1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 165 * address on 64bit arches : cf INET_MATCH() 166 */ 167 union { 168 __addrpair skc_addrpair; 169 struct { 170 __be32 skc_daddr; 171 __be32 skc_rcv_saddr; 172 }; 173 }; 174 union { 175 unsigned int skc_hash; 176 __u16 skc_u16hashes[2]; 177 }; 178 /* skc_dport && skc_num must be grouped as well */ 179 union { 180 __portpair skc_portpair; 181 struct { 182 __be16 skc_dport; 183 __u16 skc_num; 184 }; 185 }; 186 187 unsigned short skc_family; 188 volatile unsigned char skc_state; 189 unsigned char skc_reuse:4; 190 unsigned char skc_reuseport:1; 191 unsigned char skc_ipv6only:1; 192 unsigned char skc_net_refcnt:1; 193 int skc_bound_dev_if; 194 union { 195 struct hlist_node skc_bind_node; 196 struct hlist_node skc_portaddr_node; 197 }; 198 struct proto *skc_prot; 199 possible_net_t skc_net; 200 201 #if IS_ENABLED(CONFIG_IPV6) 202 struct in6_addr skc_v6_daddr; 203 struct in6_addr skc_v6_rcv_saddr; 204 #endif 205 206 atomic64_t skc_cookie; 207 208 /* following fields are padding to force 209 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 210 * assuming IPV6 is enabled. We use this padding differently 211 * for different kind of 'sockets' 212 */ 213 union { 214 unsigned long skc_flags; 215 struct sock *skc_listener; /* request_sock */ 216 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 217 }; 218 /* 219 * fields between dontcopy_begin/dontcopy_end 220 * are not copied in sock_copy() 221 */ 222 /* private: */ 223 int skc_dontcopy_begin[0]; 224 /* public: */ 225 union { 226 struct hlist_node skc_node; 227 struct hlist_nulls_node skc_nulls_node; 228 }; 229 unsigned short skc_tx_queue_mapping; 230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 231 unsigned short skc_rx_queue_mapping; 232 #endif 233 union { 234 int skc_incoming_cpu; 235 u32 skc_rcv_wnd; 236 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 237 }; 238 239 refcount_t skc_refcnt; 240 /* private: */ 241 int skc_dontcopy_end[0]; 242 union { 243 u32 skc_rxhash; 244 u32 skc_window_clamp; 245 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 246 }; 247 /* public: */ 248 }; 249 250 struct bpf_local_storage; 251 struct sk_filter; 252 253 /** 254 * struct sock - network layer representation of sockets 255 * @__sk_common: shared layout with inet_timewait_sock 256 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 257 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 258 * @sk_lock: synchronizer 259 * @sk_kern_sock: True if sock is using kernel lock classes 260 * @sk_rcvbuf: size of receive buffer in bytes 261 * @sk_wq: sock wait queue and async head 262 * @sk_rx_dst: receive input route used by early demux 263 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 264 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 265 * @sk_dst_cache: destination cache 266 * @sk_dst_pending_confirm: need to confirm neighbour 267 * @sk_policy: flow policy 268 * @sk_receive_queue: incoming packets 269 * @sk_wmem_alloc: transmit queue bytes committed 270 * @sk_tsq_flags: TCP Small Queues flags 271 * @sk_write_queue: Packet sending queue 272 * @sk_omem_alloc: "o" is "option" or "other" 273 * @sk_wmem_queued: persistent queue size 274 * @sk_forward_alloc: space allocated forward 275 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 276 * @sk_napi_id: id of the last napi context to receive data for sk 277 * @sk_ll_usec: usecs to busypoll when there is no data 278 * @sk_allocation: allocation mode 279 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 280 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 281 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 282 * @sk_sndbuf: size of send buffer in bytes 283 * @__sk_flags_offset: empty field used to determine location of bitfield 284 * @sk_padding: unused element for alignment 285 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 286 * @sk_no_check_rx: allow zero checksum in RX packets 287 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 288 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 289 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 290 * @sk_gso_max_size: Maximum GSO segment size to build 291 * @sk_gso_max_segs: Maximum number of GSO segments 292 * @sk_pacing_shift: scaling factor for TCP Small Queues 293 * @sk_lingertime: %SO_LINGER l_linger setting 294 * @sk_backlog: always used with the per-socket spinlock held 295 * @sk_callback_lock: used with the callbacks in the end of this struct 296 * @sk_error_queue: rarely used 297 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 298 * IPV6_ADDRFORM for instance) 299 * @sk_err: last error 300 * @sk_err_soft: errors that don't cause failure but are the cause of a 301 * persistent failure not just 'timed out' 302 * @sk_drops: raw/udp drops counter 303 * @sk_ack_backlog: current listen backlog 304 * @sk_max_ack_backlog: listen backlog set in listen() 305 * @sk_uid: user id of owner 306 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 307 * @sk_busy_poll_budget: napi processing budget when busypolling 308 * @sk_priority: %SO_PRIORITY setting 309 * @sk_type: socket type (%SOCK_STREAM, etc) 310 * @sk_protocol: which protocol this socket belongs in this network family 311 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 312 * @sk_peer_pid: &struct pid for this socket's peer 313 * @sk_peer_cred: %SO_PEERCRED setting 314 * @sk_rcvlowat: %SO_RCVLOWAT setting 315 * @sk_rcvtimeo: %SO_RCVTIMEO setting 316 * @sk_sndtimeo: %SO_SNDTIMEO setting 317 * @sk_txhash: computed flow hash for use on transmit 318 * @sk_txrehash: enable TX hash rethink 319 * @sk_filter: socket filtering instructions 320 * @sk_timer: sock cleanup timer 321 * @sk_stamp: time stamp of last packet received 322 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 323 * @sk_tsflags: SO_TIMESTAMPING flags 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_state_change: callback to indicate change in the state of the sock 340 * @sk_data_ready: callback to indicate there is data to be processed 341 * @sk_write_space: callback to indicate there is bf sending space available 342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 343 * @sk_backlog_rcv: callback to process the backlog 344 * @sk_validate_xmit_skb: ptr to an optional validate function 345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 346 * @sk_reuseport_cb: reuseport group container 347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 348 * @sk_rcu: used during RCU grace period 349 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 350 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 351 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 352 * @sk_txtime_unused: unused txtime flags 353 * @ns_tracker: tracker for netns reference 354 */ 355 struct sock { 356 /* 357 * Now struct inet_timewait_sock also uses sock_common, so please just 358 * don't add nothing before this first member (__sk_common) --acme 359 */ 360 struct sock_common __sk_common; 361 #define sk_node __sk_common.skc_node 362 #define sk_nulls_node __sk_common.skc_nulls_node 363 #define sk_refcnt __sk_common.skc_refcnt 364 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 365 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 366 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 367 #endif 368 369 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 370 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 371 #define sk_hash __sk_common.skc_hash 372 #define sk_portpair __sk_common.skc_portpair 373 #define sk_num __sk_common.skc_num 374 #define sk_dport __sk_common.skc_dport 375 #define sk_addrpair __sk_common.skc_addrpair 376 #define sk_daddr __sk_common.skc_daddr 377 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 378 #define sk_family __sk_common.skc_family 379 #define sk_state __sk_common.skc_state 380 #define sk_reuse __sk_common.skc_reuse 381 #define sk_reuseport __sk_common.skc_reuseport 382 #define sk_ipv6only __sk_common.skc_ipv6only 383 #define sk_net_refcnt __sk_common.skc_net_refcnt 384 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 385 #define sk_bind_node __sk_common.skc_bind_node 386 #define sk_prot __sk_common.skc_prot 387 #define sk_net __sk_common.skc_net 388 #define sk_v6_daddr __sk_common.skc_v6_daddr 389 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 390 #define sk_cookie __sk_common.skc_cookie 391 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 392 #define sk_flags __sk_common.skc_flags 393 #define sk_rxhash __sk_common.skc_rxhash 394 395 /* early demux fields */ 396 struct dst_entry __rcu *sk_rx_dst; 397 int sk_rx_dst_ifindex; 398 u32 sk_rx_dst_cookie; 399 400 socket_lock_t sk_lock; 401 atomic_t sk_drops; 402 int sk_rcvlowat; 403 struct sk_buff_head sk_error_queue; 404 struct sk_buff_head sk_receive_queue; 405 /* 406 * The backlog queue is special, it is always used with 407 * the per-socket spinlock held and requires low latency 408 * access. Therefore we special case it's implementation. 409 * Note : rmem_alloc is in this structure to fill a hole 410 * on 64bit arches, not because its logically part of 411 * backlog. 412 */ 413 struct { 414 atomic_t rmem_alloc; 415 int len; 416 struct sk_buff *head; 417 struct sk_buff *tail; 418 } sk_backlog; 419 420 #define sk_rmem_alloc sk_backlog.rmem_alloc 421 422 int sk_forward_alloc; 423 u32 sk_reserved_mem; 424 #ifdef CONFIG_NET_RX_BUSY_POLL 425 unsigned int sk_ll_usec; 426 /* ===== mostly read cache line ===== */ 427 unsigned int sk_napi_id; 428 #endif 429 int sk_rcvbuf; 430 431 struct sk_filter __rcu *sk_filter; 432 union { 433 struct socket_wq __rcu *sk_wq; 434 /* private: */ 435 struct socket_wq *sk_wq_raw; 436 /* public: */ 437 }; 438 #ifdef CONFIG_XFRM 439 struct xfrm_policy __rcu *sk_policy[2]; 440 #endif 441 442 struct dst_entry __rcu *sk_dst_cache; 443 atomic_t sk_omem_alloc; 444 int sk_sndbuf; 445 446 /* ===== cache line for TX ===== */ 447 int sk_wmem_queued; 448 refcount_t sk_wmem_alloc; 449 unsigned long sk_tsq_flags; 450 union { 451 struct sk_buff *sk_send_head; 452 struct rb_root tcp_rtx_queue; 453 }; 454 struct sk_buff_head sk_write_queue; 455 __s32 sk_peek_off; 456 int sk_write_pending; 457 __u32 sk_dst_pending_confirm; 458 u32 sk_pacing_status; /* see enum sk_pacing */ 459 long sk_sndtimeo; 460 struct timer_list sk_timer; 461 __u32 sk_priority; 462 __u32 sk_mark; 463 unsigned long sk_pacing_rate; /* bytes per second */ 464 unsigned long sk_max_pacing_rate; 465 struct page_frag sk_frag; 466 netdev_features_t sk_route_caps; 467 int sk_gso_type; 468 unsigned int sk_gso_max_size; 469 gfp_t sk_allocation; 470 __u32 sk_txhash; 471 472 /* 473 * Because of non atomicity rules, all 474 * changes are protected by socket lock. 475 */ 476 u8 sk_gso_disabled : 1, 477 sk_kern_sock : 1, 478 sk_no_check_tx : 1, 479 sk_no_check_rx : 1, 480 sk_userlocks : 4; 481 u8 sk_pacing_shift; 482 u16 sk_type; 483 u16 sk_protocol; 484 u16 sk_gso_max_segs; 485 unsigned long sk_lingertime; 486 struct proto *sk_prot_creator; 487 rwlock_t sk_callback_lock; 488 int sk_err, 489 sk_err_soft; 490 u32 sk_ack_backlog; 491 u32 sk_max_ack_backlog; 492 kuid_t sk_uid; 493 u8 sk_txrehash; 494 #ifdef CONFIG_NET_RX_BUSY_POLL 495 u8 sk_prefer_busy_poll; 496 u16 sk_busy_poll_budget; 497 #endif 498 spinlock_t sk_peer_lock; 499 int sk_bind_phc; 500 struct pid *sk_peer_pid; 501 const struct cred *sk_peer_cred; 502 503 long sk_rcvtimeo; 504 ktime_t sk_stamp; 505 #if BITS_PER_LONG==32 506 seqlock_t sk_stamp_seq; 507 #endif 508 u16 sk_tsflags; 509 u8 sk_shutdown; 510 atomic_t sk_tskey; 511 atomic_t sk_zckey; 512 513 u8 sk_clockid; 514 u8 sk_txtime_deadline_mode : 1, 515 sk_txtime_report_errors : 1, 516 sk_txtime_unused : 6; 517 518 struct socket *sk_socket; 519 void *sk_user_data; 520 #ifdef CONFIG_SECURITY 521 void *sk_security; 522 #endif 523 struct sock_cgroup_data sk_cgrp_data; 524 struct mem_cgroup *sk_memcg; 525 void (*sk_state_change)(struct sock *sk); 526 void (*sk_data_ready)(struct sock *sk); 527 void (*sk_write_space)(struct sock *sk); 528 void (*sk_error_report)(struct sock *sk); 529 int (*sk_backlog_rcv)(struct sock *sk, 530 struct sk_buff *skb); 531 #ifdef CONFIG_SOCK_VALIDATE_XMIT 532 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 533 struct net_device *dev, 534 struct sk_buff *skb); 535 #endif 536 void (*sk_destruct)(struct sock *sk); 537 struct sock_reuseport __rcu *sk_reuseport_cb; 538 #ifdef CONFIG_BPF_SYSCALL 539 struct bpf_local_storage __rcu *sk_bpf_storage; 540 #endif 541 struct rcu_head sk_rcu; 542 netns_tracker ns_tracker; 543 }; 544 545 enum sk_pacing { 546 SK_PACING_NONE = 0, 547 SK_PACING_NEEDED = 1, 548 SK_PACING_FQ = 2, 549 }; 550 551 /* Pointer stored in sk_user_data might not be suitable for copying 552 * when cloning the socket. For instance, it can point to a reference 553 * counted object. sk_user_data bottom bit is set if pointer must not 554 * be copied. 555 */ 556 #define SK_USER_DATA_NOCOPY 1UL 557 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 558 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 559 560 /** 561 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 562 * @sk: socket 563 */ 564 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 565 { 566 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 567 } 568 569 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 570 571 #define rcu_dereference_sk_user_data(sk) \ 572 ({ \ 573 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 574 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 575 }) 576 #define rcu_assign_sk_user_data(sk, ptr) \ 577 ({ \ 578 uintptr_t __tmp = (uintptr_t)(ptr); \ 579 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 580 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 581 }) 582 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 583 ({ \ 584 uintptr_t __tmp = (uintptr_t)(ptr); \ 585 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 586 rcu_assign_pointer(__sk_user_data((sk)), \ 587 __tmp | SK_USER_DATA_NOCOPY); \ 588 }) 589 590 static inline 591 struct net *sock_net(const struct sock *sk) 592 { 593 return read_pnet(&sk->sk_net); 594 } 595 596 static inline 597 void sock_net_set(struct sock *sk, struct net *net) 598 { 599 write_pnet(&sk->sk_net, net); 600 } 601 602 /* 603 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 604 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 605 * on a socket means that the socket will reuse everybody else's port 606 * without looking at the other's sk_reuse value. 607 */ 608 609 #define SK_NO_REUSE 0 610 #define SK_CAN_REUSE 1 611 #define SK_FORCE_REUSE 2 612 613 int sk_set_peek_off(struct sock *sk, int val); 614 615 static inline int sk_peek_offset(struct sock *sk, int flags) 616 { 617 if (unlikely(flags & MSG_PEEK)) { 618 return READ_ONCE(sk->sk_peek_off); 619 } 620 621 return 0; 622 } 623 624 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 625 { 626 s32 off = READ_ONCE(sk->sk_peek_off); 627 628 if (unlikely(off >= 0)) { 629 off = max_t(s32, off - val, 0); 630 WRITE_ONCE(sk->sk_peek_off, off); 631 } 632 } 633 634 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 635 { 636 sk_peek_offset_bwd(sk, -val); 637 } 638 639 /* 640 * Hashed lists helper routines 641 */ 642 static inline struct sock *sk_entry(const struct hlist_node *node) 643 { 644 return hlist_entry(node, struct sock, sk_node); 645 } 646 647 static inline struct sock *__sk_head(const struct hlist_head *head) 648 { 649 return hlist_entry(head->first, struct sock, sk_node); 650 } 651 652 static inline struct sock *sk_head(const struct hlist_head *head) 653 { 654 return hlist_empty(head) ? NULL : __sk_head(head); 655 } 656 657 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 658 { 659 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 660 } 661 662 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 663 { 664 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 665 } 666 667 static inline struct sock *sk_next(const struct sock *sk) 668 { 669 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 670 } 671 672 static inline struct sock *sk_nulls_next(const struct sock *sk) 673 { 674 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 675 hlist_nulls_entry(sk->sk_nulls_node.next, 676 struct sock, sk_nulls_node) : 677 NULL; 678 } 679 680 static inline bool sk_unhashed(const struct sock *sk) 681 { 682 return hlist_unhashed(&sk->sk_node); 683 } 684 685 static inline bool sk_hashed(const struct sock *sk) 686 { 687 return !sk_unhashed(sk); 688 } 689 690 static inline void sk_node_init(struct hlist_node *node) 691 { 692 node->pprev = NULL; 693 } 694 695 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 696 { 697 node->pprev = NULL; 698 } 699 700 static inline void __sk_del_node(struct sock *sk) 701 { 702 __hlist_del(&sk->sk_node); 703 } 704 705 /* NB: equivalent to hlist_del_init_rcu */ 706 static inline bool __sk_del_node_init(struct sock *sk) 707 { 708 if (sk_hashed(sk)) { 709 __sk_del_node(sk); 710 sk_node_init(&sk->sk_node); 711 return true; 712 } 713 return false; 714 } 715 716 /* Grab socket reference count. This operation is valid only 717 when sk is ALREADY grabbed f.e. it is found in hash table 718 or a list and the lookup is made under lock preventing hash table 719 modifications. 720 */ 721 722 static __always_inline void sock_hold(struct sock *sk) 723 { 724 refcount_inc(&sk->sk_refcnt); 725 } 726 727 /* Ungrab socket in the context, which assumes that socket refcnt 728 cannot hit zero, f.e. it is true in context of any socketcall. 729 */ 730 static __always_inline void __sock_put(struct sock *sk) 731 { 732 refcount_dec(&sk->sk_refcnt); 733 } 734 735 static inline bool sk_del_node_init(struct sock *sk) 736 { 737 bool rc = __sk_del_node_init(sk); 738 739 if (rc) { 740 /* paranoid for a while -acme */ 741 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 742 __sock_put(sk); 743 } 744 return rc; 745 } 746 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 747 748 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 749 { 750 if (sk_hashed(sk)) { 751 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 752 return true; 753 } 754 return false; 755 } 756 757 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 758 { 759 bool rc = __sk_nulls_del_node_init_rcu(sk); 760 761 if (rc) { 762 /* paranoid for a while -acme */ 763 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 764 __sock_put(sk); 765 } 766 return rc; 767 } 768 769 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 770 { 771 hlist_add_head(&sk->sk_node, list); 772 } 773 774 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 775 { 776 sock_hold(sk); 777 __sk_add_node(sk, list); 778 } 779 780 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 781 { 782 sock_hold(sk); 783 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 784 sk->sk_family == AF_INET6) 785 hlist_add_tail_rcu(&sk->sk_node, list); 786 else 787 hlist_add_head_rcu(&sk->sk_node, list); 788 } 789 790 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 791 { 792 sock_hold(sk); 793 hlist_add_tail_rcu(&sk->sk_node, list); 794 } 795 796 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 797 { 798 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 799 } 800 801 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 802 { 803 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 804 } 805 806 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 807 { 808 sock_hold(sk); 809 __sk_nulls_add_node_rcu(sk, list); 810 } 811 812 static inline void __sk_del_bind_node(struct sock *sk) 813 { 814 __hlist_del(&sk->sk_bind_node); 815 } 816 817 static inline void sk_add_bind_node(struct sock *sk, 818 struct hlist_head *list) 819 { 820 hlist_add_head(&sk->sk_bind_node, list); 821 } 822 823 #define sk_for_each(__sk, list) \ 824 hlist_for_each_entry(__sk, list, sk_node) 825 #define sk_for_each_rcu(__sk, list) \ 826 hlist_for_each_entry_rcu(__sk, list, sk_node) 827 #define sk_nulls_for_each(__sk, node, list) \ 828 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 829 #define sk_nulls_for_each_rcu(__sk, node, list) \ 830 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 831 #define sk_for_each_from(__sk) \ 832 hlist_for_each_entry_from(__sk, sk_node) 833 #define sk_nulls_for_each_from(__sk, node) \ 834 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 835 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 836 #define sk_for_each_safe(__sk, tmp, list) \ 837 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 838 #define sk_for_each_bound(__sk, list) \ 839 hlist_for_each_entry(__sk, list, sk_bind_node) 840 841 /** 842 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 843 * @tpos: the type * to use as a loop cursor. 844 * @pos: the &struct hlist_node to use as a loop cursor. 845 * @head: the head for your list. 846 * @offset: offset of hlist_node within the struct. 847 * 848 */ 849 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 850 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 851 pos != NULL && \ 852 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 853 pos = rcu_dereference(hlist_next_rcu(pos))) 854 855 static inline struct user_namespace *sk_user_ns(struct sock *sk) 856 { 857 /* Careful only use this in a context where these parameters 858 * can not change and must all be valid, such as recvmsg from 859 * userspace. 860 */ 861 return sk->sk_socket->file->f_cred->user_ns; 862 } 863 864 /* Sock flags */ 865 enum sock_flags { 866 SOCK_DEAD, 867 SOCK_DONE, 868 SOCK_URGINLINE, 869 SOCK_KEEPOPEN, 870 SOCK_LINGER, 871 SOCK_DESTROY, 872 SOCK_BROADCAST, 873 SOCK_TIMESTAMP, 874 SOCK_ZAPPED, 875 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 876 SOCK_DBG, /* %SO_DEBUG setting */ 877 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 878 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 879 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 880 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 881 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 882 SOCK_FASYNC, /* fasync() active */ 883 SOCK_RXQ_OVFL, 884 SOCK_ZEROCOPY, /* buffers from userspace */ 885 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 886 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 887 * Will use last 4 bytes of packet sent from 888 * user-space instead. 889 */ 890 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 891 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 892 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 893 SOCK_TXTIME, 894 SOCK_XDP, /* XDP is attached */ 895 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 896 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 897 }; 898 899 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 900 901 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 902 { 903 nsk->sk_flags = osk->sk_flags; 904 } 905 906 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 907 { 908 __set_bit(flag, &sk->sk_flags); 909 } 910 911 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 912 { 913 __clear_bit(flag, &sk->sk_flags); 914 } 915 916 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 917 int valbool) 918 { 919 if (valbool) 920 sock_set_flag(sk, bit); 921 else 922 sock_reset_flag(sk, bit); 923 } 924 925 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 926 { 927 return test_bit(flag, &sk->sk_flags); 928 } 929 930 #ifdef CONFIG_NET 931 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 932 static inline int sk_memalloc_socks(void) 933 { 934 return static_branch_unlikely(&memalloc_socks_key); 935 } 936 937 void __receive_sock(struct file *file); 938 #else 939 940 static inline int sk_memalloc_socks(void) 941 { 942 return 0; 943 } 944 945 static inline void __receive_sock(struct file *file) 946 { } 947 #endif 948 949 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 950 { 951 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 952 } 953 954 static inline void sk_acceptq_removed(struct sock *sk) 955 { 956 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 957 } 958 959 static inline void sk_acceptq_added(struct sock *sk) 960 { 961 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 962 } 963 964 /* Note: If you think the test should be: 965 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 966 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 967 */ 968 static inline bool sk_acceptq_is_full(const struct sock *sk) 969 { 970 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 971 } 972 973 /* 974 * Compute minimal free write space needed to queue new packets. 975 */ 976 static inline int sk_stream_min_wspace(const struct sock *sk) 977 { 978 return READ_ONCE(sk->sk_wmem_queued) >> 1; 979 } 980 981 static inline int sk_stream_wspace(const struct sock *sk) 982 { 983 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 984 } 985 986 static inline void sk_wmem_queued_add(struct sock *sk, int val) 987 { 988 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 989 } 990 991 void sk_stream_write_space(struct sock *sk); 992 993 /* OOB backlog add */ 994 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 995 { 996 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 997 skb_dst_force(skb); 998 999 if (!sk->sk_backlog.tail) 1000 WRITE_ONCE(sk->sk_backlog.head, skb); 1001 else 1002 sk->sk_backlog.tail->next = skb; 1003 1004 WRITE_ONCE(sk->sk_backlog.tail, skb); 1005 skb->next = NULL; 1006 } 1007 1008 /* 1009 * Take into account size of receive queue and backlog queue 1010 * Do not take into account this skb truesize, 1011 * to allow even a single big packet to come. 1012 */ 1013 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1014 { 1015 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1016 1017 return qsize > limit; 1018 } 1019 1020 /* The per-socket spinlock must be held here. */ 1021 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1022 unsigned int limit) 1023 { 1024 if (sk_rcvqueues_full(sk, limit)) 1025 return -ENOBUFS; 1026 1027 /* 1028 * If the skb was allocated from pfmemalloc reserves, only 1029 * allow SOCK_MEMALLOC sockets to use it as this socket is 1030 * helping free memory 1031 */ 1032 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1033 return -ENOMEM; 1034 1035 __sk_add_backlog(sk, skb); 1036 sk->sk_backlog.len += skb->truesize; 1037 return 0; 1038 } 1039 1040 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1041 1042 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1043 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1044 1045 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1046 { 1047 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1048 return __sk_backlog_rcv(sk, skb); 1049 1050 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1051 tcp_v6_do_rcv, 1052 tcp_v4_do_rcv, 1053 sk, skb); 1054 } 1055 1056 static inline void sk_incoming_cpu_update(struct sock *sk) 1057 { 1058 int cpu = raw_smp_processor_id(); 1059 1060 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1061 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1062 } 1063 1064 static inline void sock_rps_record_flow_hash(__u32 hash) 1065 { 1066 #ifdef CONFIG_RPS 1067 struct rps_sock_flow_table *sock_flow_table; 1068 1069 rcu_read_lock(); 1070 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1071 rps_record_sock_flow(sock_flow_table, hash); 1072 rcu_read_unlock(); 1073 #endif 1074 } 1075 1076 static inline void sock_rps_record_flow(const struct sock *sk) 1077 { 1078 #ifdef CONFIG_RPS 1079 if (static_branch_unlikely(&rfs_needed)) { 1080 /* Reading sk->sk_rxhash might incur an expensive cache line 1081 * miss. 1082 * 1083 * TCP_ESTABLISHED does cover almost all states where RFS 1084 * might be useful, and is cheaper [1] than testing : 1085 * IPv4: inet_sk(sk)->inet_daddr 1086 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1087 * OR an additional socket flag 1088 * [1] : sk_state and sk_prot are in the same cache line. 1089 */ 1090 if (sk->sk_state == TCP_ESTABLISHED) 1091 sock_rps_record_flow_hash(sk->sk_rxhash); 1092 } 1093 #endif 1094 } 1095 1096 static inline void sock_rps_save_rxhash(struct sock *sk, 1097 const struct sk_buff *skb) 1098 { 1099 #ifdef CONFIG_RPS 1100 if (unlikely(sk->sk_rxhash != skb->hash)) 1101 sk->sk_rxhash = skb->hash; 1102 #endif 1103 } 1104 1105 static inline void sock_rps_reset_rxhash(struct sock *sk) 1106 { 1107 #ifdef CONFIG_RPS 1108 sk->sk_rxhash = 0; 1109 #endif 1110 } 1111 1112 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1113 ({ int __rc; \ 1114 release_sock(__sk); \ 1115 __rc = __condition; \ 1116 if (!__rc) { \ 1117 *(__timeo) = wait_woken(__wait, \ 1118 TASK_INTERRUPTIBLE, \ 1119 *(__timeo)); \ 1120 } \ 1121 sched_annotate_sleep(); \ 1122 lock_sock(__sk); \ 1123 __rc = __condition; \ 1124 __rc; \ 1125 }) 1126 1127 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1128 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1129 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1130 int sk_stream_error(struct sock *sk, int flags, int err); 1131 void sk_stream_kill_queues(struct sock *sk); 1132 void sk_set_memalloc(struct sock *sk); 1133 void sk_clear_memalloc(struct sock *sk); 1134 1135 void __sk_flush_backlog(struct sock *sk); 1136 1137 static inline bool sk_flush_backlog(struct sock *sk) 1138 { 1139 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1140 __sk_flush_backlog(sk); 1141 return true; 1142 } 1143 return false; 1144 } 1145 1146 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1147 1148 struct request_sock_ops; 1149 struct timewait_sock_ops; 1150 struct inet_hashinfo; 1151 struct raw_hashinfo; 1152 struct smc_hashinfo; 1153 struct module; 1154 struct sk_psock; 1155 1156 /* 1157 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1158 * un-modified. Special care is taken when initializing object to zero. 1159 */ 1160 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1161 { 1162 if (offsetof(struct sock, sk_node.next) != 0) 1163 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1164 memset(&sk->sk_node.pprev, 0, 1165 size - offsetof(struct sock, sk_node.pprev)); 1166 } 1167 1168 /* Networking protocol blocks we attach to sockets. 1169 * socket layer -> transport layer interface 1170 */ 1171 struct proto { 1172 void (*close)(struct sock *sk, 1173 long timeout); 1174 int (*pre_connect)(struct sock *sk, 1175 struct sockaddr *uaddr, 1176 int addr_len); 1177 int (*connect)(struct sock *sk, 1178 struct sockaddr *uaddr, 1179 int addr_len); 1180 int (*disconnect)(struct sock *sk, int flags); 1181 1182 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1183 bool kern); 1184 1185 int (*ioctl)(struct sock *sk, int cmd, 1186 unsigned long arg); 1187 int (*init)(struct sock *sk); 1188 void (*destroy)(struct sock *sk); 1189 void (*shutdown)(struct sock *sk, int how); 1190 int (*setsockopt)(struct sock *sk, int level, 1191 int optname, sockptr_t optval, 1192 unsigned int optlen); 1193 int (*getsockopt)(struct sock *sk, int level, 1194 int optname, char __user *optval, 1195 int __user *option); 1196 void (*keepalive)(struct sock *sk, int valbool); 1197 #ifdef CONFIG_COMPAT 1198 int (*compat_ioctl)(struct sock *sk, 1199 unsigned int cmd, unsigned long arg); 1200 #endif 1201 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1202 size_t len); 1203 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1204 size_t len, int flags, int *addr_len); 1205 int (*sendpage)(struct sock *sk, struct page *page, 1206 int offset, size_t size, int flags); 1207 int (*bind)(struct sock *sk, 1208 struct sockaddr *addr, int addr_len); 1209 int (*bind_add)(struct sock *sk, 1210 struct sockaddr *addr, int addr_len); 1211 1212 int (*backlog_rcv) (struct sock *sk, 1213 struct sk_buff *skb); 1214 bool (*bpf_bypass_getsockopt)(int level, 1215 int optname); 1216 1217 void (*release_cb)(struct sock *sk); 1218 1219 /* Keeping track of sk's, looking them up, and port selection methods. */ 1220 int (*hash)(struct sock *sk); 1221 void (*unhash)(struct sock *sk); 1222 void (*rehash)(struct sock *sk); 1223 int (*get_port)(struct sock *sk, unsigned short snum); 1224 void (*put_port)(struct sock *sk); 1225 #ifdef CONFIG_BPF_SYSCALL 1226 int (*psock_update_sk_prot)(struct sock *sk, 1227 struct sk_psock *psock, 1228 bool restore); 1229 #endif 1230 1231 /* Keeping track of sockets in use */ 1232 #ifdef CONFIG_PROC_FS 1233 unsigned int inuse_idx; 1234 #endif 1235 1236 #if IS_ENABLED(CONFIG_MPTCP) 1237 int (*forward_alloc_get)(const struct sock *sk); 1238 #endif 1239 1240 bool (*stream_memory_free)(const struct sock *sk, int wake); 1241 bool (*sock_is_readable)(struct sock *sk); 1242 /* Memory pressure */ 1243 void (*enter_memory_pressure)(struct sock *sk); 1244 void (*leave_memory_pressure)(struct sock *sk); 1245 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1246 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1247 1248 /* 1249 * Pressure flag: try to collapse. 1250 * Technical note: it is used by multiple contexts non atomically. 1251 * All the __sk_mem_schedule() is of this nature: accounting 1252 * is strict, actions are advisory and have some latency. 1253 */ 1254 unsigned long *memory_pressure; 1255 long *sysctl_mem; 1256 1257 int *sysctl_wmem; 1258 int *sysctl_rmem; 1259 u32 sysctl_wmem_offset; 1260 u32 sysctl_rmem_offset; 1261 1262 int max_header; 1263 bool no_autobind; 1264 1265 struct kmem_cache *slab; 1266 unsigned int obj_size; 1267 slab_flags_t slab_flags; 1268 unsigned int useroffset; /* Usercopy region offset */ 1269 unsigned int usersize; /* Usercopy region size */ 1270 1271 unsigned int __percpu *orphan_count; 1272 1273 struct request_sock_ops *rsk_prot; 1274 struct timewait_sock_ops *twsk_prot; 1275 1276 union { 1277 struct inet_hashinfo *hashinfo; 1278 struct udp_table *udp_table; 1279 struct raw_hashinfo *raw_hash; 1280 struct smc_hashinfo *smc_hash; 1281 } h; 1282 1283 struct module *owner; 1284 1285 char name[32]; 1286 1287 struct list_head node; 1288 #ifdef SOCK_REFCNT_DEBUG 1289 atomic_t socks; 1290 #endif 1291 int (*diag_destroy)(struct sock *sk, int err); 1292 } __randomize_layout; 1293 1294 int proto_register(struct proto *prot, int alloc_slab); 1295 void proto_unregister(struct proto *prot); 1296 int sock_load_diag_module(int family, int protocol); 1297 1298 #ifdef SOCK_REFCNT_DEBUG 1299 static inline void sk_refcnt_debug_inc(struct sock *sk) 1300 { 1301 atomic_inc(&sk->sk_prot->socks); 1302 } 1303 1304 static inline void sk_refcnt_debug_dec(struct sock *sk) 1305 { 1306 atomic_dec(&sk->sk_prot->socks); 1307 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1308 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1309 } 1310 1311 static inline void sk_refcnt_debug_release(const struct sock *sk) 1312 { 1313 if (refcount_read(&sk->sk_refcnt) != 1) 1314 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1315 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1316 } 1317 #else /* SOCK_REFCNT_DEBUG */ 1318 #define sk_refcnt_debug_inc(sk) do { } while (0) 1319 #define sk_refcnt_debug_dec(sk) do { } while (0) 1320 #define sk_refcnt_debug_release(sk) do { } while (0) 1321 #endif /* SOCK_REFCNT_DEBUG */ 1322 1323 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1324 1325 static inline int sk_forward_alloc_get(const struct sock *sk) 1326 { 1327 #if IS_ENABLED(CONFIG_MPTCP) 1328 if (sk->sk_prot->forward_alloc_get) 1329 return sk->sk_prot->forward_alloc_get(sk); 1330 #endif 1331 return sk->sk_forward_alloc; 1332 } 1333 1334 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1335 { 1336 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1337 return false; 1338 1339 return sk->sk_prot->stream_memory_free ? 1340 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1341 tcp_stream_memory_free, sk, wake) : true; 1342 } 1343 1344 static inline bool sk_stream_memory_free(const struct sock *sk) 1345 { 1346 return __sk_stream_memory_free(sk, 0); 1347 } 1348 1349 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1350 { 1351 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1352 __sk_stream_memory_free(sk, wake); 1353 } 1354 1355 static inline bool sk_stream_is_writeable(const struct sock *sk) 1356 { 1357 return __sk_stream_is_writeable(sk, 0); 1358 } 1359 1360 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1361 struct cgroup *ancestor) 1362 { 1363 #ifdef CONFIG_SOCK_CGROUP_DATA 1364 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1365 ancestor); 1366 #else 1367 return -ENOTSUPP; 1368 #endif 1369 } 1370 1371 static inline bool sk_has_memory_pressure(const struct sock *sk) 1372 { 1373 return sk->sk_prot->memory_pressure != NULL; 1374 } 1375 1376 static inline bool sk_under_memory_pressure(const struct sock *sk) 1377 { 1378 if (!sk->sk_prot->memory_pressure) 1379 return false; 1380 1381 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1382 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1383 return true; 1384 1385 return !!*sk->sk_prot->memory_pressure; 1386 } 1387 1388 static inline long 1389 sk_memory_allocated(const struct sock *sk) 1390 { 1391 return atomic_long_read(sk->sk_prot->memory_allocated); 1392 } 1393 1394 static inline long 1395 sk_memory_allocated_add(struct sock *sk, int amt) 1396 { 1397 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1398 } 1399 1400 static inline void 1401 sk_memory_allocated_sub(struct sock *sk, int amt) 1402 { 1403 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1404 } 1405 1406 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1407 1408 static inline void sk_sockets_allocated_dec(struct sock *sk) 1409 { 1410 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1411 SK_ALLOC_PERCPU_COUNTER_BATCH); 1412 } 1413 1414 static inline void sk_sockets_allocated_inc(struct sock *sk) 1415 { 1416 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1417 SK_ALLOC_PERCPU_COUNTER_BATCH); 1418 } 1419 1420 static inline u64 1421 sk_sockets_allocated_read_positive(struct sock *sk) 1422 { 1423 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1424 } 1425 1426 static inline int 1427 proto_sockets_allocated_sum_positive(struct proto *prot) 1428 { 1429 return percpu_counter_sum_positive(prot->sockets_allocated); 1430 } 1431 1432 static inline long 1433 proto_memory_allocated(struct proto *prot) 1434 { 1435 return atomic_long_read(prot->memory_allocated); 1436 } 1437 1438 static inline bool 1439 proto_memory_pressure(struct proto *prot) 1440 { 1441 if (!prot->memory_pressure) 1442 return false; 1443 return !!*prot->memory_pressure; 1444 } 1445 1446 1447 #ifdef CONFIG_PROC_FS 1448 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1449 struct prot_inuse { 1450 int all; 1451 int val[PROTO_INUSE_NR]; 1452 }; 1453 1454 static inline void sock_prot_inuse_add(const struct net *net, 1455 const struct proto *prot, int val) 1456 { 1457 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1458 } 1459 1460 static inline void sock_inuse_add(const struct net *net, int val) 1461 { 1462 this_cpu_add(net->core.prot_inuse->all, val); 1463 } 1464 1465 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1466 int sock_inuse_get(struct net *net); 1467 #else 1468 static inline void sock_prot_inuse_add(const struct net *net, 1469 const struct proto *prot, int val) 1470 { 1471 } 1472 1473 static inline void sock_inuse_add(const struct net *net, int val) 1474 { 1475 } 1476 #endif 1477 1478 1479 /* With per-bucket locks this operation is not-atomic, so that 1480 * this version is not worse. 1481 */ 1482 static inline int __sk_prot_rehash(struct sock *sk) 1483 { 1484 sk->sk_prot->unhash(sk); 1485 return sk->sk_prot->hash(sk); 1486 } 1487 1488 /* About 10 seconds */ 1489 #define SOCK_DESTROY_TIME (10*HZ) 1490 1491 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1492 #define PROT_SOCK 1024 1493 1494 #define SHUTDOWN_MASK 3 1495 #define RCV_SHUTDOWN 1 1496 #define SEND_SHUTDOWN 2 1497 1498 #define SOCK_BINDADDR_LOCK 4 1499 #define SOCK_BINDPORT_LOCK 8 1500 1501 struct socket_alloc { 1502 struct socket socket; 1503 struct inode vfs_inode; 1504 }; 1505 1506 static inline struct socket *SOCKET_I(struct inode *inode) 1507 { 1508 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1509 } 1510 1511 static inline struct inode *SOCK_INODE(struct socket *socket) 1512 { 1513 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1514 } 1515 1516 /* 1517 * Functions for memory accounting 1518 */ 1519 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1520 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1521 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1522 void __sk_mem_reclaim(struct sock *sk, int amount); 1523 1524 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1525 * do not necessarily have 16x time more memory than 4KB ones. 1526 */ 1527 #define SK_MEM_QUANTUM 4096 1528 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1529 #define SK_MEM_SEND 0 1530 #define SK_MEM_RECV 1 1531 1532 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1533 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1534 { 1535 long val = sk->sk_prot->sysctl_mem[index]; 1536 1537 #if PAGE_SIZE > SK_MEM_QUANTUM 1538 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1539 #elif PAGE_SIZE < SK_MEM_QUANTUM 1540 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1541 #endif 1542 return val; 1543 } 1544 1545 static inline int sk_mem_pages(int amt) 1546 { 1547 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1548 } 1549 1550 static inline bool sk_has_account(struct sock *sk) 1551 { 1552 /* return true if protocol supports memory accounting */ 1553 return !!sk->sk_prot->memory_allocated; 1554 } 1555 1556 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1557 { 1558 if (!sk_has_account(sk)) 1559 return true; 1560 return size <= sk->sk_forward_alloc || 1561 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1562 } 1563 1564 static inline bool 1565 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1566 { 1567 if (!sk_has_account(sk)) 1568 return true; 1569 return size <= sk->sk_forward_alloc || 1570 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1571 skb_pfmemalloc(skb); 1572 } 1573 1574 static inline int sk_unused_reserved_mem(const struct sock *sk) 1575 { 1576 int unused_mem; 1577 1578 if (likely(!sk->sk_reserved_mem)) 1579 return 0; 1580 1581 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1582 atomic_read(&sk->sk_rmem_alloc); 1583 1584 return unused_mem > 0 ? unused_mem : 0; 1585 } 1586 1587 static inline void sk_mem_reclaim(struct sock *sk) 1588 { 1589 int reclaimable; 1590 1591 if (!sk_has_account(sk)) 1592 return; 1593 1594 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1595 1596 if (reclaimable >= SK_MEM_QUANTUM) 1597 __sk_mem_reclaim(sk, reclaimable); 1598 } 1599 1600 static inline void sk_mem_reclaim_final(struct sock *sk) 1601 { 1602 sk->sk_reserved_mem = 0; 1603 sk_mem_reclaim(sk); 1604 } 1605 1606 static inline void sk_mem_reclaim_partial(struct sock *sk) 1607 { 1608 int reclaimable; 1609 1610 if (!sk_has_account(sk)) 1611 return; 1612 1613 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1614 1615 if (reclaimable > SK_MEM_QUANTUM) 1616 __sk_mem_reclaim(sk, reclaimable - 1); 1617 } 1618 1619 static inline void sk_mem_charge(struct sock *sk, int size) 1620 { 1621 if (!sk_has_account(sk)) 1622 return; 1623 sk->sk_forward_alloc -= size; 1624 } 1625 1626 /* the following macros control memory reclaiming in sk_mem_uncharge() 1627 */ 1628 #define SK_RECLAIM_THRESHOLD (1 << 21) 1629 #define SK_RECLAIM_CHUNK (1 << 20) 1630 1631 static inline void sk_mem_uncharge(struct sock *sk, int size) 1632 { 1633 int reclaimable; 1634 1635 if (!sk_has_account(sk)) 1636 return; 1637 sk->sk_forward_alloc += size; 1638 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1639 1640 /* Avoid a possible overflow. 1641 * TCP send queues can make this happen, if sk_mem_reclaim() 1642 * is not called and more than 2 GBytes are released at once. 1643 * 1644 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1645 * no need to hold that much forward allocation anyway. 1646 */ 1647 if (unlikely(reclaimable >= SK_RECLAIM_THRESHOLD)) 1648 __sk_mem_reclaim(sk, SK_RECLAIM_CHUNK); 1649 } 1650 1651 /* 1652 * Macro so as to not evaluate some arguments when 1653 * lockdep is not enabled. 1654 * 1655 * Mark both the sk_lock and the sk_lock.slock as a 1656 * per-address-family lock class. 1657 */ 1658 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1659 do { \ 1660 sk->sk_lock.owned = 0; \ 1661 init_waitqueue_head(&sk->sk_lock.wq); \ 1662 spin_lock_init(&(sk)->sk_lock.slock); \ 1663 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1664 sizeof((sk)->sk_lock)); \ 1665 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1666 (skey), (sname)); \ 1667 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1668 } while (0) 1669 1670 static inline bool lockdep_sock_is_held(const struct sock *sk) 1671 { 1672 return lockdep_is_held(&sk->sk_lock) || 1673 lockdep_is_held(&sk->sk_lock.slock); 1674 } 1675 1676 void lock_sock_nested(struct sock *sk, int subclass); 1677 1678 static inline void lock_sock(struct sock *sk) 1679 { 1680 lock_sock_nested(sk, 0); 1681 } 1682 1683 void __lock_sock(struct sock *sk); 1684 void __release_sock(struct sock *sk); 1685 void release_sock(struct sock *sk); 1686 1687 /* BH context may only use the following locking interface. */ 1688 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1689 #define bh_lock_sock_nested(__sk) \ 1690 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1691 SINGLE_DEPTH_NESTING) 1692 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1693 1694 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1695 1696 /** 1697 * lock_sock_fast - fast version of lock_sock 1698 * @sk: socket 1699 * 1700 * This version should be used for very small section, where process wont block 1701 * return false if fast path is taken: 1702 * 1703 * sk_lock.slock locked, owned = 0, BH disabled 1704 * 1705 * return true if slow path is taken: 1706 * 1707 * sk_lock.slock unlocked, owned = 1, BH enabled 1708 */ 1709 static inline bool lock_sock_fast(struct sock *sk) 1710 { 1711 /* The sk_lock has mutex_lock() semantics here. */ 1712 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1713 1714 return __lock_sock_fast(sk); 1715 } 1716 1717 /* fast socket lock variant for caller already holding a [different] socket lock */ 1718 static inline bool lock_sock_fast_nested(struct sock *sk) 1719 { 1720 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1721 1722 return __lock_sock_fast(sk); 1723 } 1724 1725 /** 1726 * unlock_sock_fast - complement of lock_sock_fast 1727 * @sk: socket 1728 * @slow: slow mode 1729 * 1730 * fast unlock socket for user context. 1731 * If slow mode is on, we call regular release_sock() 1732 */ 1733 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1734 __releases(&sk->sk_lock.slock) 1735 { 1736 if (slow) { 1737 release_sock(sk); 1738 __release(&sk->sk_lock.slock); 1739 } else { 1740 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1741 spin_unlock_bh(&sk->sk_lock.slock); 1742 } 1743 } 1744 1745 /* Used by processes to "lock" a socket state, so that 1746 * interrupts and bottom half handlers won't change it 1747 * from under us. It essentially blocks any incoming 1748 * packets, so that we won't get any new data or any 1749 * packets that change the state of the socket. 1750 * 1751 * While locked, BH processing will add new packets to 1752 * the backlog queue. This queue is processed by the 1753 * owner of the socket lock right before it is released. 1754 * 1755 * Since ~2.3.5 it is also exclusive sleep lock serializing 1756 * accesses from user process context. 1757 */ 1758 1759 static inline void sock_owned_by_me(const struct sock *sk) 1760 { 1761 #ifdef CONFIG_LOCKDEP 1762 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1763 #endif 1764 } 1765 1766 static inline bool sock_owned_by_user(const struct sock *sk) 1767 { 1768 sock_owned_by_me(sk); 1769 return sk->sk_lock.owned; 1770 } 1771 1772 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1773 { 1774 return sk->sk_lock.owned; 1775 } 1776 1777 static inline void sock_release_ownership(struct sock *sk) 1778 { 1779 if (sock_owned_by_user_nocheck(sk)) { 1780 sk->sk_lock.owned = 0; 1781 1782 /* The sk_lock has mutex_unlock() semantics: */ 1783 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1784 } 1785 } 1786 1787 /* no reclassification while locks are held */ 1788 static inline bool sock_allow_reclassification(const struct sock *csk) 1789 { 1790 struct sock *sk = (struct sock *)csk; 1791 1792 return !sock_owned_by_user_nocheck(sk) && 1793 !spin_is_locked(&sk->sk_lock.slock); 1794 } 1795 1796 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1797 struct proto *prot, int kern); 1798 void sk_free(struct sock *sk); 1799 void sk_destruct(struct sock *sk); 1800 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1801 void sk_free_unlock_clone(struct sock *sk); 1802 1803 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1804 gfp_t priority); 1805 void __sock_wfree(struct sk_buff *skb); 1806 void sock_wfree(struct sk_buff *skb); 1807 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1808 gfp_t priority); 1809 void skb_orphan_partial(struct sk_buff *skb); 1810 void sock_rfree(struct sk_buff *skb); 1811 void sock_efree(struct sk_buff *skb); 1812 #ifdef CONFIG_INET 1813 void sock_edemux(struct sk_buff *skb); 1814 void sock_pfree(struct sk_buff *skb); 1815 #else 1816 #define sock_edemux sock_efree 1817 #endif 1818 1819 int sock_setsockopt(struct socket *sock, int level, int op, 1820 sockptr_t optval, unsigned int optlen); 1821 1822 int sock_getsockopt(struct socket *sock, int level, int op, 1823 char __user *optval, int __user *optlen); 1824 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1825 bool timeval, bool time32); 1826 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1827 unsigned long data_len, int noblock, 1828 int *errcode, int max_page_order); 1829 1830 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1831 unsigned long size, 1832 int noblock, int *errcode) 1833 { 1834 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1835 } 1836 1837 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1838 void sock_kfree_s(struct sock *sk, void *mem, int size); 1839 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1840 void sk_send_sigurg(struct sock *sk); 1841 1842 struct sockcm_cookie { 1843 u64 transmit_time; 1844 u32 mark; 1845 u16 tsflags; 1846 }; 1847 1848 static inline void sockcm_init(struct sockcm_cookie *sockc, 1849 const struct sock *sk) 1850 { 1851 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1852 } 1853 1854 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1855 struct sockcm_cookie *sockc); 1856 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1857 struct sockcm_cookie *sockc); 1858 1859 /* 1860 * Functions to fill in entries in struct proto_ops when a protocol 1861 * does not implement a particular function. 1862 */ 1863 int sock_no_bind(struct socket *, struct sockaddr *, int); 1864 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1865 int sock_no_socketpair(struct socket *, struct socket *); 1866 int sock_no_accept(struct socket *, struct socket *, int, bool); 1867 int sock_no_getname(struct socket *, struct sockaddr *, int); 1868 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1869 int sock_no_listen(struct socket *, int); 1870 int sock_no_shutdown(struct socket *, int); 1871 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1872 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1873 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1874 int sock_no_mmap(struct file *file, struct socket *sock, 1875 struct vm_area_struct *vma); 1876 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1877 size_t size, int flags); 1878 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1879 int offset, size_t size, int flags); 1880 1881 /* 1882 * Functions to fill in entries in struct proto_ops when a protocol 1883 * uses the inet style. 1884 */ 1885 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1886 char __user *optval, int __user *optlen); 1887 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1888 int flags); 1889 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1890 sockptr_t optval, unsigned int optlen); 1891 1892 void sk_common_release(struct sock *sk); 1893 1894 /* 1895 * Default socket callbacks and setup code 1896 */ 1897 1898 /* Initialise core socket variables */ 1899 void sock_init_data(struct socket *sock, struct sock *sk); 1900 1901 /* 1902 * Socket reference counting postulates. 1903 * 1904 * * Each user of socket SHOULD hold a reference count. 1905 * * Each access point to socket (an hash table bucket, reference from a list, 1906 * running timer, skb in flight MUST hold a reference count. 1907 * * When reference count hits 0, it means it will never increase back. 1908 * * When reference count hits 0, it means that no references from 1909 * outside exist to this socket and current process on current CPU 1910 * is last user and may/should destroy this socket. 1911 * * sk_free is called from any context: process, BH, IRQ. When 1912 * it is called, socket has no references from outside -> sk_free 1913 * may release descendant resources allocated by the socket, but 1914 * to the time when it is called, socket is NOT referenced by any 1915 * hash tables, lists etc. 1916 * * Packets, delivered from outside (from network or from another process) 1917 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1918 * when they sit in queue. Otherwise, packets will leak to hole, when 1919 * socket is looked up by one cpu and unhasing is made by another CPU. 1920 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1921 * (leak to backlog). Packet socket does all the processing inside 1922 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1923 * use separate SMP lock, so that they are prone too. 1924 */ 1925 1926 /* Ungrab socket and destroy it, if it was the last reference. */ 1927 static inline void sock_put(struct sock *sk) 1928 { 1929 if (refcount_dec_and_test(&sk->sk_refcnt)) 1930 sk_free(sk); 1931 } 1932 /* Generic version of sock_put(), dealing with all sockets 1933 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1934 */ 1935 void sock_gen_put(struct sock *sk); 1936 1937 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1938 unsigned int trim_cap, bool refcounted); 1939 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1940 const int nested) 1941 { 1942 return __sk_receive_skb(sk, skb, nested, 1, true); 1943 } 1944 1945 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1946 { 1947 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1948 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1949 return; 1950 sk->sk_tx_queue_mapping = tx_queue; 1951 } 1952 1953 #define NO_QUEUE_MAPPING USHRT_MAX 1954 1955 static inline void sk_tx_queue_clear(struct sock *sk) 1956 { 1957 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1958 } 1959 1960 static inline int sk_tx_queue_get(const struct sock *sk) 1961 { 1962 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1963 return sk->sk_tx_queue_mapping; 1964 1965 return -1; 1966 } 1967 1968 static inline void __sk_rx_queue_set(struct sock *sk, 1969 const struct sk_buff *skb, 1970 bool force_set) 1971 { 1972 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1973 if (skb_rx_queue_recorded(skb)) { 1974 u16 rx_queue = skb_get_rx_queue(skb); 1975 1976 if (force_set || 1977 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 1978 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 1979 } 1980 #endif 1981 } 1982 1983 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1984 { 1985 __sk_rx_queue_set(sk, skb, true); 1986 } 1987 1988 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 1989 { 1990 __sk_rx_queue_set(sk, skb, false); 1991 } 1992 1993 static inline void sk_rx_queue_clear(struct sock *sk) 1994 { 1995 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1996 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 1997 #endif 1998 } 1999 2000 static inline int sk_rx_queue_get(const struct sock *sk) 2001 { 2002 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2003 if (sk) { 2004 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2005 2006 if (res != NO_QUEUE_MAPPING) 2007 return res; 2008 } 2009 #endif 2010 2011 return -1; 2012 } 2013 2014 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2015 { 2016 sk->sk_socket = sock; 2017 } 2018 2019 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2020 { 2021 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2022 return &rcu_dereference_raw(sk->sk_wq)->wait; 2023 } 2024 /* Detach socket from process context. 2025 * Announce socket dead, detach it from wait queue and inode. 2026 * Note that parent inode held reference count on this struct sock, 2027 * we do not release it in this function, because protocol 2028 * probably wants some additional cleanups or even continuing 2029 * to work with this socket (TCP). 2030 */ 2031 static inline void sock_orphan(struct sock *sk) 2032 { 2033 write_lock_bh(&sk->sk_callback_lock); 2034 sock_set_flag(sk, SOCK_DEAD); 2035 sk_set_socket(sk, NULL); 2036 sk->sk_wq = NULL; 2037 write_unlock_bh(&sk->sk_callback_lock); 2038 } 2039 2040 static inline void sock_graft(struct sock *sk, struct socket *parent) 2041 { 2042 WARN_ON(parent->sk); 2043 write_lock_bh(&sk->sk_callback_lock); 2044 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2045 parent->sk = sk; 2046 sk_set_socket(sk, parent); 2047 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2048 security_sock_graft(sk, parent); 2049 write_unlock_bh(&sk->sk_callback_lock); 2050 } 2051 2052 kuid_t sock_i_uid(struct sock *sk); 2053 unsigned long sock_i_ino(struct sock *sk); 2054 2055 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2056 { 2057 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2058 } 2059 2060 static inline u32 net_tx_rndhash(void) 2061 { 2062 u32 v = prandom_u32(); 2063 2064 return v ?: 1; 2065 } 2066 2067 static inline void sk_set_txhash(struct sock *sk) 2068 { 2069 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2070 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2071 } 2072 2073 static inline bool sk_rethink_txhash(struct sock *sk) 2074 { 2075 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2076 sk_set_txhash(sk); 2077 return true; 2078 } 2079 return false; 2080 } 2081 2082 static inline struct dst_entry * 2083 __sk_dst_get(struct sock *sk) 2084 { 2085 return rcu_dereference_check(sk->sk_dst_cache, 2086 lockdep_sock_is_held(sk)); 2087 } 2088 2089 static inline struct dst_entry * 2090 sk_dst_get(struct sock *sk) 2091 { 2092 struct dst_entry *dst; 2093 2094 rcu_read_lock(); 2095 dst = rcu_dereference(sk->sk_dst_cache); 2096 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2097 dst = NULL; 2098 rcu_read_unlock(); 2099 return dst; 2100 } 2101 2102 static inline void __dst_negative_advice(struct sock *sk) 2103 { 2104 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2105 2106 if (dst && dst->ops->negative_advice) { 2107 ndst = dst->ops->negative_advice(dst); 2108 2109 if (ndst != dst) { 2110 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2111 sk_tx_queue_clear(sk); 2112 sk->sk_dst_pending_confirm = 0; 2113 } 2114 } 2115 } 2116 2117 static inline void dst_negative_advice(struct sock *sk) 2118 { 2119 sk_rethink_txhash(sk); 2120 __dst_negative_advice(sk); 2121 } 2122 2123 static inline void 2124 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2125 { 2126 struct dst_entry *old_dst; 2127 2128 sk_tx_queue_clear(sk); 2129 sk->sk_dst_pending_confirm = 0; 2130 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2131 lockdep_sock_is_held(sk)); 2132 rcu_assign_pointer(sk->sk_dst_cache, dst); 2133 dst_release(old_dst); 2134 } 2135 2136 static inline void 2137 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2138 { 2139 struct dst_entry *old_dst; 2140 2141 sk_tx_queue_clear(sk); 2142 sk->sk_dst_pending_confirm = 0; 2143 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2144 dst_release(old_dst); 2145 } 2146 2147 static inline void 2148 __sk_dst_reset(struct sock *sk) 2149 { 2150 __sk_dst_set(sk, NULL); 2151 } 2152 2153 static inline void 2154 sk_dst_reset(struct sock *sk) 2155 { 2156 sk_dst_set(sk, NULL); 2157 } 2158 2159 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2160 2161 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2162 2163 static inline void sk_dst_confirm(struct sock *sk) 2164 { 2165 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2166 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2167 } 2168 2169 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2170 { 2171 if (skb_get_dst_pending_confirm(skb)) { 2172 struct sock *sk = skb->sk; 2173 2174 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2175 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2176 neigh_confirm(n); 2177 } 2178 } 2179 2180 bool sk_mc_loop(struct sock *sk); 2181 2182 static inline bool sk_can_gso(const struct sock *sk) 2183 { 2184 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2185 } 2186 2187 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2188 2189 static inline void sk_gso_disable(struct sock *sk) 2190 { 2191 sk->sk_gso_disabled = 1; 2192 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2193 } 2194 2195 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2196 struct iov_iter *from, char *to, 2197 int copy, int offset) 2198 { 2199 if (skb->ip_summed == CHECKSUM_NONE) { 2200 __wsum csum = 0; 2201 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2202 return -EFAULT; 2203 skb->csum = csum_block_add(skb->csum, csum, offset); 2204 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2205 if (!copy_from_iter_full_nocache(to, copy, from)) 2206 return -EFAULT; 2207 } else if (!copy_from_iter_full(to, copy, from)) 2208 return -EFAULT; 2209 2210 return 0; 2211 } 2212 2213 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2214 struct iov_iter *from, int copy) 2215 { 2216 int err, offset = skb->len; 2217 2218 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2219 copy, offset); 2220 if (err) 2221 __skb_trim(skb, offset); 2222 2223 return err; 2224 } 2225 2226 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2227 struct sk_buff *skb, 2228 struct page *page, 2229 int off, int copy) 2230 { 2231 int err; 2232 2233 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2234 copy, skb->len); 2235 if (err) 2236 return err; 2237 2238 skb->len += copy; 2239 skb->data_len += copy; 2240 skb->truesize += copy; 2241 sk_wmem_queued_add(sk, copy); 2242 sk_mem_charge(sk, copy); 2243 return 0; 2244 } 2245 2246 /** 2247 * sk_wmem_alloc_get - returns write allocations 2248 * @sk: socket 2249 * 2250 * Return: sk_wmem_alloc minus initial offset of one 2251 */ 2252 static inline int sk_wmem_alloc_get(const struct sock *sk) 2253 { 2254 return refcount_read(&sk->sk_wmem_alloc) - 1; 2255 } 2256 2257 /** 2258 * sk_rmem_alloc_get - returns read allocations 2259 * @sk: socket 2260 * 2261 * Return: sk_rmem_alloc 2262 */ 2263 static inline int sk_rmem_alloc_get(const struct sock *sk) 2264 { 2265 return atomic_read(&sk->sk_rmem_alloc); 2266 } 2267 2268 /** 2269 * sk_has_allocations - check if allocations are outstanding 2270 * @sk: socket 2271 * 2272 * Return: true if socket has write or read allocations 2273 */ 2274 static inline bool sk_has_allocations(const struct sock *sk) 2275 { 2276 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2277 } 2278 2279 /** 2280 * skwq_has_sleeper - check if there are any waiting processes 2281 * @wq: struct socket_wq 2282 * 2283 * Return: true if socket_wq has waiting processes 2284 * 2285 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2286 * barrier call. They were added due to the race found within the tcp code. 2287 * 2288 * Consider following tcp code paths:: 2289 * 2290 * CPU1 CPU2 2291 * sys_select receive packet 2292 * ... ... 2293 * __add_wait_queue update tp->rcv_nxt 2294 * ... ... 2295 * tp->rcv_nxt check sock_def_readable 2296 * ... { 2297 * schedule rcu_read_lock(); 2298 * wq = rcu_dereference(sk->sk_wq); 2299 * if (wq && waitqueue_active(&wq->wait)) 2300 * wake_up_interruptible(&wq->wait) 2301 * ... 2302 * } 2303 * 2304 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2305 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2306 * could then endup calling schedule and sleep forever if there are no more 2307 * data on the socket. 2308 * 2309 */ 2310 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2311 { 2312 return wq && wq_has_sleeper(&wq->wait); 2313 } 2314 2315 /** 2316 * sock_poll_wait - place memory barrier behind the poll_wait call. 2317 * @filp: file 2318 * @sock: socket to wait on 2319 * @p: poll_table 2320 * 2321 * See the comments in the wq_has_sleeper function. 2322 */ 2323 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2324 poll_table *p) 2325 { 2326 if (!poll_does_not_wait(p)) { 2327 poll_wait(filp, &sock->wq.wait, p); 2328 /* We need to be sure we are in sync with the 2329 * socket flags modification. 2330 * 2331 * This memory barrier is paired in the wq_has_sleeper. 2332 */ 2333 smp_mb(); 2334 } 2335 } 2336 2337 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2338 { 2339 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2340 u32 txhash = READ_ONCE(sk->sk_txhash); 2341 2342 if (txhash) { 2343 skb->l4_hash = 1; 2344 skb->hash = txhash; 2345 } 2346 } 2347 2348 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2349 2350 /* 2351 * Queue a received datagram if it will fit. Stream and sequenced 2352 * protocols can't normally use this as they need to fit buffers in 2353 * and play with them. 2354 * 2355 * Inlined as it's very short and called for pretty much every 2356 * packet ever received. 2357 */ 2358 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2359 { 2360 skb_orphan(skb); 2361 skb->sk = sk; 2362 skb->destructor = sock_rfree; 2363 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2364 sk_mem_charge(sk, skb->truesize); 2365 } 2366 2367 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2368 { 2369 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2370 skb_orphan(skb); 2371 skb->destructor = sock_efree; 2372 skb->sk = sk; 2373 return true; 2374 } 2375 return false; 2376 } 2377 2378 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2379 { 2380 if (skb->destructor != sock_wfree) { 2381 skb_orphan(skb); 2382 return; 2383 } 2384 skb->slow_gro = 1; 2385 } 2386 2387 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2388 unsigned long expires); 2389 2390 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2391 2392 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2393 2394 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2395 struct sk_buff *skb, unsigned int flags, 2396 void (*destructor)(struct sock *sk, 2397 struct sk_buff *skb)); 2398 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2399 2400 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2401 enum skb_drop_reason *reason); 2402 2403 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2404 { 2405 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2406 } 2407 2408 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2409 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2410 2411 /* 2412 * Recover an error report and clear atomically 2413 */ 2414 2415 static inline int sock_error(struct sock *sk) 2416 { 2417 int err; 2418 2419 /* Avoid an atomic operation for the common case. 2420 * This is racy since another cpu/thread can change sk_err under us. 2421 */ 2422 if (likely(data_race(!sk->sk_err))) 2423 return 0; 2424 2425 err = xchg(&sk->sk_err, 0); 2426 return -err; 2427 } 2428 2429 void sk_error_report(struct sock *sk); 2430 2431 static inline unsigned long sock_wspace(struct sock *sk) 2432 { 2433 int amt = 0; 2434 2435 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2436 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2437 if (amt < 0) 2438 amt = 0; 2439 } 2440 return amt; 2441 } 2442 2443 /* Note: 2444 * We use sk->sk_wq_raw, from contexts knowing this 2445 * pointer is not NULL and cannot disappear/change. 2446 */ 2447 static inline void sk_set_bit(int nr, struct sock *sk) 2448 { 2449 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2450 !sock_flag(sk, SOCK_FASYNC)) 2451 return; 2452 2453 set_bit(nr, &sk->sk_wq_raw->flags); 2454 } 2455 2456 static inline void sk_clear_bit(int nr, struct sock *sk) 2457 { 2458 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2459 !sock_flag(sk, SOCK_FASYNC)) 2460 return; 2461 2462 clear_bit(nr, &sk->sk_wq_raw->flags); 2463 } 2464 2465 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2466 { 2467 if (sock_flag(sk, SOCK_FASYNC)) { 2468 rcu_read_lock(); 2469 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2470 rcu_read_unlock(); 2471 } 2472 } 2473 2474 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2475 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2476 * Note: for send buffers, TCP works better if we can build two skbs at 2477 * minimum. 2478 */ 2479 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2480 2481 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2482 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2483 2484 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2485 { 2486 u32 val; 2487 2488 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2489 return; 2490 2491 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2492 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2493 2494 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2495 } 2496 2497 /** 2498 * sk_page_frag - return an appropriate page_frag 2499 * @sk: socket 2500 * 2501 * Use the per task page_frag instead of the per socket one for 2502 * optimization when we know that we're in process context and own 2503 * everything that's associated with %current. 2504 * 2505 * Both direct reclaim and page faults can nest inside other 2506 * socket operations and end up recursing into sk_page_frag() 2507 * while it's already in use: explicitly avoid task page_frag 2508 * usage if the caller is potentially doing any of them. 2509 * This assumes that page fault handlers use the GFP_NOFS flags. 2510 * 2511 * Return: a per task page_frag if context allows that, 2512 * otherwise a per socket one. 2513 */ 2514 static inline struct page_frag *sk_page_frag(struct sock *sk) 2515 { 2516 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) == 2517 (__GFP_DIRECT_RECLAIM | __GFP_FS)) 2518 return ¤t->task_frag; 2519 2520 return &sk->sk_frag; 2521 } 2522 2523 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2524 2525 /* 2526 * Default write policy as shown to user space via poll/select/SIGIO 2527 */ 2528 static inline bool sock_writeable(const struct sock *sk) 2529 { 2530 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2531 } 2532 2533 static inline gfp_t gfp_any(void) 2534 { 2535 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2536 } 2537 2538 static inline gfp_t gfp_memcg_charge(void) 2539 { 2540 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2541 } 2542 2543 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2544 { 2545 return noblock ? 0 : sk->sk_rcvtimeo; 2546 } 2547 2548 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2549 { 2550 return noblock ? 0 : sk->sk_sndtimeo; 2551 } 2552 2553 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2554 { 2555 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2556 2557 return v ?: 1; 2558 } 2559 2560 /* Alas, with timeout socket operations are not restartable. 2561 * Compare this to poll(). 2562 */ 2563 static inline int sock_intr_errno(long timeo) 2564 { 2565 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2566 } 2567 2568 struct sock_skb_cb { 2569 u32 dropcount; 2570 }; 2571 2572 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2573 * using skb->cb[] would keep using it directly and utilize its 2574 * alignement guarantee. 2575 */ 2576 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2577 sizeof(struct sock_skb_cb))) 2578 2579 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2580 SOCK_SKB_CB_OFFSET)) 2581 2582 #define sock_skb_cb_check_size(size) \ 2583 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2584 2585 static inline void 2586 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2587 { 2588 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2589 atomic_read(&sk->sk_drops) : 0; 2590 } 2591 2592 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2593 { 2594 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2595 2596 atomic_add(segs, &sk->sk_drops); 2597 } 2598 2599 static inline ktime_t sock_read_timestamp(struct sock *sk) 2600 { 2601 #if BITS_PER_LONG==32 2602 unsigned int seq; 2603 ktime_t kt; 2604 2605 do { 2606 seq = read_seqbegin(&sk->sk_stamp_seq); 2607 kt = sk->sk_stamp; 2608 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2609 2610 return kt; 2611 #else 2612 return READ_ONCE(sk->sk_stamp); 2613 #endif 2614 } 2615 2616 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2617 { 2618 #if BITS_PER_LONG==32 2619 write_seqlock(&sk->sk_stamp_seq); 2620 sk->sk_stamp = kt; 2621 write_sequnlock(&sk->sk_stamp_seq); 2622 #else 2623 WRITE_ONCE(sk->sk_stamp, kt); 2624 #endif 2625 } 2626 2627 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2628 struct sk_buff *skb); 2629 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2630 struct sk_buff *skb); 2631 2632 static inline void 2633 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2634 { 2635 ktime_t kt = skb->tstamp; 2636 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2637 2638 /* 2639 * generate control messages if 2640 * - receive time stamping in software requested 2641 * - software time stamp available and wanted 2642 * - hardware time stamps available and wanted 2643 */ 2644 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2645 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2646 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2647 (hwtstamps->hwtstamp && 2648 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2649 __sock_recv_timestamp(msg, sk, skb); 2650 else 2651 sock_write_timestamp(sk, kt); 2652 2653 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2654 __sock_recv_wifi_status(msg, sk, skb); 2655 } 2656 2657 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2658 struct sk_buff *skb); 2659 2660 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2661 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2662 struct sk_buff *skb) 2663 { 2664 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2665 (1UL << SOCK_RCVTSTAMP) | \ 2666 (1UL << SOCK_RCVMARK)) 2667 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2668 SOF_TIMESTAMPING_RAW_HARDWARE) 2669 2670 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2671 __sock_recv_cmsgs(msg, sk, skb); 2672 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2673 sock_write_timestamp(sk, skb->tstamp); 2674 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2675 sock_write_timestamp(sk, 0); 2676 } 2677 2678 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2679 2680 /** 2681 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2682 * @sk: socket sending this packet 2683 * @tsflags: timestamping flags to use 2684 * @tx_flags: completed with instructions for time stamping 2685 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2686 * 2687 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2688 */ 2689 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2690 __u8 *tx_flags, __u32 *tskey) 2691 { 2692 if (unlikely(tsflags)) { 2693 __sock_tx_timestamp(tsflags, tx_flags); 2694 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2695 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2696 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2697 } 2698 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2699 *tx_flags |= SKBTX_WIFI_STATUS; 2700 } 2701 2702 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2703 __u8 *tx_flags) 2704 { 2705 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2706 } 2707 2708 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2709 { 2710 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2711 &skb_shinfo(skb)->tskey); 2712 } 2713 2714 static inline bool sk_is_tcp(const struct sock *sk) 2715 { 2716 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2717 } 2718 2719 /** 2720 * sk_eat_skb - Release a skb if it is no longer needed 2721 * @sk: socket to eat this skb from 2722 * @skb: socket buffer to eat 2723 * 2724 * This routine must be called with interrupts disabled or with the socket 2725 * locked so that the sk_buff queue operation is ok. 2726 */ 2727 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2728 { 2729 __skb_unlink(skb, &sk->sk_receive_queue); 2730 __kfree_skb(skb); 2731 } 2732 2733 static inline bool 2734 skb_sk_is_prefetched(struct sk_buff *skb) 2735 { 2736 #ifdef CONFIG_INET 2737 return skb->destructor == sock_pfree; 2738 #else 2739 return false; 2740 #endif /* CONFIG_INET */ 2741 } 2742 2743 /* This helper checks if a socket is a full socket, 2744 * ie _not_ a timewait or request socket. 2745 */ 2746 static inline bool sk_fullsock(const struct sock *sk) 2747 { 2748 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2749 } 2750 2751 static inline bool 2752 sk_is_refcounted(struct sock *sk) 2753 { 2754 /* Only full sockets have sk->sk_flags. */ 2755 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2756 } 2757 2758 /** 2759 * skb_steal_sock - steal a socket from an sk_buff 2760 * @skb: sk_buff to steal the socket from 2761 * @refcounted: is set to true if the socket is reference-counted 2762 */ 2763 static inline struct sock * 2764 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2765 { 2766 if (skb->sk) { 2767 struct sock *sk = skb->sk; 2768 2769 *refcounted = true; 2770 if (skb_sk_is_prefetched(skb)) 2771 *refcounted = sk_is_refcounted(sk); 2772 skb->destructor = NULL; 2773 skb->sk = NULL; 2774 return sk; 2775 } 2776 *refcounted = false; 2777 return NULL; 2778 } 2779 2780 /* Checks if this SKB belongs to an HW offloaded socket 2781 * and whether any SW fallbacks are required based on dev. 2782 * Check decrypted mark in case skb_orphan() cleared socket. 2783 */ 2784 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2785 struct net_device *dev) 2786 { 2787 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2788 struct sock *sk = skb->sk; 2789 2790 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2791 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2792 #ifdef CONFIG_TLS_DEVICE 2793 } else if (unlikely(skb->decrypted)) { 2794 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2795 kfree_skb(skb); 2796 skb = NULL; 2797 #endif 2798 } 2799 #endif 2800 2801 return skb; 2802 } 2803 2804 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2805 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2806 */ 2807 static inline bool sk_listener(const struct sock *sk) 2808 { 2809 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2810 } 2811 2812 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2813 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2814 int type); 2815 2816 bool sk_ns_capable(const struct sock *sk, 2817 struct user_namespace *user_ns, int cap); 2818 bool sk_capable(const struct sock *sk, int cap); 2819 bool sk_net_capable(const struct sock *sk, int cap); 2820 2821 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2822 2823 /* Take into consideration the size of the struct sk_buff overhead in the 2824 * determination of these values, since that is non-constant across 2825 * platforms. This makes socket queueing behavior and performance 2826 * not depend upon such differences. 2827 */ 2828 #define _SK_MEM_PACKETS 256 2829 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2830 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2831 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2832 2833 extern __u32 sysctl_wmem_max; 2834 extern __u32 sysctl_rmem_max; 2835 2836 extern int sysctl_tstamp_allow_data; 2837 extern int sysctl_optmem_max; 2838 2839 extern __u32 sysctl_wmem_default; 2840 extern __u32 sysctl_rmem_default; 2841 2842 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2843 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2844 2845 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2846 { 2847 /* Does this proto have per netns sysctl_wmem ? */ 2848 if (proto->sysctl_wmem_offset) 2849 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2850 2851 return *proto->sysctl_wmem; 2852 } 2853 2854 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2855 { 2856 /* Does this proto have per netns sysctl_rmem ? */ 2857 if (proto->sysctl_rmem_offset) 2858 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2859 2860 return *proto->sysctl_rmem; 2861 } 2862 2863 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2864 * Some wifi drivers need to tweak it to get more chunks. 2865 * They can use this helper from their ndo_start_xmit() 2866 */ 2867 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2868 { 2869 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2870 return; 2871 WRITE_ONCE(sk->sk_pacing_shift, val); 2872 } 2873 2874 /* if a socket is bound to a device, check that the given device 2875 * index is either the same or that the socket is bound to an L3 2876 * master device and the given device index is also enslaved to 2877 * that L3 master 2878 */ 2879 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2880 { 2881 int mdif; 2882 2883 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2884 return true; 2885 2886 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2887 if (mdif && mdif == sk->sk_bound_dev_if) 2888 return true; 2889 2890 return false; 2891 } 2892 2893 void sock_def_readable(struct sock *sk); 2894 2895 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2896 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2897 int sock_set_timestamping(struct sock *sk, int optname, 2898 struct so_timestamping timestamping); 2899 2900 void sock_enable_timestamps(struct sock *sk); 2901 void sock_no_linger(struct sock *sk); 2902 void sock_set_keepalive(struct sock *sk); 2903 void sock_set_priority(struct sock *sk, u32 priority); 2904 void sock_set_rcvbuf(struct sock *sk, int val); 2905 void sock_set_mark(struct sock *sk, u32 val); 2906 void sock_set_reuseaddr(struct sock *sk); 2907 void sock_set_reuseport(struct sock *sk); 2908 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2909 2910 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2911 2912 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2913 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2914 sockptr_t optval, int optlen, bool old_timeval); 2915 2916 static inline bool sk_is_readable(struct sock *sk) 2917 { 2918 if (sk->sk_prot->sock_is_readable) 2919 return sk->sk_prot->sock_is_readable(sk); 2920 return false; 2921 } 2922 #endif /* _SOCK_H */ 2923