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