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