndtimeo; 2495 } 2496 2497 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2498 { 2499 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2500 2501 return v ?: 1; 2502 } 2503 2504 /* Alas, with timeout socket operations are not restartable. 2505 * Compare this to poll(). 2506 */ 2507 static inline int sock_intr_errno(long timeo) 2508 { 2509 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2510 } 2511 2512 struct sock_skb_cb { 2513 u32 dropcount; 2514 }; 2515 2516 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2517 * using skb->cb[] would keep using it directly and utilize its 2518 * alignement guarantee. 2519 */ 2520 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2521 sizeof(struct sock_skb_cb))) 2522 2523 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2524 SOCK_SKB_CB_OFFSET)) 2525 2526 #define sock_skb_cb_check_size(size) \ 2527 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2528 2529 static inline void 2530 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2531 { 2532 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2533 atomic_read(&sk->sk_drops) : 0; 2534 } 2535 2536 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2537 { 2538 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2539 2540 atomic_add(segs, &sk->sk_drops); 2541 } 2542 2543 static inline ktime_t sock_read_timestamp(struct sock *sk) 2544 { 2545 #if BITS_PER_LONG==32 2546 unsigned int seq; 2547 ktime_t kt; 2548 2549 do { 2550 seq = read_seqbegin(&sk->sk_stamp_seq); 2551 kt = sk->sk_stamp; 2552 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2553 2554 return kt; 2555 #else 2556 return READ_ONCE(sk->sk_stamp); 2557 #endif 2558 } 2559 2560 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2561 { 2562 #if BITS_PER_LONG==32 2563 write_seqlock(&sk->sk_stamp_seq); 2564 sk->sk_stamp = kt; 2565 write_sequnlock(&sk->sk_stamp_seq); 2566 #else 2567 WRITE_ONCE(sk->sk_stamp, kt); 2568 #endif 2569 } 2570 2571 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2572 struct sk_buff *skb); 2573 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2574 struct sk_buff *skb); 2575 2576 static inline void 2577 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2578 { 2579 ktime_t kt = skb->tstamp; 2580 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2581 2582 /* 2583 * generate control messages if 2584 * - receive time stamping in software requested 2585 * - software time stamp available and wanted 2586 * - hardware time stamps available and wanted 2587 */ 2588 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2589 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2590 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2591 (hwtstamps->hwtstamp && 2592 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2593 __sock_recv_timestamp(msg, sk, skb); 2594 else 2595 sock_write_timestamp(sk, kt); 2596 2597 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2598 __sock_recv_wifi_status(msg, sk, skb); 2599 } 2600 2601 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2602 struct sk_buff *skb); 2603 2604 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2605 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2606 struct sk_buff *skb) 2607 { 2608 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2609 (1UL << SOCK_RCVTSTAMP)) 2610 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2611 SOF_TIMESTAMPING_RAW_HARDWARE) 2612 2613 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2614 __sock_recv_ts_and_drops(msg, sk, skb); 2615 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2616 sock_write_timestamp(sk, skb->tstamp); 2617 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2618 sock_write_timestamp(sk, 0); 2619 } 2620 2621 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2622 2623 /** 2624 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2625 * @sk: socket sending this packet 2626 * @tsflags: timestamping flags to use 2627 * @tx_flags: completed with instructions for time stamping 2628 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2629 * 2630 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2631 */ 2632 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2633 __u8 *tx_flags, __u32 *tskey) 2634 { 2635 if (unlikely(tsflags)) { 2636 __sock_tx_timestamp(tsflags, tx_flags); 2637 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2638 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2639 *tskey = sk->sk_tskey++; 2640 } 2641 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2642 *tx_flags |= SKBTX_WIFI_STATUS; 2643 } 2644 2645 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2646 __u8 *tx_flags) 2647 { 2648 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2649 } 2650 2651 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2652 { 2653 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2654 &skb_shinfo(skb)->tskey); 2655 } 2656 2657 /** 2658 * sk_eat_skb - Release a skb if it is no longer needed 2659 * @sk: socket to eat this skb from 2660 * @skb: socket buffer to eat 2661 * 2662 * This routine must be called with interrupts disabled or with the socket 2663 * locked so that the sk_buff queue operation is ok. 2664 */ 2665 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2666 { 2667 __skb_unlink(skb, &sk->sk_receive_queue); 2668 __kfree_skb(skb); 2669 } 2670 2671 static inline 2672 struct net *sock_net(const struct sock *sk) 2673 { 2674 return read_pnet(&sk->sk_net); 2675 } 2676 2677 static inline 2678 void sock_net_set(struct sock *sk, struct net *net) 2679 { 2680 write_pnet(&sk->sk_net, net); 2681 } 2682 2683 static inline bool 2684 skb_sk_is_prefetched(struct sk_buff *skb) 2685 { 2686 #ifdef CONFIG_INET 2687 return skb->destructor == sock_pfree; 2688 #else 2689 return false; 2690 #endif /* CONFIG_INET */ 2691 } 2692 2693 /* This helper checks if a socket is a full socket, 2694 * ie _not_ a timewait or request socket. 2695 */ 2696 static inline bool sk_fullsock(const struct sock *sk) 2697 { 2698 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2699 } 2700 2701 static inline bool 2702 sk_is_refcounted(struct sock *sk) 2703 { 2704 /* Only full sockets have sk->sk_flags. */ 2705 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2706 } 2707 2708 /** 2709 * skb_steal_sock - steal a socket from an sk_buff 2710 * @skb: sk_buff to steal the socket from 2711 * @refcounted: is set to true if the socket is reference-counted 2712 */ 2713 static inline struct sock * 2714 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2715 { 2716 if (skb->sk) { 2717 struct sock *sk = skb->sk; 2718 2719 *refcounted = true; 2720 if (skb_sk_is_prefetched(skb)) 2721 *refcounted = sk_is_refcounted(sk); 2722 skb->destructor = NULL; 2723 skb->sk = NULL; 2724 return sk; 2725 } 2726 *refcounted = false; 2727 return NULL; 2728 } 2729 2730 /* Checks if this SKB belongs to an HW offloaded socket 2731 * and whether any SW fallbacks are required based on dev. 2732 * Check decrypted mark in case skb_orphan() cleared socket. 2733 */ 2734 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2735 struct net_device *dev) 2736 { 2737 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2738 struct sock *sk = skb->sk; 2739 2740 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2741 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2742 #ifdef CONFIG_TLS_DEVICE 2743 } else if (unlikely(skb->decrypted)) { 2744 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2745 kfree_skb(skb); 2746 skb = NULL; 2747 #endif 2748 } 2749 #endif 2750 2751 return skb; 2752 } 2753 2754 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2755 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2756 */ 2757 static inline bool sk_listener(const struct sock *sk) 2758 { 2759 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2760 } 2761 2762 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2763 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2764 int type); 2765 2766 bool sk_ns_capable(const struct sock *sk, 2767 struct user_namespace *user_ns, int cap); 2768 bool sk_capable(const struct sock *sk, int cap); 2769 bool sk_net_capable(const struct sock *sk, int cap); 2770 2771 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2772 2773 /* Take into consideration the size of the struct sk_buff overhead in the 2774 * determination of these values, since that is non-constant across 2775 * platforms. This makes socket queueing behavior and performance 2776 * not depend upon such differences. 2777 */ 2778 #define _SK_MEM_PACKETS 256 2779 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2780 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2781 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2782 2783 extern __u32 sysctl_wmem_max; 2784 extern __u32 sysctl_rmem_max; 2785 2786 extern int sysctl_tstamp_allow_data; 2787 extern int sysctl_optmem_max; 2788 2789 extern __u32 sysctl_wmem_default; 2790 extern __u32 sysctl_rmem_default; 2791 2792 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2793 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2794 2795 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2796 { 2797 /* Does this proto have per netns sysctl_wmem ? */ 2798 if (proto->sysctl_wmem_offset) 2799 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2800 2801 return *proto->sysctl_wmem; 2802 } 2803 2804 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2805 { 2806 /* Does this proto have per netns sysctl_rmem ? */ 2807 if (proto->sysctl_rmem_offset) 2808 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2809 2810 return *proto->sysctl_rmem; 2811 } 2812 2813 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2814 * Some wifi drivers need to tweak it to get more chunks. 2815 * They can use this helper from their ndo_start_xmit() 2816 */ 2817 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2818 { 2819 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2820 return; 2821 WRITE_ONCE(sk->sk_pacing_shift, val); 2822 } 2823 2824 /* if a socket is bound to a device, check that the given device 2825 * index is either the same or that the socket is bound to an L3 2826 * master device and the given device index is also enslaved to 2827 * that L3 master 2828 */ 2829 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2830 { 2831 int mdif; 2832 2833 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2834 return true; 2835 2836 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2837 if (mdif && mdif == sk->sk_bound_dev_if) 2838 return true; 2839 2840 return false; 2841 } 2842 2843 void sock_def_readable(struct sock *sk); 2844 2845 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2846 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2847 int sock_set_timestamping(struct sock *sk, int optname, 2848 struct so_timestamping timestamping); 2849 2850 void sock_enable_timestamps(struct sock *sk); 2851 void sock_no_linger(struct sock *sk); 2852 void sock_set_keepalive(struct sock *sk); 2853 void sock_set_priority(struct sock *sk, u32 priority); 2854 void sock_set_rcvbuf(struct sock *sk, int val); 2855 void sock_set_mark(struct sock *sk, u32 val); 2856 void sock_set_reuseaddr(struct sock *sk); 2857 void sock_set_reuseport(struct sock *sk); 2858 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2859 2860 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2861 2862 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2863 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2864 sockptr_t optval, int optlen, bool old_timeval); 2865 2866 static inline bool sk_is_readable(struct sock *sk) 2867 { 2868 if (sk->sk_prot->sock_is_readable) 2869 return sk->sk_prot->sock_is_readable(sk); 2870 return false; 2871 } 2872 #endif /* _SOCK_H */ 2873