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