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 union { 165 __addrpair skc_addrpair; 166 struct { 167 __be32 skc_daddr; 168 __be32 skc_rcv_saddr; 169 }; 170 }; 171 union { 172 unsigned int skc_hash; 173 __u16 skc_u16hashes[2]; 174 }; 175 /* skc_dport && skc_num must be grouped as well */ 176 union { 177 __portpair skc_portpair; 178 struct { 179 __be16 skc_dport; 180 __u16 skc_num; 181 }; 182 }; 183 184 unsigned short skc_family; 185 volatile unsigned char skc_state; 186 unsigned char skc_reuse:4; 187 unsigned char skc_reuseport:1; 188 unsigned char skc_ipv6only:1; 189 unsigned char skc_net_refcnt:1; 190 int skc_bound_dev_if; 191 union { 192 struct hlist_node skc_bind_node; 193 struct hlist_node skc_portaddr_node; 194 }; 195 struct proto *skc_prot; 196 possible_net_t skc_net; 197 198 #if IS_ENABLED(CONFIG_IPV6) 199 struct in6_addr skc_v6_daddr; 200 struct in6_addr skc_v6_rcv_saddr; 201 #endif 202 203 atomic64_t skc_cookie; 204 205 /* following fields are padding to force 206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 207 * assuming IPV6 is enabled. We use this padding differently 208 * for different kind of 'sockets' 209 */ 210 union { 211 unsigned long skc_flags; 212 struct sock *skc_listener; /* request_sock */ 213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 214 }; 215 /* 216 * fields between dontcopy_begin/dontcopy_end 217 * are not copied in sock_copy() 218 */ 219 /* private: */ 220 int skc_dontcopy_begin[0]; 221 /* public: */ 222 union { 223 struct hlist_node skc_node; 224 struct hlist_nulls_node skc_nulls_node; 225 }; 226 unsigned short skc_tx_queue_mapping; 227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 228 unsigned short skc_rx_queue_mapping; 229 #endif 230 union { 231 int skc_incoming_cpu; 232 u32 skc_rcv_wnd; 233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 234 }; 235 236 refcount_t skc_refcnt; 237 /* private: */ 238 int skc_dontcopy_end[0]; 239 union { 240 u32 skc_rxhash; 241 u32 skc_window_clamp; 242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 243 }; 244 /* public: */ 245 }; 246 247 struct bpf_local_storage; 248 struct sk_filter; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 304 * @sk_busy_poll_budget: napi processing budget when busypolling 305 * @sk_priority: %SO_PRIORITY setting 306 * @sk_type: socket type (%SOCK_STREAM, etc) 307 * @sk_protocol: which protocol this socket belongs in this network family 308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_txrehash: enable TX hash rethink 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 322 * for timestamping 323 * @sk_tskey: counter to disambiguate concurrent tstamp requests 324 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 325 * @sk_socket: Identd and reporting IO signals 326 * @sk_user_data: RPC layer private data 327 * @sk_frag: cached page frag 328 * @sk_peek_off: current peek_offset value 329 * @sk_send_head: front of stuff to transmit 330 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 331 * @sk_security: used by security modules 332 * @sk_mark: generic packet mark 333 * @sk_cgrp_data: cgroup data for this cgroup 334 * @sk_memcg: this socket's memory cgroup association 335 * @sk_write_pending: a write to stream socket waits to start 336 * @sk_state_change: callback to indicate change in the state of the sock 337 * @sk_data_ready: callback to indicate there is data to be processed 338 * @sk_write_space: callback to indicate there is bf sending space available 339 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 340 * @sk_backlog_rcv: callback to process the backlog 341 * @sk_validate_xmit_skb: ptr to an optional validate function 342 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 343 * @sk_reuseport_cb: reuseport group container 344 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 345 * @sk_rcu: used during RCU grace period 346 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 347 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 348 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 349 * @sk_txtime_unused: unused txtime flags 350 * @ns_tracker: tracker for netns reference 351 */ 352 struct sock { 353 /* 354 * Now struct inet_timewait_sock also uses sock_common, so please just 355 * don't add nothing before this first member (__sk_common) --acme 356 */ 357 struct sock_common __sk_common; 358 #define sk_node __sk_common.skc_node 359 #define sk_nulls_node __sk_common.skc_nulls_node 360 #define sk_refcnt __sk_common.skc_refcnt 361 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 362 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 363 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 364 #endif 365 366 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 367 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 368 #define sk_hash __sk_common.skc_hash 369 #define sk_portpair __sk_common.skc_portpair 370 #define sk_num __sk_common.skc_num 371 #define sk_dport __sk_common.skc_dport 372 #define sk_addrpair __sk_common.skc_addrpair 373 #define sk_daddr __sk_common.skc_daddr 374 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 375 #define sk_family __sk_common.skc_family 376 #define sk_state __sk_common.skc_state 377 #define sk_reuse __sk_common.skc_reuse 378 #define sk_reuseport __sk_common.skc_reuseport 379 #define sk_ipv6only __sk_common.skc_ipv6only 380 #define sk_net_refcnt __sk_common.skc_net_refcnt 381 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 382 #define sk_bind_node __sk_common.skc_bind_node 383 #define sk_prot __sk_common.skc_prot 384 #define sk_net __sk_common.skc_net 385 #define sk_v6_daddr __sk_common.skc_v6_daddr 386 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 387 #define sk_cookie __sk_common.skc_cookie 388 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 389 #define sk_flags __sk_common.skc_flags 390 #define sk_rxhash __sk_common.skc_rxhash 391 392 /* early demux fields */ 393 struct dst_entry __rcu *sk_rx_dst; 394 int sk_rx_dst_ifindex; 395 u32 sk_rx_dst_cookie; 396 397 socket_lock_t sk_lock; 398 atomic_t sk_drops; 399 int sk_rcvlowat; 400 struct sk_buff_head sk_error_queue; 401 struct sk_buff_head sk_receive_queue; 402 /* 403 * The backlog queue is special, it is always used with 404 * the per-socket spinlock held and requires low latency 405 * access. Therefore we special case it's implementation. 406 * Note : rmem_alloc is in this structure to fill a hole 407 * on 64bit arches, not because its logically part of 408 * backlog. 409 */ 410 struct { 411 atomic_t rmem_alloc; 412 int len; 413 struct sk_buff *head; 414 struct sk_buff *tail; 415 } sk_backlog; 416 417 #define sk_rmem_alloc sk_backlog.rmem_alloc 418 419 int sk_forward_alloc; 420 u32 sk_reserved_mem; 421 #ifdef CONFIG_NET_RX_BUSY_POLL 422 unsigned int sk_ll_usec; 423 /* ===== mostly read cache line ===== */ 424 unsigned int sk_napi_id; 425 #endif 426 int sk_rcvbuf; 427 428 struct sk_filter __rcu *sk_filter; 429 union { 430 struct socket_wq __rcu *sk_wq; 431 /* private: */ 432 struct socket_wq *sk_wq_raw; 433 /* public: */ 434 }; 435 #ifdef CONFIG_XFRM 436 struct xfrm_policy __rcu *sk_policy[2]; 437 #endif 438 439 struct dst_entry __rcu *sk_dst_cache; 440 atomic_t sk_omem_alloc; 441 int sk_sndbuf; 442 443 /* ===== cache line for TX ===== */ 444 int sk_wmem_queued; 445 refcount_t sk_wmem_alloc; 446 unsigned long sk_tsq_flags; 447 union { 448 struct sk_buff *sk_send_head; 449 struct rb_root tcp_rtx_queue; 450 }; 451 struct sk_buff_head sk_write_queue; 452 __s32 sk_peek_off; 453 int sk_write_pending; 454 __u32 sk_dst_pending_confirm; 455 u32 sk_pacing_status; /* see enum sk_pacing */ 456 long sk_sndtimeo; 457 struct timer_list sk_timer; 458 __u32 sk_priority; 459 __u32 sk_mark; 460 unsigned long sk_pacing_rate; /* bytes per second */ 461 unsigned long sk_max_pacing_rate; 462 struct page_frag sk_frag; 463 netdev_features_t sk_route_caps; 464 int sk_gso_type; 465 unsigned int sk_gso_max_size; 466 gfp_t sk_allocation; 467 __u32 sk_txhash; 468 469 /* 470 * Because of non atomicity rules, all 471 * changes are protected by socket lock. 472 */ 473 u8 sk_gso_disabled : 1, 474 sk_kern_sock : 1, 475 sk_no_check_tx : 1, 476 sk_no_check_rx : 1, 477 sk_userlocks : 4; 478 u8 sk_pacing_shift; 479 u16 sk_type; 480 u16 sk_protocol; 481 u16 sk_gso_max_segs; 482 unsigned long sk_lingertime; 483 struct proto *sk_prot_creator; 484 rwlock_t sk_callback_lock; 485 int sk_err, 486 sk_err_soft; 487 u32 sk_ack_backlog; 488 u32 sk_max_ack_backlog; 489 kuid_t sk_uid; 490 u8 sk_txrehash; 491 #ifdef CONFIG_NET_RX_BUSY_POLL 492 u8 sk_prefer_busy_poll; 493 u16 sk_busy_poll_budget; 494 #endif 495 spinlock_t sk_peer_lock; 496 int sk_bind_phc; 497 struct pid *sk_peer_pid; 498 const struct cred *sk_peer_cred; 499 500 long sk_rcvtimeo; 501 ktime_t sk_stamp; 502 #if BITS_PER_LONG==32 503 seqlock_t sk_stamp_seq; 504 #endif 505 u16 sk_tsflags; 506 u8 sk_shutdown; 507 atomic_t sk_tskey; 508 atomic_t sk_zckey; 509 510 u8 sk_clockid; 511 u8 sk_txtime_deadline_mode : 1, 512 sk_txtime_report_errors : 1, 513 sk_txtime_unused : 6; 514 515 struct socket *sk_socket; 516 void *sk_user_data; 517 #ifdef CONFIG_SECURITY 518 void *sk_security; 519 #endif 520 struct sock_cgroup_data sk_cgrp_data; 521 struct mem_cgroup *sk_memcg; 522 void (*sk_state_change)(struct sock *sk); 523 void (*sk_data_ready)(struct sock *sk); 524 void (*sk_write_space)(struct sock *sk); 525 void (*sk_error_report)(struct sock *sk); 526 int (*sk_backlog_rcv)(struct sock *sk, 527 struct sk_buff *skb); 528 #ifdef CONFIG_SOCK_VALIDATE_XMIT 529 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 530 struct net_device *dev, 531 struct sk_buff *skb); 532 #endif 533 void (*sk_destruct)(struct sock *sk); 534 struct sock_reuseport __rcu *sk_reuseport_cb; 535 #ifdef CONFIG_BPF_SYSCALL 536 struct bpf_local_storage __rcu *sk_bpf_storage; 537 #endif 538 struct rcu_head sk_rcu; 539 netns_tracker ns_tracker; 540 }; 541 542 enum sk_pacing { 543 SK_PACING_NONE = 0, 544 SK_PACING_NEEDED = 1, 545 SK_PACING_FQ = 2, 546 }; 547 548 /* flag bits in sk_user_data 549 * 550 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 551 * not be suitable for copying when cloning the socket. For instance, 552 * it can point to a reference counted object. sk_user_data bottom 553 * bit is set if pointer must not be copied. 554 * 555 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 556 * managed/owned by a BPF reuseport array. This bit should be set 557 * when sk_user_data's sk is added to the bpf's reuseport_array. 558 * 559 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 560 * sk_user_data points to psock type. This bit should be set 561 * when sk_user_data is assigned to a psock object. 562 */ 563 #define SK_USER_DATA_NOCOPY 1UL 564 #define SK_USER_DATA_BPF 2UL 565 #define SK_USER_DATA_PSOCK 4UL 566 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 567 SK_USER_DATA_PSOCK) 568 569 /** 570 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 571 * @sk: socket 572 */ 573 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 574 { 575 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 576 } 577 578 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 579 580 /** 581 * __rcu_dereference_sk_user_data_with_flags - return the pointer 582 * only if argument flags all has been set in sk_user_data. Otherwise 583 * return NULL 584 * 585 * @sk: socket 586 * @flags: flag bits 587 */ 588 static inline void * 589 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 590 uintptr_t flags) 591 { 592 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 593 594 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 595 596 if ((sk_user_data & flags) == flags) 597 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 598 return NULL; 599 } 600 601 #define rcu_dereference_sk_user_data(sk) \ 602 __rcu_dereference_sk_user_data_with_flags(sk, 0) 603 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 604 ({ \ 605 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 606 __tmp2 = (uintptr_t)(flags); \ 607 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 608 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 609 rcu_assign_pointer(__sk_user_data((sk)), \ 610 __tmp1 | __tmp2); \ 611 }) 612 #define rcu_assign_sk_user_data(sk, ptr) \ 613 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 614 615 static inline 616 struct net *sock_net(const struct sock *sk) 617 { 618 return read_pnet(&sk->sk_net); 619 } 620 621 static inline 622 void sock_net_set(struct sock *sk, struct net *net) 623 { 624 write_pnet(&sk->sk_net, net); 625 } 626 627 /* 628 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 629 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 630 * on a socket means that the socket will reuse everybody else's port 631 * without looking at the other's sk_reuse value. 632 */ 633 634 #define SK_NO_REUSE 0 635 #define SK_CAN_REUSE 1 636 #define SK_FORCE_REUSE 2 637 638 int sk_set_peek_off(struct sock *sk, int val); 639 640 static inline int sk_peek_offset(const struct sock *sk, int flags) 641 { 642 if (unlikely(flags & MSG_PEEK)) { 643 return READ_ONCE(sk->sk_peek_off); 644 } 645 646 return 0; 647 } 648 649 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 650 { 651 s32 off = READ_ONCE(sk->sk_peek_off); 652 653 if (unlikely(off >= 0)) { 654 off = max_t(s32, off - val, 0); 655 WRITE_ONCE(sk->sk_peek_off, off); 656 } 657 } 658 659 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 660 { 661 sk_peek_offset_bwd(sk, -val); 662 } 663 664 /* 665 * Hashed lists helper routines 666 */ 667 static inline struct sock *sk_entry(const struct hlist_node *node) 668 { 669 return hlist_entry(node, struct sock, sk_node); 670 } 671 672 static inline struct sock *__sk_head(const struct hlist_head *head) 673 { 674 return hlist_entry(head->first, struct sock, sk_node); 675 } 676 677 static inline struct sock *sk_head(const struct hlist_head *head) 678 { 679 return hlist_empty(head) ? NULL : __sk_head(head); 680 } 681 682 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 683 { 684 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 685 } 686 687 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 688 { 689 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 690 } 691 692 static inline struct sock *sk_next(const struct sock *sk) 693 { 694 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 695 } 696 697 static inline struct sock *sk_nulls_next(const struct sock *sk) 698 { 699 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 700 hlist_nulls_entry(sk->sk_nulls_node.next, 701 struct sock, sk_nulls_node) : 702 NULL; 703 } 704 705 static inline bool sk_unhashed(const struct sock *sk) 706 { 707 return hlist_unhashed(&sk->sk_node); 708 } 709 710 static inline bool sk_hashed(const struct sock *sk) 711 { 712 return !sk_unhashed(sk); 713 } 714 715 static inline void sk_node_init(struct hlist_node *node) 716 { 717 node->pprev = NULL; 718 } 719 720 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 721 { 722 node->pprev = NULL; 723 } 724 725 static inline void __sk_del_node(struct sock *sk) 726 { 727 __hlist_del(&sk->sk_node); 728 } 729 730 /* NB: equivalent to hlist_del_init_rcu */ 731 static inline bool __sk_del_node_init(struct sock *sk) 732 { 733 if (sk_hashed(sk)) { 734 __sk_del_node(sk); 735 sk_node_init(&sk->sk_node); 736 return true; 737 } 738 return false; 739 } 740 741 /* Grab socket reference count. This operation is valid only 742 when sk is ALREADY grabbed f.e. it is found in hash table 743 or a list and the lookup is made under lock preventing hash table 744 modifications. 745 */ 746 747 static __always_inline void sock_hold(struct sock *sk) 748 { 749 refcount_inc(&sk->sk_refcnt); 750 } 751 752 /* Ungrab socket in the context, which assumes that socket refcnt 753 cannot hit zero, f.e. it is true in context of any socketcall. 754 */ 755 static __always_inline void __sock_put(struct sock *sk) 756 { 757 refcount_dec(&sk->sk_refcnt); 758 } 759 760 static inline bool sk_del_node_init(struct sock *sk) 761 { 762 bool rc = __sk_del_node_init(sk); 763 764 if (rc) { 765 /* paranoid for a while -acme */ 766 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 767 __sock_put(sk); 768 } 769 return rc; 770 } 771 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 772 773 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 774 { 775 if (sk_hashed(sk)) { 776 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 777 return true; 778 } 779 return false; 780 } 781 782 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 783 { 784 bool rc = __sk_nulls_del_node_init_rcu(sk); 785 786 if (rc) { 787 /* paranoid for a while -acme */ 788 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 789 __sock_put(sk); 790 } 791 return rc; 792 } 793 794 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 795 { 796 hlist_add_head(&sk->sk_node, list); 797 } 798 799 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 800 { 801 sock_hold(sk); 802 __sk_add_node(sk, list); 803 } 804 805 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 806 { 807 sock_hold(sk); 808 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 809 sk->sk_family == AF_INET6) 810 hlist_add_tail_rcu(&sk->sk_node, list); 811 else 812 hlist_add_head_rcu(&sk->sk_node, list); 813 } 814 815 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 816 { 817 sock_hold(sk); 818 hlist_add_tail_rcu(&sk->sk_node, list); 819 } 820 821 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 822 { 823 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 824 } 825 826 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 827 { 828 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 829 } 830 831 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 832 { 833 sock_hold(sk); 834 __sk_nulls_add_node_rcu(sk, list); 835 } 836 837 static inline void __sk_del_bind_node(struct sock *sk) 838 { 839 __hlist_del(&sk->sk_bind_node); 840 } 841 842 static inline void sk_add_bind_node(struct sock *sk, 843 struct hlist_head *list) 844 { 845 hlist_add_head(&sk->sk_bind_node, list); 846 } 847 848 #define sk_for_each(__sk, list) \ 849 hlist_for_each_entry(__sk, list, sk_node) 850 #define sk_for_each_rcu(__sk, list) \ 851 hlist_for_each_entry_rcu(__sk, list, sk_node) 852 #define sk_nulls_for_each(__sk, node, list) \ 853 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 854 #define sk_nulls_for_each_rcu(__sk, node, list) \ 855 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 856 #define sk_for_each_from(__sk) \ 857 hlist_for_each_entry_from(__sk, sk_node) 858 #define sk_nulls_for_each_from(__sk, node) \ 859 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 860 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 861 #define sk_for_each_safe(__sk, tmp, list) \ 862 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 863 #define sk_for_each_bound(__sk, list) \ 864 hlist_for_each_entry(__sk, list, sk_bind_node) 865 866 /** 867 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 868 * @tpos: the type * to use as a loop cursor. 869 * @pos: the &struct hlist_node to use as a loop cursor. 870 * @head: the head for your list. 871 * @offset: offset of hlist_node within the struct. 872 * 873 */ 874 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 875 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 876 pos != NULL && \ 877 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 878 pos = rcu_dereference(hlist_next_rcu(pos))) 879 880 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 881 { 882 /* Careful only use this in a context where these parameters 883 * can not change and must all be valid, such as recvmsg from 884 * userspace. 885 */ 886 return sk->sk_socket->file->f_cred->user_ns; 887 } 888 889 /* Sock flags */ 890 enum sock_flags { 891 SOCK_DEAD, 892 SOCK_DONE, 893 SOCK_URGINLINE, 894 SOCK_KEEPOPEN, 895 SOCK_LINGER, 896 SOCK_DESTROY, 897 SOCK_BROADCAST, 898 SOCK_TIMESTAMP, 899 SOCK_ZAPPED, 900 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 901 SOCK_DBG, /* %SO_DEBUG setting */ 902 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 903 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 904 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 905 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 906 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 907 SOCK_FASYNC, /* fasync() active */ 908 SOCK_RXQ_OVFL, 909 SOCK_ZEROCOPY, /* buffers from userspace */ 910 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 911 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 912 * Will use last 4 bytes of packet sent from 913 * user-space instead. 914 */ 915 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 916 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 917 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 918 SOCK_TXTIME, 919 SOCK_XDP, /* XDP is attached */ 920 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 921 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 922 }; 923 924 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 925 926 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 927 { 928 nsk->sk_flags = osk->sk_flags; 929 } 930 931 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 932 { 933 __set_bit(flag, &sk->sk_flags); 934 } 935 936 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 937 { 938 __clear_bit(flag, &sk->sk_flags); 939 } 940 941 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 942 int valbool) 943 { 944 if (valbool) 945 sock_set_flag(sk, bit); 946 else 947 sock_reset_flag(sk, bit); 948 } 949 950 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 951 { 952 return test_bit(flag, &sk->sk_flags); 953 } 954 955 #ifdef CONFIG_NET 956 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 957 static inline int sk_memalloc_socks(void) 958 { 959 return static_branch_unlikely(&memalloc_socks_key); 960 } 961 962 void __receive_sock(struct file *file); 963 #else 964 965 static inline int sk_memalloc_socks(void) 966 { 967 return 0; 968 } 969 970 static inline void __receive_sock(struct file *file) 971 { } 972 #endif 973 974 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 975 { 976 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 977 } 978 979 static inline void sk_acceptq_removed(struct sock *sk) 980 { 981 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 982 } 983 984 static inline void sk_acceptq_added(struct sock *sk) 985 { 986 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 987 } 988 989 /* Note: If you think the test should be: 990 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 991 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 992 */ 993 static inline bool sk_acceptq_is_full(const struct sock *sk) 994 { 995 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 996 } 997 998 /* 999 * Compute minimal free write space needed to queue new packets. 1000 */ 1001 static inline int sk_stream_min_wspace(const struct sock *sk) 1002 { 1003 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1004 } 1005 1006 static inline int sk_stream_wspace(const struct sock *sk) 1007 { 1008 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1009 } 1010 1011 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1012 { 1013 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1014 } 1015 1016 void sk_stream_write_space(struct sock *sk); 1017 1018 /* OOB backlog add */ 1019 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1020 { 1021 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1022 skb_dst_force(skb); 1023 1024 if (!sk->sk_backlog.tail) 1025 WRITE_ONCE(sk->sk_backlog.head, skb); 1026 else 1027 sk->sk_backlog.tail->next = skb; 1028 1029 WRITE_ONCE(sk->sk_backlog.tail, skb); 1030 skb->next = NULL; 1031 } 1032 1033 /* 1034 * Take into account size of receive queue and backlog queue 1035 * Do not take into account this skb truesize, 1036 * to allow even a single big packet to come. 1037 */ 1038 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1039 { 1040 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1041 1042 return qsize > limit; 1043 } 1044 1045 /* The per-socket spinlock must be held here. */ 1046 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1047 unsigned int limit) 1048 { 1049 if (sk_rcvqueues_full(sk, limit)) 1050 return -ENOBUFS; 1051 1052 /* 1053 * If the skb was allocated from pfmemalloc reserves, only 1054 * allow SOCK_MEMALLOC sockets to use it as this socket is 1055 * helping free memory 1056 */ 1057 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1058 return -ENOMEM; 1059 1060 __sk_add_backlog(sk, skb); 1061 sk->sk_backlog.len += skb->truesize; 1062 return 0; 1063 } 1064 1065 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1066 1067 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1068 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1069 1070 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1071 { 1072 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1073 return __sk_backlog_rcv(sk, skb); 1074 1075 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1076 tcp_v6_do_rcv, 1077 tcp_v4_do_rcv, 1078 sk, skb); 1079 } 1080 1081 static inline void sk_incoming_cpu_update(struct sock *sk) 1082 { 1083 int cpu = raw_smp_processor_id(); 1084 1085 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1086 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1087 } 1088 1089 static inline void sock_rps_record_flow_hash(__u32 hash) 1090 { 1091 #ifdef CONFIG_RPS 1092 struct rps_sock_flow_table *sock_flow_table; 1093 1094 rcu_read_lock(); 1095 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1096 rps_record_sock_flow(sock_flow_table, hash); 1097 rcu_read_unlock(); 1098 #endif 1099 } 1100 1101 static inline void sock_rps_record_flow(const struct sock *sk) 1102 { 1103 #ifdef CONFIG_RPS 1104 if (static_branch_unlikely(&rfs_needed)) { 1105 /* Reading sk->sk_rxhash might incur an expensive cache line 1106 * miss. 1107 * 1108 * TCP_ESTABLISHED does cover almost all states where RFS 1109 * might be useful, and is cheaper [1] than testing : 1110 * IPv4: inet_sk(sk)->inet_daddr 1111 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1112 * OR an additional socket flag 1113 * [1] : sk_state and sk_prot are in the same cache line. 1114 */ 1115 if (sk->sk_state == TCP_ESTABLISHED) 1116 sock_rps_record_flow_hash(sk->sk_rxhash); 1117 } 1118 #endif 1119 } 1120 1121 static inline void sock_rps_save_rxhash(struct sock *sk, 1122 const struct sk_buff *skb) 1123 { 1124 #ifdef CONFIG_RPS 1125 if (unlikely(sk->sk_rxhash != skb->hash)) 1126 sk->sk_rxhash = skb->hash; 1127 #endif 1128 } 1129 1130 static inline void sock_rps_reset_rxhash(struct sock *sk) 1131 { 1132 #ifdef CONFIG_RPS 1133 sk->sk_rxhash = 0; 1134 #endif 1135 } 1136 1137 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1138 ({ int __rc; \ 1139 release_sock(__sk); \ 1140 __rc = __condition; \ 1141 if (!__rc) { \ 1142 *(__timeo) = wait_woken(__wait, \ 1143 TASK_INTERRUPTIBLE, \ 1144 *(__timeo)); \ 1145 } \ 1146 sched_annotate_sleep(); \ 1147 lock_sock(__sk); \ 1148 __rc = __condition; \ 1149 __rc; \ 1150 }) 1151 1152 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1153 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1154 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1155 int sk_stream_error(struct sock *sk, int flags, int err); 1156 void sk_stream_kill_queues(struct sock *sk); 1157 void sk_set_memalloc(struct sock *sk); 1158 void sk_clear_memalloc(struct sock *sk); 1159 1160 void __sk_flush_backlog(struct sock *sk); 1161 1162 static inline bool sk_flush_backlog(struct sock *sk) 1163 { 1164 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1165 __sk_flush_backlog(sk); 1166 return true; 1167 } 1168 return false; 1169 } 1170 1171 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1172 1173 struct request_sock_ops; 1174 struct timewait_sock_ops; 1175 struct inet_hashinfo; 1176 struct raw_hashinfo; 1177 struct smc_hashinfo; 1178 struct module; 1179 struct sk_psock; 1180 1181 /* 1182 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1183 * un-modified. Special care is taken when initializing object to zero. 1184 */ 1185 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1186 { 1187 if (offsetof(struct sock, sk_node.next) != 0) 1188 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1189 memset(&sk->sk_node.pprev, 0, 1190 size - offsetof(struct sock, sk_node.pprev)); 1191 } 1192 1193 /* Networking protocol blocks we attach to sockets. 1194 * socket layer -> transport layer interface 1195 */ 1196 struct proto { 1197 void (*close)(struct sock *sk, 1198 long timeout); 1199 int (*pre_connect)(struct sock *sk, 1200 struct sockaddr *uaddr, 1201 int addr_len); 1202 int (*connect)(struct sock *sk, 1203 struct sockaddr *uaddr, 1204 int addr_len); 1205 int (*disconnect)(struct sock *sk, int flags); 1206 1207 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1208 bool kern); 1209 1210 int (*ioctl)(struct sock *sk, int cmd, 1211 unsigned long arg); 1212 int (*init)(struct sock *sk); 1213 void (*destroy)(struct sock *sk); 1214 void (*shutdown)(struct sock *sk, int how); 1215 int (*setsockopt)(struct sock *sk, int level, 1216 int optname, sockptr_t optval, 1217 unsigned int optlen); 1218 int (*getsockopt)(struct sock *sk, int level, 1219 int optname, char __user *optval, 1220 int __user *option); 1221 void (*keepalive)(struct sock *sk, int valbool); 1222 #ifdef CONFIG_COMPAT 1223 int (*compat_ioctl)(struct sock *sk, 1224 unsigned int cmd, unsigned long arg); 1225 #endif 1226 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1227 size_t len); 1228 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1229 size_t len, int flags, int *addr_len); 1230 int (*sendpage)(struct sock *sk, struct page *page, 1231 int offset, size_t size, int flags); 1232 int (*bind)(struct sock *sk, 1233 struct sockaddr *addr, int addr_len); 1234 int (*bind_add)(struct sock *sk, 1235 struct sockaddr *addr, int addr_len); 1236 1237 int (*backlog_rcv) (struct sock *sk, 1238 struct sk_buff *skb); 1239 bool (*bpf_bypass_getsockopt)(int level, 1240 int optname); 1241 1242 void (*release_cb)(struct sock *sk); 1243 1244 /* Keeping track of sk's, looking them up, and port selection methods. */ 1245 int (*hash)(struct sock *sk); 1246 void (*unhash)(struct sock *sk); 1247 void (*rehash)(struct sock *sk); 1248 int (*get_port)(struct sock *sk, unsigned short snum); 1249 void (*put_port)(struct sock *sk); 1250 #ifdef CONFIG_BPF_SYSCALL 1251 int (*psock_update_sk_prot)(struct sock *sk, 1252 struct sk_psock *psock, 1253 bool restore); 1254 #endif 1255 1256 /* Keeping track of sockets in use */ 1257 #ifdef CONFIG_PROC_FS 1258 unsigned int inuse_idx; 1259 #endif 1260 1261 #if IS_ENABLED(CONFIG_MPTCP) 1262 int (*forward_alloc_get)(const struct sock *sk); 1263 #endif 1264 1265 bool (*stream_memory_free)(const struct sock *sk, int wake); 1266 bool (*sock_is_readable)(struct sock *sk); 1267 /* Memory pressure */ 1268 void (*enter_memory_pressure)(struct sock *sk); 1269 void (*leave_memory_pressure)(struct sock *sk); 1270 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1271 int __percpu *per_cpu_fw_alloc; 1272 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1273 1274 /* 1275 * Pressure flag: try to collapse. 1276 * Technical note: it is used by multiple contexts non atomically. 1277 * All the __sk_mem_schedule() is of this nature: accounting 1278 * is strict, actions are advisory and have some latency. 1279 */ 1280 unsigned long *memory_pressure; 1281 long *sysctl_mem; 1282 1283 int *sysctl_wmem; 1284 int *sysctl_rmem; 1285 u32 sysctl_wmem_offset; 1286 u32 sysctl_rmem_offset; 1287 1288 int max_header; 1289 bool no_autobind; 1290 1291 struct kmem_cache *slab; 1292 unsigned int obj_size; 1293 slab_flags_t slab_flags; 1294 unsigned int useroffset; /* Usercopy region offset */ 1295 unsigned int usersize; /* Usercopy region size */ 1296 1297 unsigned int __percpu *orphan_count; 1298 1299 struct request_sock_ops *rsk_prot; 1300 struct timewait_sock_ops *twsk_prot; 1301 1302 union { 1303 struct inet_hashinfo *hashinfo; 1304 struct udp_table *udp_table; 1305 struct raw_hashinfo *raw_hash; 1306 struct smc_hashinfo *smc_hash; 1307 } h; 1308 1309 struct module *owner; 1310 1311 char name[32]; 1312 1313 struct list_head node; 1314 #ifdef SOCK_REFCNT_DEBUG 1315 atomic_t socks; 1316 #endif 1317 int (*diag_destroy)(struct sock *sk, int err); 1318 } __randomize_layout; 1319 1320 int proto_register(struct proto *prot, int alloc_slab); 1321 void proto_unregister(struct proto *prot); 1322 int sock_load_diag_module(int family, int protocol); 1323 1324 #ifdef SOCK_REFCNT_DEBUG 1325 static inline void sk_refcnt_debug_inc(struct sock *sk) 1326 { 1327 atomic_inc(&sk->sk_prot->socks); 1328 } 1329 1330 static inline void sk_refcnt_debug_dec(struct sock *sk) 1331 { 1332 atomic_dec(&sk->sk_prot->socks); 1333 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1334 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1335 } 1336 1337 static inline void sk_refcnt_debug_release(const struct sock *sk) 1338 { 1339 if (refcount_read(&sk->sk_refcnt) != 1) 1340 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1341 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1342 } 1343 #else /* SOCK_REFCNT_DEBUG */ 1344 #define sk_refcnt_debug_inc(sk) do { } while (0) 1345 #define sk_refcnt_debug_dec(sk) do { } while (0) 1346 #define sk_refcnt_debug_release(sk) do { } while (0) 1347 #endif /* SOCK_REFCNT_DEBUG */ 1348 1349 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1350 1351 static inline int sk_forward_alloc_get(const struct sock *sk) 1352 { 1353 #if IS_ENABLED(CONFIG_MPTCP) 1354 if (sk->sk_prot->forward_alloc_get) 1355 return sk->sk_prot->forward_alloc_get(sk); 1356 #endif 1357 return sk->sk_forward_alloc; 1358 } 1359 1360 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1361 { 1362 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1363 return false; 1364 1365 return sk->sk_prot->stream_memory_free ? 1366 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1367 tcp_stream_memory_free, sk, wake) : true; 1368 } 1369 1370 static inline bool sk_stream_memory_free(const struct sock *sk) 1371 { 1372 return __sk_stream_memory_free(sk, 0); 1373 } 1374 1375 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1376 { 1377 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1378 __sk_stream_memory_free(sk, wake); 1379 } 1380 1381 static inline bool sk_stream_is_writeable(const struct sock *sk) 1382 { 1383 return __sk_stream_is_writeable(sk, 0); 1384 } 1385 1386 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1387 struct cgroup *ancestor) 1388 { 1389 #ifdef CONFIG_SOCK_CGROUP_DATA 1390 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1391 ancestor); 1392 #else 1393 return -ENOTSUPP; 1394 #endif 1395 } 1396 1397 static inline bool sk_has_memory_pressure(const struct sock *sk) 1398 { 1399 return sk->sk_prot->memory_pressure != NULL; 1400 } 1401 1402 static inline bool sk_under_memory_pressure(const struct sock *sk) 1403 { 1404 if (!sk->sk_prot->memory_pressure) 1405 return false; 1406 1407 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1408 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1409 return true; 1410 1411 return !!*sk->sk_prot->memory_pressure; 1412 } 1413 1414 static inline long 1415 proto_memory_allocated(const struct proto *prot) 1416 { 1417 return max(0L, atomic_long_read(prot->memory_allocated)); 1418 } 1419 1420 static inline long 1421 sk_memory_allocated(const struct sock *sk) 1422 { 1423 return proto_memory_allocated(sk->sk_prot); 1424 } 1425 1426 /* 1 MB per cpu, in page units */ 1427 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1428 1429 static inline void 1430 sk_memory_allocated_add(struct sock *sk, int amt) 1431 { 1432 int local_reserve; 1433 1434 preempt_disable(); 1435 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1436 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) { 1437 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1438 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1439 } 1440 preempt_enable(); 1441 } 1442 1443 static inline void 1444 sk_memory_allocated_sub(struct sock *sk, int amt) 1445 { 1446 int local_reserve; 1447 1448 preempt_disable(); 1449 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1450 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) { 1451 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1452 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1453 } 1454 preempt_enable(); 1455 } 1456 1457 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1458 1459 static inline void sk_sockets_allocated_dec(struct sock *sk) 1460 { 1461 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1462 SK_ALLOC_PERCPU_COUNTER_BATCH); 1463 } 1464 1465 static inline void sk_sockets_allocated_inc(struct sock *sk) 1466 { 1467 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1468 SK_ALLOC_PERCPU_COUNTER_BATCH); 1469 } 1470 1471 static inline u64 1472 sk_sockets_allocated_read_positive(struct sock *sk) 1473 { 1474 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1475 } 1476 1477 static inline int 1478 proto_sockets_allocated_sum_positive(struct proto *prot) 1479 { 1480 return percpu_counter_sum_positive(prot->sockets_allocated); 1481 } 1482 1483 static inline bool 1484 proto_memory_pressure(struct proto *prot) 1485 { 1486 if (!prot->memory_pressure) 1487 return false; 1488 return !!*prot->memory_pressure; 1489 } 1490 1491 1492 #ifdef CONFIG_PROC_FS 1493 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1494 struct prot_inuse { 1495 int all; 1496 int val[PROTO_INUSE_NR]; 1497 }; 1498 1499 static inline void sock_prot_inuse_add(const struct net *net, 1500 const struct proto *prot, int val) 1501 { 1502 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1503 } 1504 1505 static inline void sock_inuse_add(const struct net *net, int val) 1506 { 1507 this_cpu_add(net->core.prot_inuse->all, val); 1508 } 1509 1510 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1511 int sock_inuse_get(struct net *net); 1512 #else 1513 static inline void sock_prot_inuse_add(const struct net *net, 1514 const struct proto *prot, int val) 1515 { 1516 } 1517 1518 static inline void sock_inuse_add(const struct net *net, int val) 1519 { 1520 } 1521 #endif 1522 1523 1524 /* With per-bucket locks this operation is not-atomic, so that 1525 * this version is not worse. 1526 */ 1527 static inline int __sk_prot_rehash(struct sock *sk) 1528 { 1529 sk->sk_prot->unhash(sk); 1530 return sk->sk_prot->hash(sk); 1531 } 1532 1533 /* About 10 seconds */ 1534 #define SOCK_DESTROY_TIME (10*HZ) 1535 1536 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1537 #define PROT_SOCK 1024 1538 1539 #define SHUTDOWN_MASK 3 1540 #define RCV_SHUTDOWN 1 1541 #define SEND_SHUTDOWN 2 1542 1543 #define SOCK_BINDADDR_LOCK 4 1544 #define SOCK_BINDPORT_LOCK 8 1545 1546 struct socket_alloc { 1547 struct socket socket; 1548 struct inode vfs_inode; 1549 }; 1550 1551 static inline struct socket *SOCKET_I(struct inode *inode) 1552 { 1553 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1554 } 1555 1556 static inline struct inode *SOCK_INODE(struct socket *socket) 1557 { 1558 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1559 } 1560 1561 /* 1562 * Functions for memory accounting 1563 */ 1564 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1565 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1566 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1567 void __sk_mem_reclaim(struct sock *sk, int amount); 1568 1569 #define SK_MEM_SEND 0 1570 #define SK_MEM_RECV 1 1571 1572 /* sysctl_mem values are in pages */ 1573 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1574 { 1575 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1576 } 1577 1578 static inline int sk_mem_pages(int amt) 1579 { 1580 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1581 } 1582 1583 static inline bool sk_has_account(struct sock *sk) 1584 { 1585 /* return true if protocol supports memory accounting */ 1586 return !!sk->sk_prot->memory_allocated; 1587 } 1588 1589 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1590 { 1591 int delta; 1592 1593 if (!sk_has_account(sk)) 1594 return true; 1595 delta = size - sk->sk_forward_alloc; 1596 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1597 } 1598 1599 static inline bool 1600 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1601 { 1602 int delta; 1603 1604 if (!sk_has_account(sk)) 1605 return true; 1606 delta = size - sk->sk_forward_alloc; 1607 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1608 skb_pfmemalloc(skb); 1609 } 1610 1611 static inline int sk_unused_reserved_mem(const struct sock *sk) 1612 { 1613 int unused_mem; 1614 1615 if (likely(!sk->sk_reserved_mem)) 1616 return 0; 1617 1618 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1619 atomic_read(&sk->sk_rmem_alloc); 1620 1621 return unused_mem > 0 ? unused_mem : 0; 1622 } 1623 1624 static inline void sk_mem_reclaim(struct sock *sk) 1625 { 1626 int reclaimable; 1627 1628 if (!sk_has_account(sk)) 1629 return; 1630 1631 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1632 1633 if (reclaimable >= (int)PAGE_SIZE) 1634 __sk_mem_reclaim(sk, reclaimable); 1635 } 1636 1637 static inline void sk_mem_reclaim_final(struct sock *sk) 1638 { 1639 sk->sk_reserved_mem = 0; 1640 sk_mem_reclaim(sk); 1641 } 1642 1643 static inline void sk_mem_charge(struct sock *sk, int size) 1644 { 1645 if (!sk_has_account(sk)) 1646 return; 1647 sk->sk_forward_alloc -= size; 1648 } 1649 1650 static inline void sk_mem_uncharge(struct sock *sk, int size) 1651 { 1652 if (!sk_has_account(sk)) 1653 return; 1654 sk->sk_forward_alloc += size; 1655 sk_mem_reclaim(sk); 1656 } 1657 1658 /* 1659 * Macro so as to not evaluate some arguments when 1660 * lockdep is not enabled. 1661 * 1662 * Mark both the sk_lock and the sk_lock.slock as a 1663 * per-address-family lock class. 1664 */ 1665 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1666 do { \ 1667 sk->sk_lock.owned = 0; \ 1668 init_waitqueue_head(&sk->sk_lock.wq); \ 1669 spin_lock_init(&(sk)->sk_lock.slock); \ 1670 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1671 sizeof((sk)->sk_lock)); \ 1672 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1673 (skey), (sname)); \ 1674 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1675 } while (0) 1676 1677 static inline bool lockdep_sock_is_held(const struct sock *sk) 1678 { 1679 return lockdep_is_held(&sk->sk_lock) || 1680 lockdep_is_held(&sk->sk_lock.slock); 1681 } 1682 1683 void lock_sock_nested(struct sock *sk, int subclass); 1684 1685 static inline void lock_sock(struct sock *sk) 1686 { 1687 lock_sock_nested(sk, 0); 1688 } 1689 1690 void __lock_sock(struct sock *sk); 1691 void __release_sock(struct sock *sk); 1692 void release_sock(struct sock *sk); 1693 1694 /* BH context may only use the following locking interface. */ 1695 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1696 #define bh_lock_sock_nested(__sk) \ 1697 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1698 SINGLE_DEPTH_NESTING) 1699 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1700 1701 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1702 1703 /** 1704 * lock_sock_fast - fast version of lock_sock 1705 * @sk: socket 1706 * 1707 * This version should be used for very small section, where process wont block 1708 * return false if fast path is taken: 1709 * 1710 * sk_lock.slock locked, owned = 0, BH disabled 1711 * 1712 * return true if slow path is taken: 1713 * 1714 * sk_lock.slock unlocked, owned = 1, BH enabled 1715 */ 1716 static inline bool lock_sock_fast(struct sock *sk) 1717 { 1718 /* The sk_lock has mutex_lock() semantics here. */ 1719 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1720 1721 return __lock_sock_fast(sk); 1722 } 1723 1724 /* fast socket lock variant for caller already holding a [different] socket lock */ 1725 static inline bool lock_sock_fast_nested(struct sock *sk) 1726 { 1727 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1728 1729 return __lock_sock_fast(sk); 1730 } 1731 1732 /** 1733 * unlock_sock_fast - complement of lock_sock_fast 1734 * @sk: socket 1735 * @slow: slow mode 1736 * 1737 * fast unlock socket for user context. 1738 * If slow mode is on, we call regular release_sock() 1739 */ 1740 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1741 __releases(&sk->sk_lock.slock) 1742 { 1743 if (slow) { 1744 release_sock(sk); 1745 __release(&sk->sk_lock.slock); 1746 } else { 1747 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1748 spin_unlock_bh(&sk->sk_lock.slock); 1749 } 1750 } 1751 1752 /* Used by processes to "lock" a socket state, so that 1753 * interrupts and bottom half handlers won't change it 1754 * from under us. It essentially blocks any incoming 1755 * packets, so that we won't get any new data or any 1756 * packets that change the state of the socket. 1757 * 1758 * While locked, BH processing will add new packets to 1759 * the backlog queue. This queue is processed by the 1760 * owner of the socket lock right before it is released. 1761 * 1762 * Since ~2.3.5 it is also exclusive sleep lock serializing 1763 * accesses from user process context. 1764 */ 1765 1766 static inline void sock_owned_by_me(const struct sock *sk) 1767 { 1768 #ifdef CONFIG_LOCKDEP 1769 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1770 #endif 1771 } 1772 1773 static inline bool sock_owned_by_user(const struct sock *sk) 1774 { 1775 sock_owned_by_me(sk); 1776 return sk->sk_lock.owned; 1777 } 1778 1779 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1780 { 1781 return sk->sk_lock.owned; 1782 } 1783 1784 static inline void sock_release_ownership(struct sock *sk) 1785 { 1786 if (sock_owned_by_user_nocheck(sk)) { 1787 sk->sk_lock.owned = 0; 1788 1789 /* The sk_lock has mutex_unlock() semantics: */ 1790 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1791 } 1792 } 1793 1794 /* no reclassification while locks are held */ 1795 static inline bool sock_allow_reclassification(const struct sock *csk) 1796 { 1797 struct sock *sk = (struct sock *)csk; 1798 1799 return !sock_owned_by_user_nocheck(sk) && 1800 !spin_is_locked(&sk->sk_lock.slock); 1801 } 1802 1803 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1804 struct proto *prot, int kern); 1805 void sk_free(struct sock *sk); 1806 void sk_destruct(struct sock *sk); 1807 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1808 void sk_free_unlock_clone(struct sock *sk); 1809 1810 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1811 gfp_t priority); 1812 void __sock_wfree(struct sk_buff *skb); 1813 void sock_wfree(struct sk_buff *skb); 1814 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1815 gfp_t priority); 1816 void skb_orphan_partial(struct sk_buff *skb); 1817 void sock_rfree(struct sk_buff *skb); 1818 void sock_efree(struct sk_buff *skb); 1819 #ifdef CONFIG_INET 1820 void sock_edemux(struct sk_buff *skb); 1821 void sock_pfree(struct sk_buff *skb); 1822 #else 1823 #define sock_edemux sock_efree 1824 #endif 1825 1826 int sock_setsockopt(struct socket *sock, int level, int op, 1827 sockptr_t optval, unsigned int optlen); 1828 1829 int sock_getsockopt(struct socket *sock, int level, int op, 1830 char __user *optval, int __user *optlen); 1831 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1832 bool timeval, bool time32); 1833 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1834 unsigned long data_len, int noblock, 1835 int *errcode, int max_page_order); 1836 1837 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1838 unsigned long size, 1839 int noblock, int *errcode) 1840 { 1841 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1842 } 1843 1844 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1845 void sock_kfree_s(struct sock *sk, void *mem, int size); 1846 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1847 void sk_send_sigurg(struct sock *sk); 1848 1849 struct sockcm_cookie { 1850 u64 transmit_time; 1851 u32 mark; 1852 u16 tsflags; 1853 }; 1854 1855 static inline void sockcm_init(struct sockcm_cookie *sockc, 1856 const struct sock *sk) 1857 { 1858 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1859 } 1860 1861 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1862 struct sockcm_cookie *sockc); 1863 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1864 struct sockcm_cookie *sockc); 1865 1866 /* 1867 * Functions to fill in entries in struct proto_ops when a protocol 1868 * does not implement a particular function. 1869 */ 1870 int sock_no_bind(struct socket *, struct sockaddr *, int); 1871 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1872 int sock_no_socketpair(struct socket *, struct socket *); 1873 int sock_no_accept(struct socket *, struct socket *, int, bool); 1874 int sock_no_getname(struct socket *, struct sockaddr *, int); 1875 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1876 int sock_no_listen(struct socket *, int); 1877 int sock_no_shutdown(struct socket *, int); 1878 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1879 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1880 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1881 int sock_no_mmap(struct file *file, struct socket *sock, 1882 struct vm_area_struct *vma); 1883 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1884 size_t size, int flags); 1885 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1886 int offset, size_t size, int flags); 1887 1888 /* 1889 * Functions to fill in entries in struct proto_ops when a protocol 1890 * uses the inet style. 1891 */ 1892 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1893 char __user *optval, int __user *optlen); 1894 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1895 int flags); 1896 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1897 sockptr_t optval, unsigned int optlen); 1898 1899 void sk_common_release(struct sock *sk); 1900 1901 /* 1902 * Default socket callbacks and setup code 1903 */ 1904 1905 /* Initialise core socket variables */ 1906 void sock_init_data(struct socket *sock, struct sock *sk); 1907 1908 /* 1909 * Socket reference counting postulates. 1910 * 1911 * * Each user of socket SHOULD hold a reference count. 1912 * * Each access point to socket (an hash table bucket, reference from a list, 1913 * running timer, skb in flight MUST hold a reference count. 1914 * * When reference count hits 0, it means it will never increase back. 1915 * * When reference count hits 0, it means that no references from 1916 * outside exist to this socket and current process on current CPU 1917 * is last user and may/should destroy this socket. 1918 * * sk_free is called from any context: process, BH, IRQ. When 1919 * it is called, socket has no references from outside -> sk_free 1920 * may release descendant resources allocated by the socket, but 1921 * to the time when it is called, socket is NOT referenced by any 1922 * hash tables, lists etc. 1923 * * Packets, delivered from outside (from network or from another process) 1924 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1925 * when they sit in queue. Otherwise, packets will leak to hole, when 1926 * socket is looked up by one cpu and unhasing is made by another CPU. 1927 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1928 * (leak to backlog). Packet socket does all the processing inside 1929 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1930 * use separate SMP lock, so that they are prone too. 1931 */ 1932 1933 /* Ungrab socket and destroy it, if it was the last reference. */ 1934 static inline void sock_put(struct sock *sk) 1935 { 1936 if (refcount_dec_and_test(&sk->sk_refcnt)) 1937 sk_free(sk); 1938 } 1939 /* Generic version of sock_put(), dealing with all sockets 1940 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1941 */ 1942 void sock_gen_put(struct sock *sk); 1943 1944 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1945 unsigned int trim_cap, bool refcounted); 1946 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1947 const int nested) 1948 { 1949 return __sk_receive_skb(sk, skb, nested, 1, true); 1950 } 1951 1952 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1953 { 1954 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1955 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1956 return; 1957 sk->sk_tx_queue_mapping = tx_queue; 1958 } 1959 1960 #define NO_QUEUE_MAPPING USHRT_MAX 1961 1962 static inline void sk_tx_queue_clear(struct sock *sk) 1963 { 1964 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1965 } 1966 1967 static inline int sk_tx_queue_get(const struct sock *sk) 1968 { 1969 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1970 return sk->sk_tx_queue_mapping; 1971 1972 return -1; 1973 } 1974 1975 static inline void __sk_rx_queue_set(struct sock *sk, 1976 const struct sk_buff *skb, 1977 bool force_set) 1978 { 1979 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 1980 if (skb_rx_queue_recorded(skb)) { 1981 u16 rx_queue = skb_get_rx_queue(skb); 1982 1983 if (force_set || 1984 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 1985 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 1986 } 1987 #endif 1988 } 1989 1990 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 1991 { 1992 __sk_rx_queue_set(sk, skb, true); 1993 } 1994 1995 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 1996 { 1997 __sk_rx_queue_set(sk, skb, false); 1998 } 1999 2000 static inline void sk_rx_queue_clear(struct sock *sk) 2001 { 2002 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2003 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2004 #endif 2005 } 2006 2007 static inline int sk_rx_queue_get(const struct sock *sk) 2008 { 2009 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2010 if (sk) { 2011 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2012 2013 if (res != NO_QUEUE_MAPPING) 2014 return res; 2015 } 2016 #endif 2017 2018 return -1; 2019 } 2020 2021 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2022 { 2023 sk->sk_socket = sock; 2024 } 2025 2026 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2027 { 2028 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2029 return &rcu_dereference_raw(sk->sk_wq)->wait; 2030 } 2031 /* Detach socket from process context. 2032 * Announce socket dead, detach it from wait queue and inode. 2033 * Note that parent inode held reference count on this struct sock, 2034 * we do not release it in this function, because protocol 2035 * probably wants some additional cleanups or even continuing 2036 * to work with this socket (TCP). 2037 */ 2038 static inline void sock_orphan(struct sock *sk) 2039 { 2040 write_lock_bh(&sk->sk_callback_lock); 2041 sock_set_flag(sk, SOCK_DEAD); 2042 sk_set_socket(sk, NULL); 2043 sk->sk_wq = NULL; 2044 write_unlock_bh(&sk->sk_callback_lock); 2045 } 2046 2047 static inline void sock_graft(struct sock *sk, struct socket *parent) 2048 { 2049 WARN_ON(parent->sk); 2050 write_lock_bh(&sk->sk_callback_lock); 2051 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2052 parent->sk = sk; 2053 sk_set_socket(sk, parent); 2054 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2055 security_sock_graft(sk, parent); 2056 write_unlock_bh(&sk->sk_callback_lock); 2057 } 2058 2059 kuid_t sock_i_uid(struct sock *sk); 2060 unsigned long sock_i_ino(struct sock *sk); 2061 2062 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2063 { 2064 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2065 } 2066 2067 static inline u32 net_tx_rndhash(void) 2068 { 2069 u32 v = prandom_u32(); 2070 2071 return v ?: 1; 2072 } 2073 2074 static inline void sk_set_txhash(struct sock *sk) 2075 { 2076 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2077 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2078 } 2079 2080 static inline bool sk_rethink_txhash(struct sock *sk) 2081 { 2082 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2083 sk_set_txhash(sk); 2084 return true; 2085 } 2086 return false; 2087 } 2088 2089 static inline struct dst_entry * 2090 __sk_dst_get(struct sock *sk) 2091 { 2092 return rcu_dereference_check(sk->sk_dst_cache, 2093 lockdep_sock_is_held(sk)); 2094 } 2095 2096 static inline struct dst_entry * 2097 sk_dst_get(struct sock *sk) 2098 { 2099 struct dst_entry *dst; 2100 2101 rcu_read_lock(); 2102 dst = rcu_dereference(sk->sk_dst_cache); 2103 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2104 dst = NULL; 2105 rcu_read_unlock(); 2106 return dst; 2107 } 2108 2109 static inline void __dst_negative_advice(struct sock *sk) 2110 { 2111 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2112 2113 if (dst && dst->ops->negative_advice) { 2114 ndst = dst->ops->negative_advice(dst); 2115 2116 if (ndst != dst) { 2117 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2118 sk_tx_queue_clear(sk); 2119 sk->sk_dst_pending_confirm = 0; 2120 } 2121 } 2122 } 2123 2124 static inline void dst_negative_advice(struct sock *sk) 2125 { 2126 sk_rethink_txhash(sk); 2127 __dst_negative_advice(sk); 2128 } 2129 2130 static inline void 2131 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2132 { 2133 struct dst_entry *old_dst; 2134 2135 sk_tx_queue_clear(sk); 2136 sk->sk_dst_pending_confirm = 0; 2137 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2138 lockdep_sock_is_held(sk)); 2139 rcu_assign_pointer(sk->sk_dst_cache, dst); 2140 dst_release(old_dst); 2141 } 2142 2143 static inline void 2144 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2145 { 2146 struct dst_entry *old_dst; 2147 2148 sk_tx_queue_clear(sk); 2149 sk->sk_dst_pending_confirm = 0; 2150 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2151 dst_release(old_dst); 2152 } 2153 2154 static inline void 2155 __sk_dst_reset(struct sock *sk) 2156 { 2157 __sk_dst_set(sk, NULL); 2158 } 2159 2160 static inline void 2161 sk_dst_reset(struct sock *sk) 2162 { 2163 sk_dst_set(sk, NULL); 2164 } 2165 2166 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2167 2168 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2169 2170 static inline void sk_dst_confirm(struct sock *sk) 2171 { 2172 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2173 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2174 } 2175 2176 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2177 { 2178 if (skb_get_dst_pending_confirm(skb)) { 2179 struct sock *sk = skb->sk; 2180 2181 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2182 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2183 neigh_confirm(n); 2184 } 2185 } 2186 2187 bool sk_mc_loop(struct sock *sk); 2188 2189 static inline bool sk_can_gso(const struct sock *sk) 2190 { 2191 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2192 } 2193 2194 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2195 2196 static inline void sk_gso_disable(struct sock *sk) 2197 { 2198 sk->sk_gso_disabled = 1; 2199 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2200 } 2201 2202 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2203 struct iov_iter *from, char *to, 2204 int copy, int offset) 2205 { 2206 if (skb->ip_summed == CHECKSUM_NONE) { 2207 __wsum csum = 0; 2208 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2209 return -EFAULT; 2210 skb->csum = csum_block_add(skb->csum, csum, offset); 2211 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2212 if (!copy_from_iter_full_nocache(to, copy, from)) 2213 return -EFAULT; 2214 } else if (!copy_from_iter_full(to, copy, from)) 2215 return -EFAULT; 2216 2217 return 0; 2218 } 2219 2220 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2221 struct iov_iter *from, int copy) 2222 { 2223 int err, offset = skb->len; 2224 2225 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2226 copy, offset); 2227 if (err) 2228 __skb_trim(skb, offset); 2229 2230 return err; 2231 } 2232 2233 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2234 struct sk_buff *skb, 2235 struct page *page, 2236 int off, int copy) 2237 { 2238 int err; 2239 2240 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2241 copy, skb->len); 2242 if (err) 2243 return err; 2244 2245 skb_len_add(skb, copy); 2246 sk_wmem_queued_add(sk, copy); 2247 sk_mem_charge(sk, copy); 2248 return 0; 2249 } 2250 2251 /** 2252 * sk_wmem_alloc_get - returns write allocations 2253 * @sk: socket 2254 * 2255 * Return: sk_wmem_alloc minus initial offset of one 2256 */ 2257 static inline int sk_wmem_alloc_get(const struct sock *sk) 2258 { 2259 return refcount_read(&sk->sk_wmem_alloc) - 1; 2260 } 2261 2262 /** 2263 * sk_rmem_alloc_get - returns read allocations 2264 * @sk: socket 2265 * 2266 * Return: sk_rmem_alloc 2267 */ 2268 static inline int sk_rmem_alloc_get(const struct sock *sk) 2269 { 2270 return atomic_read(&sk->sk_rmem_alloc); 2271 } 2272 2273 /** 2274 * sk_has_allocations - check if allocations are outstanding 2275 * @sk: socket 2276 * 2277 * Return: true if socket has write or read allocations 2278 */ 2279 static inline bool sk_has_allocations(const struct sock *sk) 2280 { 2281 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2282 } 2283 2284 /** 2285 * skwq_has_sleeper - check if there are any waiting processes 2286 * @wq: struct socket_wq 2287 * 2288 * Return: true if socket_wq has waiting processes 2289 * 2290 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2291 * barrier call. They were added due to the race found within the tcp code. 2292 * 2293 * Consider following tcp code paths:: 2294 * 2295 * CPU1 CPU2 2296 * sys_select receive packet 2297 * ... ... 2298 * __add_wait_queue update tp->rcv_nxt 2299 * ... ... 2300 * tp->rcv_nxt check sock_def_readable 2301 * ... { 2302 * schedule rcu_read_lock(); 2303 * wq = rcu_dereference(sk->sk_wq); 2304 * if (wq && waitqueue_active(&wq->wait)) 2305 * wake_up_interruptible(&wq->wait) 2306 * ... 2307 * } 2308 * 2309 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2310 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2311 * could then endup calling schedule and sleep forever if there are no more 2312 * data on the socket. 2313 * 2314 */ 2315 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2316 { 2317 return wq && wq_has_sleeper(&wq->wait); 2318 } 2319 2320 /** 2321 * sock_poll_wait - place memory barrier behind the poll_wait call. 2322 * @filp: file 2323 * @sock: socket to wait on 2324 * @p: poll_table 2325 * 2326 * See the comments in the wq_has_sleeper function. 2327 */ 2328 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2329 poll_table *p) 2330 { 2331 if (!poll_does_not_wait(p)) { 2332 poll_wait(filp, &sock->wq.wait, p); 2333 /* We need to be sure we are in sync with the 2334 * socket flags modification. 2335 * 2336 * This memory barrier is paired in the wq_has_sleeper. 2337 */ 2338 smp_mb(); 2339 } 2340 } 2341 2342 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2343 { 2344 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2345 u32 txhash = READ_ONCE(sk->sk_txhash); 2346 2347 if (txhash) { 2348 skb->l4_hash = 1; 2349 skb->hash = txhash; 2350 } 2351 } 2352 2353 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2354 2355 /* 2356 * Queue a received datagram if it will fit. Stream and sequenced 2357 * protocols can't normally use this as they need to fit buffers in 2358 * and play with them. 2359 * 2360 * Inlined as it's very short and called for pretty much every 2361 * packet ever received. 2362 */ 2363 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2364 { 2365 skb_orphan(skb); 2366 skb->sk = sk; 2367 skb->destructor = sock_rfree; 2368 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2369 sk_mem_charge(sk, skb->truesize); 2370 } 2371 2372 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2373 { 2374 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2375 skb_orphan(skb); 2376 skb->destructor = sock_efree; 2377 skb->sk = sk; 2378 return true; 2379 } 2380 return false; 2381 } 2382 2383 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2384 { 2385 if (skb->destructor != sock_wfree) { 2386 skb_orphan(skb); 2387 return; 2388 } 2389 skb->slow_gro = 1; 2390 } 2391 2392 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2393 unsigned long expires); 2394 2395 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2396 2397 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2398 2399 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2400 struct sk_buff *skb, unsigned int flags, 2401 void (*destructor)(struct sock *sk, 2402 struct sk_buff *skb)); 2403 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2404 2405 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2406 enum skb_drop_reason *reason); 2407 2408 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2409 { 2410 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2411 } 2412 2413 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2414 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2415 2416 /* 2417 * Recover an error report and clear atomically 2418 */ 2419 2420 static inline int sock_error(struct sock *sk) 2421 { 2422 int err; 2423 2424 /* Avoid an atomic operation for the common case. 2425 * This is racy since another cpu/thread can change sk_err under us. 2426 */ 2427 if (likely(data_race(!sk->sk_err))) 2428 return 0; 2429 2430 err = xchg(&sk->sk_err, 0); 2431 return -err; 2432 } 2433 2434 void sk_error_report(struct sock *sk); 2435 2436 static inline unsigned long sock_wspace(struct sock *sk) 2437 { 2438 int amt = 0; 2439 2440 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2441 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2442 if (amt < 0) 2443 amt = 0; 2444 } 2445 return amt; 2446 } 2447 2448 /* Note: 2449 * We use sk->sk_wq_raw, from contexts knowing this 2450 * pointer is not NULL and cannot disappear/change. 2451 */ 2452 static inline void sk_set_bit(int nr, struct sock *sk) 2453 { 2454 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2455 !sock_flag(sk, SOCK_FASYNC)) 2456 return; 2457 2458 set_bit(nr, &sk->sk_wq_raw->flags); 2459 } 2460 2461 static inline void sk_clear_bit(int nr, struct sock *sk) 2462 { 2463 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2464 !sock_flag(sk, SOCK_FASYNC)) 2465 return; 2466 2467 clear_bit(nr, &sk->sk_wq_raw->flags); 2468 } 2469 2470 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2471 { 2472 if (sock_flag(sk, SOCK_FASYNC)) { 2473 rcu_read_lock(); 2474 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2475 rcu_read_unlock(); 2476 } 2477 } 2478 2479 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2480 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2481 * Note: for send buffers, TCP works better if we can build two skbs at 2482 * minimum. 2483 */ 2484 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2485 2486 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2487 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2488 2489 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2490 { 2491 u32 val; 2492 2493 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2494 return; 2495 2496 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2497 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2498 2499 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2500 } 2501 2502 /** 2503 * sk_page_frag - return an appropriate page_frag 2504 * @sk: socket 2505 * 2506 * Use the per task page_frag instead of the per socket one for 2507 * optimization when we know that we're in process context and own 2508 * everything that's associated with %current. 2509 * 2510 * Both direct reclaim and page faults can nest inside other 2511 * socket operations and end up recursing into sk_page_frag() 2512 * while it's already in use: explicitly avoid task page_frag 2513 * usage if the caller is potentially doing any of them. 2514 * This assumes that page fault handlers use the GFP_NOFS flags. 2515 * 2516 * Return: a per task page_frag if context allows that, 2517 * otherwise a per socket one. 2518 */ 2519 static inline struct page_frag *sk_page_frag(struct sock *sk) 2520 { 2521 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) == 2522 (__GFP_DIRECT_RECLAIM | __GFP_FS)) 2523 return ¤t->task_frag; 2524 2525 return &sk->sk_frag; 2526 } 2527 2528 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2529 2530 /* 2531 * Default write policy as shown to user space via poll/select/SIGIO 2532 */ 2533 static inline bool sock_writeable(const struct sock *sk) 2534 { 2535 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2536 } 2537 2538 static inline gfp_t gfp_any(void) 2539 { 2540 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2541 } 2542 2543 static inline gfp_t gfp_memcg_charge(void) 2544 { 2545 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2546 } 2547 2548 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2549 { 2550 return noblock ? 0 : sk->sk_rcvtimeo; 2551 } 2552 2553 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2554 { 2555 return noblock ? 0 : sk->sk_sndtimeo; 2556 } 2557 2558 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2559 { 2560 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2561 2562 return v ?: 1; 2563 } 2564 2565 /* Alas, with timeout socket operations are not restartable. 2566 * Compare this to poll(). 2567 */ 2568 static inline int sock_intr_errno(long timeo) 2569 { 2570 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2571 } 2572 2573 struct sock_skb_cb { 2574 u32 dropcount; 2575 }; 2576 2577 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2578 * using skb->cb[] would keep using it directly and utilize its 2579 * alignement guarantee. 2580 */ 2581 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2582 sizeof(struct sock_skb_cb))) 2583 2584 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2585 SOCK_SKB_CB_OFFSET)) 2586 2587 #define sock_skb_cb_check_size(size) \ 2588 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2589 2590 static inline void 2591 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2592 { 2593 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2594 atomic_read(&sk->sk_drops) : 0; 2595 } 2596 2597 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2598 { 2599 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2600 2601 atomic_add(segs, &sk->sk_drops); 2602 } 2603 2604 static inline ktime_t sock_read_timestamp(struct sock *sk) 2605 { 2606 #if BITS_PER_LONG==32 2607 unsigned int seq; 2608 ktime_t kt; 2609 2610 do { 2611 seq = read_seqbegin(&sk->sk_stamp_seq); 2612 kt = sk->sk_stamp; 2613 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2614 2615 return kt; 2616 #else 2617 return READ_ONCE(sk->sk_stamp); 2618 #endif 2619 } 2620 2621 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2622 { 2623 #if BITS_PER_LONG==32 2624 write_seqlock(&sk->sk_stamp_seq); 2625 sk->sk_stamp = kt; 2626 write_sequnlock(&sk->sk_stamp_seq); 2627 #else 2628 WRITE_ONCE(sk->sk_stamp, kt); 2629 #endif 2630 } 2631 2632 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2633 struct sk_buff *skb); 2634 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2635 struct sk_buff *skb); 2636 2637 static inline void 2638 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2639 { 2640 ktime_t kt = skb->tstamp; 2641 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2642 2643 /* 2644 * generate control messages if 2645 * - receive time stamping in software requested 2646 * - software time stamp available and wanted 2647 * - hardware time stamps available and wanted 2648 */ 2649 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2650 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2651 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2652 (hwtstamps->hwtstamp && 2653 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2654 __sock_recv_timestamp(msg, sk, skb); 2655 else 2656 sock_write_timestamp(sk, kt); 2657 2658 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2659 __sock_recv_wifi_status(msg, sk, skb); 2660 } 2661 2662 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2663 struct sk_buff *skb); 2664 2665 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2666 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2667 struct sk_buff *skb) 2668 { 2669 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2670 (1UL << SOCK_RCVTSTAMP) | \ 2671 (1UL << SOCK_RCVMARK)) 2672 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2673 SOF_TIMESTAMPING_RAW_HARDWARE) 2674 2675 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2676 __sock_recv_cmsgs(msg, sk, skb); 2677 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2678 sock_write_timestamp(sk, skb->tstamp); 2679 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2680 sock_write_timestamp(sk, 0); 2681 } 2682 2683 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2684 2685 /** 2686 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2687 * @sk: socket sending this packet 2688 * @tsflags: timestamping flags to use 2689 * @tx_flags: completed with instructions for time stamping 2690 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2691 * 2692 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2693 */ 2694 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2695 __u8 *tx_flags, __u32 *tskey) 2696 { 2697 if (unlikely(tsflags)) { 2698 __sock_tx_timestamp(tsflags, tx_flags); 2699 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2700 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2701 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2702 } 2703 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2704 *tx_flags |= SKBTX_WIFI_STATUS; 2705 } 2706 2707 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2708 __u8 *tx_flags) 2709 { 2710 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2711 } 2712 2713 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2714 { 2715 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2716 &skb_shinfo(skb)->tskey); 2717 } 2718 2719 static inline bool sk_is_tcp(const struct sock *sk) 2720 { 2721 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2722 } 2723 2724 /** 2725 * sk_eat_skb - Release a skb if it is no longer needed 2726 * @sk: socket to eat this skb from 2727 * @skb: socket buffer to eat 2728 * 2729 * This routine must be called with interrupts disabled or with the socket 2730 * locked so that the sk_buff queue operation is ok. 2731 */ 2732 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2733 { 2734 __skb_unlink(skb, &sk->sk_receive_queue); 2735 __kfree_skb(skb); 2736 } 2737 2738 static inline bool 2739 skb_sk_is_prefetched(struct sk_buff *skb) 2740 { 2741 #ifdef CONFIG_INET 2742 return skb->destructor == sock_pfree; 2743 #else 2744 return false; 2745 #endif /* CONFIG_INET */ 2746 } 2747 2748 /* This helper checks if a socket is a full socket, 2749 * ie _not_ a timewait or request socket. 2750 */ 2751 static inline bool sk_fullsock(const struct sock *sk) 2752 { 2753 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2754 } 2755 2756 static inline bool 2757 sk_is_refcounted(struct sock *sk) 2758 { 2759 /* Only full sockets have sk->sk_flags. */ 2760 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2761 } 2762 2763 /** 2764 * skb_steal_sock - steal a socket from an sk_buff 2765 * @skb: sk_buff to steal the socket from 2766 * @refcounted: is set to true if the socket is reference-counted 2767 */ 2768 static inline struct sock * 2769 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2770 { 2771 if (skb->sk) { 2772 struct sock *sk = skb->sk; 2773 2774 *refcounted = true; 2775 if (skb_sk_is_prefetched(skb)) 2776 *refcounted = sk_is_refcounted(sk); 2777 skb->destructor = NULL; 2778 skb->sk = NULL; 2779 return sk; 2780 } 2781 *refcounted = false; 2782 return NULL; 2783 } 2784 2785 /* Checks if this SKB belongs to an HW offloaded socket 2786 * and whether any SW fallbacks are required based on dev. 2787 * Check decrypted mark in case skb_orphan() cleared socket. 2788 */ 2789 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2790 struct net_device *dev) 2791 { 2792 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2793 struct sock *sk = skb->sk; 2794 2795 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2796 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2797 #ifdef CONFIG_TLS_DEVICE 2798 } else if (unlikely(skb->decrypted)) { 2799 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2800 kfree_skb(skb); 2801 skb = NULL; 2802 #endif 2803 } 2804 #endif 2805 2806 return skb; 2807 } 2808 2809 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2810 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2811 */ 2812 static inline bool sk_listener(const struct sock *sk) 2813 { 2814 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2815 } 2816 2817 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2818 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2819 int type); 2820 2821 bool sk_ns_capable(const struct sock *sk, 2822 struct user_namespace *user_ns, int cap); 2823 bool sk_capable(const struct sock *sk, int cap); 2824 bool sk_net_capable(const struct sock *sk, int cap); 2825 2826 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2827 2828 /* Take into consideration the size of the struct sk_buff overhead in the 2829 * determination of these values, since that is non-constant across 2830 * platforms. This makes socket queueing behavior and performance 2831 * not depend upon such differences. 2832 */ 2833 #define _SK_MEM_PACKETS 256 2834 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2835 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2836 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2837 2838 extern __u32 sysctl_wmem_max; 2839 extern __u32 sysctl_rmem_max; 2840 2841 extern int sysctl_tstamp_allow_data; 2842 extern int sysctl_optmem_max; 2843 2844 extern __u32 sysctl_wmem_default; 2845 extern __u32 sysctl_rmem_default; 2846 2847 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2848 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2849 2850 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2851 { 2852 /* Does this proto have per netns sysctl_wmem ? */ 2853 if (proto->sysctl_wmem_offset) 2854 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2855 2856 return READ_ONCE(*proto->sysctl_wmem); 2857 } 2858 2859 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2860 { 2861 /* Does this proto have per netns sysctl_rmem ? */ 2862 if (proto->sysctl_rmem_offset) 2863 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2864 2865 return READ_ONCE(*proto->sysctl_rmem); 2866 } 2867 2868 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2869 * Some wifi drivers need to tweak it to get more chunks. 2870 * They can use this helper from their ndo_start_xmit() 2871 */ 2872 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2873 { 2874 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2875 return; 2876 WRITE_ONCE(sk->sk_pacing_shift, val); 2877 } 2878 2879 /* if a socket is bound to a device, check that the given device 2880 * index is either the same or that the socket is bound to an L3 2881 * master device and the given device index is also enslaved to 2882 * that L3 master 2883 */ 2884 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2885 { 2886 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2887 int mdif; 2888 2889 if (!bound_dev_if || bound_dev_if == dif) 2890 return true; 2891 2892 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2893 if (mdif && mdif == bound_dev_if) 2894 return true; 2895 2896 return false; 2897 } 2898 2899 void sock_def_readable(struct sock *sk); 2900 2901 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2902 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2903 int sock_set_timestamping(struct sock *sk, int optname, 2904 struct so_timestamping timestamping); 2905 2906 void sock_enable_timestamps(struct sock *sk); 2907 void sock_no_linger(struct sock *sk); 2908 void sock_set_keepalive(struct sock *sk); 2909 void sock_set_priority(struct sock *sk, u32 priority); 2910 void sock_set_rcvbuf(struct sock *sk, int val); 2911 void sock_set_mark(struct sock *sk, u32 val); 2912 void sock_set_reuseaddr(struct sock *sk); 2913 void sock_set_reuseport(struct sock *sk); 2914 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2915 2916 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2917 2918 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2919 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2920 sockptr_t optval, int optlen, bool old_timeval); 2921 2922 static inline bool sk_is_readable(struct sock *sk) 2923 { 2924 if (sk->sk_prot->sock_is_readable) 2925 return sk->sk_prot->sock_is_readable(sk); 2926 return false; 2927 } 2928 #endif /* _SOCK_H */ 2929