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_use_task_frag: allow sk_page_frag() to use current->task_frag. 322 * Sockets that can be used under memory reclaim should 323 * set this to false. 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_disconnects: number of disconnect operations performed on this sock 340 * @sk_state_change: callback to indicate change in the state of the sock 341 * @sk_data_ready: callback to indicate there is data to be processed 342 * @sk_write_space: callback to indicate there is bf sending space available 343 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 344 * @sk_backlog_rcv: callback to process the backlog 345 * @sk_validate_xmit_skb: ptr to an optional validate function 346 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 347 * @sk_reuseport_cb: reuseport group container 348 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 349 * @sk_rcu: used during RCU grace period 350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 353 * @sk_txtime_unused: unused txtime flags 354 * @ns_tracker: tracker for netns reference 355 * @sk_bind2_node: bind node in the bhash2 table 356 */ 357 struct sock { 358 /* 359 * Now struct inet_timewait_sock also uses sock_common, so please just 360 * don't add nothing before this first member (__sk_common) --acme 361 */ 362 struct sock_common __sk_common; 363 #define sk_node __sk_common.skc_node 364 #define sk_nulls_node __sk_common.skc_nulls_node 365 #define sk_refcnt __sk_common.skc_refcnt 366 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 367 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 368 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 369 #endif 370 371 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 372 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 373 #define sk_hash __sk_common.skc_hash 374 #define sk_portpair __sk_common.skc_portpair 375 #define sk_num __sk_common.skc_num 376 #define sk_dport __sk_common.skc_dport 377 #define sk_addrpair __sk_common.skc_addrpair 378 #define sk_daddr __sk_common.skc_daddr 379 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 380 #define sk_family __sk_common.skc_family 381 #define sk_state __sk_common.skc_state 382 #define sk_reuse __sk_common.skc_reuse 383 #define sk_reuseport __sk_common.skc_reuseport 384 #define sk_ipv6only __sk_common.skc_ipv6only 385 #define sk_net_refcnt __sk_common.skc_net_refcnt 386 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 387 #define sk_bind_node __sk_common.skc_bind_node 388 #define sk_prot __sk_common.skc_prot 389 #define sk_net __sk_common.skc_net 390 #define sk_v6_daddr __sk_common.skc_v6_daddr 391 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 392 #define sk_cookie __sk_common.skc_cookie 393 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 394 #define sk_flags __sk_common.skc_flags 395 #define sk_rxhash __sk_common.skc_rxhash 396 397 /* early demux fields */ 398 struct dst_entry __rcu *sk_rx_dst; 399 int sk_rx_dst_ifindex; 400 u32 sk_rx_dst_cookie; 401 402 socket_lock_t sk_lock; 403 atomic_t sk_drops; 404 int sk_rcvlowat; 405 struct sk_buff_head sk_error_queue; 406 struct sk_buff_head sk_receive_queue; 407 /* 408 * The backlog queue is special, it is always used with 409 * the per-socket spinlock held and requires low latency 410 * access. Therefore we special case it's implementation. 411 * Note : rmem_alloc is in this structure to fill a hole 412 * on 64bit arches, not because its logically part of 413 * backlog. 414 */ 415 struct { 416 atomic_t rmem_alloc; 417 int len; 418 struct sk_buff *head; 419 struct sk_buff *tail; 420 } sk_backlog; 421 422 #define sk_rmem_alloc sk_backlog.rmem_alloc 423 424 int sk_forward_alloc; 425 u32 sk_reserved_mem; 426 #ifdef CONFIG_NET_RX_BUSY_POLL 427 unsigned int sk_ll_usec; 428 /* ===== mostly read cache line ===== */ 429 unsigned int sk_napi_id; 430 #endif 431 int sk_rcvbuf; 432 int sk_disconnects; 433 434 struct sk_filter __rcu *sk_filter; 435 union { 436 struct socket_wq __rcu *sk_wq; 437 /* private: */ 438 struct socket_wq *sk_wq_raw; 439 /* public: */ 440 }; 441 #ifdef CONFIG_XFRM 442 struct xfrm_policy __rcu *sk_policy[2]; 443 #endif 444 445 struct dst_entry __rcu *sk_dst_cache; 446 atomic_t sk_omem_alloc; 447 int sk_sndbuf; 448 449 /* ===== cache line for TX ===== */ 450 int sk_wmem_queued; 451 refcount_t sk_wmem_alloc; 452 unsigned long sk_tsq_flags; 453 union { 454 struct sk_buff *sk_send_head; 455 struct rb_root tcp_rtx_queue; 456 }; 457 struct sk_buff_head sk_write_queue; 458 __s32 sk_peek_off; 459 int sk_write_pending; 460 __u32 sk_dst_pending_confirm; 461 u32 sk_pacing_status; /* see enum sk_pacing */ 462 long sk_sndtimeo; 463 struct timer_list sk_timer; 464 __u32 sk_priority; 465 __u32 sk_mark; 466 unsigned long sk_pacing_rate; /* bytes per second */ 467 unsigned long sk_max_pacing_rate; 468 struct page_frag sk_frag; 469 netdev_features_t sk_route_caps; 470 int sk_gso_type; 471 unsigned int sk_gso_max_size; 472 gfp_t sk_allocation; 473 __u32 sk_txhash; 474 475 /* 476 * Because of non atomicity rules, all 477 * changes are protected by socket lock. 478 */ 479 u8 sk_gso_disabled : 1, 480 sk_kern_sock : 1, 481 sk_no_check_tx : 1, 482 sk_no_check_rx : 1, 483 sk_userlocks : 4; 484 u8 sk_pacing_shift; 485 u16 sk_type; 486 u16 sk_protocol; 487 u16 sk_gso_max_segs; 488 unsigned long sk_lingertime; 489 struct proto *sk_prot_creator; 490 rwlock_t sk_callback_lock; 491 int sk_err, 492 sk_err_soft; 493 u32 sk_ack_backlog; 494 u32 sk_max_ack_backlog; 495 kuid_t sk_uid; 496 u8 sk_txrehash; 497 #ifdef CONFIG_NET_RX_BUSY_POLL 498 u8 sk_prefer_busy_poll; 499 u16 sk_busy_poll_budget; 500 #endif 501 spinlock_t sk_peer_lock; 502 int sk_bind_phc; 503 struct pid *sk_peer_pid; 504 const struct cred *sk_peer_cred; 505 506 long sk_rcvtimeo; 507 ktime_t sk_stamp; 508 #if BITS_PER_LONG==32 509 seqlock_t sk_stamp_seq; 510 #endif 511 atomic_t sk_tskey; 512 atomic_t sk_zckey; 513 u32 sk_tsflags; 514 u8 sk_shutdown; 515 516 u8 sk_clockid; 517 u8 sk_txtime_deadline_mode : 1, 518 sk_txtime_report_errors : 1, 519 sk_txtime_unused : 6; 520 bool sk_use_task_frag; 521 522 struct socket *sk_socket; 523 void *sk_user_data; 524 #ifdef CONFIG_SECURITY 525 void *sk_security; 526 #endif 527 struct sock_cgroup_data sk_cgrp_data; 528 struct mem_cgroup *sk_memcg; 529 void (*sk_state_change)(struct sock *sk); 530 void (*sk_data_ready)(struct sock *sk); 531 void (*sk_write_space)(struct sock *sk); 532 void (*sk_error_report)(struct sock *sk); 533 int (*sk_backlog_rcv)(struct sock *sk, 534 struct sk_buff *skb); 535 #ifdef CONFIG_SOCK_VALIDATE_XMIT 536 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 537 struct net_device *dev, 538 struct sk_buff *skb); 539 #endif 540 void (*sk_destruct)(struct sock *sk); 541 struct sock_reuseport __rcu *sk_reuseport_cb; 542 #ifdef CONFIG_BPF_SYSCALL 543 struct bpf_local_storage __rcu *sk_bpf_storage; 544 #endif 545 struct rcu_head sk_rcu; 546 netns_tracker ns_tracker; 547 struct hlist_node sk_bind2_node; 548 }; 549 550 enum sk_pacing { 551 SK_PACING_NONE = 0, 552 SK_PACING_NEEDED = 1, 553 SK_PACING_FQ = 2, 554 }; 555 556 /* flag bits in sk_user_data 557 * 558 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 559 * not be suitable for copying when cloning the socket. For instance, 560 * it can point to a reference counted object. sk_user_data bottom 561 * bit is set if pointer must not be copied. 562 * 563 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 564 * managed/owned by a BPF reuseport array. This bit should be set 565 * when sk_user_data's sk is added to the bpf's reuseport_array. 566 * 567 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 568 * sk_user_data points to psock type. This bit should be set 569 * when sk_user_data is assigned to a psock object. 570 */ 571 #define SK_USER_DATA_NOCOPY 1UL 572 #define SK_USER_DATA_BPF 2UL 573 #define SK_USER_DATA_PSOCK 4UL 574 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 575 SK_USER_DATA_PSOCK) 576 577 /** 578 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 579 * @sk: socket 580 */ 581 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 582 { 583 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 584 } 585 586 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 587 588 /** 589 * __locked_read_sk_user_data_with_flags - return the pointer 590 * only if argument flags all has been set in sk_user_data. Otherwise 591 * return NULL 592 * 593 * @sk: socket 594 * @flags: flag bits 595 * 596 * The caller must be holding sk->sk_callback_lock. 597 */ 598 static inline void * 599 __locked_read_sk_user_data_with_flags(const struct sock *sk, 600 uintptr_t flags) 601 { 602 uintptr_t sk_user_data = 603 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 604 lockdep_is_held(&sk->sk_callback_lock)); 605 606 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 607 608 if ((sk_user_data & flags) == flags) 609 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 610 return NULL; 611 } 612 613 /** 614 * __rcu_dereference_sk_user_data_with_flags - return the pointer 615 * only if argument flags all has been set in sk_user_data. Otherwise 616 * return NULL 617 * 618 * @sk: socket 619 * @flags: flag bits 620 */ 621 static inline void * 622 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 623 uintptr_t flags) 624 { 625 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 626 627 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 628 629 if ((sk_user_data & flags) == flags) 630 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 631 return NULL; 632 } 633 634 #define rcu_dereference_sk_user_data(sk) \ 635 __rcu_dereference_sk_user_data_with_flags(sk, 0) 636 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 637 ({ \ 638 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 639 __tmp2 = (uintptr_t)(flags); \ 640 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 641 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 642 rcu_assign_pointer(__sk_user_data((sk)), \ 643 __tmp1 | __tmp2); \ 644 }) 645 #define rcu_assign_sk_user_data(sk, ptr) \ 646 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 647 648 static inline 649 struct net *sock_net(const struct sock *sk) 650 { 651 return read_pnet(&sk->sk_net); 652 } 653 654 static inline 655 void sock_net_set(struct sock *sk, struct net *net) 656 { 657 write_pnet(&sk->sk_net, net); 658 } 659 660 /* 661 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 662 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 663 * on a socket means that the socket will reuse everybody else's port 664 * without looking at the other's sk_reuse value. 665 */ 666 667 #define SK_NO_REUSE 0 668 #define SK_CAN_REUSE 1 669 #define SK_FORCE_REUSE 2 670 671 int sk_set_peek_off(struct sock *sk, int val); 672 673 static inline int sk_peek_offset(const struct sock *sk, int flags) 674 { 675 if (unlikely(flags & MSG_PEEK)) { 676 return READ_ONCE(sk->sk_peek_off); 677 } 678 679 return 0; 680 } 681 682 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 683 { 684 s32 off = READ_ONCE(sk->sk_peek_off); 685 686 if (unlikely(off >= 0)) { 687 off = max_t(s32, off - val, 0); 688 WRITE_ONCE(sk->sk_peek_off, off); 689 } 690 } 691 692 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 693 { 694 sk_peek_offset_bwd(sk, -val); 695 } 696 697 /* 698 * Hashed lists helper routines 699 */ 700 static inline struct sock *sk_entry(const struct hlist_node *node) 701 { 702 return hlist_entry(node, struct sock, sk_node); 703 } 704 705 static inline struct sock *__sk_head(const struct hlist_head *head) 706 { 707 return hlist_entry(head->first, struct sock, sk_node); 708 } 709 710 static inline struct sock *sk_head(const struct hlist_head *head) 711 { 712 return hlist_empty(head) ? NULL : __sk_head(head); 713 } 714 715 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 716 { 717 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 718 } 719 720 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 721 { 722 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 723 } 724 725 static inline struct sock *sk_next(const struct sock *sk) 726 { 727 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 728 } 729 730 static inline struct sock *sk_nulls_next(const struct sock *sk) 731 { 732 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 733 hlist_nulls_entry(sk->sk_nulls_node.next, 734 struct sock, sk_nulls_node) : 735 NULL; 736 } 737 738 static inline bool sk_unhashed(const struct sock *sk) 739 { 740 return hlist_unhashed(&sk->sk_node); 741 } 742 743 static inline bool sk_hashed(const struct sock *sk) 744 { 745 return !sk_unhashed(sk); 746 } 747 748 static inline void sk_node_init(struct hlist_node *node) 749 { 750 node->pprev = NULL; 751 } 752 753 static inline void __sk_del_node(struct sock *sk) 754 { 755 __hlist_del(&sk->sk_node); 756 } 757 758 /* NB: equivalent to hlist_del_init_rcu */ 759 static inline bool __sk_del_node_init(struct sock *sk) 760 { 761 if (sk_hashed(sk)) { 762 __sk_del_node(sk); 763 sk_node_init(&sk->sk_node); 764 return true; 765 } 766 return false; 767 } 768 769 /* Grab socket reference count. This operation is valid only 770 when sk is ALREADY grabbed f.e. it is found in hash table 771 or a list and the lookup is made under lock preventing hash table 772 modifications. 773 */ 774 775 static __always_inline void sock_hold(struct sock *sk) 776 { 777 refcount_inc(&sk->sk_refcnt); 778 } 779 780 /* Ungrab socket in the context, which assumes that socket refcnt 781 cannot hit zero, f.e. it is true in context of any socketcall. 782 */ 783 static __always_inline void __sock_put(struct sock *sk) 784 { 785 refcount_dec(&sk->sk_refcnt); 786 } 787 788 static inline bool sk_del_node_init(struct sock *sk) 789 { 790 bool rc = __sk_del_node_init(sk); 791 792 if (rc) { 793 /* paranoid for a while -acme */ 794 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 795 __sock_put(sk); 796 } 797 return rc; 798 } 799 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 800 801 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 802 { 803 if (sk_hashed(sk)) { 804 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 805 return true; 806 } 807 return false; 808 } 809 810 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 811 { 812 bool rc = __sk_nulls_del_node_init_rcu(sk); 813 814 if (rc) { 815 /* paranoid for a while -acme */ 816 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 817 __sock_put(sk); 818 } 819 return rc; 820 } 821 822 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 823 { 824 hlist_add_head(&sk->sk_node, list); 825 } 826 827 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 828 { 829 sock_hold(sk); 830 __sk_add_node(sk, list); 831 } 832 833 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 834 { 835 sock_hold(sk); 836 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 837 sk->sk_family == AF_INET6) 838 hlist_add_tail_rcu(&sk->sk_node, list); 839 else 840 hlist_add_head_rcu(&sk->sk_node, list); 841 } 842 843 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 844 { 845 sock_hold(sk); 846 hlist_add_tail_rcu(&sk->sk_node, list); 847 } 848 849 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 850 { 851 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 852 } 853 854 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 855 { 856 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 857 } 858 859 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 860 { 861 sock_hold(sk); 862 __sk_nulls_add_node_rcu(sk, list); 863 } 864 865 static inline void __sk_del_bind_node(struct sock *sk) 866 { 867 __hlist_del(&sk->sk_bind_node); 868 } 869 870 static inline void sk_add_bind_node(struct sock *sk, 871 struct hlist_head *list) 872 { 873 hlist_add_head(&sk->sk_bind_node, list); 874 } 875 876 static inline void __sk_del_bind2_node(struct sock *sk) 877 { 878 __hlist_del(&sk->sk_bind2_node); 879 } 880 881 static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list) 882 { 883 hlist_add_head(&sk->sk_bind2_node, list); 884 } 885 886 #define sk_for_each(__sk, list) \ 887 hlist_for_each_entry(__sk, list, sk_node) 888 #define sk_for_each_rcu(__sk, list) \ 889 hlist_for_each_entry_rcu(__sk, list, sk_node) 890 #define sk_nulls_for_each(__sk, node, list) \ 891 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 892 #define sk_nulls_for_each_rcu(__sk, node, list) \ 893 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 894 #define sk_for_each_from(__sk) \ 895 hlist_for_each_entry_from(__sk, sk_node) 896 #define sk_nulls_for_each_from(__sk, node) \ 897 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 898 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 899 #define sk_for_each_safe(__sk, tmp, list) \ 900 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 901 #define sk_for_each_bound(__sk, list) \ 902 hlist_for_each_entry(__sk, list, sk_bind_node) 903 #define sk_for_each_bound_bhash2(__sk, list) \ 904 hlist_for_each_entry(__sk, list, sk_bind2_node) 905 906 /** 907 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 908 * @tpos: the type * to use as a loop cursor. 909 * @pos: the &struct hlist_node to use as a loop cursor. 910 * @head: the head for your list. 911 * @offset: offset of hlist_node within the struct. 912 * 913 */ 914 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 915 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 916 pos != NULL && \ 917 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 918 pos = rcu_dereference(hlist_next_rcu(pos))) 919 920 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 921 { 922 /* Careful only use this in a context where these parameters 923 * can not change and must all be valid, such as recvmsg from 924 * userspace. 925 */ 926 return sk->sk_socket->file->f_cred->user_ns; 927 } 928 929 /* Sock flags */ 930 enum sock_flags { 931 SOCK_DEAD, 932 SOCK_DONE, 933 SOCK_URGINLINE, 934 SOCK_KEEPOPEN, 935 SOCK_LINGER, 936 SOCK_DESTROY, 937 SOCK_BROADCAST, 938 SOCK_TIMESTAMP, 939 SOCK_ZAPPED, 940 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 941 SOCK_DBG, /* %SO_DEBUG setting */ 942 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 943 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 944 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 945 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 946 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 947 SOCK_FASYNC, /* fasync() active */ 948 SOCK_RXQ_OVFL, 949 SOCK_ZEROCOPY, /* buffers from userspace */ 950 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 951 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 952 * Will use last 4 bytes of packet sent from 953 * user-space instead. 954 */ 955 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 956 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 957 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 958 SOCK_TXTIME, 959 SOCK_XDP, /* XDP is attached */ 960 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 961 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 962 }; 963 964 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 965 966 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 967 { 968 nsk->sk_flags = osk->sk_flags; 969 } 970 971 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 972 { 973 __set_bit(flag, &sk->sk_flags); 974 } 975 976 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 977 { 978 __clear_bit(flag, &sk->sk_flags); 979 } 980 981 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 982 int valbool) 983 { 984 if (valbool) 985 sock_set_flag(sk, bit); 986 else 987 sock_reset_flag(sk, bit); 988 } 989 990 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 991 { 992 return test_bit(flag, &sk->sk_flags); 993 } 994 995 #ifdef CONFIG_NET 996 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 997 static inline int sk_memalloc_socks(void) 998 { 999 return static_branch_unlikely(&memalloc_socks_key); 1000 } 1001 1002 void __receive_sock(struct file *file); 1003 #else 1004 1005 static inline int sk_memalloc_socks(void) 1006 { 1007 return 0; 1008 } 1009 1010 static inline void __receive_sock(struct file *file) 1011 { } 1012 #endif 1013 1014 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1015 { 1016 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1017 } 1018 1019 static inline void sk_acceptq_removed(struct sock *sk) 1020 { 1021 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1022 } 1023 1024 static inline void sk_acceptq_added(struct sock *sk) 1025 { 1026 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1027 } 1028 1029 /* Note: If you think the test should be: 1030 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1031 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1032 */ 1033 static inline bool sk_acceptq_is_full(const struct sock *sk) 1034 { 1035 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1036 } 1037 1038 /* 1039 * Compute minimal free write space needed to queue new packets. 1040 */ 1041 static inline int sk_stream_min_wspace(const struct sock *sk) 1042 { 1043 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1044 } 1045 1046 static inline int sk_stream_wspace(const struct sock *sk) 1047 { 1048 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1049 } 1050 1051 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1052 { 1053 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1054 } 1055 1056 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1057 { 1058 /* Paired with lockless reads of sk->sk_forward_alloc */ 1059 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1060 } 1061 1062 void sk_stream_write_space(struct sock *sk); 1063 1064 /* OOB backlog add */ 1065 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1066 { 1067 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1068 skb_dst_force(skb); 1069 1070 if (!sk->sk_backlog.tail) 1071 WRITE_ONCE(sk->sk_backlog.head, skb); 1072 else 1073 sk->sk_backlog.tail->next = skb; 1074 1075 WRITE_ONCE(sk->sk_backlog.tail, skb); 1076 skb->next = NULL; 1077 } 1078 1079 /* 1080 * Take into account size of receive queue and backlog queue 1081 * Do not take into account this skb truesize, 1082 * to allow even a single big packet to come. 1083 */ 1084 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1085 { 1086 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1087 1088 return qsize > limit; 1089 } 1090 1091 /* The per-socket spinlock must be held here. */ 1092 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1093 unsigned int limit) 1094 { 1095 if (sk_rcvqueues_full(sk, limit)) 1096 return -ENOBUFS; 1097 1098 /* 1099 * If the skb was allocated from pfmemalloc reserves, only 1100 * allow SOCK_MEMALLOC sockets to use it as this socket is 1101 * helping free memory 1102 */ 1103 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1104 return -ENOMEM; 1105 1106 __sk_add_backlog(sk, skb); 1107 sk->sk_backlog.len += skb->truesize; 1108 return 0; 1109 } 1110 1111 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1112 1113 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1114 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1115 1116 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1117 { 1118 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1119 return __sk_backlog_rcv(sk, skb); 1120 1121 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1122 tcp_v6_do_rcv, 1123 tcp_v4_do_rcv, 1124 sk, skb); 1125 } 1126 1127 static inline void sk_incoming_cpu_update(struct sock *sk) 1128 { 1129 int cpu = raw_smp_processor_id(); 1130 1131 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1132 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1133 } 1134 1135 static inline void sock_rps_record_flow_hash(__u32 hash) 1136 { 1137 #ifdef CONFIG_RPS 1138 struct rps_sock_flow_table *sock_flow_table; 1139 1140 rcu_read_lock(); 1141 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1142 rps_record_sock_flow(sock_flow_table, hash); 1143 rcu_read_unlock(); 1144 #endif 1145 } 1146 1147 static inline void sock_rps_record_flow(const struct sock *sk) 1148 { 1149 #ifdef CONFIG_RPS 1150 if (static_branch_unlikely(&rfs_needed)) { 1151 /* Reading sk->sk_rxhash might incur an expensive cache line 1152 * miss. 1153 * 1154 * TCP_ESTABLISHED does cover almost all states where RFS 1155 * might be useful, and is cheaper [1] than testing : 1156 * IPv4: inet_sk(sk)->inet_daddr 1157 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1158 * OR an additional socket flag 1159 * [1] : sk_state and sk_prot are in the same cache line. 1160 */ 1161 if (sk->sk_state == TCP_ESTABLISHED) { 1162 /* This READ_ONCE() is paired with the WRITE_ONCE() 1163 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash(). 1164 */ 1165 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash)); 1166 } 1167 } 1168 #endif 1169 } 1170 1171 static inline void sock_rps_save_rxhash(struct sock *sk, 1172 const struct sk_buff *skb) 1173 { 1174 #ifdef CONFIG_RPS 1175 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1176 * here, and another one in sock_rps_record_flow(). 1177 */ 1178 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1179 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1180 #endif 1181 } 1182 1183 static inline void sock_rps_reset_rxhash(struct sock *sk) 1184 { 1185 #ifdef CONFIG_RPS 1186 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1187 WRITE_ONCE(sk->sk_rxhash, 0); 1188 #endif 1189 } 1190 1191 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1192 ({ int __rc, __dis = __sk->sk_disconnects; \ 1193 release_sock(__sk); \ 1194 __rc = __condition; \ 1195 if (!__rc) { \ 1196 *(__timeo) = wait_woken(__wait, \ 1197 TASK_INTERRUPTIBLE, \ 1198 *(__timeo)); \ 1199 } \ 1200 sched_annotate_sleep(); \ 1201 lock_sock(__sk); \ 1202 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1203 __rc; \ 1204 }) 1205 1206 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1207 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1208 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1209 int sk_stream_error(struct sock *sk, int flags, int err); 1210 void sk_stream_kill_queues(struct sock *sk); 1211 void sk_set_memalloc(struct sock *sk); 1212 void sk_clear_memalloc(struct sock *sk); 1213 1214 void __sk_flush_backlog(struct sock *sk); 1215 1216 static inline bool sk_flush_backlog(struct sock *sk) 1217 { 1218 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1219 __sk_flush_backlog(sk); 1220 return true; 1221 } 1222 return false; 1223 } 1224 1225 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1226 1227 struct request_sock_ops; 1228 struct timewait_sock_ops; 1229 struct inet_hashinfo; 1230 struct raw_hashinfo; 1231 struct smc_hashinfo; 1232 struct module; 1233 struct sk_psock; 1234 1235 /* 1236 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1237 * un-modified. Special care is taken when initializing object to zero. 1238 */ 1239 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1240 { 1241 if (offsetof(struct sock, sk_node.next) != 0) 1242 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1243 memset(&sk->sk_node.pprev, 0, 1244 size - offsetof(struct sock, sk_node.pprev)); 1245 } 1246 1247 /* Networking protocol blocks we attach to sockets. 1248 * socket layer -> transport layer interface 1249 */ 1250 struct proto { 1251 void (*close)(struct sock *sk, 1252 long timeout); 1253 int (*pre_connect)(struct sock *sk, 1254 struct sockaddr *uaddr, 1255 int addr_len); 1256 int (*connect)(struct sock *sk, 1257 struct sockaddr *uaddr, 1258 int addr_len); 1259 int (*disconnect)(struct sock *sk, int flags); 1260 1261 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1262 bool kern); 1263 1264 int (*ioctl)(struct sock *sk, int cmd, 1265 int *karg); 1266 int (*init)(struct sock *sk); 1267 void (*destroy)(struct sock *sk); 1268 void (*shutdown)(struct sock *sk, int how); 1269 int (*setsockopt)(struct sock *sk, int level, 1270 int optname, sockptr_t optval, 1271 unsigned int optlen); 1272 int (*getsockopt)(struct sock *sk, int level, 1273 int optname, char __user *optval, 1274 int __user *option); 1275 void (*keepalive)(struct sock *sk, int valbool); 1276 #ifdef CONFIG_COMPAT 1277 int (*compat_ioctl)(struct sock *sk, 1278 unsigned int cmd, unsigned long arg); 1279 #endif 1280 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1281 size_t len); 1282 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1283 size_t len, int flags, int *addr_len); 1284 void (*splice_eof)(struct socket *sock); 1285 int (*bind)(struct sock *sk, 1286 struct sockaddr *addr, int addr_len); 1287 int (*bind_add)(struct sock *sk, 1288 struct sockaddr *addr, int addr_len); 1289 1290 int (*backlog_rcv) (struct sock *sk, 1291 struct sk_buff *skb); 1292 bool (*bpf_bypass_getsockopt)(int level, 1293 int optname); 1294 1295 void (*release_cb)(struct sock *sk); 1296 1297 /* Keeping track of sk's, looking them up, and port selection methods. */ 1298 int (*hash)(struct sock *sk); 1299 void (*unhash)(struct sock *sk); 1300 void (*rehash)(struct sock *sk); 1301 int (*get_port)(struct sock *sk, unsigned short snum); 1302 void (*put_port)(struct sock *sk); 1303 #ifdef CONFIG_BPF_SYSCALL 1304 int (*psock_update_sk_prot)(struct sock *sk, 1305 struct sk_psock *psock, 1306 bool restore); 1307 #endif 1308 1309 /* Keeping track of sockets in use */ 1310 #ifdef CONFIG_PROC_FS 1311 unsigned int inuse_idx; 1312 #endif 1313 1314 #if IS_ENABLED(CONFIG_MPTCP) 1315 int (*forward_alloc_get)(const struct sock *sk); 1316 #endif 1317 1318 bool (*stream_memory_free)(const struct sock *sk, int wake); 1319 bool (*sock_is_readable)(struct sock *sk); 1320 /* Memory pressure */ 1321 void (*enter_memory_pressure)(struct sock *sk); 1322 void (*leave_memory_pressure)(struct sock *sk); 1323 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1324 int __percpu *per_cpu_fw_alloc; 1325 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1326 1327 /* 1328 * Pressure flag: try to collapse. 1329 * Technical note: it is used by multiple contexts non atomically. 1330 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1331 * All the __sk_mem_schedule() is of this nature: accounting 1332 * is strict, actions are advisory and have some latency. 1333 */ 1334 unsigned long *memory_pressure; 1335 long *sysctl_mem; 1336 1337 int *sysctl_wmem; 1338 int *sysctl_rmem; 1339 u32 sysctl_wmem_offset; 1340 u32 sysctl_rmem_offset; 1341 1342 int max_header; 1343 bool no_autobind; 1344 1345 struct kmem_cache *slab; 1346 unsigned int obj_size; 1347 unsigned int ipv6_pinfo_offset; 1348 slab_flags_t slab_flags; 1349 unsigned int useroffset; /* Usercopy region offset */ 1350 unsigned int usersize; /* Usercopy region size */ 1351 1352 unsigned int __percpu *orphan_count; 1353 1354 struct request_sock_ops *rsk_prot; 1355 struct timewait_sock_ops *twsk_prot; 1356 1357 union { 1358 struct inet_hashinfo *hashinfo; 1359 struct udp_table *udp_table; 1360 struct raw_hashinfo *raw_hash; 1361 struct smc_hashinfo *smc_hash; 1362 } h; 1363 1364 struct module *owner; 1365 1366 char name[32]; 1367 1368 struct list_head node; 1369 int (*diag_destroy)(struct sock *sk, int err); 1370 } __randomize_layout; 1371 1372 int proto_register(struct proto *prot, int alloc_slab); 1373 void proto_unregister(struct proto *prot); 1374 int sock_load_diag_module(int family, int protocol); 1375 1376 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1377 1378 static inline int sk_forward_alloc_get(const struct sock *sk) 1379 { 1380 #if IS_ENABLED(CONFIG_MPTCP) 1381 if (sk->sk_prot->forward_alloc_get) 1382 return sk->sk_prot->forward_alloc_get(sk); 1383 #endif 1384 return READ_ONCE(sk->sk_forward_alloc); 1385 } 1386 1387 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1388 { 1389 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1390 return false; 1391 1392 return sk->sk_prot->stream_memory_free ? 1393 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1394 tcp_stream_memory_free, sk, wake) : true; 1395 } 1396 1397 static inline bool sk_stream_memory_free(const struct sock *sk) 1398 { 1399 return __sk_stream_memory_free(sk, 0); 1400 } 1401 1402 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1403 { 1404 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1405 __sk_stream_memory_free(sk, wake); 1406 } 1407 1408 static inline bool sk_stream_is_writeable(const struct sock *sk) 1409 { 1410 return __sk_stream_is_writeable(sk, 0); 1411 } 1412 1413 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1414 struct cgroup *ancestor) 1415 { 1416 #ifdef CONFIG_SOCK_CGROUP_DATA 1417 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1418 ancestor); 1419 #else 1420 return -ENOTSUPP; 1421 #endif 1422 } 1423 1424 static inline bool sk_has_memory_pressure(const struct sock *sk) 1425 { 1426 return sk->sk_prot->memory_pressure != NULL; 1427 } 1428 1429 static inline bool sk_under_global_memory_pressure(const struct sock *sk) 1430 { 1431 return sk->sk_prot->memory_pressure && 1432 !!READ_ONCE(*sk->sk_prot->memory_pressure); 1433 } 1434 1435 static inline bool sk_under_memory_pressure(const struct sock *sk) 1436 { 1437 if (!sk->sk_prot->memory_pressure) 1438 return false; 1439 1440 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1441 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1442 return true; 1443 1444 return !!READ_ONCE(*sk->sk_prot->memory_pressure); 1445 } 1446 1447 static inline long 1448 proto_memory_allocated(const struct proto *prot) 1449 { 1450 return max(0L, atomic_long_read(prot->memory_allocated)); 1451 } 1452 1453 static inline long 1454 sk_memory_allocated(const struct sock *sk) 1455 { 1456 return proto_memory_allocated(sk->sk_prot); 1457 } 1458 1459 /* 1 MB per cpu, in page units */ 1460 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1461 extern int sysctl_mem_pcpu_rsv; 1462 1463 static inline void proto_memory_pcpu_drain(struct proto *proto) 1464 { 1465 int val = this_cpu_xchg(*proto->per_cpu_fw_alloc, 0); 1466 1467 if (val) 1468 atomic_long_add(val, proto->memory_allocated); 1469 } 1470 1471 static inline void 1472 sk_memory_allocated_add(const struct sock *sk, int val) 1473 { 1474 struct proto *proto = sk->sk_prot; 1475 1476 val = this_cpu_add_return(*proto->per_cpu_fw_alloc, val); 1477 1478 if (unlikely(val >= READ_ONCE(sysctl_mem_pcpu_rsv))) 1479 proto_memory_pcpu_drain(proto); 1480 } 1481 1482 static inline void 1483 sk_memory_allocated_sub(const struct sock *sk, int val) 1484 { 1485 struct proto *proto = sk->sk_prot; 1486 1487 val = this_cpu_sub_return(*proto->per_cpu_fw_alloc, val); 1488 1489 if (unlikely(val <= -READ_ONCE(sysctl_mem_pcpu_rsv))) 1490 proto_memory_pcpu_drain(proto); 1491 } 1492 1493 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1494 1495 static inline void sk_sockets_allocated_dec(struct sock *sk) 1496 { 1497 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1498 SK_ALLOC_PERCPU_COUNTER_BATCH); 1499 } 1500 1501 static inline void sk_sockets_allocated_inc(struct sock *sk) 1502 { 1503 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1504 SK_ALLOC_PERCPU_COUNTER_BATCH); 1505 } 1506 1507 static inline u64 1508 sk_sockets_allocated_read_positive(struct sock *sk) 1509 { 1510 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1511 } 1512 1513 static inline int 1514 proto_sockets_allocated_sum_positive(struct proto *prot) 1515 { 1516 return percpu_counter_sum_positive(prot->sockets_allocated); 1517 } 1518 1519 static inline bool 1520 proto_memory_pressure(struct proto *prot) 1521 { 1522 if (!prot->memory_pressure) 1523 return false; 1524 return !!READ_ONCE(*prot->memory_pressure); 1525 } 1526 1527 1528 #ifdef CONFIG_PROC_FS 1529 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1530 struct prot_inuse { 1531 int all; 1532 int val[PROTO_INUSE_NR]; 1533 }; 1534 1535 static inline void sock_prot_inuse_add(const struct net *net, 1536 const struct proto *prot, int val) 1537 { 1538 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1539 } 1540 1541 static inline void sock_inuse_add(const struct net *net, int val) 1542 { 1543 this_cpu_add(net->core.prot_inuse->all, val); 1544 } 1545 1546 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1547 int sock_inuse_get(struct net *net); 1548 #else 1549 static inline void sock_prot_inuse_add(const struct net *net, 1550 const struct proto *prot, int val) 1551 { 1552 } 1553 1554 static inline void sock_inuse_add(const struct net *net, int val) 1555 { 1556 } 1557 #endif 1558 1559 1560 /* With per-bucket locks this operation is not-atomic, so that 1561 * this version is not worse. 1562 */ 1563 static inline int __sk_prot_rehash(struct sock *sk) 1564 { 1565 sk->sk_prot->unhash(sk); 1566 return sk->sk_prot->hash(sk); 1567 } 1568 1569 /* About 10 seconds */ 1570 #define SOCK_DESTROY_TIME (10*HZ) 1571 1572 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1573 #define PROT_SOCK 1024 1574 1575 #define SHUTDOWN_MASK 3 1576 #define RCV_SHUTDOWN 1 1577 #define SEND_SHUTDOWN 2 1578 1579 #define SOCK_BINDADDR_LOCK 4 1580 #define SOCK_BINDPORT_LOCK 8 1581 1582 struct socket_alloc { 1583 struct socket socket; 1584 struct inode vfs_inode; 1585 }; 1586 1587 static inline struct socket *SOCKET_I(struct inode *inode) 1588 { 1589 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1590 } 1591 1592 static inline struct inode *SOCK_INODE(struct socket *socket) 1593 { 1594 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1595 } 1596 1597 /* 1598 * Functions for memory accounting 1599 */ 1600 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1601 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1602 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1603 void __sk_mem_reclaim(struct sock *sk, int amount); 1604 1605 #define SK_MEM_SEND 0 1606 #define SK_MEM_RECV 1 1607 1608 /* sysctl_mem values are in pages */ 1609 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1610 { 1611 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1612 } 1613 1614 static inline int sk_mem_pages(int amt) 1615 { 1616 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1617 } 1618 1619 static inline bool sk_has_account(struct sock *sk) 1620 { 1621 /* return true if protocol supports memory accounting */ 1622 return !!sk->sk_prot->memory_allocated; 1623 } 1624 1625 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1626 { 1627 int delta; 1628 1629 if (!sk_has_account(sk)) 1630 return true; 1631 delta = size - sk->sk_forward_alloc; 1632 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1633 } 1634 1635 static inline bool 1636 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1637 { 1638 int delta; 1639 1640 if (!sk_has_account(sk)) 1641 return true; 1642 delta = size - sk->sk_forward_alloc; 1643 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1644 skb_pfmemalloc(skb); 1645 } 1646 1647 static inline int sk_unused_reserved_mem(const struct sock *sk) 1648 { 1649 int unused_mem; 1650 1651 if (likely(!sk->sk_reserved_mem)) 1652 return 0; 1653 1654 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1655 atomic_read(&sk->sk_rmem_alloc); 1656 1657 return unused_mem > 0 ? unused_mem : 0; 1658 } 1659 1660 static inline void sk_mem_reclaim(struct sock *sk) 1661 { 1662 int reclaimable; 1663 1664 if (!sk_has_account(sk)) 1665 return; 1666 1667 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1668 1669 if (reclaimable >= (int)PAGE_SIZE) 1670 __sk_mem_reclaim(sk, reclaimable); 1671 } 1672 1673 static inline void sk_mem_reclaim_final(struct sock *sk) 1674 { 1675 sk->sk_reserved_mem = 0; 1676 sk_mem_reclaim(sk); 1677 } 1678 1679 static inline void sk_mem_charge(struct sock *sk, int size) 1680 { 1681 if (!sk_has_account(sk)) 1682 return; 1683 sk_forward_alloc_add(sk, -size); 1684 } 1685 1686 static inline void sk_mem_uncharge(struct sock *sk, int size) 1687 { 1688 if (!sk_has_account(sk)) 1689 return; 1690 sk_forward_alloc_add(sk, size); 1691 sk_mem_reclaim(sk); 1692 } 1693 1694 /* 1695 * Macro so as to not evaluate some arguments when 1696 * lockdep is not enabled. 1697 * 1698 * Mark both the sk_lock and the sk_lock.slock as a 1699 * per-address-family lock class. 1700 */ 1701 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1702 do { \ 1703 sk->sk_lock.owned = 0; \ 1704 init_waitqueue_head(&sk->sk_lock.wq); \ 1705 spin_lock_init(&(sk)->sk_lock.slock); \ 1706 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1707 sizeof((sk)->sk_lock)); \ 1708 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1709 (skey), (sname)); \ 1710 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1711 } while (0) 1712 1713 static inline bool lockdep_sock_is_held(const struct sock *sk) 1714 { 1715 return lockdep_is_held(&sk->sk_lock) || 1716 lockdep_is_held(&sk->sk_lock.slock); 1717 } 1718 1719 void lock_sock_nested(struct sock *sk, int subclass); 1720 1721 static inline void lock_sock(struct sock *sk) 1722 { 1723 lock_sock_nested(sk, 0); 1724 } 1725 1726 void __lock_sock(struct sock *sk); 1727 void __release_sock(struct sock *sk); 1728 void release_sock(struct sock *sk); 1729 1730 /* BH context may only use the following locking interface. */ 1731 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1732 #define bh_lock_sock_nested(__sk) \ 1733 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1734 SINGLE_DEPTH_NESTING) 1735 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1736 1737 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1738 1739 /** 1740 * lock_sock_fast - fast version of lock_sock 1741 * @sk: socket 1742 * 1743 * This version should be used for very small section, where process wont block 1744 * return false if fast path is taken: 1745 * 1746 * sk_lock.slock locked, owned = 0, BH disabled 1747 * 1748 * return true if slow path is taken: 1749 * 1750 * sk_lock.slock unlocked, owned = 1, BH enabled 1751 */ 1752 static inline bool lock_sock_fast(struct sock *sk) 1753 { 1754 /* The sk_lock has mutex_lock() semantics here. */ 1755 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1756 1757 return __lock_sock_fast(sk); 1758 } 1759 1760 /* fast socket lock variant for caller already holding a [different] socket lock */ 1761 static inline bool lock_sock_fast_nested(struct sock *sk) 1762 { 1763 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1764 1765 return __lock_sock_fast(sk); 1766 } 1767 1768 /** 1769 * unlock_sock_fast - complement of lock_sock_fast 1770 * @sk: socket 1771 * @slow: slow mode 1772 * 1773 * fast unlock socket for user context. 1774 * If slow mode is on, we call regular release_sock() 1775 */ 1776 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1777 __releases(&sk->sk_lock.slock) 1778 { 1779 if (slow) { 1780 release_sock(sk); 1781 __release(&sk->sk_lock.slock); 1782 } else { 1783 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1784 spin_unlock_bh(&sk->sk_lock.slock); 1785 } 1786 } 1787 1788 void sockopt_lock_sock(struct sock *sk); 1789 void sockopt_release_sock(struct sock *sk); 1790 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1791 bool sockopt_capable(int cap); 1792 1793 /* Used by processes to "lock" a socket state, so that 1794 * interrupts and bottom half handlers won't change it 1795 * from under us. It essentially blocks any incoming 1796 * packets, so that we won't get any new data or any 1797 * packets that change the state of the socket. 1798 * 1799 * While locked, BH processing will add new packets to 1800 * the backlog queue. This queue is processed by the 1801 * owner of the socket lock right before it is released. 1802 * 1803 * Since ~2.3.5 it is also exclusive sleep lock serializing 1804 * accesses from user process context. 1805 */ 1806 1807 static inline void sock_owned_by_me(const struct sock *sk) 1808 { 1809 #ifdef CONFIG_LOCKDEP 1810 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1811 #endif 1812 } 1813 1814 static inline void sock_not_owned_by_me(const struct sock *sk) 1815 { 1816 #ifdef CONFIG_LOCKDEP 1817 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1818 #endif 1819 } 1820 1821 static inline bool sock_owned_by_user(const struct sock *sk) 1822 { 1823 sock_owned_by_me(sk); 1824 return sk->sk_lock.owned; 1825 } 1826 1827 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1828 { 1829 return sk->sk_lock.owned; 1830 } 1831 1832 static inline void sock_release_ownership(struct sock *sk) 1833 { 1834 if (sock_owned_by_user_nocheck(sk)) { 1835 sk->sk_lock.owned = 0; 1836 1837 /* The sk_lock has mutex_unlock() semantics: */ 1838 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1839 } 1840 } 1841 1842 /* no reclassification while locks are held */ 1843 static inline bool sock_allow_reclassification(const struct sock *csk) 1844 { 1845 struct sock *sk = (struct sock *)csk; 1846 1847 return !sock_owned_by_user_nocheck(sk) && 1848 !spin_is_locked(&sk->sk_lock.slock); 1849 } 1850 1851 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1852 struct proto *prot, int kern); 1853 void sk_free(struct sock *sk); 1854 void sk_destruct(struct sock *sk); 1855 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1856 void sk_free_unlock_clone(struct sock *sk); 1857 1858 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1859 gfp_t priority); 1860 void __sock_wfree(struct sk_buff *skb); 1861 void sock_wfree(struct sk_buff *skb); 1862 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1863 gfp_t priority); 1864 void skb_orphan_partial(struct sk_buff *skb); 1865 void sock_rfree(struct sk_buff *skb); 1866 void sock_efree(struct sk_buff *skb); 1867 #ifdef CONFIG_INET 1868 void sock_edemux(struct sk_buff *skb); 1869 void sock_pfree(struct sk_buff *skb); 1870 #else 1871 #define sock_edemux sock_efree 1872 #endif 1873 1874 int sk_setsockopt(struct sock *sk, int level, int optname, 1875 sockptr_t optval, unsigned int optlen); 1876 int sock_setsockopt(struct socket *sock, int level, int op, 1877 sockptr_t optval, unsigned int optlen); 1878 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1879 int optname, sockptr_t optval, int optlen); 1880 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1881 int optname, sockptr_t optval, sockptr_t optlen); 1882 1883 int sk_getsockopt(struct sock *sk, int level, int optname, 1884 sockptr_t optval, sockptr_t optlen); 1885 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1886 bool timeval, bool time32); 1887 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1888 unsigned long data_len, int noblock, 1889 int *errcode, int max_page_order); 1890 1891 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1892 unsigned long size, 1893 int noblock, int *errcode) 1894 { 1895 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1896 } 1897 1898 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1899 void sock_kfree_s(struct sock *sk, void *mem, int size); 1900 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1901 void sk_send_sigurg(struct sock *sk); 1902 1903 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1904 { 1905 if (sk->sk_socket) 1906 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1907 WRITE_ONCE(sk->sk_prot, proto); 1908 } 1909 1910 struct sockcm_cookie { 1911 u64 transmit_time; 1912 u32 mark; 1913 u32 tsflags; 1914 }; 1915 1916 static inline void sockcm_init(struct sockcm_cookie *sockc, 1917 const struct sock *sk) 1918 { 1919 *sockc = (struct sockcm_cookie) { 1920 .tsflags = READ_ONCE(sk->sk_tsflags) 1921 }; 1922 } 1923 1924 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1925 struct sockcm_cookie *sockc); 1926 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1927 struct sockcm_cookie *sockc); 1928 1929 /* 1930 * Functions to fill in entries in struct proto_ops when a protocol 1931 * does not implement a particular function. 1932 */ 1933 int sock_no_bind(struct socket *, struct sockaddr *, int); 1934 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1935 int sock_no_socketpair(struct socket *, struct socket *); 1936 int sock_no_accept(struct socket *, struct socket *, int, bool); 1937 int sock_no_getname(struct socket *, struct sockaddr *, int); 1938 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1939 int sock_no_listen(struct socket *, int); 1940 int sock_no_shutdown(struct socket *, int); 1941 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1942 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1943 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1944 int sock_no_mmap(struct file *file, struct socket *sock, 1945 struct vm_area_struct *vma); 1946 1947 /* 1948 * Functions to fill in entries in struct proto_ops when a protocol 1949 * uses the inet style. 1950 */ 1951 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1952 char __user *optval, int __user *optlen); 1953 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1954 int flags); 1955 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1956 sockptr_t optval, unsigned int optlen); 1957 1958 void sk_common_release(struct sock *sk); 1959 1960 /* 1961 * Default socket callbacks and setup code 1962 */ 1963 1964 /* Initialise core socket variables using an explicit uid. */ 1965 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1966 1967 /* Initialise core socket variables. 1968 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1969 */ 1970 void sock_init_data(struct socket *sock, struct sock *sk); 1971 1972 /* 1973 * Socket reference counting postulates. 1974 * 1975 * * Each user of socket SHOULD hold a reference count. 1976 * * Each access point to socket (an hash table bucket, reference from a list, 1977 * running timer, skb in flight MUST hold a reference count. 1978 * * When reference count hits 0, it means it will never increase back. 1979 * * When reference count hits 0, it means that no references from 1980 * outside exist to this socket and current process on current CPU 1981 * is last user and may/should destroy this socket. 1982 * * sk_free is called from any context: process, BH, IRQ. When 1983 * it is called, socket has no references from outside -> sk_free 1984 * may release descendant resources allocated by the socket, but 1985 * to the time when it is called, socket is NOT referenced by any 1986 * hash tables, lists etc. 1987 * * Packets, delivered from outside (from network or from another process) 1988 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1989 * when they sit in queue. Otherwise, packets will leak to hole, when 1990 * socket is looked up by one cpu and unhasing is made by another CPU. 1991 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1992 * (leak to backlog). Packet socket does all the processing inside 1993 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1994 * use separate SMP lock, so that they are prone too. 1995 */ 1996 1997 /* Ungrab socket and destroy it, if it was the last reference. */ 1998 static inline void sock_put(struct sock *sk) 1999 { 2000 if (refcount_dec_and_test(&sk->sk_refcnt)) 2001 sk_free(sk); 2002 } 2003 /* Generic version of sock_put(), dealing with all sockets 2004 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 2005 */ 2006 void sock_gen_put(struct sock *sk); 2007 2008 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 2009 unsigned int trim_cap, bool refcounted); 2010 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 2011 const int nested) 2012 { 2013 return __sk_receive_skb(sk, skb, nested, 1, true); 2014 } 2015 2016 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 2017 { 2018 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2019 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2020 return; 2021 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2022 * other WRITE_ONCE() because socket lock might be not held. 2023 */ 2024 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 2025 } 2026 2027 #define NO_QUEUE_MAPPING USHRT_MAX 2028 2029 static inline void sk_tx_queue_clear(struct sock *sk) 2030 { 2031 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2032 * other WRITE_ONCE() because socket lock might be not held. 2033 */ 2034 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 2035 } 2036 2037 static inline int sk_tx_queue_get(const struct sock *sk) 2038 { 2039 if (sk) { 2040 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 2041 * and sk_tx_queue_set(). 2042 */ 2043 int val = READ_ONCE(sk->sk_tx_queue_mapping); 2044 2045 if (val != NO_QUEUE_MAPPING) 2046 return val; 2047 } 2048 return -1; 2049 } 2050 2051 static inline void __sk_rx_queue_set(struct sock *sk, 2052 const struct sk_buff *skb, 2053 bool force_set) 2054 { 2055 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2056 if (skb_rx_queue_recorded(skb)) { 2057 u16 rx_queue = skb_get_rx_queue(skb); 2058 2059 if (force_set || 2060 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2061 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2062 } 2063 #endif 2064 } 2065 2066 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2067 { 2068 __sk_rx_queue_set(sk, skb, true); 2069 } 2070 2071 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2072 { 2073 __sk_rx_queue_set(sk, skb, false); 2074 } 2075 2076 static inline void sk_rx_queue_clear(struct sock *sk) 2077 { 2078 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2079 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2080 #endif 2081 } 2082 2083 static inline int sk_rx_queue_get(const struct sock *sk) 2084 { 2085 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2086 if (sk) { 2087 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2088 2089 if (res != NO_QUEUE_MAPPING) 2090 return res; 2091 } 2092 #endif 2093 2094 return -1; 2095 } 2096 2097 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2098 { 2099 sk->sk_socket = sock; 2100 } 2101 2102 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2103 { 2104 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2105 return &rcu_dereference_raw(sk->sk_wq)->wait; 2106 } 2107 /* Detach socket from process context. 2108 * Announce socket dead, detach it from wait queue and inode. 2109 * Note that parent inode held reference count on this struct sock, 2110 * we do not release it in this function, because protocol 2111 * probably wants some additional cleanups or even continuing 2112 * to work with this socket (TCP). 2113 */ 2114 static inline void sock_orphan(struct sock *sk) 2115 { 2116 write_lock_bh(&sk->sk_callback_lock); 2117 sock_set_flag(sk, SOCK_DEAD); 2118 sk_set_socket(sk, NULL); 2119 sk->sk_wq = NULL; 2120 write_unlock_bh(&sk->sk_callback_lock); 2121 } 2122 2123 static inline void sock_graft(struct sock *sk, struct socket *parent) 2124 { 2125 WARN_ON(parent->sk); 2126 write_lock_bh(&sk->sk_callback_lock); 2127 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2128 parent->sk = sk; 2129 sk_set_socket(sk, parent); 2130 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2131 security_sock_graft(sk, parent); 2132 write_unlock_bh(&sk->sk_callback_lock); 2133 } 2134 2135 kuid_t sock_i_uid(struct sock *sk); 2136 unsigned long __sock_i_ino(struct sock *sk); 2137 unsigned long sock_i_ino(struct sock *sk); 2138 2139 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2140 { 2141 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2142 } 2143 2144 static inline u32 net_tx_rndhash(void) 2145 { 2146 u32 v = get_random_u32(); 2147 2148 return v ?: 1; 2149 } 2150 2151 static inline void sk_set_txhash(struct sock *sk) 2152 { 2153 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2154 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2155 } 2156 2157 static inline bool sk_rethink_txhash(struct sock *sk) 2158 { 2159 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2160 sk_set_txhash(sk); 2161 return true; 2162 } 2163 return false; 2164 } 2165 2166 static inline struct dst_entry * 2167 __sk_dst_get(const struct sock *sk) 2168 { 2169 return rcu_dereference_check(sk->sk_dst_cache, 2170 lockdep_sock_is_held(sk)); 2171 } 2172 2173 static inline struct dst_entry * 2174 sk_dst_get(const struct sock *sk) 2175 { 2176 struct dst_entry *dst; 2177 2178 rcu_read_lock(); 2179 dst = rcu_dereference(sk->sk_dst_cache); 2180 if (dst && !rcuref_get(&dst->__rcuref)) 2181 dst = NULL; 2182 rcu_read_unlock(); 2183 return dst; 2184 } 2185 2186 static inline void __dst_negative_advice(struct sock *sk) 2187 { 2188 struct dst_entry *dst = __sk_dst_get(sk); 2189 2190 if (dst && dst->ops->negative_advice) 2191 dst->ops->negative_advice(sk, dst); 2192 } 2193 2194 static inline void dst_negative_advice(struct sock *sk) 2195 { 2196 sk_rethink_txhash(sk); 2197 __dst_negative_advice(sk); 2198 } 2199 2200 static inline void 2201 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2202 { 2203 struct dst_entry *old_dst; 2204 2205 sk_tx_queue_clear(sk); 2206 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2207 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2208 lockdep_sock_is_held(sk)); 2209 rcu_assign_pointer(sk->sk_dst_cache, dst); 2210 dst_release(old_dst); 2211 } 2212 2213 static inline void 2214 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2215 { 2216 struct dst_entry *old_dst; 2217 2218 sk_tx_queue_clear(sk); 2219 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2220 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2221 dst_release(old_dst); 2222 } 2223 2224 static inline void 2225 __sk_dst_reset(struct sock *sk) 2226 { 2227 __sk_dst_set(sk, NULL); 2228 } 2229 2230 static inline void 2231 sk_dst_reset(struct sock *sk) 2232 { 2233 sk_dst_set(sk, NULL); 2234 } 2235 2236 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2237 2238 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2239 2240 static inline void sk_dst_confirm(struct sock *sk) 2241 { 2242 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2243 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2244 } 2245 2246 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2247 { 2248 if (skb_get_dst_pending_confirm(skb)) { 2249 struct sock *sk = skb->sk; 2250 2251 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2252 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2253 neigh_confirm(n); 2254 } 2255 } 2256 2257 bool sk_mc_loop(struct sock *sk); 2258 2259 static inline bool sk_can_gso(const struct sock *sk) 2260 { 2261 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2262 } 2263 2264 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2265 2266 static inline void sk_gso_disable(struct sock *sk) 2267 { 2268 sk->sk_gso_disabled = 1; 2269 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2270 } 2271 2272 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2273 struct iov_iter *from, char *to, 2274 int copy, int offset) 2275 { 2276 if (skb->ip_summed == CHECKSUM_NONE) { 2277 __wsum csum = 0; 2278 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2279 return -EFAULT; 2280 skb->csum = csum_block_add(skb->csum, csum, offset); 2281 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2282 if (!copy_from_iter_full_nocache(to, copy, from)) 2283 return -EFAULT; 2284 } else if (!copy_from_iter_full(to, copy, from)) 2285 return -EFAULT; 2286 2287 return 0; 2288 } 2289 2290 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2291 struct iov_iter *from, int copy) 2292 { 2293 int err, offset = skb->len; 2294 2295 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2296 copy, offset); 2297 if (err) 2298 __skb_trim(skb, offset); 2299 2300 return err; 2301 } 2302 2303 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2304 struct sk_buff *skb, 2305 struct page *page, 2306 int off, int copy) 2307 { 2308 int err; 2309 2310 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2311 copy, skb->len); 2312 if (err) 2313 return err; 2314 2315 skb_len_add(skb, copy); 2316 sk_wmem_queued_add(sk, copy); 2317 sk_mem_charge(sk, copy); 2318 return 0; 2319 } 2320 2321 /** 2322 * sk_wmem_alloc_get - returns write allocations 2323 * @sk: socket 2324 * 2325 * Return: sk_wmem_alloc minus initial offset of one 2326 */ 2327 static inline int sk_wmem_alloc_get(const struct sock *sk) 2328 { 2329 return refcount_read(&sk->sk_wmem_alloc) - 1; 2330 } 2331 2332 /** 2333 * sk_rmem_alloc_get - returns read allocations 2334 * @sk: socket 2335 * 2336 * Return: sk_rmem_alloc 2337 */ 2338 static inline int sk_rmem_alloc_get(const struct sock *sk) 2339 { 2340 return atomic_read(&sk->sk_rmem_alloc); 2341 } 2342 2343 /** 2344 * sk_has_allocations - check if allocations are outstanding 2345 * @sk: socket 2346 * 2347 * Return: true if socket has write or read allocations 2348 */ 2349 static inline bool sk_has_allocations(const struct sock *sk) 2350 { 2351 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2352 } 2353 2354 /** 2355 * skwq_has_sleeper - check if there are any waiting processes 2356 * @wq: struct socket_wq 2357 * 2358 * Return: true if socket_wq has waiting processes 2359 * 2360 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2361 * barrier call. They were added due to the race found within the tcp code. 2362 * 2363 * Consider following tcp code paths:: 2364 * 2365 * CPU1 CPU2 2366 * sys_select receive packet 2367 * ... ... 2368 * __add_wait_queue update tp->rcv_nxt 2369 * ... ... 2370 * tp->rcv_nxt check sock_def_readable 2371 * ... { 2372 * schedule rcu_read_lock(); 2373 * wq = rcu_dereference(sk->sk_wq); 2374 * if (wq && waitqueue_active(&wq->wait)) 2375 * wake_up_interruptible(&wq->wait) 2376 * ... 2377 * } 2378 * 2379 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2380 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2381 * could then endup calling schedule and sleep forever if there are no more 2382 * data on the socket. 2383 * 2384 */ 2385 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2386 { 2387 return wq && wq_has_sleeper(&wq->wait); 2388 } 2389 2390 /** 2391 * sock_poll_wait - place memory barrier behind the poll_wait call. 2392 * @filp: file 2393 * @sock: socket to wait on 2394 * @p: poll_table 2395 * 2396 * See the comments in the wq_has_sleeper function. 2397 */ 2398 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2399 poll_table *p) 2400 { 2401 if (!poll_does_not_wait(p)) { 2402 poll_wait(filp, &sock->wq.wait, p); 2403 /* We need to be sure we are in sync with the 2404 * socket flags modification. 2405 * 2406 * This memory barrier is paired in the wq_has_sleeper. 2407 */ 2408 smp_mb(); 2409 } 2410 } 2411 2412 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2413 { 2414 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2415 u32 txhash = READ_ONCE(sk->sk_txhash); 2416 2417 if (txhash) { 2418 skb->l4_hash = 1; 2419 skb->hash = txhash; 2420 } 2421 } 2422 2423 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2424 2425 /* 2426 * Queue a received datagram if it will fit. Stream and sequenced 2427 * protocols can't normally use this as they need to fit buffers in 2428 * and play with them. 2429 * 2430 * Inlined as it's very short and called for pretty much every 2431 * packet ever received. 2432 */ 2433 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2434 { 2435 skb_orphan(skb); 2436 skb->sk = sk; 2437 skb->destructor = sock_rfree; 2438 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2439 sk_mem_charge(sk, skb->truesize); 2440 } 2441 2442 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2443 { 2444 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2445 skb_orphan(skb); 2446 skb->destructor = sock_efree; 2447 skb->sk = sk; 2448 return true; 2449 } 2450 return false; 2451 } 2452 2453 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2454 { 2455 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2456 if (skb) { 2457 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2458 skb_set_owner_r(skb, sk); 2459 return skb; 2460 } 2461 __kfree_skb(skb); 2462 } 2463 return NULL; 2464 } 2465 2466 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2467 { 2468 if (skb->destructor != sock_wfree) { 2469 skb_orphan(skb); 2470 return; 2471 } 2472 skb->slow_gro = 1; 2473 } 2474 2475 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2476 unsigned long expires); 2477 2478 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2479 2480 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2481 2482 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2483 struct sk_buff *skb, unsigned int flags, 2484 void (*destructor)(struct sock *sk, 2485 struct sk_buff *skb)); 2486 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2487 2488 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2489 enum skb_drop_reason *reason); 2490 2491 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2492 { 2493 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2494 } 2495 2496 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2497 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2498 2499 /* 2500 * Recover an error report and clear atomically 2501 */ 2502 2503 static inline int sock_error(struct sock *sk) 2504 { 2505 int err; 2506 2507 /* Avoid an atomic operation for the common case. 2508 * This is racy since another cpu/thread can change sk_err under us. 2509 */ 2510 if (likely(data_race(!sk->sk_err))) 2511 return 0; 2512 2513 err = xchg(&sk->sk_err, 0); 2514 return -err; 2515 } 2516 2517 void sk_error_report(struct sock *sk); 2518 2519 static inline unsigned long sock_wspace(struct sock *sk) 2520 { 2521 int amt = 0; 2522 2523 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2524 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2525 if (amt < 0) 2526 amt = 0; 2527 } 2528 return amt; 2529 } 2530 2531 /* Note: 2532 * We use sk->sk_wq_raw, from contexts knowing this 2533 * pointer is not NULL and cannot disappear/change. 2534 */ 2535 static inline void sk_set_bit(int nr, struct sock *sk) 2536 { 2537 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2538 !sock_flag(sk, SOCK_FASYNC)) 2539 return; 2540 2541 set_bit(nr, &sk->sk_wq_raw->flags); 2542 } 2543 2544 static inline void sk_clear_bit(int nr, struct sock *sk) 2545 { 2546 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2547 !sock_flag(sk, SOCK_FASYNC)) 2548 return; 2549 2550 clear_bit(nr, &sk->sk_wq_raw->flags); 2551 } 2552 2553 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2554 { 2555 if (sock_flag(sk, SOCK_FASYNC)) { 2556 rcu_read_lock(); 2557 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2558 rcu_read_unlock(); 2559 } 2560 } 2561 2562 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2563 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2564 * Note: for send buffers, TCP works better if we can build two skbs at 2565 * minimum. 2566 */ 2567 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2568 2569 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2570 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2571 2572 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2573 { 2574 u32 val; 2575 2576 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2577 return; 2578 2579 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2580 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2581 2582 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2583 } 2584 2585 /** 2586 * sk_page_frag - return an appropriate page_frag 2587 * @sk: socket 2588 * 2589 * Use the per task page_frag instead of the per socket one for 2590 * optimization when we know that we're in process context and own 2591 * everything that's associated with %current. 2592 * 2593 * Both direct reclaim and page faults can nest inside other 2594 * socket operations and end up recursing into sk_page_frag() 2595 * while it's already in use: explicitly avoid task page_frag 2596 * when users disable sk_use_task_frag. 2597 * 2598 * Return: a per task page_frag if context allows that, 2599 * otherwise a per socket one. 2600 */ 2601 static inline struct page_frag *sk_page_frag(struct sock *sk) 2602 { 2603 if (sk->sk_use_task_frag) 2604 return ¤t->task_frag; 2605 2606 return &sk->sk_frag; 2607 } 2608 2609 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2610 2611 /* 2612 * Default write policy as shown to user space via poll/select/SIGIO 2613 */ 2614 static inline bool sock_writeable(const struct sock *sk) 2615 { 2616 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2617 } 2618 2619 static inline gfp_t gfp_any(void) 2620 { 2621 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2622 } 2623 2624 static inline gfp_t gfp_memcg_charge(void) 2625 { 2626 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2627 } 2628 2629 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2630 { 2631 return noblock ? 0 : sk->sk_rcvtimeo; 2632 } 2633 2634 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2635 { 2636 return noblock ? 0 : sk->sk_sndtimeo; 2637 } 2638 2639 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2640 { 2641 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2642 2643 return v ?: 1; 2644 } 2645 2646 /* Alas, with timeout socket operations are not restartable. 2647 * Compare this to poll(). 2648 */ 2649 static inline int sock_intr_errno(long timeo) 2650 { 2651 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2652 } 2653 2654 struct sock_skb_cb { 2655 u32 dropcount; 2656 }; 2657 2658 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2659 * using skb->cb[] would keep using it directly and utilize its 2660 * alignement guarantee. 2661 */ 2662 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2663 sizeof(struct sock_skb_cb))) 2664 2665 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2666 SOCK_SKB_CB_OFFSET)) 2667 2668 #define sock_skb_cb_check_size(size) \ 2669 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2670 2671 static inline void 2672 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2673 { 2674 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2675 atomic_read(&sk->sk_drops) : 0; 2676 } 2677 2678 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2679 { 2680 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2681 2682 atomic_add(segs, &sk->sk_drops); 2683 } 2684 2685 static inline ktime_t sock_read_timestamp(struct sock *sk) 2686 { 2687 #if BITS_PER_LONG==32 2688 unsigned int seq; 2689 ktime_t kt; 2690 2691 do { 2692 seq = read_seqbegin(&sk->sk_stamp_seq); 2693 kt = sk->sk_stamp; 2694 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2695 2696 return kt; 2697 #else 2698 return READ_ONCE(sk->sk_stamp); 2699 #endif 2700 } 2701 2702 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2703 { 2704 #if BITS_PER_LONG==32 2705 write_seqlock(&sk->sk_stamp_seq); 2706 sk->sk_stamp = kt; 2707 write_sequnlock(&sk->sk_stamp_seq); 2708 #else 2709 WRITE_ONCE(sk->sk_stamp, kt); 2710 #endif 2711 } 2712 2713 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2714 struct sk_buff *skb); 2715 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2716 struct sk_buff *skb); 2717 2718 static inline void 2719 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2720 { 2721 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2722 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2723 ktime_t kt = skb->tstamp; 2724 /* 2725 * generate control messages if 2726 * - receive time stamping in software requested 2727 * - software time stamp available and wanted 2728 * - hardware time stamps available and wanted 2729 */ 2730 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2731 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2732 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2733 (hwtstamps->hwtstamp && 2734 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2735 __sock_recv_timestamp(msg, sk, skb); 2736 else 2737 sock_write_timestamp(sk, kt); 2738 2739 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2740 __sock_recv_wifi_status(msg, sk, skb); 2741 } 2742 2743 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2744 struct sk_buff *skb); 2745 2746 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2747 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2748 struct sk_buff *skb) 2749 { 2750 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2751 (1UL << SOCK_RCVTSTAMP) | \ 2752 (1UL << SOCK_RCVMARK)) 2753 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2754 SOF_TIMESTAMPING_RAW_HARDWARE) 2755 2756 if (sk->sk_flags & FLAGS_RECV_CMSGS || 2757 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY) 2758 __sock_recv_cmsgs(msg, sk, skb); 2759 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2760 sock_write_timestamp(sk, skb->tstamp); 2761 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2762 sock_write_timestamp(sk, 0); 2763 } 2764 2765 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2766 2767 /** 2768 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2769 * @sk: socket sending this packet 2770 * @tsflags: timestamping flags to use 2771 * @tx_flags: completed with instructions for time stamping 2772 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2773 * 2774 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2775 */ 2776 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2777 __u8 *tx_flags, __u32 *tskey) 2778 { 2779 if (unlikely(tsflags)) { 2780 __sock_tx_timestamp(tsflags, tx_flags); 2781 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2782 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2783 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2784 } 2785 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2786 *tx_flags |= SKBTX_WIFI_STATUS; 2787 } 2788 2789 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2790 __u8 *tx_flags) 2791 { 2792 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2793 } 2794 2795 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2796 { 2797 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2798 &skb_shinfo(skb)->tskey); 2799 } 2800 2801 static inline bool sk_is_inet(const struct sock *sk) 2802 { 2803 int family = READ_ONCE(sk->sk_family); 2804 2805 return family == AF_INET || family == AF_INET6; 2806 } 2807 2808 static inline bool sk_is_tcp(const struct sock *sk) 2809 { 2810 return sk_is_inet(sk) && 2811 sk->sk_type == SOCK_STREAM && 2812 sk->sk_protocol == IPPROTO_TCP; 2813 } 2814 2815 static inline bool sk_is_udp(const struct sock *sk) 2816 { 2817 return sk_is_inet(sk) && 2818 sk->sk_type == SOCK_DGRAM && 2819 sk->sk_protocol == IPPROTO_UDP; 2820 } 2821 2822 static inline bool sk_is_stream_unix(const struct sock *sk) 2823 { 2824 return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM; 2825 } 2826 2827 /** 2828 * sk_eat_skb - Release a skb if it is no longer needed 2829 * @sk: socket to eat this skb from 2830 * @skb: socket buffer to eat 2831 * 2832 * This routine must be called with interrupts disabled or with the socket 2833 * locked so that the sk_buff queue operation is ok. 2834 */ 2835 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2836 { 2837 __skb_unlink(skb, &sk->sk_receive_queue); 2838 __kfree_skb(skb); 2839 } 2840 2841 static inline bool 2842 skb_sk_is_prefetched(struct sk_buff *skb) 2843 { 2844 #ifdef CONFIG_INET 2845 return skb->destructor == sock_pfree; 2846 #else 2847 return false; 2848 #endif /* CONFIG_INET */ 2849 } 2850 2851 /* This helper checks if a socket is a full socket, 2852 * ie _not_ a timewait or request socket. 2853 */ 2854 static inline bool sk_fullsock(const struct sock *sk) 2855 { 2856 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2857 } 2858 2859 static inline bool 2860 sk_is_refcounted(struct sock *sk) 2861 { 2862 /* Only full sockets have sk->sk_flags. */ 2863 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2864 } 2865 2866 /** 2867 * skb_steal_sock - steal a socket from an sk_buff 2868 * @skb: sk_buff to steal the socket from 2869 * @refcounted: is set to true if the socket is reference-counted 2870 * @prefetched: is set to true if the socket was assigned from bpf 2871 */ 2872 static inline struct sock * 2873 skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched) 2874 { 2875 if (skb->sk) { 2876 struct sock *sk = skb->sk; 2877 2878 *refcounted = true; 2879 *prefetched = skb_sk_is_prefetched(skb); 2880 if (*prefetched) 2881 *refcounted = sk_is_refcounted(sk); 2882 skb->destructor = NULL; 2883 skb->sk = NULL; 2884 return sk; 2885 } 2886 *prefetched = false; 2887 *refcounted = false; 2888 return NULL; 2889 } 2890 2891 /* Checks if this SKB belongs to an HW offloaded socket 2892 * and whether any SW fallbacks are required based on dev. 2893 * Check decrypted mark in case skb_orphan() cleared socket. 2894 */ 2895 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2896 struct net_device *dev) 2897 { 2898 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2899 struct sock *sk = skb->sk; 2900 2901 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2902 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2903 #ifdef CONFIG_TLS_DEVICE 2904 } else if (unlikely(skb->decrypted)) { 2905 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2906 kfree_skb(skb); 2907 skb = NULL; 2908 #endif 2909 } 2910 #endif 2911 2912 return skb; 2913 } 2914 2915 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2916 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2917 */ 2918 static inline bool sk_listener(const struct sock *sk) 2919 { 2920 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2921 } 2922 2923 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2924 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2925 int type); 2926 2927 bool sk_ns_capable(const struct sock *sk, 2928 struct user_namespace *user_ns, int cap); 2929 bool sk_capable(const struct sock *sk, int cap); 2930 bool sk_net_capable(const struct sock *sk, int cap); 2931 2932 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2933 2934 /* Take into consideration the size of the struct sk_buff overhead in the 2935 * determination of these values, since that is non-constant across 2936 * platforms. This makes socket queueing behavior and performance 2937 * not depend upon such differences. 2938 */ 2939 #define _SK_MEM_PACKETS 256 2940 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2941 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2942 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2943 2944 extern __u32 sysctl_wmem_max; 2945 extern __u32 sysctl_rmem_max; 2946 2947 extern int sysctl_tstamp_allow_data; 2948 extern int sysctl_optmem_max; 2949 2950 extern __u32 sysctl_wmem_default; 2951 extern __u32 sysctl_rmem_default; 2952 2953 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2954 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2955 2956 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2957 { 2958 /* Does this proto have per netns sysctl_wmem ? */ 2959 if (proto->sysctl_wmem_offset) 2960 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2961 2962 return READ_ONCE(*proto->sysctl_wmem); 2963 } 2964 2965 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2966 { 2967 /* Does this proto have per netns sysctl_rmem ? */ 2968 if (proto->sysctl_rmem_offset) 2969 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2970 2971 return READ_ONCE(*proto->sysctl_rmem); 2972 } 2973 2974 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2975 * Some wifi drivers need to tweak it to get more chunks. 2976 * They can use this helper from their ndo_start_xmit() 2977 */ 2978 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2979 { 2980 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2981 return; 2982 WRITE_ONCE(sk->sk_pacing_shift, val); 2983 } 2984 2985 /* if a socket is bound to a device, check that the given device 2986 * index is either the same or that the socket is bound to an L3 2987 * master device and the given device index is also enslaved to 2988 * that L3 master 2989 */ 2990 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2991 { 2992 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2993 int mdif; 2994 2995 if (!bound_dev_if || bound_dev_if == dif) 2996 return true; 2997 2998 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2999 if (mdif && mdif == bound_dev_if) 3000 return true; 3001 3002 return false; 3003 } 3004 3005 void sock_def_readable(struct sock *sk); 3006 3007 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 3008 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 3009 int sock_set_timestamping(struct sock *sk, int optname, 3010 struct so_timestamping timestamping); 3011 3012 void sock_enable_timestamps(struct sock *sk); 3013 void sock_no_linger(struct sock *sk); 3014 void sock_set_keepalive(struct sock *sk); 3015 void sock_set_priority(struct sock *sk, u32 priority); 3016 void sock_set_rcvbuf(struct sock *sk, int val); 3017 void sock_set_mark(struct sock *sk, u32 val); 3018 void sock_set_reuseaddr(struct sock *sk); 3019 void sock_set_reuseport(struct sock *sk); 3020 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3021 3022 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 3023 3024 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3025 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3026 sockptr_t optval, int optlen, bool old_timeval); 3027 3028 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3029 void __user *arg, void *karg, size_t size); 3030 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3031 static inline bool sk_is_readable(struct sock *sk) 3032 { 3033 if (sk->sk_prot->sock_is_readable) 3034 return sk->sk_prot->sock_is_readable(sk); 3035 return false; 3036 } 3037 #endif /* _SOCK_H */ 3038