1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/filter.h> 60 #include <linux/rculist_nulls.h> 61 #include <linux/poll.h> 62 #include <linux/sockptr.h> 63 #include <linux/indirect_call_wrapper.h> 64 #include <linux/atomic.h> 65 #include <linux/refcount.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned 165 * address on 64bit arches : cf INET_MATCH() 166 */ 167 union { 168 __addrpair skc_addrpair; 169 struct { 170 __be32 skc_daddr; 171 __be32 skc_rcv_saddr; 172 }; 173 }; 174 union { 175 unsigned int skc_hash; 176 __u16 skc_u16hashes[2]; 177 }; 178 /* skc_dport && skc_num must be grouped as well */ 179 union { 180 __portpair skc_portpair; 181 struct { 182 __be16 skc_dport; 183 __u16 skc_num; 184 }; 185 }; 186 187 unsigned short skc_family; 188 volatile unsigned char skc_state; 189 unsigned char skc_reuse:4; 190 unsigned char skc_reuseport:1; 191 unsigned char skc_ipv6only:1; 192 unsigned char skc_net_refcnt:1; 193 int skc_bound_dev_if; 194 union { 195 struct hlist_node skc_bind_node; 196 struct hlist_node skc_portaddr_node; 197 }; 198 struct proto *skc_prot; 199 possible_net_t skc_net; 200 201 #if IS_ENABLED(CONFIG_IPV6) 202 struct in6_addr skc_v6_daddr; 203 struct in6_addr skc_v6_rcv_saddr; 204 #endif 205 206 atomic64_t skc_cookie; 207 208 /* following fields are padding to force 209 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 210 * assuming IPV6 is enabled. We use this padding differently 211 * for different kind of 'sockets' 212 */ 213 union { 214 unsigned long skc_flags; 215 struct sock *skc_listener; /* request_sock */ 216 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 217 }; 218 /* 219 * fields between dontcopy_begin/dontcopy_end 220 * are not copied in sock_copy() 221 */ 222 /* private: */ 223 int skc_dontcopy_begin[0]; 224 /* public: */ 225 union { 226 struct hlist_node skc_node; 227 struct hlist_nulls_node skc_nulls_node; 228 }; 229 unsigned short skc_tx_queue_mapping; 230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 231 unsigned short skc_rx_queue_mapping; 232 #endif 233 union { 234 int skc_incoming_cpu; 235 u32 skc_rcv_wnd; 236 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 237 }; 238 239 refcount_t skc_refcnt; 240 /* private: */ 241 int skc_dontcopy_end[0]; 242 union { 243 u32 skc_rxhash; 244 u32 skc_window_clamp; 245 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 246 }; 247 /* public: */ 248 }; 249 250 struct bpf_local_storage; 251 252 /** 253 * struct sock - network layer representation of sockets 254 * @__sk_common: shared layout with inet_timewait_sock 255 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 256 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 257 * @sk_lock: synchronizer 258 * @sk_kern_sock: True if sock is using kernel lock classes 259 * @sk_rcvbuf: size of receive buffer in bytes 260 * @sk_wq: sock wait queue and async head 261 * @sk_rx_dst: receive input route used by early demux 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_rx_skb_cache: cache copy of recently accessed RX skb 266 * @sk_receive_queue: incoming packets 267 * @sk_wmem_alloc: transmit queue bytes committed 268 * @sk_tsq_flags: TCP Small Queues flags 269 * @sk_write_queue: Packet sending queue 270 * @sk_omem_alloc: "o" is "option" or "other" 271 * @sk_wmem_queued: persistent queue size 272 * @sk_forward_alloc: space allocated forward 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK) 286 * @sk_route_forced_caps: static, forced route capabilities 287 * (set in tcp_init_sock()) 288 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 289 * @sk_gso_max_size: Maximum GSO segment size to build 290 * @sk_gso_max_segs: Maximum number of GSO segments 291 * @sk_pacing_shift: scaling factor for TCP Small Queues 292 * @sk_lingertime: %SO_LINGER l_linger setting 293 * @sk_backlog: always used with the per-socket spinlock held 294 * @sk_callback_lock: used with the callbacks in the end of this struct 295 * @sk_error_queue: rarely used 296 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 297 * IPV6_ADDRFORM for instance) 298 * @sk_err: last error 299 * @sk_err_soft: errors that don't cause failure but are the cause of a 300 * persistent failure not just 'timed out' 301 * @sk_drops: raw/udp drops counter 302 * @sk_ack_backlog: current listen backlog 303 * @sk_max_ack_backlog: listen backlog set in listen() 304 * @sk_uid: user id of owner 305 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 306 * @sk_busy_poll_budget: napi processing budget when busypolling 307 * @sk_priority: %SO_PRIORITY setting 308 * @sk_type: socket type (%SOCK_STREAM, etc) 309 * @sk_protocol: which protocol this socket belongs in this network family 310 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 311 * @sk_peer_pid: &struct pid for this socket's peer 312 * @sk_peer_cred: %SO_PEERCRED setting 313 * @sk_rcvlowat: %SO_RCVLOWAT setting 314 * @sk_rcvtimeo: %SO_RCVTIMEO setting 315 * @sk_sndtimeo: %SO_SNDTIMEO setting 316 * @sk_txhash: computed flow hash for use on transmit 317 * @sk_filter: socket filtering instructions 318 * @sk_timer: sock cleanup timer 319 * @sk_stamp: time stamp of last packet received 320 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 321 * @sk_tsflags: SO_TIMESTAMPING flags 322 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 323 * for timestamping 324 * @sk_tskey: counter to disambiguate concurrent tstamp requests 325 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 326 * @sk_socket: Identd and reporting IO signals 327 * @sk_user_data: RPC layer private data 328 * @sk_frag: cached page frag 329 * @sk_peek_off: current peek_offset value 330 * @sk_send_head: front of stuff to transmit 331 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 332 * @sk_tx_skb_cache: cache copy of recently accessed TX skb 333 * @sk_security: used by security modules 334 * @sk_mark: generic packet mark 335 * @sk_cgrp_data: cgroup data for this cgroup 336 * @sk_memcg: this socket's memory cgroup association 337 * @sk_write_pending: a write to stream socket waits to start 338 * @sk_state_change: callback to indicate change in the state of the sock 339 * @sk_data_ready: callback to indicate there is data to be processed 340 * @sk_write_space: callback to indicate there is bf sending space available 341 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 342 * @sk_backlog_rcv: callback to process the backlog 343 * @sk_validate_xmit_skb: ptr to an optional validate function 344 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 345 * @sk_reuseport_cb: reuseport group container 346 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 347 * @sk_rcu: used during RCU grace period 348 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 349 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 350 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 351 * @sk_txtime_unused: unused txtime flags 352 */ 353 struct sock { 354 /* 355 * Now struct inet_timewait_sock also uses sock_common, so please just 356 * don't add nothing before this first member (__sk_common) --acme 357 */ 358 struct sock_common __sk_common; 359 #define sk_node __sk_common.skc_node 360 #define sk_nulls_node __sk_common.skc_nulls_node 361 #define sk_refcnt __sk_common.skc_refcnt 362 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 363 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 364 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 365 #endif 366 367 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 368 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 369 #define sk_hash __sk_common.skc_hash 370 #define sk_portpair __sk_common.skc_portpair 371 #define sk_num __sk_common.skc_num 372 #define sk_dport __sk_common.skc_dport 373 #define sk_addrpair __sk_common.skc_addrpair 374 #define sk_daddr __sk_common.skc_daddr 375 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 376 #define sk_family __sk_common.skc_family 377 #define sk_state __sk_common.skc_state 378 #define sk_reuse __sk_common.skc_reuse 379 #define sk_reuseport __sk_common.skc_reuseport 380 #define sk_ipv6only __sk_common.skc_ipv6only 381 #define sk_net_refcnt __sk_common.skc_net_refcnt 382 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 383 #define sk_bind_node __sk_common.skc_bind_node 384 #define sk_prot __sk_common.skc_prot 385 #define sk_net __sk_common.skc_net 386 #define sk_v6_daddr __sk_common.skc_v6_daddr 387 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 388 #define sk_cookie __sk_common.skc_cookie 389 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 390 #define sk_flags __sk_common.skc_flags 391 #define sk_rxhash __sk_common.skc_rxhash 392 393 socket_lock_t sk_lock; 394 atomic_t sk_drops; 395 int sk_rcvlowat; 396 struct sk_buff_head sk_error_queue; 397 struct sk_buff *sk_rx_skb_cache; 398 struct sk_buff_head sk_receive_queue; 399 /* 400 * The backlog queue is special, it is always used with 401 * the per-socket spinlock held and requires low latency 402 * access. Therefore we special case it's implementation. 403 * Note : rmem_alloc is in this structure to fill a hole 404 * on 64bit arches, not because its logically part of 405 * backlog. 406 */ 407 struct { 408 atomic_t rmem_alloc; 409 int len; 410 struct sk_buff *head; 411 struct sk_buff *tail; 412 } sk_backlog; 413 #define sk_rmem_alloc sk_backlog.rmem_alloc 414 415 int sk_forward_alloc; 416 #ifdef CONFIG_NET_RX_BUSY_POLL 417 unsigned int sk_ll_usec; 418 /* ===== mostly read cache line ===== */ 419 unsigned int sk_napi_id; 420 #endif 421 int sk_rcvbuf; 422 423 struct sk_filter __rcu *sk_filter; 424 union { 425 struct socket_wq __rcu *sk_wq; 426 /* private: */ 427 struct socket_wq *sk_wq_raw; 428 /* public: */ 429 }; 430 #ifdef CONFIG_XFRM 431 struct xfrm_policy __rcu *sk_policy[2]; 432 #endif 433 struct dst_entry *sk_rx_dst; 434 struct dst_entry __rcu *sk_dst_cache; 435 atomic_t sk_omem_alloc; 436 int sk_sndbuf; 437 438 /* ===== cache line for TX ===== */ 439 int sk_wmem_queued; 440 refcount_t sk_wmem_alloc; 441 unsigned long sk_tsq_flags; 442 union { 443 struct sk_buff *sk_send_head; 444 struct rb_root tcp_rtx_queue; 445 }; 446 struct sk_buff *sk_tx_skb_cache; 447 struct sk_buff_head sk_write_queue; 448 __s32 sk_peek_off; 449 int sk_write_pending; 450 __u32 sk_dst_pending_confirm; 451 u32 sk_pacing_status; /* see enum sk_pacing */ 452 long sk_sndtimeo; 453 struct timer_list sk_timer; 454 __u32 sk_priority; 455 __u32 sk_mark; 456 unsigned long sk_pacing_rate; /* bytes per second */ 457 unsigned long sk_max_pacing_rate; 458 struct page_frag sk_frag; 459 netdev_features_t sk_route_caps; 460 netdev_features_t sk_route_nocaps; 461 netdev_features_t sk_route_forced_caps; 462 int sk_gso_type; 463 unsigned int sk_gso_max_size; 464 gfp_t sk_allocation; 465 __u32 sk_txhash; 466 467 /* 468 * Because of non atomicity rules, all 469 * changes are protected by socket lock. 470 */ 471 u8 sk_padding : 1, 472 sk_kern_sock : 1, 473 sk_no_check_tx : 1, 474 sk_no_check_rx : 1, 475 sk_userlocks : 4; 476 u8 sk_pacing_shift; 477 u16 sk_type; 478 u16 sk_protocol; 479 u16 sk_gso_max_segs; 480 unsigned long sk_lingertime; 481 struct proto *sk_prot_creator; 482 rwlock_t sk_callback_lock; 483 int sk_err, 484 sk_err_soft; 485 u32 sk_ack_backlog; 486 u32 sk_max_ack_backlog; 487 kuid_t sk_uid; 488 #ifdef CONFIG_NET_RX_BUSY_POLL 489 u8 sk_prefer_busy_poll; 490 u16 sk_busy_poll_budget; 491 #endif 492 spinlock_t sk_peer_lock; 493 struct pid *sk_peer_pid; 494 const struct cred *sk_peer_cred; 495 496 long sk_rcvtimeo; 497 ktime_t sk_stamp; 498 #if BITS_PER_LONG==32 499 seqlock_t sk_stamp_seq; 500 #endif 501 u16 sk_tsflags; 502 int sk_bind_phc; 503 u8 sk_shutdown; 504 u32 sk_tskey; 505 atomic_t sk_zckey; 506 507 u8 sk_clockid; 508 u8 sk_txtime_deadline_mode : 1, 509 sk_txtime_report_errors : 1, 510 sk_txtime_unused : 6; 511 512 struct socket *sk_socket; 513 void *sk_user_data; 514 #ifdef CONFIG_SECURITY 515 void *sk_security; 516 #endif 517 struct sock_cgroup_data sk_cgrp_data; 518 struct mem_cgroup *sk_memcg; 519 void (*sk_state_change)(struct sock *sk); 520 void (*sk_data_ready)(struct sock *sk); 521 void (*sk_write_space)(struct sock *sk); 522 void (*sk_error_report)(struct sock *sk); 523 int (*sk_backlog_rcv)(struct sock *sk, 524 struct sk_buff *skb); 525 #ifdef CONFIG_SOCK_VALIDATE_XMIT 526 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 527 struct net_device *dev, 528 struct sk_buff *skb); 529 #endif 530 void (*sk_destruct)(struct sock *sk); 531 struct sock_reuseport __rcu *sk_reuseport_cb; 532 #ifdef CONFIG_BPF_SYSCALL 533 struct bpf_local_storage __rcu *sk_bpf_storage; 534 #endif 535 struct rcu_head sk_rcu; 536 }; 537 538 enum sk_pacing { 539 SK_PACING_NONE = 0, 540 SK_PACING_NEEDED = 1, 541 SK_PACING_FQ = 2, 542 }; 543 544 /* Pointer stored in sk_user_data might not be suitable for copying 545 * when cloning the socket. For instance, it can point to a reference 546 * counted object. sk_user_data bottom bit is set if pointer must not 547 * be copied. 548 */ 549 #define SK_USER_DATA_NOCOPY 1UL 550 #define SK_USER_DATA_BPF 2UL /* Managed by BPF */ 551 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF) 552 553 /** 554 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 555 * @sk: socket 556 */ 557 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 558 { 559 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 560 } 561 562 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 563 564 #define rcu_dereference_sk_user_data(sk) \ 565 ({ \ 566 void *__tmp = rcu_dereference(__sk_user_data((sk))); \ 567 (void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK); \ 568 }) 569 #define rcu_assign_sk_user_data(sk, ptr) \ 570 ({ \ 571 uintptr_t __tmp = (uintptr_t)(ptr); \ 572 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 573 rcu_assign_pointer(__sk_user_data((sk)), __tmp); \ 574 }) 575 #define rcu_assign_sk_user_data_nocopy(sk, ptr) \ 576 ({ \ 577 uintptr_t __tmp = (uintptr_t)(ptr); \ 578 WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK); \ 579 rcu_assign_pointer(__sk_user_data((sk)), \ 580 __tmp | SK_USER_DATA_NOCOPY); \ 581 }) 582 583 /* 584 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 585 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 586 * on a socket means that the socket will reuse everybody else's port 587 * without looking at the other's sk_reuse value. 588 */ 589 590 #define SK_NO_REUSE 0 591 #define SK_CAN_REUSE 1 592 #define SK_FORCE_REUSE 2 593 594 int sk_set_peek_off(struct sock *sk, int val); 595 596 static inline int sk_peek_offset(struct sock *sk, int flags) 597 { 598 if (unlikely(flags & MSG_PEEK)) { 599 return READ_ONCE(sk->sk_peek_off); 600 } 601 602 return 0; 603 } 604 605 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 606 { 607 s32 off = READ_ONCE(sk->sk_peek_off); 608 609 if (unlikely(off >= 0)) { 610 off = max_t(s32, off - val, 0); 611 WRITE_ONCE(sk->sk_peek_off, off); 612 } 613 } 614 615 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 616 { 617 sk_peek_offset_bwd(sk, -val); 618 } 619 620 /* 621 * Hashed lists helper routines 622 */ 623 static inline struct sock *sk_entry(const struct hlist_node *node) 624 { 625 return hlist_entry(node, struct sock, sk_node); 626 } 627 628 static inline struct sock *__sk_head(const struct hlist_head *head) 629 { 630 return hlist_entry(head->first, struct sock, sk_node); 631 } 632 633 static inline struct sock *sk_head(const struct hlist_head *head) 634 { 635 return hlist_empty(head) ? NULL : __sk_head(head); 636 } 637 638 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 639 { 640 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 641 } 642 643 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 644 { 645 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 646 } 647 648 static inline struct sock *sk_next(const struct sock *sk) 649 { 650 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 651 } 652 653 static inline struct sock *sk_nulls_next(const struct sock *sk) 654 { 655 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 656 hlist_nulls_entry(sk->sk_nulls_node.next, 657 struct sock, sk_nulls_node) : 658 NULL; 659 } 660 661 static inline bool sk_unhashed(const struct sock *sk) 662 { 663 return hlist_unhashed(&sk->sk_node); 664 } 665 666 static inline bool sk_hashed(const struct sock *sk) 667 { 668 return !sk_unhashed(sk); 669 } 670 671 static inline void sk_node_init(struct hlist_node *node) 672 { 673 node->pprev = NULL; 674 } 675 676 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 677 { 678 node->pprev = NULL; 679 } 680 681 static inline void __sk_del_node(struct sock *sk) 682 { 683 __hlist_del(&sk->sk_node); 684 } 685 686 /* NB: equivalent to hlist_del_init_rcu */ 687 static inline bool __sk_del_node_init(struct sock *sk) 688 { 689 if (sk_hashed(sk)) { 690 __sk_del_node(sk); 691 sk_node_init(&sk->sk_node); 692 return true; 693 } 694 return false; 695 } 696 697 /* Grab socket reference count. This operation is valid only 698 when sk is ALREADY grabbed f.e. it is found in hash table 699 or a list and the lookup is made under lock preventing hash table 700 modifications. 701 */ 702 703 static __always_inline void sock_hold(struct sock *sk) 704 { 705 refcount_inc(&sk->sk_refcnt); 706 } 707 708 /* Ungrab socket in the context, which assumes that socket refcnt 709 cannot hit zero, f.e. it is true in context of any socketcall. 710 */ 711 static __always_inline void __sock_put(struct sock *sk) 712 { 713 refcount_dec(&sk->sk_refcnt); 714 } 715 716 static inline bool sk_del_node_init(struct sock *sk) 717 { 718 bool rc = __sk_del_node_init(sk); 719 720 if (rc) { 721 /* paranoid for a while -acme */ 722 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 723 __sock_put(sk); 724 } 725 return rc; 726 } 727 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 728 729 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 730 { 731 if (sk_hashed(sk)) { 732 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 733 return true; 734 } 735 return false; 736 } 737 738 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 739 { 740 bool rc = __sk_nulls_del_node_init_rcu(sk); 741 742 if (rc) { 743 /* paranoid for a while -acme */ 744 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 745 __sock_put(sk); 746 } 747 return rc; 748 } 749 750 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 751 { 752 hlist_add_head(&sk->sk_node, list); 753 } 754 755 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 756 { 757 sock_hold(sk); 758 __sk_add_node(sk, list); 759 } 760 761 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 762 { 763 sock_hold(sk); 764 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 765 sk->sk_family == AF_INET6) 766 hlist_add_tail_rcu(&sk->sk_node, list); 767 else 768 hlist_add_head_rcu(&sk->sk_node, list); 769 } 770 771 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 772 { 773 sock_hold(sk); 774 hlist_add_tail_rcu(&sk->sk_node, list); 775 } 776 777 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 778 { 779 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 780 } 781 782 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 783 { 784 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 785 } 786 787 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 788 { 789 sock_hold(sk); 790 __sk_nulls_add_node_rcu(sk, list); 791 } 792 793 static inline void __sk_del_bind_node(struct sock *sk) 794 { 795 __hlist_del(&sk->sk_bind_node); 796 } 797 798 static inline void sk_add_bind_node(struct sock *sk, 799 struct hlist_head *list) 800 { 801 hlist_add_head(&sk->sk_bind_node, list); 802 } 803 804 #define sk_for_each(__sk, list) \ 805 hlist_for_each_entry(__sk, list, sk_node) 806 #define sk_for_each_rcu(__sk, list) \ 807 hlist_for_each_entry_rcu(__sk, list, sk_node) 808 #define sk_nulls_for_each(__sk, node, list) \ 809 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 810 #define sk_nulls_for_each_rcu(__sk, node, list) \ 811 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 812 #define sk_for_each_from(__sk) \ 813 hlist_for_each_entry_from(__sk, sk_node) 814 #define sk_nulls_for_each_from(__sk, node) \ 815 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 816 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 817 #define sk_for_each_safe(__sk, tmp, list) \ 818 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 819 #define sk_for_each_bound(__sk, list) \ 820 hlist_for_each_entry(__sk, list, sk_bind_node) 821 822 /** 823 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 824 * @tpos: the type * to use as a loop cursor. 825 * @pos: the &struct hlist_node to use as a loop cursor. 826 * @head: the head for your list. 827 * @offset: offset of hlist_node within the struct. 828 * 829 */ 830 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 831 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 832 pos != NULL && \ 833 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 834 pos = rcu_dereference(hlist_next_rcu(pos))) 835 836 static inline struct user_namespace *sk_user_ns(struct sock *sk) 837 { 838 /* Careful only use this in a context where these parameters 839 * can not change and must all be valid, such as recvmsg from 840 * userspace. 841 */ 842 return sk->sk_socket->file->f_cred->user_ns; 843 } 844 845 /* Sock flags */ 846 enum sock_flags { 847 SOCK_DEAD, 848 SOCK_DONE, 849 SOCK_URGINLINE, 850 SOCK_KEEPOPEN, 851 SOCK_LINGER, 852 SOCK_DESTROY, 853 SOCK_BROADCAST, 854 SOCK_TIMESTAMP, 855 SOCK_ZAPPED, 856 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 857 SOCK_DBG, /* %SO_DEBUG setting */ 858 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 859 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 860 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 861 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 862 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 863 SOCK_FASYNC, /* fasync() active */ 864 SOCK_RXQ_OVFL, 865 SOCK_ZEROCOPY, /* buffers from userspace */ 866 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 867 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 868 * Will use last 4 bytes of packet sent from 869 * user-space instead. 870 */ 871 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 872 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 873 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 874 SOCK_TXTIME, 875 SOCK_XDP, /* XDP is attached */ 876 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 877 }; 878 879 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 880 881 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk) 882 { 883 nsk->sk_flags = osk->sk_flags; 884 } 885 886 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 887 { 888 __set_bit(flag, &sk->sk_flags); 889 } 890 891 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 892 { 893 __clear_bit(flag, &sk->sk_flags); 894 } 895 896 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 897 int valbool) 898 { 899 if (valbool) 900 sock_set_flag(sk, bit); 901 else 902 sock_reset_flag(sk, bit); 903 } 904 905 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 906 { 907 return test_bit(flag, &sk->sk_flags); 908 } 909 910 #ifdef CONFIG_NET 911 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 912 static inline int sk_memalloc_socks(void) 913 { 914 return static_branch_unlikely(&memalloc_socks_key); 915 } 916 917 void __receive_sock(struct file *file); 918 #else 919 920 static inline int sk_memalloc_socks(void) 921 { 922 return 0; 923 } 924 925 static inline void __receive_sock(struct file *file) 926 { } 927 #endif 928 929 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 930 { 931 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 932 } 933 934 static inline void sk_acceptq_removed(struct sock *sk) 935 { 936 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 937 } 938 939 static inline void sk_acceptq_added(struct sock *sk) 940 { 941 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 942 } 943 944 /* Note: If you think the test should be: 945 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 946 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 947 */ 948 static inline bool sk_acceptq_is_full(const struct sock *sk) 949 { 950 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 951 } 952 953 /* 954 * Compute minimal free write space needed to queue new packets. 955 */ 956 static inline int sk_stream_min_wspace(const struct sock *sk) 957 { 958 return READ_ONCE(sk->sk_wmem_queued) >> 1; 959 } 960 961 static inline int sk_stream_wspace(const struct sock *sk) 962 { 963 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 964 } 965 966 static inline void sk_wmem_queued_add(struct sock *sk, int val) 967 { 968 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 969 } 970 971 void sk_stream_write_space(struct sock *sk); 972 973 /* OOB backlog add */ 974 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 975 { 976 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 977 skb_dst_force(skb); 978 979 if (!sk->sk_backlog.tail) 980 WRITE_ONCE(sk->sk_backlog.head, skb); 981 else 982 sk->sk_backlog.tail->next = skb; 983 984 WRITE_ONCE(sk->sk_backlog.tail, skb); 985 skb->next = NULL; 986 } 987 988 /* 989 * Take into account size of receive queue and backlog queue 990 * Do not take into account this skb truesize, 991 * to allow even a single big packet to come. 992 */ 993 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 994 { 995 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 996 997 return qsize > limit; 998 } 999 1000 /* The per-socket spinlock must be held here. */ 1001 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1002 unsigned int limit) 1003 { 1004 if (sk_rcvqueues_full(sk, limit)) 1005 return -ENOBUFS; 1006 1007 /* 1008 * If the skb was allocated from pfmemalloc reserves, only 1009 * allow SOCK_MEMALLOC sockets to use it as this socket is 1010 * helping free memory 1011 */ 1012 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1013 return -ENOMEM; 1014 1015 __sk_add_backlog(sk, skb); 1016 sk->sk_backlog.len += skb->truesize; 1017 return 0; 1018 } 1019 1020 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1021 1022 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1023 { 1024 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1025 return __sk_backlog_rcv(sk, skb); 1026 1027 return sk->sk_backlog_rcv(sk, skb); 1028 } 1029 1030 static inline void sk_incoming_cpu_update(struct sock *sk) 1031 { 1032 int cpu = raw_smp_processor_id(); 1033 1034 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1035 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1036 } 1037 1038 static inline void sock_rps_record_flow_hash(__u32 hash) 1039 { 1040 #ifdef CONFIG_RPS 1041 struct rps_sock_flow_table *sock_flow_table; 1042 1043 rcu_read_lock(); 1044 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1045 rps_record_sock_flow(sock_flow_table, hash); 1046 rcu_read_unlock(); 1047 #endif 1048 } 1049 1050 static inline void sock_rps_record_flow(const struct sock *sk) 1051 { 1052 #ifdef CONFIG_RPS 1053 if (static_branch_unlikely(&rfs_needed)) { 1054 /* Reading sk->sk_rxhash might incur an expensive cache line 1055 * miss. 1056 * 1057 * TCP_ESTABLISHED does cover almost all states where RFS 1058 * might be useful, and is cheaper [1] than testing : 1059 * IPv4: inet_sk(sk)->inet_daddr 1060 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1061 * OR an additional socket flag 1062 * [1] : sk_state and sk_prot are in the same cache line. 1063 */ 1064 if (sk->sk_state == TCP_ESTABLISHED) 1065 sock_rps_record_flow_hash(sk->sk_rxhash); 1066 } 1067 #endif 1068 } 1069 1070 static inline void sock_rps_save_rxhash(struct sock *sk, 1071 const struct sk_buff *skb) 1072 { 1073 #ifdef CONFIG_RPS 1074 if (unlikely(sk->sk_rxhash != skb->hash)) 1075 sk->sk_rxhash = skb->hash; 1076 #endif 1077 } 1078 1079 static inline void sock_rps_reset_rxhash(struct sock *sk) 1080 { 1081 #ifdef CONFIG_RPS 1082 sk->sk_rxhash = 0; 1083 #endif 1084 } 1085 1086 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1087 ({ int __rc; \ 1088 release_sock(__sk); \ 1089 __rc = __condition; \ 1090 if (!__rc) { \ 1091 *(__timeo) = wait_woken(__wait, \ 1092 TASK_INTERRUPTIBLE, \ 1093 *(__timeo)); \ 1094 } \ 1095 sched_annotate_sleep(); \ 1096 lock_sock(__sk); \ 1097 __rc = __condition; \ 1098 __rc; \ 1099 }) 1100 1101 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1102 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1103 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1104 int sk_stream_error(struct sock *sk, int flags, int err); 1105 void sk_stream_kill_queues(struct sock *sk); 1106 void sk_set_memalloc(struct sock *sk); 1107 void sk_clear_memalloc(struct sock *sk); 1108 1109 void __sk_flush_backlog(struct sock *sk); 1110 1111 static inline bool sk_flush_backlog(struct sock *sk) 1112 { 1113 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1114 __sk_flush_backlog(sk); 1115 return true; 1116 } 1117 return false; 1118 } 1119 1120 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1121 1122 struct request_sock_ops; 1123 struct timewait_sock_ops; 1124 struct inet_hashinfo; 1125 struct raw_hashinfo; 1126 struct smc_hashinfo; 1127 struct module; 1128 struct sk_psock; 1129 1130 /* 1131 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1132 * un-modified. Special care is taken when initializing object to zero. 1133 */ 1134 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1135 { 1136 if (offsetof(struct sock, sk_node.next) != 0) 1137 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1138 memset(&sk->sk_node.pprev, 0, 1139 size - offsetof(struct sock, sk_node.pprev)); 1140 } 1141 1142 /* Networking protocol blocks we attach to sockets. 1143 * socket layer -> transport layer interface 1144 */ 1145 struct proto { 1146 void (*close)(struct sock *sk, 1147 long timeout); 1148 int (*pre_connect)(struct sock *sk, 1149 struct sockaddr *uaddr, 1150 int addr_len); 1151 int (*connect)(struct sock *sk, 1152 struct sockaddr *uaddr, 1153 int addr_len); 1154 int (*disconnect)(struct sock *sk, int flags); 1155 1156 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1157 bool kern); 1158 1159 int (*ioctl)(struct sock *sk, int cmd, 1160 unsigned long arg); 1161 int (*init)(struct sock *sk); 1162 void (*destroy)(struct sock *sk); 1163 void (*shutdown)(struct sock *sk, int how); 1164 int (*setsockopt)(struct sock *sk, int level, 1165 int optname, sockptr_t optval, 1166 unsigned int optlen); 1167 int (*getsockopt)(struct sock *sk, int level, 1168 int optname, char __user *optval, 1169 int __user *option); 1170 void (*keepalive)(struct sock *sk, int valbool); 1171 #ifdef CONFIG_COMPAT 1172 int (*compat_ioctl)(struct sock *sk, 1173 unsigned int cmd, unsigned long arg); 1174 #endif 1175 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1176 size_t len); 1177 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1178 size_t len, int noblock, int flags, 1179 int *addr_len); 1180 int (*sendpage)(struct sock *sk, struct page *page, 1181 int offset, size_t size, int flags); 1182 int (*bind)(struct sock *sk, 1183 struct sockaddr *addr, int addr_len); 1184 int (*bind_add)(struct sock *sk, 1185 struct sockaddr *addr, int addr_len); 1186 1187 int (*backlog_rcv) (struct sock *sk, 1188 struct sk_buff *skb); 1189 bool (*bpf_bypass_getsockopt)(int level, 1190 int optname); 1191 1192 void (*release_cb)(struct sock *sk); 1193 1194 /* Keeping track of sk's, looking them up, and port selection methods. */ 1195 int (*hash)(struct sock *sk); 1196 void (*unhash)(struct sock *sk); 1197 void (*rehash)(struct sock *sk); 1198 int (*get_port)(struct sock *sk, unsigned short snum); 1199 #ifdef CONFIG_BPF_SYSCALL 1200 int (*psock_update_sk_prot)(struct sock *sk, 1201 struct sk_psock *psock, 1202 bool restore); 1203 #endif 1204 1205 /* Keeping track of sockets in use */ 1206 #ifdef CONFIG_PROC_FS 1207 unsigned int inuse_idx; 1208 #endif 1209 1210 bool (*stream_memory_free)(const struct sock *sk, int wake); 1211 bool (*stream_memory_read)(const struct sock *sk); 1212 /* Memory pressure */ 1213 void (*enter_memory_pressure)(struct sock *sk); 1214 void (*leave_memory_pressure)(struct sock *sk); 1215 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1216 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1217 /* 1218 * Pressure flag: try to collapse. 1219 * Technical note: it is used by multiple contexts non atomically. 1220 * All the __sk_mem_schedule() is of this nature: accounting 1221 * is strict, actions are advisory and have some latency. 1222 */ 1223 unsigned long *memory_pressure; 1224 long *sysctl_mem; 1225 1226 int *sysctl_wmem; 1227 int *sysctl_rmem; 1228 u32 sysctl_wmem_offset; 1229 u32 sysctl_rmem_offset; 1230 1231 int max_header; 1232 bool no_autobind; 1233 1234 struct kmem_cache *slab; 1235 unsigned int obj_size; 1236 slab_flags_t slab_flags; 1237 unsigned int useroffset; /* Usercopy region offset */ 1238 unsigned int usersize; /* Usercopy region size */ 1239 1240 struct percpu_counter *orphan_count; 1241 1242 struct request_sock_ops *rsk_prot; 1243 struct timewait_sock_ops *twsk_prot; 1244 1245 union { 1246 struct inet_hashinfo *hashinfo; 1247 struct udp_table *udp_table; 1248 struct raw_hashinfo *raw_hash; 1249 struct smc_hashinfo *smc_hash; 1250 } h; 1251 1252 struct module *owner; 1253 1254 char name[32]; 1255 1256 struct list_head node; 1257 #ifdef SOCK_REFCNT_DEBUG 1258 atomic_t socks; 1259 #endif 1260 int (*diag_destroy)(struct sock *sk, int err); 1261 } __randomize_layout; 1262 1263 int proto_register(struct proto *prot, int alloc_slab); 1264 void proto_unregister(struct proto *prot); 1265 int sock_load_diag_module(int family, int protocol); 1266 1267 #ifdef SOCK_REFCNT_DEBUG 1268 static inline void sk_refcnt_debug_inc(struct sock *sk) 1269 { 1270 atomic_inc(&sk->sk_prot->socks); 1271 } 1272 1273 static inline void sk_refcnt_debug_dec(struct sock *sk) 1274 { 1275 atomic_dec(&sk->sk_prot->socks); 1276 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1277 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1278 } 1279 1280 static inline void sk_refcnt_debug_release(const struct sock *sk) 1281 { 1282 if (refcount_read(&sk->sk_refcnt) != 1) 1283 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1284 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1285 } 1286 #else /* SOCK_REFCNT_DEBUG */ 1287 #define sk_refcnt_debug_inc(sk) do { } while (0) 1288 #define sk_refcnt_debug_dec(sk) do { } while (0) 1289 #define sk_refcnt_debug_release(sk) do { } while (0) 1290 #endif /* SOCK_REFCNT_DEBUG */ 1291 1292 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1293 1294 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1295 { 1296 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1297 return false; 1298 1299 #ifdef CONFIG_INET 1300 return sk->sk_prot->stream_memory_free ? 1301 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free, 1302 tcp_stream_memory_free, 1303 sk, wake) : true; 1304 #else 1305 return sk->sk_prot->stream_memory_free ? 1306 sk->sk_prot->stream_memory_free(sk, wake) : true; 1307 #endif 1308 } 1309 1310 static inline bool sk_stream_memory_free(const struct sock *sk) 1311 { 1312 return __sk_stream_memory_free(sk, 0); 1313 } 1314 1315 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1316 { 1317 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1318 __sk_stream_memory_free(sk, wake); 1319 } 1320 1321 static inline bool sk_stream_is_writeable(const struct sock *sk) 1322 { 1323 return __sk_stream_is_writeable(sk, 0); 1324 } 1325 1326 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1327 struct cgroup *ancestor) 1328 { 1329 #ifdef CONFIG_SOCK_CGROUP_DATA 1330 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1331 ancestor); 1332 #else 1333 return -ENOTSUPP; 1334 #endif 1335 } 1336 1337 static inline bool sk_has_memory_pressure(const struct sock *sk) 1338 { 1339 return sk->sk_prot->memory_pressure != NULL; 1340 } 1341 1342 static inline bool sk_under_memory_pressure(const struct sock *sk) 1343 { 1344 if (!sk->sk_prot->memory_pressure) 1345 return false; 1346 1347 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1348 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1349 return true; 1350 1351 return !!*sk->sk_prot->memory_pressure; 1352 } 1353 1354 static inline long 1355 sk_memory_allocated(const struct sock *sk) 1356 { 1357 return atomic_long_read(sk->sk_prot->memory_allocated); 1358 } 1359 1360 static inline long 1361 sk_memory_allocated_add(struct sock *sk, int amt) 1362 { 1363 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated); 1364 } 1365 1366 static inline void 1367 sk_memory_allocated_sub(struct sock *sk, int amt) 1368 { 1369 atomic_long_sub(amt, sk->sk_prot->memory_allocated); 1370 } 1371 1372 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1373 1374 static inline void sk_sockets_allocated_dec(struct sock *sk) 1375 { 1376 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1377 SK_ALLOC_PERCPU_COUNTER_BATCH); 1378 } 1379 1380 static inline void sk_sockets_allocated_inc(struct sock *sk) 1381 { 1382 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1383 SK_ALLOC_PERCPU_COUNTER_BATCH); 1384 } 1385 1386 static inline u64 1387 sk_sockets_allocated_read_positive(struct sock *sk) 1388 { 1389 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1390 } 1391 1392 static inline int 1393 proto_sockets_allocated_sum_positive(struct proto *prot) 1394 { 1395 return percpu_counter_sum_positive(prot->sockets_allocated); 1396 } 1397 1398 static inline long 1399 proto_memory_allocated(struct proto *prot) 1400 { 1401 return atomic_long_read(prot->memory_allocated); 1402 } 1403 1404 static inline bool 1405 proto_memory_pressure(struct proto *prot) 1406 { 1407 if (!prot->memory_pressure) 1408 return false; 1409 return !!*prot->memory_pressure; 1410 } 1411 1412 1413 #ifdef CONFIG_PROC_FS 1414 /* Called with local bh disabled */ 1415 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc); 1416 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1417 int sock_inuse_get(struct net *net); 1418 #else 1419 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot, 1420 int inc) 1421 { 1422 } 1423 #endif 1424 1425 1426 /* With per-bucket locks this operation is not-atomic, so that 1427 * this version is not worse. 1428 */ 1429 static inline int __sk_prot_rehash(struct sock *sk) 1430 { 1431 sk->sk_prot->unhash(sk); 1432 return sk->sk_prot->hash(sk); 1433 } 1434 1435 /* About 10 seconds */ 1436 #define SOCK_DESTROY_TIME (10*HZ) 1437 1438 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1439 #define PROT_SOCK 1024 1440 1441 #define SHUTDOWN_MASK 3 1442 #define RCV_SHUTDOWN 1 1443 #define SEND_SHUTDOWN 2 1444 1445 #define SOCK_BINDADDR_LOCK 4 1446 #define SOCK_BINDPORT_LOCK 8 1447 1448 struct socket_alloc { 1449 struct socket socket; 1450 struct inode vfs_inode; 1451 }; 1452 1453 static inline struct socket *SOCKET_I(struct inode *inode) 1454 { 1455 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1456 } 1457 1458 static inline struct inode *SOCK_INODE(struct socket *socket) 1459 { 1460 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1461 } 1462 1463 /* 1464 * Functions for memory accounting 1465 */ 1466 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1467 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1468 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1469 void __sk_mem_reclaim(struct sock *sk, int amount); 1470 1471 /* We used to have PAGE_SIZE here, but systems with 64KB pages 1472 * do not necessarily have 16x time more memory than 4KB ones. 1473 */ 1474 #define SK_MEM_QUANTUM 4096 1475 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM) 1476 #define SK_MEM_SEND 0 1477 #define SK_MEM_RECV 1 1478 1479 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */ 1480 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1481 { 1482 long val = sk->sk_prot->sysctl_mem[index]; 1483 1484 #if PAGE_SIZE > SK_MEM_QUANTUM 1485 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT; 1486 #elif PAGE_SIZE < SK_MEM_QUANTUM 1487 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT; 1488 #endif 1489 return val; 1490 } 1491 1492 static inline int sk_mem_pages(int amt) 1493 { 1494 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT; 1495 } 1496 1497 static inline bool sk_has_account(struct sock *sk) 1498 { 1499 /* return true if protocol supports memory accounting */ 1500 return !!sk->sk_prot->memory_allocated; 1501 } 1502 1503 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1504 { 1505 if (!sk_has_account(sk)) 1506 return true; 1507 return size <= sk->sk_forward_alloc || 1508 __sk_mem_schedule(sk, size, SK_MEM_SEND); 1509 } 1510 1511 static inline bool 1512 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1513 { 1514 if (!sk_has_account(sk)) 1515 return true; 1516 return size <= sk->sk_forward_alloc || 1517 __sk_mem_schedule(sk, size, SK_MEM_RECV) || 1518 skb_pfmemalloc(skb); 1519 } 1520 1521 static inline void sk_mem_reclaim(struct sock *sk) 1522 { 1523 if (!sk_has_account(sk)) 1524 return; 1525 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM) 1526 __sk_mem_reclaim(sk, sk->sk_forward_alloc); 1527 } 1528 1529 static inline void sk_mem_reclaim_partial(struct sock *sk) 1530 { 1531 if (!sk_has_account(sk)) 1532 return; 1533 if (sk->sk_forward_alloc > SK_MEM_QUANTUM) 1534 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1); 1535 } 1536 1537 static inline void sk_mem_charge(struct sock *sk, int size) 1538 { 1539 if (!sk_has_account(sk)) 1540 return; 1541 sk->sk_forward_alloc -= size; 1542 } 1543 1544 static inline void sk_mem_uncharge(struct sock *sk, int size) 1545 { 1546 if (!sk_has_account(sk)) 1547 return; 1548 sk->sk_forward_alloc += size; 1549 1550 /* Avoid a possible overflow. 1551 * TCP send queues can make this happen, if sk_mem_reclaim() 1552 * is not called and more than 2 GBytes are released at once. 1553 * 1554 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is 1555 * no need to hold that much forward allocation anyway. 1556 */ 1557 if (unlikely(sk->sk_forward_alloc >= 1 << 21)) 1558 __sk_mem_reclaim(sk, 1 << 20); 1559 } 1560 1561 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key); 1562 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb) 1563 { 1564 sk_wmem_queued_add(sk, -skb->truesize); 1565 sk_mem_uncharge(sk, skb->truesize); 1566 if (static_branch_unlikely(&tcp_tx_skb_cache_key) && 1567 !sk->sk_tx_skb_cache && !skb_cloned(skb)) { 1568 skb_ext_reset(skb); 1569 skb_zcopy_clear(skb, true); 1570 sk->sk_tx_skb_cache = skb; 1571 return; 1572 } 1573 __kfree_skb(skb); 1574 } 1575 1576 static inline void sock_release_ownership(struct sock *sk) 1577 { 1578 if (sk->sk_lock.owned) { 1579 sk->sk_lock.owned = 0; 1580 1581 /* The sk_lock has mutex_unlock() semantics: */ 1582 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1583 } 1584 } 1585 1586 /* 1587 * Macro so as to not evaluate some arguments when 1588 * lockdep is not enabled. 1589 * 1590 * Mark both the sk_lock and the sk_lock.slock as a 1591 * per-address-family lock class. 1592 */ 1593 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1594 do { \ 1595 sk->sk_lock.owned = 0; \ 1596 init_waitqueue_head(&sk->sk_lock.wq); \ 1597 spin_lock_init(&(sk)->sk_lock.slock); \ 1598 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1599 sizeof((sk)->sk_lock)); \ 1600 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1601 (skey), (sname)); \ 1602 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1603 } while (0) 1604 1605 static inline bool lockdep_sock_is_held(const struct sock *sk) 1606 { 1607 return lockdep_is_held(&sk->sk_lock) || 1608 lockdep_is_held(&sk->sk_lock.slock); 1609 } 1610 1611 void lock_sock_nested(struct sock *sk, int subclass); 1612 1613 static inline void lock_sock(struct sock *sk) 1614 { 1615 lock_sock_nested(sk, 0); 1616 } 1617 1618 void __lock_sock(struct sock *sk); 1619 void __release_sock(struct sock *sk); 1620 void release_sock(struct sock *sk); 1621 1622 /* BH context may only use the following locking interface. */ 1623 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1624 #define bh_lock_sock_nested(__sk) \ 1625 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1626 SINGLE_DEPTH_NESTING) 1627 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1628 1629 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1630 1631 /** 1632 * lock_sock_fast - fast version of lock_sock 1633 * @sk: socket 1634 * 1635 * This version should be used for very small section, where process wont block 1636 * return false if fast path is taken: 1637 * 1638 * sk_lock.slock locked, owned = 0, BH disabled 1639 * 1640 * return true if slow path is taken: 1641 * 1642 * sk_lock.slock unlocked, owned = 1, BH enabled 1643 */ 1644 static inline bool lock_sock_fast(struct sock *sk) 1645 { 1646 /* The sk_lock has mutex_lock() semantics here. */ 1647 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1648 1649 return __lock_sock_fast(sk); 1650 } 1651 1652 /* fast socket lock variant for caller already holding a [different] socket lock */ 1653 static inline bool lock_sock_fast_nested(struct sock *sk) 1654 { 1655 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1656 1657 return __lock_sock_fast(sk); 1658 } 1659 1660 /** 1661 * unlock_sock_fast - complement of lock_sock_fast 1662 * @sk: socket 1663 * @slow: slow mode 1664 * 1665 * fast unlock socket for user context. 1666 * If slow mode is on, we call regular release_sock() 1667 */ 1668 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1669 __releases(&sk->sk_lock.slock) 1670 { 1671 if (slow) { 1672 release_sock(sk); 1673 __release(&sk->sk_lock.slock); 1674 } else { 1675 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1676 spin_unlock_bh(&sk->sk_lock.slock); 1677 } 1678 } 1679 1680 /* Used by processes to "lock" a socket state, so that 1681 * interrupts and bottom half handlers won't change it 1682 * from under us. It essentially blocks any incoming 1683 * packets, so that we won't get any new data or any 1684 * packets that change the state of the socket. 1685 * 1686 * While locked, BH processing will add new packets to 1687 * the backlog queue. This queue is processed by the 1688 * owner of the socket lock right before it is released. 1689 * 1690 * Since ~2.3.5 it is also exclusive sleep lock serializing 1691 * accesses from user process context. 1692 */ 1693 1694 static inline void sock_owned_by_me(const struct sock *sk) 1695 { 1696 #ifdef CONFIG_LOCKDEP 1697 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1698 #endif 1699 } 1700 1701 static inline bool sock_owned_by_user(const struct sock *sk) 1702 { 1703 sock_owned_by_me(sk); 1704 return sk->sk_lock.owned; 1705 } 1706 1707 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1708 { 1709 return sk->sk_lock.owned; 1710 } 1711 1712 /* no reclassification while locks are held */ 1713 static inline bool sock_allow_reclassification(const struct sock *csk) 1714 { 1715 struct sock *sk = (struct sock *)csk; 1716 1717 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock); 1718 } 1719 1720 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1721 struct proto *prot, int kern); 1722 void sk_free(struct sock *sk); 1723 void sk_destruct(struct sock *sk); 1724 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1725 void sk_free_unlock_clone(struct sock *sk); 1726 1727 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1728 gfp_t priority); 1729 void __sock_wfree(struct sk_buff *skb); 1730 void sock_wfree(struct sk_buff *skb); 1731 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1732 gfp_t priority); 1733 void skb_orphan_partial(struct sk_buff *skb); 1734 void sock_rfree(struct sk_buff *skb); 1735 void sock_efree(struct sk_buff *skb); 1736 #ifdef CONFIG_INET 1737 void sock_edemux(struct sk_buff *skb); 1738 void sock_pfree(struct sk_buff *skb); 1739 #else 1740 #define sock_edemux sock_efree 1741 #endif 1742 1743 int sock_setsockopt(struct socket *sock, int level, int op, 1744 sockptr_t optval, unsigned int optlen); 1745 1746 int sock_getsockopt(struct socket *sock, int level, int op, 1747 char __user *optval, int __user *optlen); 1748 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1749 bool timeval, bool time32); 1750 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size, 1751 int noblock, int *errcode); 1752 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1753 unsigned long data_len, int noblock, 1754 int *errcode, int max_page_order); 1755 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1756 void sock_kfree_s(struct sock *sk, void *mem, int size); 1757 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1758 void sk_send_sigurg(struct sock *sk); 1759 1760 struct sockcm_cookie { 1761 u64 transmit_time; 1762 u32 mark; 1763 u16 tsflags; 1764 }; 1765 1766 static inline void sockcm_init(struct sockcm_cookie *sockc, 1767 const struct sock *sk) 1768 { 1769 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1770 } 1771 1772 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1773 struct sockcm_cookie *sockc); 1774 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1775 struct sockcm_cookie *sockc); 1776 1777 /* 1778 * Functions to fill in entries in struct proto_ops when a protocol 1779 * does not implement a particular function. 1780 */ 1781 int sock_no_bind(struct socket *, struct sockaddr *, int); 1782 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1783 int sock_no_socketpair(struct socket *, struct socket *); 1784 int sock_no_accept(struct socket *, struct socket *, int, bool); 1785 int sock_no_getname(struct socket *, struct sockaddr *, int); 1786 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1787 int sock_no_listen(struct socket *, int); 1788 int sock_no_shutdown(struct socket *, int); 1789 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1790 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1791 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1792 int sock_no_mmap(struct file *file, struct socket *sock, 1793 struct vm_area_struct *vma); 1794 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1795 size_t size, int flags); 1796 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1797 int offset, size_t size, int flags); 1798 1799 /* 1800 * Functions to fill in entries in struct proto_ops when a protocol 1801 * uses the inet style. 1802 */ 1803 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1804 char __user *optval, int __user *optlen); 1805 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1806 int flags); 1807 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1808 sockptr_t optval, unsigned int optlen); 1809 1810 void sk_common_release(struct sock *sk); 1811 1812 /* 1813 * Default socket callbacks and setup code 1814 */ 1815 1816 /* Initialise core socket variables */ 1817 void sock_init_data(struct socket *sock, struct sock *sk); 1818 1819 /* 1820 * Socket reference counting postulates. 1821 * 1822 * * Each user of socket SHOULD hold a reference count. 1823 * * Each access point to socket (an hash table bucket, reference from a list, 1824 * running timer, skb in flight MUST hold a reference count. 1825 * * When reference count hits 0, it means it will never increase back. 1826 * * When reference count hits 0, it means that no references from 1827 * outside exist to this socket and current process on current CPU 1828 * is last user and may/should destroy this socket. 1829 * * sk_free is called from any context: process, BH, IRQ. When 1830 * it is called, socket has no references from outside -> sk_free 1831 * may release descendant resources allocated by the socket, but 1832 * to the time when it is called, socket is NOT referenced by any 1833 * hash tables, lists etc. 1834 * * Packets, delivered from outside (from network or from another process) 1835 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1836 * when they sit in queue. Otherwise, packets will leak to hole, when 1837 * socket is looked up by one cpu and unhasing is made by another CPU. 1838 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1839 * (leak to backlog). Packet socket does all the processing inside 1840 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1841 * use separate SMP lock, so that they are prone too. 1842 */ 1843 1844 /* Ungrab socket and destroy it, if it was the last reference. */ 1845 static inline void sock_put(struct sock *sk) 1846 { 1847 if (refcount_dec_and_test(&sk->sk_refcnt)) 1848 sk_free(sk); 1849 } 1850 /* Generic version of sock_put(), dealing with all sockets 1851 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1852 */ 1853 void sock_gen_put(struct sock *sk); 1854 1855 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1856 unsigned int trim_cap, bool refcounted); 1857 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1858 const int nested) 1859 { 1860 return __sk_receive_skb(sk, skb, nested, 1, true); 1861 } 1862 1863 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1864 { 1865 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1866 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1867 return; 1868 sk->sk_tx_queue_mapping = tx_queue; 1869 } 1870 1871 #define NO_QUEUE_MAPPING USHRT_MAX 1872 1873 static inline void sk_tx_queue_clear(struct sock *sk) 1874 { 1875 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1876 } 1877 1878 static inline int sk_tx_queue_get(const struct sock *sk) 1879 { 1880 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1881 return sk->sk_tx_queue_mapping; 1882 1883 return -1; 1884 } 1885 1886 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1887 { 1888 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1889 if (skb_rx_queue_recorded(skb)) { 1890 u16 rx_queue = skb_get_rx_queue(skb); 1891 1892 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING)) 1893 return; 1894 1895 sk->sk_rx_queue_mapping = rx_queue; 1896 } 1897 #endif 1898 } 1899 1900 static inline void sk_rx_queue_clear(struct sock *sk) 1901 { 1902 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1903 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING; 1904 #endif 1905 } 1906 1907 static inline int sk_rx_queue_get(const struct sock *sk) 1908 { 1909 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1910 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING) 1911 return sk->sk_rx_queue_mapping; 1912 #endif 1913 1914 return -1; 1915 } 1916 1917 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 1918 { 1919 sk->sk_socket = sock; 1920 } 1921 1922 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 1923 { 1924 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 1925 return &rcu_dereference_raw(sk->sk_wq)->wait; 1926 } 1927 /* Detach socket from process context. 1928 * Announce socket dead, detach it from wait queue and inode. 1929 * Note that parent inode held reference count on this struct sock, 1930 * we do not release it in this function, because protocol 1931 * probably wants some additional cleanups or even continuing 1932 * to work with this socket (TCP). 1933 */ 1934 static inline void sock_orphan(struct sock *sk) 1935 { 1936 write_lock_bh(&sk->sk_callback_lock); 1937 sock_set_flag(sk, SOCK_DEAD); 1938 sk_set_socket(sk, NULL); 1939 sk->sk_wq = NULL; 1940 write_unlock_bh(&sk->sk_callback_lock); 1941 } 1942 1943 static inline void sock_graft(struct sock *sk, struct socket *parent) 1944 { 1945 WARN_ON(parent->sk); 1946 write_lock_bh(&sk->sk_callback_lock); 1947 rcu_assign_pointer(sk->sk_wq, &parent->wq); 1948 parent->sk = sk; 1949 sk_set_socket(sk, parent); 1950 sk->sk_uid = SOCK_INODE(parent)->i_uid; 1951 security_sock_graft(sk, parent); 1952 write_unlock_bh(&sk->sk_callback_lock); 1953 } 1954 1955 kuid_t sock_i_uid(struct sock *sk); 1956 unsigned long sock_i_ino(struct sock *sk); 1957 1958 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 1959 { 1960 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 1961 } 1962 1963 static inline u32 net_tx_rndhash(void) 1964 { 1965 u32 v = prandom_u32(); 1966 1967 return v ?: 1; 1968 } 1969 1970 static inline void sk_set_txhash(struct sock *sk) 1971 { 1972 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 1973 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 1974 } 1975 1976 static inline bool sk_rethink_txhash(struct sock *sk) 1977 { 1978 if (sk->sk_txhash) { 1979 sk_set_txhash(sk); 1980 return true; 1981 } 1982 return false; 1983 } 1984 1985 static inline struct dst_entry * 1986 __sk_dst_get(struct sock *sk) 1987 { 1988 return rcu_dereference_check(sk->sk_dst_cache, 1989 lockdep_sock_is_held(sk)); 1990 } 1991 1992 static inline struct dst_entry * 1993 sk_dst_get(struct sock *sk) 1994 { 1995 struct dst_entry *dst; 1996 1997 rcu_read_lock(); 1998 dst = rcu_dereference(sk->sk_dst_cache); 1999 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2000 dst = NULL; 2001 rcu_read_unlock(); 2002 return dst; 2003 } 2004 2005 static inline void __dst_negative_advice(struct sock *sk) 2006 { 2007 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2008 2009 if (dst && dst->ops->negative_advice) { 2010 ndst = dst->ops->negative_advice(dst); 2011 2012 if (ndst != dst) { 2013 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2014 sk_tx_queue_clear(sk); 2015 sk->sk_dst_pending_confirm = 0; 2016 } 2017 } 2018 } 2019 2020 static inline void dst_negative_advice(struct sock *sk) 2021 { 2022 sk_rethink_txhash(sk); 2023 __dst_negative_advice(sk); 2024 } 2025 2026 static inline void 2027 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2028 { 2029 struct dst_entry *old_dst; 2030 2031 sk_tx_queue_clear(sk); 2032 sk->sk_dst_pending_confirm = 0; 2033 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2034 lockdep_sock_is_held(sk)); 2035 rcu_assign_pointer(sk->sk_dst_cache, dst); 2036 dst_release(old_dst); 2037 } 2038 2039 static inline void 2040 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2041 { 2042 struct dst_entry *old_dst; 2043 2044 sk_tx_queue_clear(sk); 2045 sk->sk_dst_pending_confirm = 0; 2046 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2047 dst_release(old_dst); 2048 } 2049 2050 static inline void 2051 __sk_dst_reset(struct sock *sk) 2052 { 2053 __sk_dst_set(sk, NULL); 2054 } 2055 2056 static inline void 2057 sk_dst_reset(struct sock *sk) 2058 { 2059 sk_dst_set(sk, NULL); 2060 } 2061 2062 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2063 2064 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2065 2066 static inline void sk_dst_confirm(struct sock *sk) 2067 { 2068 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2069 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2070 } 2071 2072 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2073 { 2074 if (skb_get_dst_pending_confirm(skb)) { 2075 struct sock *sk = skb->sk; 2076 unsigned long now = jiffies; 2077 2078 /* avoid dirtying neighbour */ 2079 if (READ_ONCE(n->confirmed) != now) 2080 WRITE_ONCE(n->confirmed, now); 2081 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2082 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2083 } 2084 } 2085 2086 bool sk_mc_loop(struct sock *sk); 2087 2088 static inline bool sk_can_gso(const struct sock *sk) 2089 { 2090 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2091 } 2092 2093 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2094 2095 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags) 2096 { 2097 sk->sk_route_nocaps |= flags; 2098 sk->sk_route_caps &= ~flags; 2099 } 2100 2101 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2102 struct iov_iter *from, char *to, 2103 int copy, int offset) 2104 { 2105 if (skb->ip_summed == CHECKSUM_NONE) { 2106 __wsum csum = 0; 2107 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2108 return -EFAULT; 2109 skb->csum = csum_block_add(skb->csum, csum, offset); 2110 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2111 if (!copy_from_iter_full_nocache(to, copy, from)) 2112 return -EFAULT; 2113 } else if (!copy_from_iter_full(to, copy, from)) 2114 return -EFAULT; 2115 2116 return 0; 2117 } 2118 2119 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2120 struct iov_iter *from, int copy) 2121 { 2122 int err, offset = skb->len; 2123 2124 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2125 copy, offset); 2126 if (err) 2127 __skb_trim(skb, offset); 2128 2129 return err; 2130 } 2131 2132 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2133 struct sk_buff *skb, 2134 struct page *page, 2135 int off, int copy) 2136 { 2137 int err; 2138 2139 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2140 copy, skb->len); 2141 if (err) 2142 return err; 2143 2144 skb->len += copy; 2145 skb->data_len += copy; 2146 skb->truesize += copy; 2147 sk_wmem_queued_add(sk, copy); 2148 sk_mem_charge(sk, copy); 2149 return 0; 2150 } 2151 2152 /** 2153 * sk_wmem_alloc_get - returns write allocations 2154 * @sk: socket 2155 * 2156 * Return: sk_wmem_alloc minus initial offset of one 2157 */ 2158 static inline int sk_wmem_alloc_get(const struct sock *sk) 2159 { 2160 return refcount_read(&sk->sk_wmem_alloc) - 1; 2161 } 2162 2163 /** 2164 * sk_rmem_alloc_get - returns read allocations 2165 * @sk: socket 2166 * 2167 * Return: sk_rmem_alloc 2168 */ 2169 static inline int sk_rmem_alloc_get(const struct sock *sk) 2170 { 2171 return atomic_read(&sk->sk_rmem_alloc); 2172 } 2173 2174 /** 2175 * sk_has_allocations - check if allocations are outstanding 2176 * @sk: socket 2177 * 2178 * Return: true if socket has write or read allocations 2179 */ 2180 static inline bool sk_has_allocations(const struct sock *sk) 2181 { 2182 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2183 } 2184 2185 /** 2186 * skwq_has_sleeper - check if there are any waiting processes 2187 * @wq: struct socket_wq 2188 * 2189 * Return: true if socket_wq has waiting processes 2190 * 2191 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2192 * barrier call. They were added due to the race found within the tcp code. 2193 * 2194 * Consider following tcp code paths:: 2195 * 2196 * CPU1 CPU2 2197 * sys_select receive packet 2198 * ... ... 2199 * __add_wait_queue update tp->rcv_nxt 2200 * ... ... 2201 * tp->rcv_nxt check sock_def_readable 2202 * ... { 2203 * schedule rcu_read_lock(); 2204 * wq = rcu_dereference(sk->sk_wq); 2205 * if (wq && waitqueue_active(&wq->wait)) 2206 * wake_up_interruptible(&wq->wait) 2207 * ... 2208 * } 2209 * 2210 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2211 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2212 * could then endup calling schedule and sleep forever if there are no more 2213 * data on the socket. 2214 * 2215 */ 2216 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2217 { 2218 return wq && wq_has_sleeper(&wq->wait); 2219 } 2220 2221 /** 2222 * sock_poll_wait - place memory barrier behind the poll_wait call. 2223 * @filp: file 2224 * @sock: socket to wait on 2225 * @p: poll_table 2226 * 2227 * See the comments in the wq_has_sleeper function. 2228 */ 2229 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2230 poll_table *p) 2231 { 2232 if (!poll_does_not_wait(p)) { 2233 poll_wait(filp, &sock->wq.wait, p); 2234 /* We need to be sure we are in sync with the 2235 * socket flags modification. 2236 * 2237 * This memory barrier is paired in the wq_has_sleeper. 2238 */ 2239 smp_mb(); 2240 } 2241 } 2242 2243 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2244 { 2245 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2246 u32 txhash = READ_ONCE(sk->sk_txhash); 2247 2248 if (txhash) { 2249 skb->l4_hash = 1; 2250 skb->hash = txhash; 2251 } 2252 } 2253 2254 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2255 2256 /* 2257 * Queue a received datagram if it will fit. Stream and sequenced 2258 * protocols can't normally use this as they need to fit buffers in 2259 * and play with them. 2260 * 2261 * Inlined as it's very short and called for pretty much every 2262 * packet ever received. 2263 */ 2264 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2265 { 2266 skb_orphan(skb); 2267 skb->sk = sk; 2268 skb->destructor = sock_rfree; 2269 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2270 sk_mem_charge(sk, skb->truesize); 2271 } 2272 2273 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2274 { 2275 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2276 skb_orphan(skb); 2277 skb->destructor = sock_efree; 2278 skb->sk = sk; 2279 return true; 2280 } 2281 return false; 2282 } 2283 2284 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2285 { 2286 if (skb->destructor != sock_wfree) { 2287 skb_orphan(skb); 2288 return; 2289 } 2290 skb->slow_gro = 1; 2291 } 2292 2293 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2294 unsigned long expires); 2295 2296 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2297 2298 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2299 2300 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2301 struct sk_buff *skb, unsigned int flags, 2302 void (*destructor)(struct sock *sk, 2303 struct sk_buff *skb)); 2304 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2305 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2306 2307 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2308 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2309 2310 /* 2311 * Recover an error report and clear atomically 2312 */ 2313 2314 static inline int sock_error(struct sock *sk) 2315 { 2316 int err; 2317 2318 /* Avoid an atomic operation for the common case. 2319 * This is racy since another cpu/thread can change sk_err under us. 2320 */ 2321 if (likely(data_race(!sk->sk_err))) 2322 return 0; 2323 2324 err = xchg(&sk->sk_err, 0); 2325 return -err; 2326 } 2327 2328 void sk_error_report(struct sock *sk); 2329 2330 static inline unsigned long sock_wspace(struct sock *sk) 2331 { 2332 int amt = 0; 2333 2334 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2335 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2336 if (amt < 0) 2337 amt = 0; 2338 } 2339 return amt; 2340 } 2341 2342 /* Note: 2343 * We use sk->sk_wq_raw, from contexts knowing this 2344 * pointer is not NULL and cannot disappear/change. 2345 */ 2346 static inline void sk_set_bit(int nr, struct sock *sk) 2347 { 2348 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2349 !sock_flag(sk, SOCK_FASYNC)) 2350 return; 2351 2352 set_bit(nr, &sk->sk_wq_raw->flags); 2353 } 2354 2355 static inline void sk_clear_bit(int nr, struct sock *sk) 2356 { 2357 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2358 !sock_flag(sk, SOCK_FASYNC)) 2359 return; 2360 2361 clear_bit(nr, &sk->sk_wq_raw->flags); 2362 } 2363 2364 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2365 { 2366 if (sock_flag(sk, SOCK_FASYNC)) { 2367 rcu_read_lock(); 2368 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2369 rcu_read_unlock(); 2370 } 2371 } 2372 2373 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2374 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2375 * Note: for send buffers, TCP works better if we can build two skbs at 2376 * minimum. 2377 */ 2378 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2379 2380 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2381 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2382 2383 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2384 { 2385 u32 val; 2386 2387 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2388 return; 2389 2390 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2391 2392 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2393 } 2394 2395 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp, 2396 bool force_schedule); 2397 2398 /** 2399 * sk_page_frag - return an appropriate page_frag 2400 * @sk: socket 2401 * 2402 * Use the per task page_frag instead of the per socket one for 2403 * optimization when we know that we're in the normal context and owns 2404 * everything that's associated with %current. 2405 * 2406 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest 2407 * inside other socket operations and end up recursing into sk_page_frag() 2408 * while it's already in use. 2409 * 2410 * Return: a per task page_frag if context allows that, 2411 * otherwise a per socket one. 2412 */ 2413 static inline struct page_frag *sk_page_frag(struct sock *sk) 2414 { 2415 if (gfpflags_normal_context(sk->sk_allocation)) 2416 return ¤t->task_frag; 2417 2418 return &sk->sk_frag; 2419 } 2420 2421 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2422 2423 /* 2424 * Default write policy as shown to user space via poll/select/SIGIO 2425 */ 2426 static inline bool sock_writeable(const struct sock *sk) 2427 { 2428 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2429 } 2430 2431 static inline gfp_t gfp_any(void) 2432 { 2433 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2434 } 2435 2436 static inline gfp_t gfp_memcg_charge(void) 2437 { 2438 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2439 } 2440 2441 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2442 { 2443 return noblock ? 0 : sk->sk_rcvtimeo; 2444 } 2445 2446 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2447 { 2448 return noblock ? 0 : sk->sk_sndtimeo; 2449 } 2450 2451 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2452 { 2453 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2454 2455 return v ?: 1; 2456 } 2457 2458 /* Alas, with timeout socket operations are not restartable. 2459 * Compare this to poll(). 2460 */ 2461 static inline int sock_intr_errno(long timeo) 2462 { 2463 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2464 } 2465 2466 struct sock_skb_cb { 2467 u32 dropcount; 2468 }; 2469 2470 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2471 * using skb->cb[] would keep using it directly and utilize its 2472 * alignement guarantee. 2473 */ 2474 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2475 sizeof(struct sock_skb_cb))) 2476 2477 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2478 SOCK_SKB_CB_OFFSET)) 2479 2480 #define sock_skb_cb_check_size(size) \ 2481 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2482 2483 static inline void 2484 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2485 { 2486 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2487 atomic_read(&sk->sk_drops) : 0; 2488 } 2489 2490 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2491 { 2492 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2493 2494 atomic_add(segs, &sk->sk_drops); 2495 } 2496 2497 static inline ktime_t sock_read_timestamp(struct sock *sk) 2498 { 2499 #if BITS_PER_LONG==32 2500 unsigned int seq; 2501 ktime_t kt; 2502 2503 do { 2504 seq = read_seqbegin(&sk->sk_stamp_seq); 2505 kt = sk->sk_stamp; 2506 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2507 2508 return kt; 2509 #else 2510 return READ_ONCE(sk->sk_stamp); 2511 #endif 2512 } 2513 2514 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2515 { 2516 #if BITS_PER_LONG==32 2517 write_seqlock(&sk->sk_stamp_seq); 2518 sk->sk_stamp = kt; 2519 write_sequnlock(&sk->sk_stamp_seq); 2520 #else 2521 WRITE_ONCE(sk->sk_stamp, kt); 2522 #endif 2523 } 2524 2525 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2526 struct sk_buff *skb); 2527 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2528 struct sk_buff *skb); 2529 2530 static inline void 2531 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2532 { 2533 ktime_t kt = skb->tstamp; 2534 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2535 2536 /* 2537 * generate control messages if 2538 * - receive time stamping in software requested 2539 * - software time stamp available and wanted 2540 * - hardware time stamps available and wanted 2541 */ 2542 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2543 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2544 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2545 (hwtstamps->hwtstamp && 2546 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2547 __sock_recv_timestamp(msg, sk, skb); 2548 else 2549 sock_write_timestamp(sk, kt); 2550 2551 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2552 __sock_recv_wifi_status(msg, sk, skb); 2553 } 2554 2555 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2556 struct sk_buff *skb); 2557 2558 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2559 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, 2560 struct sk_buff *skb) 2561 { 2562 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \ 2563 (1UL << SOCK_RCVTSTAMP)) 2564 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2565 SOF_TIMESTAMPING_RAW_HARDWARE) 2566 2567 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY) 2568 __sock_recv_ts_and_drops(msg, sk, skb); 2569 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2570 sock_write_timestamp(sk, skb->tstamp); 2571 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2572 sock_write_timestamp(sk, 0); 2573 } 2574 2575 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2576 2577 /** 2578 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2579 * @sk: socket sending this packet 2580 * @tsflags: timestamping flags to use 2581 * @tx_flags: completed with instructions for time stamping 2582 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2583 * 2584 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2585 */ 2586 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2587 __u8 *tx_flags, __u32 *tskey) 2588 { 2589 if (unlikely(tsflags)) { 2590 __sock_tx_timestamp(tsflags, tx_flags); 2591 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2592 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2593 *tskey = sk->sk_tskey++; 2594 } 2595 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2596 *tx_flags |= SKBTX_WIFI_STATUS; 2597 } 2598 2599 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2600 __u8 *tx_flags) 2601 { 2602 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2603 } 2604 2605 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2606 { 2607 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2608 &skb_shinfo(skb)->tskey); 2609 } 2610 2611 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key); 2612 /** 2613 * sk_eat_skb - Release a skb if it is no longer needed 2614 * @sk: socket to eat this skb from 2615 * @skb: socket buffer to eat 2616 * 2617 * This routine must be called with interrupts disabled or with the socket 2618 * locked so that the sk_buff queue operation is ok. 2619 */ 2620 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2621 { 2622 __skb_unlink(skb, &sk->sk_receive_queue); 2623 if (static_branch_unlikely(&tcp_rx_skb_cache_key) && 2624 !sk->sk_rx_skb_cache) { 2625 sk->sk_rx_skb_cache = skb; 2626 skb_orphan(skb); 2627 return; 2628 } 2629 __kfree_skb(skb); 2630 } 2631 2632 static inline 2633 struct net *sock_net(const struct sock *sk) 2634 { 2635 return read_pnet(&sk->sk_net); 2636 } 2637 2638 static inline 2639 void sock_net_set(struct sock *sk, struct net *net) 2640 { 2641 write_pnet(&sk->sk_net, net); 2642 } 2643 2644 static inline bool 2645 skb_sk_is_prefetched(struct sk_buff *skb) 2646 { 2647 #ifdef CONFIG_INET 2648 return skb->destructor == sock_pfree; 2649 #else 2650 return false; 2651 #endif /* CONFIG_INET */ 2652 } 2653 2654 /* This helper checks if a socket is a full socket, 2655 * ie _not_ a timewait or request socket. 2656 */ 2657 static inline bool sk_fullsock(const struct sock *sk) 2658 { 2659 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2660 } 2661 2662 static inline bool 2663 sk_is_refcounted(struct sock *sk) 2664 { 2665 /* Only full sockets have sk->sk_flags. */ 2666 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2667 } 2668 2669 /** 2670 * skb_steal_sock - steal a socket from an sk_buff 2671 * @skb: sk_buff to steal the socket from 2672 * @refcounted: is set to true if the socket is reference-counted 2673 */ 2674 static inline struct sock * 2675 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2676 { 2677 if (skb->sk) { 2678 struct sock *sk = skb->sk; 2679 2680 *refcounted = true; 2681 if (skb_sk_is_prefetched(skb)) 2682 *refcounted = sk_is_refcounted(sk); 2683 skb->destructor = NULL; 2684 skb->sk = NULL; 2685 return sk; 2686 } 2687 *refcounted = false; 2688 return NULL; 2689 } 2690 2691 /* Checks if this SKB belongs to an HW offloaded socket 2692 * and whether any SW fallbacks are required based on dev. 2693 * Check decrypted mark in case skb_orphan() cleared socket. 2694 */ 2695 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2696 struct net_device *dev) 2697 { 2698 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2699 struct sock *sk = skb->sk; 2700 2701 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2702 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2703 #ifdef CONFIG_TLS_DEVICE 2704 } else if (unlikely(skb->decrypted)) { 2705 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2706 kfree_skb(skb); 2707 skb = NULL; 2708 #endif 2709 } 2710 #endif 2711 2712 return skb; 2713 } 2714 2715 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2716 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2717 */ 2718 static inline bool sk_listener(const struct sock *sk) 2719 { 2720 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2721 } 2722 2723 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2724 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2725 int type); 2726 2727 bool sk_ns_capable(const struct sock *sk, 2728 struct user_namespace *user_ns, int cap); 2729 bool sk_capable(const struct sock *sk, int cap); 2730 bool sk_net_capable(const struct sock *sk, int cap); 2731 2732 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2733 2734 /* Take into consideration the size of the struct sk_buff overhead in the 2735 * determination of these values, since that is non-constant across 2736 * platforms. This makes socket queueing behavior and performance 2737 * not depend upon such differences. 2738 */ 2739 #define _SK_MEM_PACKETS 256 2740 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2741 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2742 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2743 2744 extern __u32 sysctl_wmem_max; 2745 extern __u32 sysctl_rmem_max; 2746 2747 extern int sysctl_tstamp_allow_data; 2748 extern int sysctl_optmem_max; 2749 2750 extern __u32 sysctl_wmem_default; 2751 extern __u32 sysctl_rmem_default; 2752 2753 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2754 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2755 2756 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2757 { 2758 /* Does this proto have per netns sysctl_wmem ? */ 2759 if (proto->sysctl_wmem_offset) 2760 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset); 2761 2762 return *proto->sysctl_wmem; 2763 } 2764 2765 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2766 { 2767 /* Does this proto have per netns sysctl_rmem ? */ 2768 if (proto->sysctl_rmem_offset) 2769 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset); 2770 2771 return *proto->sysctl_rmem; 2772 } 2773 2774 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2775 * Some wifi drivers need to tweak it to get more chunks. 2776 * They can use this helper from their ndo_start_xmit() 2777 */ 2778 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2779 { 2780 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2781 return; 2782 WRITE_ONCE(sk->sk_pacing_shift, val); 2783 } 2784 2785 /* if a socket is bound to a device, check that the given device 2786 * index is either the same or that the socket is bound to an L3 2787 * master device and the given device index is also enslaved to 2788 * that L3 master 2789 */ 2790 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2791 { 2792 int mdif; 2793 2794 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif) 2795 return true; 2796 2797 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2798 if (mdif && mdif == sk->sk_bound_dev_if) 2799 return true; 2800 2801 return false; 2802 } 2803 2804 void sock_def_readable(struct sock *sk); 2805 2806 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2807 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2808 int sock_set_timestamping(struct sock *sk, int optname, 2809 struct so_timestamping timestamping); 2810 2811 void sock_enable_timestamps(struct sock *sk); 2812 void sock_no_linger(struct sock *sk); 2813 void sock_set_keepalive(struct sock *sk); 2814 void sock_set_priority(struct sock *sk, u32 priority); 2815 void sock_set_rcvbuf(struct sock *sk, int val); 2816 void sock_set_mark(struct sock *sk, u32 val); 2817 void sock_set_reuseaddr(struct sock *sk); 2818 void sock_set_reuseport(struct sock *sk); 2819 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2820 2821 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2822 2823 #endif /* _SOCK_H */ 2824