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