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