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 /* This is the per-socket lock. The spinlock provides a synchronization 80 * between user contexts and software interrupt processing, whereas the 81 * mini-semaphore synchronizes multiple users amongst themselves. 82 */ 83 typedef struct { 84 spinlock_t slock; 85 int owned; 86 wait_queue_head_t wq; 87 /* 88 * We express the mutex-alike socket_lock semantics 89 * to the lock validator by explicitly managing 90 * the slock as a lock variant (in addition to 91 * the slock itself): 92 */ 93 #ifdef CONFIG_DEBUG_LOCK_ALLOC 94 struct lockdep_map dep_map; 95 #endif 96 } socket_lock_t; 97 98 struct sock; 99 struct proto; 100 struct net; 101 102 typedef __u32 __bitwise __portpair; 103 typedef __u64 __bitwise __addrpair; 104 105 /** 106 * struct sock_common - minimal network layer representation of sockets 107 * @skc_daddr: Foreign IPv4 addr 108 * @skc_rcv_saddr: Bound local IPv4 addr 109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 110 * @skc_hash: hash value used with various protocol lookup tables 111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 112 * @skc_dport: placeholder for inet_dport/tw_dport 113 * @skc_num: placeholder for inet_num/tw_num 114 * @skc_portpair: __u32 union of @skc_dport & @skc_num 115 * @skc_family: network address family 116 * @skc_state: Connection state 117 * @skc_reuse: %SO_REUSEADDR setting 118 * @skc_reuseport: %SO_REUSEPORT setting 119 * @skc_ipv6only: socket is IPV6 only 120 * @skc_net_refcnt: socket is using net ref counting 121 * @skc_bound_dev_if: bound device index if != 0 122 * @skc_bind_node: bind hash linkage for various protocol lookup tables 123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 124 * @skc_prot: protocol handlers inside a network family 125 * @skc_net: reference to the network namespace of this socket 126 * @skc_v6_daddr: IPV6 destination address 127 * @skc_v6_rcv_saddr: IPV6 source address 128 * @skc_cookie: socket's cookie value 129 * @skc_node: main hash linkage for various protocol lookup tables 130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 131 * @skc_tx_queue_mapping: tx queue number for this connection 132 * @skc_rx_queue_mapping: rx queue number for this connection 133 * @skc_flags: place holder for sk_flags 134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 136 * @skc_listener: connection request listener socket (aka rsk_listener) 137 * [union with @skc_flags] 138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 139 * [union with @skc_flags] 140 * @skc_incoming_cpu: record/match cpu processing incoming packets 141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 142 * [union with @skc_incoming_cpu] 143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 144 * [union with @skc_incoming_cpu] 145 * @skc_refcnt: reference count 146 * 147 * This is the minimal network layer representation of sockets, the header 148 * for struct sock and struct inet_timewait_sock. 149 */ 150 struct sock_common { 151 union { 152 __addrpair skc_addrpair; 153 struct { 154 __be32 skc_daddr; 155 __be32 skc_rcv_saddr; 156 }; 157 }; 158 union { 159 unsigned int skc_hash; 160 __u16 skc_u16hashes[2]; 161 }; 162 /* skc_dport && skc_num must be grouped as well */ 163 union { 164 __portpair skc_portpair; 165 struct { 166 __be16 skc_dport; 167 __u16 skc_num; 168 }; 169 }; 170 171 unsigned short skc_family; 172 volatile unsigned char skc_state; 173 unsigned char skc_reuse:4; 174 unsigned char skc_reuseport:1; 175 unsigned char skc_ipv6only:1; 176 unsigned char skc_net_refcnt:1; 177 int skc_bound_dev_if; 178 union { 179 struct hlist_node skc_bind_node; 180 struct hlist_node skc_portaddr_node; 181 }; 182 struct proto *skc_prot; 183 possible_net_t skc_net; 184 185 #if IS_ENABLED(CONFIG_IPV6) 186 struct in6_addr skc_v6_daddr; 187 struct in6_addr skc_v6_rcv_saddr; 188 #endif 189 190 atomic64_t skc_cookie; 191 192 /* following fields are padding to force 193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 194 * assuming IPV6 is enabled. We use this padding differently 195 * for different kind of 'sockets' 196 */ 197 union { 198 unsigned long skc_flags; 199 struct sock *skc_listener; /* request_sock */ 200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 201 }; 202 /* 203 * fields between dontcopy_begin/dontcopy_end 204 * are not copied in sock_copy() 205 */ 206 /* private: */ 207 int skc_dontcopy_begin[0]; 208 /* public: */ 209 union { 210 struct hlist_node skc_node; 211 struct hlist_nulls_node skc_nulls_node; 212 }; 213 unsigned short skc_tx_queue_mapping; 214 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 215 unsigned short skc_rx_queue_mapping; 216 #endif 217 union { 218 int skc_incoming_cpu; 219 u32 skc_rcv_wnd; 220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 221 }; 222 223 refcount_t skc_refcnt; 224 /* private: */ 225 int skc_dontcopy_end[0]; 226 union { 227 u32 skc_rxhash; 228 u32 skc_window_clamp; 229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 230 }; 231 /* public: */ 232 }; 233 234 struct bpf_local_storage; 235 struct sk_filter; 236 237 /** 238 * struct sock - network layer representation of sockets 239 * @__sk_common: shared layout with inet_timewait_sock 240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 242 * @sk_lock: synchronizer 243 * @sk_kern_sock: True if sock is using kernel lock classes 244 * @sk_rcvbuf: size of receive buffer in bytes 245 * @sk_wq: sock wait queue and async head 246 * @sk_rx_dst: receive input route used by early demux 247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 249 * @sk_dst_cache: destination cache 250 * @sk_dst_pending_confirm: need to confirm neighbour 251 * @sk_policy: flow policy 252 * @sk_receive_queue: incoming packets 253 * @sk_wmem_alloc: transmit queue bytes committed 254 * @sk_tsq_flags: TCP Small Queues flags 255 * @sk_write_queue: Packet sending queue 256 * @sk_omem_alloc: "o" is "option" or "other" 257 * @sk_wmem_queued: persistent queue size 258 * @sk_forward_alloc: space allocated forward 259 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 260 * @sk_napi_id: id of the last napi context to receive data for sk 261 * @sk_ll_usec: usecs to busypoll when there is no data 262 * @sk_allocation: allocation mode 263 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 264 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 265 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 266 * @sk_sndbuf: size of send buffer in bytes 267 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 268 * @sk_no_check_rx: allow zero checksum in RX packets 269 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 270 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 271 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 272 * @sk_gso_max_size: Maximum GSO segment size to build 273 * @sk_gso_max_segs: Maximum number of GSO segments 274 * @sk_pacing_shift: scaling factor for TCP Small Queues 275 * @sk_lingertime: %SO_LINGER l_linger setting 276 * @sk_backlog: always used with the per-socket spinlock held 277 * @sk_callback_lock: used with the callbacks in the end of this struct 278 * @sk_error_queue: rarely used 279 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 280 * IPV6_ADDRFORM for instance) 281 * @sk_err: last error 282 * @sk_err_soft: errors that don't cause failure but are the cause of a 283 * persistent failure not just 'timed out' 284 * @sk_drops: raw/udp drops counter 285 * @sk_ack_backlog: current listen backlog 286 * @sk_max_ack_backlog: listen backlog set in listen() 287 * @sk_uid: user id of owner 288 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 289 * @sk_busy_poll_budget: napi processing budget when busypolling 290 * @sk_priority: %SO_PRIORITY setting 291 * @sk_type: socket type (%SOCK_STREAM, etc) 292 * @sk_protocol: which protocol this socket belongs in this network family 293 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 294 * @sk_peer_pid: &struct pid for this socket's peer 295 * @sk_peer_cred: %SO_PEERCRED setting 296 * @sk_rcvlowat: %SO_RCVLOWAT setting 297 * @sk_rcvtimeo: %SO_RCVTIMEO setting 298 * @sk_sndtimeo: %SO_SNDTIMEO setting 299 * @sk_txhash: computed flow hash for use on transmit 300 * @sk_txrehash: enable TX hash rethink 301 * @sk_filter: socket filtering instructions 302 * @sk_timer: sock cleanup timer 303 * @sk_stamp: time stamp of last packet received 304 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 305 * @sk_tsflags: SO_TIMESTAMPING flags 306 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 307 * Sockets that can be used under memory reclaim should 308 * set this to false. 309 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 310 * for timestamping 311 * @sk_tskey: counter to disambiguate concurrent tstamp requests 312 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 313 * @sk_socket: Identd and reporting IO signals 314 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 315 * @sk_frag: cached page frag 316 * @sk_peek_off: current peek_offset value 317 * @sk_send_head: front of stuff to transmit 318 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 319 * @sk_security: used by security modules 320 * @sk_mark: generic packet mark 321 * @sk_cgrp_data: cgroup data for this cgroup 322 * @sk_memcg: this socket's memory cgroup association 323 * @sk_write_pending: a write to stream socket waits to start 324 * @sk_disconnects: number of disconnect operations performed on this sock 325 * @sk_state_change: callback to indicate change in the state of the sock 326 * @sk_data_ready: callback to indicate there is data to be processed 327 * @sk_write_space: callback to indicate there is bf sending space available 328 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 329 * @sk_backlog_rcv: callback to process the backlog 330 * @sk_validate_xmit_skb: ptr to an optional validate function 331 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 332 * @sk_reuseport_cb: reuseport group container 333 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 334 * @sk_rcu: used during RCU grace period 335 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 336 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 337 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 338 * @sk_txtime_unused: unused txtime flags 339 * @ns_tracker: tracker for netns reference 340 */ 341 struct sock { 342 /* 343 * Now struct inet_timewait_sock also uses sock_common, so please just 344 * don't add nothing before this first member (__sk_common) --acme 345 */ 346 struct sock_common __sk_common; 347 #define sk_node __sk_common.skc_node 348 #define sk_nulls_node __sk_common.skc_nulls_node 349 #define sk_refcnt __sk_common.skc_refcnt 350 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 351 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 352 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 353 #endif 354 355 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 356 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 357 #define sk_hash __sk_common.skc_hash 358 #define sk_portpair __sk_common.skc_portpair 359 #define sk_num __sk_common.skc_num 360 #define sk_dport __sk_common.skc_dport 361 #define sk_addrpair __sk_common.skc_addrpair 362 #define sk_daddr __sk_common.skc_daddr 363 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 364 #define sk_family __sk_common.skc_family 365 #define sk_state __sk_common.skc_state 366 #define sk_reuse __sk_common.skc_reuse 367 #define sk_reuseport __sk_common.skc_reuseport 368 #define sk_ipv6only __sk_common.skc_ipv6only 369 #define sk_net_refcnt __sk_common.skc_net_refcnt 370 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 371 #define sk_bind_node __sk_common.skc_bind_node 372 #define sk_prot __sk_common.skc_prot 373 #define sk_net __sk_common.skc_net 374 #define sk_v6_daddr __sk_common.skc_v6_daddr 375 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 376 #define sk_cookie __sk_common.skc_cookie 377 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 378 #define sk_flags __sk_common.skc_flags 379 #define sk_rxhash __sk_common.skc_rxhash 380 381 /* early demux fields */ 382 struct dst_entry __rcu *sk_rx_dst; 383 int sk_rx_dst_ifindex; 384 u32 sk_rx_dst_cookie; 385 386 socket_lock_t sk_lock; 387 atomic_t sk_drops; 388 int sk_rcvlowat; 389 struct sk_buff_head sk_error_queue; 390 struct sk_buff_head sk_receive_queue; 391 /* 392 * The backlog queue is special, it is always used with 393 * the per-socket spinlock held and requires low latency 394 * access. Therefore we special case it's implementation. 395 * Note : rmem_alloc is in this structure to fill a hole 396 * on 64bit arches, not because its logically part of 397 * backlog. 398 */ 399 struct { 400 atomic_t rmem_alloc; 401 int len; 402 struct sk_buff *head; 403 struct sk_buff *tail; 404 } sk_backlog; 405 406 #define sk_rmem_alloc sk_backlog.rmem_alloc 407 408 int sk_forward_alloc; 409 u32 sk_reserved_mem; 410 #ifdef CONFIG_NET_RX_BUSY_POLL 411 unsigned int sk_ll_usec; 412 /* ===== mostly read cache line ===== */ 413 unsigned int sk_napi_id; 414 #endif 415 int sk_rcvbuf; 416 int sk_disconnects; 417 418 struct sk_filter __rcu *sk_filter; 419 union { 420 struct socket_wq __rcu *sk_wq; 421 /* private: */ 422 struct socket_wq *sk_wq_raw; 423 /* public: */ 424 }; 425 #ifdef CONFIG_XFRM 426 struct xfrm_policy __rcu *sk_policy[2]; 427 #endif 428 429 struct dst_entry __rcu *sk_dst_cache; 430 atomic_t sk_omem_alloc; 431 int sk_sndbuf; 432 433 /* ===== cache line for TX ===== */ 434 int sk_wmem_queued; 435 refcount_t sk_wmem_alloc; 436 unsigned long sk_tsq_flags; 437 union { 438 struct sk_buff *sk_send_head; 439 struct rb_root tcp_rtx_queue; 440 }; 441 struct sk_buff_head sk_write_queue; 442 __s32 sk_peek_off; 443 int sk_write_pending; 444 __u32 sk_dst_pending_confirm; 445 u32 sk_pacing_status; /* see enum sk_pacing */ 446 long sk_sndtimeo; 447 struct timer_list sk_timer; 448 __u32 sk_priority; 449 __u32 sk_mark; 450 unsigned long sk_pacing_rate; /* bytes per second */ 451 unsigned long sk_max_pacing_rate; 452 struct page_frag sk_frag; 453 netdev_features_t sk_route_caps; 454 int sk_gso_type; 455 unsigned int sk_gso_max_size; 456 gfp_t sk_allocation; 457 __u32 sk_txhash; 458 459 /* 460 * Because of non atomicity rules, all 461 * changes are protected by socket lock. 462 */ 463 u8 sk_gso_disabled : 1, 464 sk_kern_sock : 1, 465 sk_no_check_tx : 1, 466 sk_no_check_rx : 1, 467 sk_userlocks : 4; 468 u8 sk_pacing_shift; 469 u16 sk_type; 470 u16 sk_protocol; 471 u16 sk_gso_max_segs; 472 unsigned long sk_lingertime; 473 struct proto *sk_prot_creator; 474 rwlock_t sk_callback_lock; 475 int sk_err, 476 sk_err_soft; 477 u32 sk_ack_backlog; 478 u32 sk_max_ack_backlog; 479 kuid_t sk_uid; 480 u8 sk_txrehash; 481 #ifdef CONFIG_NET_RX_BUSY_POLL 482 u8 sk_prefer_busy_poll; 483 u16 sk_busy_poll_budget; 484 #endif 485 spinlock_t sk_peer_lock; 486 int sk_bind_phc; 487 struct pid *sk_peer_pid; 488 const struct cred *sk_peer_cred; 489 490 long sk_rcvtimeo; 491 ktime_t sk_stamp; 492 #if BITS_PER_LONG==32 493 seqlock_t sk_stamp_seq; 494 #endif 495 atomic_t sk_tskey; 496 atomic_t sk_zckey; 497 u32 sk_tsflags; 498 u8 sk_shutdown; 499 500 u8 sk_clockid; 501 u8 sk_txtime_deadline_mode : 1, 502 sk_txtime_report_errors : 1, 503 sk_txtime_unused : 6; 504 bool sk_use_task_frag; 505 506 struct socket *sk_socket; 507 void *sk_user_data; 508 #ifdef CONFIG_SECURITY 509 void *sk_security; 510 #endif 511 struct sock_cgroup_data sk_cgrp_data; 512 struct mem_cgroup *sk_memcg; 513 void (*sk_state_change)(struct sock *sk); 514 void (*sk_data_ready)(struct sock *sk); 515 void (*sk_write_space)(struct sock *sk); 516 void (*sk_error_report)(struct sock *sk); 517 int (*sk_backlog_rcv)(struct sock *sk, 518 struct sk_buff *skb); 519 #ifdef CONFIG_SOCK_VALIDATE_XMIT 520 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 521 struct net_device *dev, 522 struct sk_buff *skb); 523 #endif 524 void (*sk_destruct)(struct sock *sk); 525 struct sock_reuseport __rcu *sk_reuseport_cb; 526 #ifdef CONFIG_BPF_SYSCALL 527 struct bpf_local_storage __rcu *sk_bpf_storage; 528 #endif 529 struct rcu_head sk_rcu; 530 netns_tracker ns_tracker; 531 }; 532 533 enum sk_pacing { 534 SK_PACING_NONE = 0, 535 SK_PACING_NEEDED = 1, 536 SK_PACING_FQ = 2, 537 }; 538 539 /* flag bits in sk_user_data 540 * 541 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 542 * not be suitable for copying when cloning the socket. For instance, 543 * it can point to a reference counted object. sk_user_data bottom 544 * bit is set if pointer must not be copied. 545 * 546 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 547 * managed/owned by a BPF reuseport array. This bit should be set 548 * when sk_user_data's sk is added to the bpf's reuseport_array. 549 * 550 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 551 * sk_user_data points to psock type. This bit should be set 552 * when sk_user_data is assigned to a psock object. 553 */ 554 #define SK_USER_DATA_NOCOPY 1UL 555 #define SK_USER_DATA_BPF 2UL 556 #define SK_USER_DATA_PSOCK 4UL 557 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 558 SK_USER_DATA_PSOCK) 559 560 /** 561 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 562 * @sk: socket 563 */ 564 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 565 { 566 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 567 } 568 569 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 570 571 /** 572 * __locked_read_sk_user_data_with_flags - return the pointer 573 * only if argument flags all has been set in sk_user_data. Otherwise 574 * return NULL 575 * 576 * @sk: socket 577 * @flags: flag bits 578 * 579 * The caller must be holding sk->sk_callback_lock. 580 */ 581 static inline void * 582 __locked_read_sk_user_data_with_flags(const struct sock *sk, 583 uintptr_t flags) 584 { 585 uintptr_t sk_user_data = 586 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 587 lockdep_is_held(&sk->sk_callback_lock)); 588 589 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 590 591 if ((sk_user_data & flags) == flags) 592 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 593 return NULL; 594 } 595 596 /** 597 * __rcu_dereference_sk_user_data_with_flags - return the pointer 598 * only if argument flags all has been set in sk_user_data. Otherwise 599 * return NULL 600 * 601 * @sk: socket 602 * @flags: flag bits 603 */ 604 static inline void * 605 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 606 uintptr_t flags) 607 { 608 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 609 610 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 611 612 if ((sk_user_data & flags) == flags) 613 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 614 return NULL; 615 } 616 617 #define rcu_dereference_sk_user_data(sk) \ 618 __rcu_dereference_sk_user_data_with_flags(sk, 0) 619 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 620 ({ \ 621 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 622 __tmp2 = (uintptr_t)(flags); \ 623 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 624 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 625 rcu_assign_pointer(__sk_user_data((sk)), \ 626 __tmp1 | __tmp2); \ 627 }) 628 #define rcu_assign_sk_user_data(sk, ptr) \ 629 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 630 631 static inline 632 struct net *sock_net(const struct sock *sk) 633 { 634 return read_pnet(&sk->sk_net); 635 } 636 637 static inline 638 void sock_net_set(struct sock *sk, struct net *net) 639 { 640 write_pnet(&sk->sk_net, net); 641 } 642 643 /* 644 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 645 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 646 * on a socket means that the socket will reuse everybody else's port 647 * without looking at the other's sk_reuse value. 648 */ 649 650 #define SK_NO_REUSE 0 651 #define SK_CAN_REUSE 1 652 #define SK_FORCE_REUSE 2 653 654 int sk_set_peek_off(struct sock *sk, int val); 655 656 static inline int sk_peek_offset(const struct sock *sk, int flags) 657 { 658 if (unlikely(flags & MSG_PEEK)) { 659 return READ_ONCE(sk->sk_peek_off); 660 } 661 662 return 0; 663 } 664 665 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 666 { 667 s32 off = READ_ONCE(sk->sk_peek_off); 668 669 if (unlikely(off >= 0)) { 670 off = max_t(s32, off - val, 0); 671 WRITE_ONCE(sk->sk_peek_off, off); 672 } 673 } 674 675 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 676 { 677 sk_peek_offset_bwd(sk, -val); 678 } 679 680 /* 681 * Hashed lists helper routines 682 */ 683 static inline struct sock *sk_entry(const struct hlist_node *node) 684 { 685 return hlist_entry(node, struct sock, sk_node); 686 } 687 688 static inline struct sock *__sk_head(const struct hlist_head *head) 689 { 690 return hlist_entry(head->first, struct sock, sk_node); 691 } 692 693 static inline struct sock *sk_head(const struct hlist_head *head) 694 { 695 return hlist_empty(head) ? NULL : __sk_head(head); 696 } 697 698 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 699 { 700 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 701 } 702 703 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 704 { 705 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 706 } 707 708 static inline struct sock *sk_next(const struct sock *sk) 709 { 710 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 711 } 712 713 static inline struct sock *sk_nulls_next(const struct sock *sk) 714 { 715 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 716 hlist_nulls_entry(sk->sk_nulls_node.next, 717 struct sock, sk_nulls_node) : 718 NULL; 719 } 720 721 static inline bool sk_unhashed(const struct sock *sk) 722 { 723 return hlist_unhashed(&sk->sk_node); 724 } 725 726 static inline bool sk_hashed(const struct sock *sk) 727 { 728 return !sk_unhashed(sk); 729 } 730 731 static inline void sk_node_init(struct hlist_node *node) 732 { 733 node->pprev = NULL; 734 } 735 736 static inline void __sk_del_node(struct sock *sk) 737 { 738 __hlist_del(&sk->sk_node); 739 } 740 741 /* NB: equivalent to hlist_del_init_rcu */ 742 static inline bool __sk_del_node_init(struct sock *sk) 743 { 744 if (sk_hashed(sk)) { 745 __sk_del_node(sk); 746 sk_node_init(&sk->sk_node); 747 return true; 748 } 749 return false; 750 } 751 752 /* Grab socket reference count. This operation is valid only 753 when sk is ALREADY grabbed f.e. it is found in hash table 754 or a list and the lookup is made under lock preventing hash table 755 modifications. 756 */ 757 758 static __always_inline void sock_hold(struct sock *sk) 759 { 760 refcount_inc(&sk->sk_refcnt); 761 } 762 763 /* Ungrab socket in the context, which assumes that socket refcnt 764 cannot hit zero, f.e. it is true in context of any socketcall. 765 */ 766 static __always_inline void __sock_put(struct sock *sk) 767 { 768 refcount_dec(&sk->sk_refcnt); 769 } 770 771 static inline bool sk_del_node_init(struct sock *sk) 772 { 773 bool rc = __sk_del_node_init(sk); 774 775 if (rc) { 776 /* paranoid for a while -acme */ 777 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 778 __sock_put(sk); 779 } 780 return rc; 781 } 782 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 783 784 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 785 { 786 if (sk_hashed(sk)) { 787 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 788 return true; 789 } 790 return false; 791 } 792 793 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 794 { 795 bool rc = __sk_nulls_del_node_init_rcu(sk); 796 797 if (rc) { 798 /* paranoid for a while -acme */ 799 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 800 __sock_put(sk); 801 } 802 return rc; 803 } 804 805 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 806 { 807 hlist_add_head(&sk->sk_node, list); 808 } 809 810 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 811 { 812 sock_hold(sk); 813 __sk_add_node(sk, list); 814 } 815 816 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 817 { 818 sock_hold(sk); 819 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 820 sk->sk_family == AF_INET6) 821 hlist_add_tail_rcu(&sk->sk_node, list); 822 else 823 hlist_add_head_rcu(&sk->sk_node, list); 824 } 825 826 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 827 { 828 sock_hold(sk); 829 hlist_add_tail_rcu(&sk->sk_node, list); 830 } 831 832 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 833 { 834 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 835 } 836 837 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 838 { 839 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 840 } 841 842 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 843 { 844 sock_hold(sk); 845 __sk_nulls_add_node_rcu(sk, list); 846 } 847 848 static inline void __sk_del_bind_node(struct sock *sk) 849 { 850 __hlist_del(&sk->sk_bind_node); 851 } 852 853 static inline void sk_add_bind_node(struct sock *sk, 854 struct hlist_head *list) 855 { 856 hlist_add_head(&sk->sk_bind_node, list); 857 } 858 859 #define sk_for_each(__sk, list) \ 860 hlist_for_each_entry(__sk, list, sk_node) 861 #define sk_for_each_rcu(__sk, list) \ 862 hlist_for_each_entry_rcu(__sk, list, sk_node) 863 #define sk_nulls_for_each(__sk, node, list) \ 864 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 865 #define sk_nulls_for_each_rcu(__sk, node, list) \ 866 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 867 #define sk_for_each_from(__sk) \ 868 hlist_for_each_entry_from(__sk, sk_node) 869 #define sk_nulls_for_each_from(__sk, node) \ 870 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 871 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 872 #define sk_for_each_safe(__sk, tmp, list) \ 873 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 874 #define sk_for_each_bound(__sk, list) \ 875 hlist_for_each_entry(__sk, list, sk_bind_node) 876 877 /** 878 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 879 * @tpos: the type * to use as a loop cursor. 880 * @pos: the &struct hlist_node to use as a loop cursor. 881 * @head: the head for your list. 882 * @offset: offset of hlist_node within the struct. 883 * 884 */ 885 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 886 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 887 pos != NULL && \ 888 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 889 pos = rcu_dereference(hlist_next_rcu(pos))) 890 891 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 892 { 893 /* Careful only use this in a context where these parameters 894 * can not change and must all be valid, such as recvmsg from 895 * userspace. 896 */ 897 return sk->sk_socket->file->f_cred->user_ns; 898 } 899 900 /* Sock flags */ 901 enum sock_flags { 902 SOCK_DEAD, 903 SOCK_DONE, 904 SOCK_URGINLINE, 905 SOCK_KEEPOPEN, 906 SOCK_LINGER, 907 SOCK_DESTROY, 908 SOCK_BROADCAST, 909 SOCK_TIMESTAMP, 910 SOCK_ZAPPED, 911 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 912 SOCK_DBG, /* %SO_DEBUG setting */ 913 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 914 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 915 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 916 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 917 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 918 SOCK_FASYNC, /* fasync() active */ 919 SOCK_RXQ_OVFL, 920 SOCK_ZEROCOPY, /* buffers from userspace */ 921 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 922 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 923 * Will use last 4 bytes of packet sent from 924 * user-space instead. 925 */ 926 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 927 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 928 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 929 SOCK_TXTIME, 930 SOCK_XDP, /* XDP is attached */ 931 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 932 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 933 }; 934 935 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 936 937 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 938 { 939 nsk->sk_flags = osk->sk_flags; 940 } 941 942 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 943 { 944 __set_bit(flag, &sk->sk_flags); 945 } 946 947 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 948 { 949 __clear_bit(flag, &sk->sk_flags); 950 } 951 952 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 953 int valbool) 954 { 955 if (valbool) 956 sock_set_flag(sk, bit); 957 else 958 sock_reset_flag(sk, bit); 959 } 960 961 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 962 { 963 return test_bit(flag, &sk->sk_flags); 964 } 965 966 #ifdef CONFIG_NET 967 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 968 static inline int sk_memalloc_socks(void) 969 { 970 return static_branch_unlikely(&memalloc_socks_key); 971 } 972 973 void __receive_sock(struct file *file); 974 #else 975 976 static inline int sk_memalloc_socks(void) 977 { 978 return 0; 979 } 980 981 static inline void __receive_sock(struct file *file) 982 { } 983 #endif 984 985 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 986 { 987 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 988 } 989 990 static inline void sk_acceptq_removed(struct sock *sk) 991 { 992 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 993 } 994 995 static inline void sk_acceptq_added(struct sock *sk) 996 { 997 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 998 } 999 1000 /* Note: If you think the test should be: 1001 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1002 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1003 */ 1004 static inline bool sk_acceptq_is_full(const struct sock *sk) 1005 { 1006 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1007 } 1008 1009 /* 1010 * Compute minimal free write space needed to queue new packets. 1011 */ 1012 static inline int sk_stream_min_wspace(const struct sock *sk) 1013 { 1014 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1015 } 1016 1017 static inline int sk_stream_wspace(const struct sock *sk) 1018 { 1019 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1020 } 1021 1022 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1023 { 1024 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1025 } 1026 1027 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1028 { 1029 /* Paired with lockless reads of sk->sk_forward_alloc */ 1030 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1031 } 1032 1033 void sk_stream_write_space(struct sock *sk); 1034 1035 /* OOB backlog add */ 1036 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1037 { 1038 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1039 skb_dst_force(skb); 1040 1041 if (!sk->sk_backlog.tail) 1042 WRITE_ONCE(sk->sk_backlog.head, skb); 1043 else 1044 sk->sk_backlog.tail->next = skb; 1045 1046 WRITE_ONCE(sk->sk_backlog.tail, skb); 1047 skb->next = NULL; 1048 } 1049 1050 /* 1051 * Take into account size of receive queue and backlog queue 1052 * Do not take into account this skb truesize, 1053 * to allow even a single big packet to come. 1054 */ 1055 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1056 { 1057 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1058 1059 return qsize > limit; 1060 } 1061 1062 /* The per-socket spinlock must be held here. */ 1063 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1064 unsigned int limit) 1065 { 1066 if (sk_rcvqueues_full(sk, limit)) 1067 return -ENOBUFS; 1068 1069 /* 1070 * If the skb was allocated from pfmemalloc reserves, only 1071 * allow SOCK_MEMALLOC sockets to use it as this socket is 1072 * helping free memory 1073 */ 1074 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1075 return -ENOMEM; 1076 1077 __sk_add_backlog(sk, skb); 1078 sk->sk_backlog.len += skb->truesize; 1079 return 0; 1080 } 1081 1082 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1083 1084 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1085 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1086 1087 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1088 { 1089 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1090 return __sk_backlog_rcv(sk, skb); 1091 1092 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1093 tcp_v6_do_rcv, 1094 tcp_v4_do_rcv, 1095 sk, skb); 1096 } 1097 1098 static inline void sk_incoming_cpu_update(struct sock *sk) 1099 { 1100 int cpu = raw_smp_processor_id(); 1101 1102 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1103 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1104 } 1105 1106 static inline void sock_rps_record_flow_hash(__u32 hash) 1107 { 1108 #ifdef CONFIG_RPS 1109 struct rps_sock_flow_table *sock_flow_table; 1110 1111 rcu_read_lock(); 1112 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1113 rps_record_sock_flow(sock_flow_table, hash); 1114 rcu_read_unlock(); 1115 #endif 1116 } 1117 1118 static inline void sock_rps_record_flow(const struct sock *sk) 1119 { 1120 #ifdef CONFIG_RPS 1121 if (static_branch_unlikely(&rfs_needed)) { 1122 /* Reading sk->sk_rxhash might incur an expensive cache line 1123 * miss. 1124 * 1125 * TCP_ESTABLISHED does cover almost all states where RFS 1126 * might be useful, and is cheaper [1] than testing : 1127 * IPv4: inet_sk(sk)->inet_daddr 1128 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1129 * OR an additional socket flag 1130 * [1] : sk_state and sk_prot are in the same cache line. 1131 */ 1132 if (sk->sk_state == TCP_ESTABLISHED) { 1133 /* This READ_ONCE() is paired with the WRITE_ONCE() 1134 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash(). 1135 */ 1136 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash)); 1137 } 1138 } 1139 #endif 1140 } 1141 1142 static inline void sock_rps_save_rxhash(struct sock *sk, 1143 const struct sk_buff *skb) 1144 { 1145 #ifdef CONFIG_RPS 1146 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1147 * here, and another one in sock_rps_record_flow(). 1148 */ 1149 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1150 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1151 #endif 1152 } 1153 1154 static inline void sock_rps_reset_rxhash(struct sock *sk) 1155 { 1156 #ifdef CONFIG_RPS 1157 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1158 WRITE_ONCE(sk->sk_rxhash, 0); 1159 #endif 1160 } 1161 1162 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1163 ({ int __rc, __dis = __sk->sk_disconnects; \ 1164 release_sock(__sk); \ 1165 __rc = __condition; \ 1166 if (!__rc) { \ 1167 *(__timeo) = wait_woken(__wait, \ 1168 TASK_INTERRUPTIBLE, \ 1169 *(__timeo)); \ 1170 } \ 1171 sched_annotate_sleep(); \ 1172 lock_sock(__sk); \ 1173 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1174 __rc; \ 1175 }) 1176 1177 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1178 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1179 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1180 int sk_stream_error(struct sock *sk, int flags, int err); 1181 void sk_stream_kill_queues(struct sock *sk); 1182 void sk_set_memalloc(struct sock *sk); 1183 void sk_clear_memalloc(struct sock *sk); 1184 1185 void __sk_flush_backlog(struct sock *sk); 1186 1187 static inline bool sk_flush_backlog(struct sock *sk) 1188 { 1189 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1190 __sk_flush_backlog(sk); 1191 return true; 1192 } 1193 return false; 1194 } 1195 1196 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1197 1198 struct request_sock_ops; 1199 struct timewait_sock_ops; 1200 struct inet_hashinfo; 1201 struct raw_hashinfo; 1202 struct smc_hashinfo; 1203 struct module; 1204 struct sk_psock; 1205 1206 /* 1207 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1208 * un-modified. Special care is taken when initializing object to zero. 1209 */ 1210 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1211 { 1212 if (offsetof(struct sock, sk_node.next) != 0) 1213 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1214 memset(&sk->sk_node.pprev, 0, 1215 size - offsetof(struct sock, sk_node.pprev)); 1216 } 1217 1218 /* Networking protocol blocks we attach to sockets. 1219 * socket layer -> transport layer interface 1220 */ 1221 struct proto { 1222 void (*close)(struct sock *sk, 1223 long timeout); 1224 int (*pre_connect)(struct sock *sk, 1225 struct sockaddr *uaddr, 1226 int addr_len); 1227 int (*connect)(struct sock *sk, 1228 struct sockaddr *uaddr, 1229 int addr_len); 1230 int (*disconnect)(struct sock *sk, int flags); 1231 1232 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1233 bool kern); 1234 1235 int (*ioctl)(struct sock *sk, int cmd, 1236 int *karg); 1237 int (*init)(struct sock *sk); 1238 void (*destroy)(struct sock *sk); 1239 void (*shutdown)(struct sock *sk, int how); 1240 int (*setsockopt)(struct sock *sk, int level, 1241 int optname, sockptr_t optval, 1242 unsigned int optlen); 1243 int (*getsockopt)(struct sock *sk, int level, 1244 int optname, char __user *optval, 1245 int __user *option); 1246 void (*keepalive)(struct sock *sk, int valbool); 1247 #ifdef CONFIG_COMPAT 1248 int (*compat_ioctl)(struct sock *sk, 1249 unsigned int cmd, unsigned long arg); 1250 #endif 1251 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1252 size_t len); 1253 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1254 size_t len, int flags, int *addr_len); 1255 void (*splice_eof)(struct socket *sock); 1256 int (*bind)(struct sock *sk, 1257 struct sockaddr *addr, int addr_len); 1258 int (*bind_add)(struct sock *sk, 1259 struct sockaddr *addr, int addr_len); 1260 1261 int (*backlog_rcv) (struct sock *sk, 1262 struct sk_buff *skb); 1263 bool (*bpf_bypass_getsockopt)(int level, 1264 int optname); 1265 1266 void (*release_cb)(struct sock *sk); 1267 1268 /* Keeping track of sk's, looking them up, and port selection methods. */ 1269 int (*hash)(struct sock *sk); 1270 void (*unhash)(struct sock *sk); 1271 void (*rehash)(struct sock *sk); 1272 int (*get_port)(struct sock *sk, unsigned short snum); 1273 void (*put_port)(struct sock *sk); 1274 #ifdef CONFIG_BPF_SYSCALL 1275 int (*psock_update_sk_prot)(struct sock *sk, 1276 struct sk_psock *psock, 1277 bool restore); 1278 #endif 1279 1280 /* Keeping track of sockets in use */ 1281 #ifdef CONFIG_PROC_FS 1282 unsigned int inuse_idx; 1283 #endif 1284 1285 #if IS_ENABLED(CONFIG_MPTCP) 1286 int (*forward_alloc_get)(const struct sock *sk); 1287 #endif 1288 1289 bool (*stream_memory_free)(const struct sock *sk, int wake); 1290 bool (*sock_is_readable)(struct sock *sk); 1291 /* Memory pressure */ 1292 void (*enter_memory_pressure)(struct sock *sk); 1293 void (*leave_memory_pressure)(struct sock *sk); 1294 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1295 int __percpu *per_cpu_fw_alloc; 1296 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1297 1298 /* 1299 * Pressure flag: try to collapse. 1300 * Technical note: it is used by multiple contexts non atomically. 1301 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1302 * All the __sk_mem_schedule() is of this nature: accounting 1303 * is strict, actions are advisory and have some latency. 1304 */ 1305 unsigned long *memory_pressure; 1306 long *sysctl_mem; 1307 1308 int *sysctl_wmem; 1309 int *sysctl_rmem; 1310 u32 sysctl_wmem_offset; 1311 u32 sysctl_rmem_offset; 1312 1313 int max_header; 1314 bool no_autobind; 1315 1316 struct kmem_cache *slab; 1317 unsigned int obj_size; 1318 unsigned int ipv6_pinfo_offset; 1319 slab_flags_t slab_flags; 1320 unsigned int useroffset; /* Usercopy region offset */ 1321 unsigned int usersize; /* Usercopy region size */ 1322 1323 unsigned int __percpu *orphan_count; 1324 1325 struct request_sock_ops *rsk_prot; 1326 struct timewait_sock_ops *twsk_prot; 1327 1328 union { 1329 struct inet_hashinfo *hashinfo; 1330 struct udp_table *udp_table; 1331 struct raw_hashinfo *raw_hash; 1332 struct smc_hashinfo *smc_hash; 1333 } h; 1334 1335 struct module *owner; 1336 1337 char name[32]; 1338 1339 struct list_head node; 1340 int (*diag_destroy)(struct sock *sk, int err); 1341 } __randomize_layout; 1342 1343 int proto_register(struct proto *prot, int alloc_slab); 1344 void proto_unregister(struct proto *prot); 1345 int sock_load_diag_module(int family, int protocol); 1346 1347 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1348 1349 static inline int sk_forward_alloc_get(const struct sock *sk) 1350 { 1351 #if IS_ENABLED(CONFIG_MPTCP) 1352 if (sk->sk_prot->forward_alloc_get) 1353 return sk->sk_prot->forward_alloc_get(sk); 1354 #endif 1355 return READ_ONCE(sk->sk_forward_alloc); 1356 } 1357 1358 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1359 { 1360 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1361 return false; 1362 1363 return sk->sk_prot->stream_memory_free ? 1364 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1365 tcp_stream_memory_free, sk, wake) : true; 1366 } 1367 1368 static inline bool sk_stream_memory_free(const struct sock *sk) 1369 { 1370 return __sk_stream_memory_free(sk, 0); 1371 } 1372 1373 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1374 { 1375 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1376 __sk_stream_memory_free(sk, wake); 1377 } 1378 1379 static inline bool sk_stream_is_writeable(const struct sock *sk) 1380 { 1381 return __sk_stream_is_writeable(sk, 0); 1382 } 1383 1384 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1385 struct cgroup *ancestor) 1386 { 1387 #ifdef CONFIG_SOCK_CGROUP_DATA 1388 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1389 ancestor); 1390 #else 1391 return -ENOTSUPP; 1392 #endif 1393 } 1394 1395 static inline bool sk_has_memory_pressure(const struct sock *sk) 1396 { 1397 return sk->sk_prot->memory_pressure != NULL; 1398 } 1399 1400 static inline bool sk_under_global_memory_pressure(const struct sock *sk) 1401 { 1402 return sk->sk_prot->memory_pressure && 1403 !!READ_ONCE(*sk->sk_prot->memory_pressure); 1404 } 1405 1406 static inline bool sk_under_memory_pressure(const struct sock *sk) 1407 { 1408 if (!sk->sk_prot->memory_pressure) 1409 return false; 1410 1411 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1412 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1413 return true; 1414 1415 return !!READ_ONCE(*sk->sk_prot->memory_pressure); 1416 } 1417 1418 static inline long 1419 proto_memory_allocated(const struct proto *prot) 1420 { 1421 return max(0L, atomic_long_read(prot->memory_allocated)); 1422 } 1423 1424 static inline long 1425 sk_memory_allocated(const struct sock *sk) 1426 { 1427 return proto_memory_allocated(sk->sk_prot); 1428 } 1429 1430 /* 1 MB per cpu, in page units */ 1431 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1432 1433 static inline void 1434 sk_memory_allocated_add(struct sock *sk, int amt) 1435 { 1436 int local_reserve; 1437 1438 preempt_disable(); 1439 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1440 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) { 1441 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1442 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1443 } 1444 preempt_enable(); 1445 } 1446 1447 static inline void 1448 sk_memory_allocated_sub(struct sock *sk, int amt) 1449 { 1450 int local_reserve; 1451 1452 preempt_disable(); 1453 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1454 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) { 1455 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1456 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1457 } 1458 preempt_enable(); 1459 } 1460 1461 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1462 1463 static inline void sk_sockets_allocated_dec(struct sock *sk) 1464 { 1465 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1466 SK_ALLOC_PERCPU_COUNTER_BATCH); 1467 } 1468 1469 static inline void sk_sockets_allocated_inc(struct sock *sk) 1470 { 1471 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1472 SK_ALLOC_PERCPU_COUNTER_BATCH); 1473 } 1474 1475 static inline u64 1476 sk_sockets_allocated_read_positive(struct sock *sk) 1477 { 1478 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1479 } 1480 1481 static inline int 1482 proto_sockets_allocated_sum_positive(struct proto *prot) 1483 { 1484 return percpu_counter_sum_positive(prot->sockets_allocated); 1485 } 1486 1487 static inline bool 1488 proto_memory_pressure(struct proto *prot) 1489 { 1490 if (!prot->memory_pressure) 1491 return false; 1492 return !!READ_ONCE(*prot->memory_pressure); 1493 } 1494 1495 1496 #ifdef CONFIG_PROC_FS 1497 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1498 struct prot_inuse { 1499 int all; 1500 int val[PROTO_INUSE_NR]; 1501 }; 1502 1503 static inline void sock_prot_inuse_add(const struct net *net, 1504 const struct proto *prot, int val) 1505 { 1506 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1507 } 1508 1509 static inline void sock_inuse_add(const struct net *net, int val) 1510 { 1511 this_cpu_add(net->core.prot_inuse->all, val); 1512 } 1513 1514 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1515 int sock_inuse_get(struct net *net); 1516 #else 1517 static inline void sock_prot_inuse_add(const struct net *net, 1518 const struct proto *prot, int val) 1519 { 1520 } 1521 1522 static inline void sock_inuse_add(const struct net *net, int val) 1523 { 1524 } 1525 #endif 1526 1527 1528 /* With per-bucket locks this operation is not-atomic, so that 1529 * this version is not worse. 1530 */ 1531 static inline int __sk_prot_rehash(struct sock *sk) 1532 { 1533 sk->sk_prot->unhash(sk); 1534 return sk->sk_prot->hash(sk); 1535 } 1536 1537 /* About 10 seconds */ 1538 #define SOCK_DESTROY_TIME (10*HZ) 1539 1540 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1541 #define PROT_SOCK 1024 1542 1543 #define SHUTDOWN_MASK 3 1544 #define RCV_SHUTDOWN 1 1545 #define SEND_SHUTDOWN 2 1546 1547 #define SOCK_BINDADDR_LOCK 4 1548 #define SOCK_BINDPORT_LOCK 8 1549 1550 struct socket_alloc { 1551 struct socket socket; 1552 struct inode vfs_inode; 1553 }; 1554 1555 static inline struct socket *SOCKET_I(struct inode *inode) 1556 { 1557 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1558 } 1559 1560 static inline struct inode *SOCK_INODE(struct socket *socket) 1561 { 1562 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1563 } 1564 1565 /* 1566 * Functions for memory accounting 1567 */ 1568 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1569 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1570 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1571 void __sk_mem_reclaim(struct sock *sk, int amount); 1572 1573 #define SK_MEM_SEND 0 1574 #define SK_MEM_RECV 1 1575 1576 /* sysctl_mem values are in pages */ 1577 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1578 { 1579 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1580 } 1581 1582 static inline int sk_mem_pages(int amt) 1583 { 1584 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1585 } 1586 1587 static inline bool sk_has_account(struct sock *sk) 1588 { 1589 /* return true if protocol supports memory accounting */ 1590 return !!sk->sk_prot->memory_allocated; 1591 } 1592 1593 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1594 { 1595 int delta; 1596 1597 if (!sk_has_account(sk)) 1598 return true; 1599 delta = size - sk->sk_forward_alloc; 1600 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1601 } 1602 1603 static inline bool 1604 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1605 { 1606 int delta; 1607 1608 if (!sk_has_account(sk)) 1609 return true; 1610 delta = size - sk->sk_forward_alloc; 1611 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1612 skb_pfmemalloc(skb); 1613 } 1614 1615 static inline int sk_unused_reserved_mem(const struct sock *sk) 1616 { 1617 int unused_mem; 1618 1619 if (likely(!sk->sk_reserved_mem)) 1620 return 0; 1621 1622 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1623 atomic_read(&sk->sk_rmem_alloc); 1624 1625 return unused_mem > 0 ? unused_mem : 0; 1626 } 1627 1628 static inline void sk_mem_reclaim(struct sock *sk) 1629 { 1630 int reclaimable; 1631 1632 if (!sk_has_account(sk)) 1633 return; 1634 1635 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1636 1637 if (reclaimable >= (int)PAGE_SIZE) 1638 __sk_mem_reclaim(sk, reclaimable); 1639 } 1640 1641 static inline void sk_mem_reclaim_final(struct sock *sk) 1642 { 1643 sk->sk_reserved_mem = 0; 1644 sk_mem_reclaim(sk); 1645 } 1646 1647 static inline void sk_mem_charge(struct sock *sk, int size) 1648 { 1649 if (!sk_has_account(sk)) 1650 return; 1651 sk_forward_alloc_add(sk, -size); 1652 } 1653 1654 static inline void sk_mem_uncharge(struct sock *sk, int size) 1655 { 1656 if (!sk_has_account(sk)) 1657 return; 1658 sk_forward_alloc_add(sk, size); 1659 sk_mem_reclaim(sk); 1660 } 1661 1662 /* 1663 * Macro so as to not evaluate some arguments when 1664 * lockdep is not enabled. 1665 * 1666 * Mark both the sk_lock and the sk_lock.slock as a 1667 * per-address-family lock class. 1668 */ 1669 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1670 do { \ 1671 sk->sk_lock.owned = 0; \ 1672 init_waitqueue_head(&sk->sk_lock.wq); \ 1673 spin_lock_init(&(sk)->sk_lock.slock); \ 1674 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1675 sizeof((sk)->sk_lock)); \ 1676 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1677 (skey), (sname)); \ 1678 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1679 } while (0) 1680 1681 static inline bool lockdep_sock_is_held(const struct sock *sk) 1682 { 1683 return lockdep_is_held(&sk->sk_lock) || 1684 lockdep_is_held(&sk->sk_lock.slock); 1685 } 1686 1687 void lock_sock_nested(struct sock *sk, int subclass); 1688 1689 static inline void lock_sock(struct sock *sk) 1690 { 1691 lock_sock_nested(sk, 0); 1692 } 1693 1694 void __lock_sock(struct sock *sk); 1695 void __release_sock(struct sock *sk); 1696 void release_sock(struct sock *sk); 1697 1698 /* BH context may only use the following locking interface. */ 1699 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1700 #define bh_lock_sock_nested(__sk) \ 1701 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1702 SINGLE_DEPTH_NESTING) 1703 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1704 1705 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1706 1707 /** 1708 * lock_sock_fast - fast version of lock_sock 1709 * @sk: socket 1710 * 1711 * This version should be used for very small section, where process wont block 1712 * return false if fast path is taken: 1713 * 1714 * sk_lock.slock locked, owned = 0, BH disabled 1715 * 1716 * return true if slow path is taken: 1717 * 1718 * sk_lock.slock unlocked, owned = 1, BH enabled 1719 */ 1720 static inline bool lock_sock_fast(struct sock *sk) 1721 { 1722 /* The sk_lock has mutex_lock() semantics here. */ 1723 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1724 1725 return __lock_sock_fast(sk); 1726 } 1727 1728 /* fast socket lock variant for caller already holding a [different] socket lock */ 1729 static inline bool lock_sock_fast_nested(struct sock *sk) 1730 { 1731 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1732 1733 return __lock_sock_fast(sk); 1734 } 1735 1736 /** 1737 * unlock_sock_fast - complement of lock_sock_fast 1738 * @sk: socket 1739 * @slow: slow mode 1740 * 1741 * fast unlock socket for user context. 1742 * If slow mode is on, we call regular release_sock() 1743 */ 1744 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1745 __releases(&sk->sk_lock.slock) 1746 { 1747 if (slow) { 1748 release_sock(sk); 1749 __release(&sk->sk_lock.slock); 1750 } else { 1751 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1752 spin_unlock_bh(&sk->sk_lock.slock); 1753 } 1754 } 1755 1756 void sockopt_lock_sock(struct sock *sk); 1757 void sockopt_release_sock(struct sock *sk); 1758 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1759 bool sockopt_capable(int cap); 1760 1761 /* Used by processes to "lock" a socket state, so that 1762 * interrupts and bottom half handlers won't change it 1763 * from under us. It essentially blocks any incoming 1764 * packets, so that we won't get any new data or any 1765 * packets that change the state of the socket. 1766 * 1767 * While locked, BH processing will add new packets to 1768 * the backlog queue. This queue is processed by the 1769 * owner of the socket lock right before it is released. 1770 * 1771 * Since ~2.3.5 it is also exclusive sleep lock serializing 1772 * accesses from user process context. 1773 */ 1774 1775 static inline void sock_owned_by_me(const struct sock *sk) 1776 { 1777 #ifdef CONFIG_LOCKDEP 1778 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1779 #endif 1780 } 1781 1782 static inline bool sock_owned_by_user(const struct sock *sk) 1783 { 1784 sock_owned_by_me(sk); 1785 return sk->sk_lock.owned; 1786 } 1787 1788 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1789 { 1790 return sk->sk_lock.owned; 1791 } 1792 1793 static inline void sock_release_ownership(struct sock *sk) 1794 { 1795 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1796 sk->sk_lock.owned = 0; 1797 1798 /* The sk_lock has mutex_unlock() semantics: */ 1799 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1800 } 1801 1802 /* no reclassification while locks are held */ 1803 static inline bool sock_allow_reclassification(const struct sock *csk) 1804 { 1805 struct sock *sk = (struct sock *)csk; 1806 1807 return !sock_owned_by_user_nocheck(sk) && 1808 !spin_is_locked(&sk->sk_lock.slock); 1809 } 1810 1811 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1812 struct proto *prot, int kern); 1813 void sk_free(struct sock *sk); 1814 void sk_destruct(struct sock *sk); 1815 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1816 void sk_free_unlock_clone(struct sock *sk); 1817 1818 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1819 gfp_t priority); 1820 void __sock_wfree(struct sk_buff *skb); 1821 void sock_wfree(struct sk_buff *skb); 1822 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1823 gfp_t priority); 1824 void skb_orphan_partial(struct sk_buff *skb); 1825 void sock_rfree(struct sk_buff *skb); 1826 void sock_efree(struct sk_buff *skb); 1827 #ifdef CONFIG_INET 1828 void sock_edemux(struct sk_buff *skb); 1829 void sock_pfree(struct sk_buff *skb); 1830 #else 1831 #define sock_edemux sock_efree 1832 #endif 1833 1834 int sk_setsockopt(struct sock *sk, int level, int optname, 1835 sockptr_t optval, unsigned int optlen); 1836 int sock_setsockopt(struct socket *sock, int level, int op, 1837 sockptr_t optval, unsigned int optlen); 1838 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1839 int optname, sockptr_t optval, int optlen); 1840 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1841 int optname, sockptr_t optval, sockptr_t optlen); 1842 1843 int sk_getsockopt(struct sock *sk, int level, int optname, 1844 sockptr_t optval, sockptr_t optlen); 1845 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1846 bool timeval, bool time32); 1847 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1848 unsigned long data_len, int noblock, 1849 int *errcode, int max_page_order); 1850 1851 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1852 unsigned long size, 1853 int noblock, int *errcode) 1854 { 1855 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1856 } 1857 1858 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1859 void sock_kfree_s(struct sock *sk, void *mem, int size); 1860 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1861 void sk_send_sigurg(struct sock *sk); 1862 1863 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1864 { 1865 if (sk->sk_socket) 1866 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1867 WRITE_ONCE(sk->sk_prot, proto); 1868 } 1869 1870 struct sockcm_cookie { 1871 u64 transmit_time; 1872 u32 mark; 1873 u32 tsflags; 1874 }; 1875 1876 static inline void sockcm_init(struct sockcm_cookie *sockc, 1877 const struct sock *sk) 1878 { 1879 *sockc = (struct sockcm_cookie) { 1880 .tsflags = READ_ONCE(sk->sk_tsflags) 1881 }; 1882 } 1883 1884 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1885 struct sockcm_cookie *sockc); 1886 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1887 struct sockcm_cookie *sockc); 1888 1889 /* 1890 * Functions to fill in entries in struct proto_ops when a protocol 1891 * does not implement a particular function. 1892 */ 1893 int sock_no_bind(struct socket *, struct sockaddr *, int); 1894 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1895 int sock_no_socketpair(struct socket *, struct socket *); 1896 int sock_no_accept(struct socket *, struct socket *, int, bool); 1897 int sock_no_getname(struct socket *, struct sockaddr *, int); 1898 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1899 int sock_no_listen(struct socket *, int); 1900 int sock_no_shutdown(struct socket *, int); 1901 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1902 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1903 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1904 int sock_no_mmap(struct file *file, struct socket *sock, 1905 struct vm_area_struct *vma); 1906 1907 /* 1908 * Functions to fill in entries in struct proto_ops when a protocol 1909 * uses the inet style. 1910 */ 1911 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1912 char __user *optval, int __user *optlen); 1913 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1914 int flags); 1915 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1916 sockptr_t optval, unsigned int optlen); 1917 1918 void sk_common_release(struct sock *sk); 1919 1920 /* 1921 * Default socket callbacks and setup code 1922 */ 1923 1924 /* Initialise core socket variables using an explicit uid. */ 1925 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1926 1927 /* Initialise core socket variables. 1928 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1929 */ 1930 void sock_init_data(struct socket *sock, struct sock *sk); 1931 1932 /* 1933 * Socket reference counting postulates. 1934 * 1935 * * Each user of socket SHOULD hold a reference count. 1936 * * Each access point to socket (an hash table bucket, reference from a list, 1937 * running timer, skb in flight MUST hold a reference count. 1938 * * When reference count hits 0, it means it will never increase back. 1939 * * When reference count hits 0, it means that no references from 1940 * outside exist to this socket and current process on current CPU 1941 * is last user and may/should destroy this socket. 1942 * * sk_free is called from any context: process, BH, IRQ. When 1943 * it is called, socket has no references from outside -> sk_free 1944 * may release descendant resources allocated by the socket, but 1945 * to the time when it is called, socket is NOT referenced by any 1946 * hash tables, lists etc. 1947 * * Packets, delivered from outside (from network or from another process) 1948 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1949 * when they sit in queue. Otherwise, packets will leak to hole, when 1950 * socket is looked up by one cpu and unhasing is made by another CPU. 1951 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1952 * (leak to backlog). Packet socket does all the processing inside 1953 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1954 * use separate SMP lock, so that they are prone too. 1955 */ 1956 1957 /* Ungrab socket and destroy it, if it was the last reference. */ 1958 static inline void sock_put(struct sock *sk) 1959 { 1960 if (refcount_dec_and_test(&sk->sk_refcnt)) 1961 sk_free(sk); 1962 } 1963 /* Generic version of sock_put(), dealing with all sockets 1964 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1965 */ 1966 void sock_gen_put(struct sock *sk); 1967 1968 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1969 unsigned int trim_cap, bool refcounted); 1970 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1971 const int nested) 1972 { 1973 return __sk_receive_skb(sk, skb, nested, 1, true); 1974 } 1975 1976 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1977 { 1978 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1979 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1980 return; 1981 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1982 * other WRITE_ONCE() because socket lock might be not held. 1983 */ 1984 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 1985 } 1986 1987 #define NO_QUEUE_MAPPING USHRT_MAX 1988 1989 static inline void sk_tx_queue_clear(struct sock *sk) 1990 { 1991 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1992 * other WRITE_ONCE() because socket lock might be not held. 1993 */ 1994 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 1995 } 1996 1997 static inline int sk_tx_queue_get(const struct sock *sk) 1998 { 1999 if (sk) { 2000 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 2001 * and sk_tx_queue_set(). 2002 */ 2003 int val = READ_ONCE(sk->sk_tx_queue_mapping); 2004 2005 if (val != NO_QUEUE_MAPPING) 2006 return val; 2007 } 2008 return -1; 2009 } 2010 2011 static inline void __sk_rx_queue_set(struct sock *sk, 2012 const struct sk_buff *skb, 2013 bool force_set) 2014 { 2015 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2016 if (skb_rx_queue_recorded(skb)) { 2017 u16 rx_queue = skb_get_rx_queue(skb); 2018 2019 if (force_set || 2020 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2021 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2022 } 2023 #endif 2024 } 2025 2026 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2027 { 2028 __sk_rx_queue_set(sk, skb, true); 2029 } 2030 2031 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2032 { 2033 __sk_rx_queue_set(sk, skb, false); 2034 } 2035 2036 static inline void sk_rx_queue_clear(struct sock *sk) 2037 { 2038 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2039 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2040 #endif 2041 } 2042 2043 static inline int sk_rx_queue_get(const struct sock *sk) 2044 { 2045 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2046 if (sk) { 2047 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2048 2049 if (res != NO_QUEUE_MAPPING) 2050 return res; 2051 } 2052 #endif 2053 2054 return -1; 2055 } 2056 2057 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2058 { 2059 sk->sk_socket = sock; 2060 } 2061 2062 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2063 { 2064 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2065 return &rcu_dereference_raw(sk->sk_wq)->wait; 2066 } 2067 /* Detach socket from process context. 2068 * Announce socket dead, detach it from wait queue and inode. 2069 * Note that parent inode held reference count on this struct sock, 2070 * we do not release it in this function, because protocol 2071 * probably wants some additional cleanups or even continuing 2072 * to work with this socket (TCP). 2073 */ 2074 static inline void sock_orphan(struct sock *sk) 2075 { 2076 write_lock_bh(&sk->sk_callback_lock); 2077 sock_set_flag(sk, SOCK_DEAD); 2078 sk_set_socket(sk, NULL); 2079 sk->sk_wq = NULL; 2080 write_unlock_bh(&sk->sk_callback_lock); 2081 } 2082 2083 static inline void sock_graft(struct sock *sk, struct socket *parent) 2084 { 2085 WARN_ON(parent->sk); 2086 write_lock_bh(&sk->sk_callback_lock); 2087 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2088 parent->sk = sk; 2089 sk_set_socket(sk, parent); 2090 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2091 security_sock_graft(sk, parent); 2092 write_unlock_bh(&sk->sk_callback_lock); 2093 } 2094 2095 kuid_t sock_i_uid(struct sock *sk); 2096 unsigned long __sock_i_ino(struct sock *sk); 2097 unsigned long sock_i_ino(struct sock *sk); 2098 2099 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2100 { 2101 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2102 } 2103 2104 static inline u32 net_tx_rndhash(void) 2105 { 2106 u32 v = get_random_u32(); 2107 2108 return v ?: 1; 2109 } 2110 2111 static inline void sk_set_txhash(struct sock *sk) 2112 { 2113 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2114 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2115 } 2116 2117 static inline bool sk_rethink_txhash(struct sock *sk) 2118 { 2119 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2120 sk_set_txhash(sk); 2121 return true; 2122 } 2123 return false; 2124 } 2125 2126 static inline struct dst_entry * 2127 __sk_dst_get(const struct sock *sk) 2128 { 2129 return rcu_dereference_check(sk->sk_dst_cache, 2130 lockdep_sock_is_held(sk)); 2131 } 2132 2133 static inline struct dst_entry * 2134 sk_dst_get(const struct sock *sk) 2135 { 2136 struct dst_entry *dst; 2137 2138 rcu_read_lock(); 2139 dst = rcu_dereference(sk->sk_dst_cache); 2140 if (dst && !rcuref_get(&dst->__rcuref)) 2141 dst = NULL; 2142 rcu_read_unlock(); 2143 return dst; 2144 } 2145 2146 static inline void __dst_negative_advice(struct sock *sk) 2147 { 2148 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2149 2150 if (dst && dst->ops->negative_advice) { 2151 ndst = dst->ops->negative_advice(dst); 2152 2153 if (ndst != dst) { 2154 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2155 sk_tx_queue_clear(sk); 2156 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2157 } 2158 } 2159 } 2160 2161 static inline void dst_negative_advice(struct sock *sk) 2162 { 2163 sk_rethink_txhash(sk); 2164 __dst_negative_advice(sk); 2165 } 2166 2167 static inline void 2168 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2169 { 2170 struct dst_entry *old_dst; 2171 2172 sk_tx_queue_clear(sk); 2173 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2174 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2175 lockdep_sock_is_held(sk)); 2176 rcu_assign_pointer(sk->sk_dst_cache, dst); 2177 dst_release(old_dst); 2178 } 2179 2180 static inline void 2181 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2182 { 2183 struct dst_entry *old_dst; 2184 2185 sk_tx_queue_clear(sk); 2186 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2187 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2188 dst_release(old_dst); 2189 } 2190 2191 static inline void 2192 __sk_dst_reset(struct sock *sk) 2193 { 2194 __sk_dst_set(sk, NULL); 2195 } 2196 2197 static inline void 2198 sk_dst_reset(struct sock *sk) 2199 { 2200 sk_dst_set(sk, NULL); 2201 } 2202 2203 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2204 2205 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2206 2207 static inline void sk_dst_confirm(struct sock *sk) 2208 { 2209 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2210 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2211 } 2212 2213 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2214 { 2215 if (skb_get_dst_pending_confirm(skb)) { 2216 struct sock *sk = skb->sk; 2217 2218 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2219 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2220 neigh_confirm(n); 2221 } 2222 } 2223 2224 bool sk_mc_loop(const struct sock *sk); 2225 2226 static inline bool sk_can_gso(const struct sock *sk) 2227 { 2228 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2229 } 2230 2231 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2232 2233 static inline void sk_gso_disable(struct sock *sk) 2234 { 2235 sk->sk_gso_disabled = 1; 2236 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2237 } 2238 2239 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2240 struct iov_iter *from, char *to, 2241 int copy, int offset) 2242 { 2243 if (skb->ip_summed == CHECKSUM_NONE) { 2244 __wsum csum = 0; 2245 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2246 return -EFAULT; 2247 skb->csum = csum_block_add(skb->csum, csum, offset); 2248 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2249 if (!copy_from_iter_full_nocache(to, copy, from)) 2250 return -EFAULT; 2251 } else if (!copy_from_iter_full(to, copy, from)) 2252 return -EFAULT; 2253 2254 return 0; 2255 } 2256 2257 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2258 struct iov_iter *from, int copy) 2259 { 2260 int err, offset = skb->len; 2261 2262 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2263 copy, offset); 2264 if (err) 2265 __skb_trim(skb, offset); 2266 2267 return err; 2268 } 2269 2270 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2271 struct sk_buff *skb, 2272 struct page *page, 2273 int off, int copy) 2274 { 2275 int err; 2276 2277 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2278 copy, skb->len); 2279 if (err) 2280 return err; 2281 2282 skb_len_add(skb, copy); 2283 sk_wmem_queued_add(sk, copy); 2284 sk_mem_charge(sk, copy); 2285 return 0; 2286 } 2287 2288 /** 2289 * sk_wmem_alloc_get - returns write allocations 2290 * @sk: socket 2291 * 2292 * Return: sk_wmem_alloc minus initial offset of one 2293 */ 2294 static inline int sk_wmem_alloc_get(const struct sock *sk) 2295 { 2296 return refcount_read(&sk->sk_wmem_alloc) - 1; 2297 } 2298 2299 /** 2300 * sk_rmem_alloc_get - returns read allocations 2301 * @sk: socket 2302 * 2303 * Return: sk_rmem_alloc 2304 */ 2305 static inline int sk_rmem_alloc_get(const struct sock *sk) 2306 { 2307 return atomic_read(&sk->sk_rmem_alloc); 2308 } 2309 2310 /** 2311 * sk_has_allocations - check if allocations are outstanding 2312 * @sk: socket 2313 * 2314 * Return: true if socket has write or read allocations 2315 */ 2316 static inline bool sk_has_allocations(const struct sock *sk) 2317 { 2318 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2319 } 2320 2321 /** 2322 * skwq_has_sleeper - check if there are any waiting processes 2323 * @wq: struct socket_wq 2324 * 2325 * Return: true if socket_wq has waiting processes 2326 * 2327 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2328 * barrier call. They were added due to the race found within the tcp code. 2329 * 2330 * Consider following tcp code paths:: 2331 * 2332 * CPU1 CPU2 2333 * sys_select receive packet 2334 * ... ... 2335 * __add_wait_queue update tp->rcv_nxt 2336 * ... ... 2337 * tp->rcv_nxt check sock_def_readable 2338 * ... { 2339 * schedule rcu_read_lock(); 2340 * wq = rcu_dereference(sk->sk_wq); 2341 * if (wq && waitqueue_active(&wq->wait)) 2342 * wake_up_interruptible(&wq->wait) 2343 * ... 2344 * } 2345 * 2346 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2347 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2348 * could then endup calling schedule and sleep forever if there are no more 2349 * data on the socket. 2350 * 2351 */ 2352 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2353 { 2354 return wq && wq_has_sleeper(&wq->wait); 2355 } 2356 2357 /** 2358 * sock_poll_wait - place memory barrier behind the poll_wait call. 2359 * @filp: file 2360 * @sock: socket to wait on 2361 * @p: poll_table 2362 * 2363 * See the comments in the wq_has_sleeper function. 2364 */ 2365 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2366 poll_table *p) 2367 { 2368 if (!poll_does_not_wait(p)) { 2369 poll_wait(filp, &sock->wq.wait, p); 2370 /* We need to be sure we are in sync with the 2371 * socket flags modification. 2372 * 2373 * This memory barrier is paired in the wq_has_sleeper. 2374 */ 2375 smp_mb(); 2376 } 2377 } 2378 2379 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2380 { 2381 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2382 u32 txhash = READ_ONCE(sk->sk_txhash); 2383 2384 if (txhash) { 2385 skb->l4_hash = 1; 2386 skb->hash = txhash; 2387 } 2388 } 2389 2390 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2391 2392 /* 2393 * Queue a received datagram if it will fit. Stream and sequenced 2394 * protocols can't normally use this as they need to fit buffers in 2395 * and play with them. 2396 * 2397 * Inlined as it's very short and called for pretty much every 2398 * packet ever received. 2399 */ 2400 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2401 { 2402 skb_orphan(skb); 2403 skb->sk = sk; 2404 skb->destructor = sock_rfree; 2405 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2406 sk_mem_charge(sk, skb->truesize); 2407 } 2408 2409 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2410 { 2411 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2412 skb_orphan(skb); 2413 skb->destructor = sock_efree; 2414 skb->sk = sk; 2415 return true; 2416 } 2417 return false; 2418 } 2419 2420 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2421 { 2422 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2423 if (skb) { 2424 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2425 skb_set_owner_r(skb, sk); 2426 return skb; 2427 } 2428 __kfree_skb(skb); 2429 } 2430 return NULL; 2431 } 2432 2433 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2434 { 2435 if (skb->destructor != sock_wfree) { 2436 skb_orphan(skb); 2437 return; 2438 } 2439 skb->slow_gro = 1; 2440 } 2441 2442 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2443 unsigned long expires); 2444 2445 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2446 2447 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2448 2449 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2450 struct sk_buff *skb, unsigned int flags, 2451 void (*destructor)(struct sock *sk, 2452 struct sk_buff *skb)); 2453 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2454 2455 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2456 enum skb_drop_reason *reason); 2457 2458 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2459 { 2460 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2461 } 2462 2463 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2464 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2465 2466 /* 2467 * Recover an error report and clear atomically 2468 */ 2469 2470 static inline int sock_error(struct sock *sk) 2471 { 2472 int err; 2473 2474 /* Avoid an atomic operation for the common case. 2475 * This is racy since another cpu/thread can change sk_err under us. 2476 */ 2477 if (likely(data_race(!sk->sk_err))) 2478 return 0; 2479 2480 err = xchg(&sk->sk_err, 0); 2481 return -err; 2482 } 2483 2484 void sk_error_report(struct sock *sk); 2485 2486 static inline unsigned long sock_wspace(struct sock *sk) 2487 { 2488 int amt = 0; 2489 2490 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2491 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2492 if (amt < 0) 2493 amt = 0; 2494 } 2495 return amt; 2496 } 2497 2498 /* Note: 2499 * We use sk->sk_wq_raw, from contexts knowing this 2500 * pointer is not NULL and cannot disappear/change. 2501 */ 2502 static inline void sk_set_bit(int nr, struct sock *sk) 2503 { 2504 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2505 !sock_flag(sk, SOCK_FASYNC)) 2506 return; 2507 2508 set_bit(nr, &sk->sk_wq_raw->flags); 2509 } 2510 2511 static inline void sk_clear_bit(int nr, struct sock *sk) 2512 { 2513 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2514 !sock_flag(sk, SOCK_FASYNC)) 2515 return; 2516 2517 clear_bit(nr, &sk->sk_wq_raw->flags); 2518 } 2519 2520 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2521 { 2522 if (sock_flag(sk, SOCK_FASYNC)) { 2523 rcu_read_lock(); 2524 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2525 rcu_read_unlock(); 2526 } 2527 } 2528 2529 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2530 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2531 * Note: for send buffers, TCP works better if we can build two skbs at 2532 * minimum. 2533 */ 2534 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2535 2536 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2537 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2538 2539 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2540 { 2541 u32 val; 2542 2543 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2544 return; 2545 2546 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2547 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2548 2549 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2550 } 2551 2552 /** 2553 * sk_page_frag - return an appropriate page_frag 2554 * @sk: socket 2555 * 2556 * Use the per task page_frag instead of the per socket one for 2557 * optimization when we know that we're in process context and own 2558 * everything that's associated with %current. 2559 * 2560 * Both direct reclaim and page faults can nest inside other 2561 * socket operations and end up recursing into sk_page_frag() 2562 * while it's already in use: explicitly avoid task page_frag 2563 * when users disable sk_use_task_frag. 2564 * 2565 * Return: a per task page_frag if context allows that, 2566 * otherwise a per socket one. 2567 */ 2568 static inline struct page_frag *sk_page_frag(struct sock *sk) 2569 { 2570 if (sk->sk_use_task_frag) 2571 return ¤t->task_frag; 2572 2573 return &sk->sk_frag; 2574 } 2575 2576 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2577 2578 /* 2579 * Default write policy as shown to user space via poll/select/SIGIO 2580 */ 2581 static inline bool sock_writeable(const struct sock *sk) 2582 { 2583 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2584 } 2585 2586 static inline gfp_t gfp_any(void) 2587 { 2588 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2589 } 2590 2591 static inline gfp_t gfp_memcg_charge(void) 2592 { 2593 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2594 } 2595 2596 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2597 { 2598 return noblock ? 0 : sk->sk_rcvtimeo; 2599 } 2600 2601 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2602 { 2603 return noblock ? 0 : sk->sk_sndtimeo; 2604 } 2605 2606 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2607 { 2608 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2609 2610 return v ?: 1; 2611 } 2612 2613 /* Alas, with timeout socket operations are not restartable. 2614 * Compare this to poll(). 2615 */ 2616 static inline int sock_intr_errno(long timeo) 2617 { 2618 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2619 } 2620 2621 struct sock_skb_cb { 2622 u32 dropcount; 2623 }; 2624 2625 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2626 * using skb->cb[] would keep using it directly and utilize its 2627 * alignement guarantee. 2628 */ 2629 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2630 sizeof(struct sock_skb_cb))) 2631 2632 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2633 SOCK_SKB_CB_OFFSET)) 2634 2635 #define sock_skb_cb_check_size(size) \ 2636 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2637 2638 static inline void 2639 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2640 { 2641 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2642 atomic_read(&sk->sk_drops) : 0; 2643 } 2644 2645 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2646 { 2647 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2648 2649 atomic_add(segs, &sk->sk_drops); 2650 } 2651 2652 static inline ktime_t sock_read_timestamp(struct sock *sk) 2653 { 2654 #if BITS_PER_LONG==32 2655 unsigned int seq; 2656 ktime_t kt; 2657 2658 do { 2659 seq = read_seqbegin(&sk->sk_stamp_seq); 2660 kt = sk->sk_stamp; 2661 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2662 2663 return kt; 2664 #else 2665 return READ_ONCE(sk->sk_stamp); 2666 #endif 2667 } 2668 2669 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2670 { 2671 #if BITS_PER_LONG==32 2672 write_seqlock(&sk->sk_stamp_seq); 2673 sk->sk_stamp = kt; 2674 write_sequnlock(&sk->sk_stamp_seq); 2675 #else 2676 WRITE_ONCE(sk->sk_stamp, kt); 2677 #endif 2678 } 2679 2680 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2681 struct sk_buff *skb); 2682 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2683 struct sk_buff *skb); 2684 2685 static inline void 2686 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2687 { 2688 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2689 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2690 ktime_t kt = skb->tstamp; 2691 /* 2692 * generate control messages if 2693 * - receive time stamping in software requested 2694 * - software time stamp available and wanted 2695 * - hardware time stamps available and wanted 2696 */ 2697 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2698 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2699 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2700 (hwtstamps->hwtstamp && 2701 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2702 __sock_recv_timestamp(msg, sk, skb); 2703 else 2704 sock_write_timestamp(sk, kt); 2705 2706 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2707 __sock_recv_wifi_status(msg, sk, skb); 2708 } 2709 2710 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2711 struct sk_buff *skb); 2712 2713 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2714 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2715 struct sk_buff *skb) 2716 { 2717 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2718 (1UL << SOCK_RCVTSTAMP) | \ 2719 (1UL << SOCK_RCVMARK)) 2720 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2721 SOF_TIMESTAMPING_RAW_HARDWARE) 2722 2723 if (sk->sk_flags & FLAGS_RECV_CMSGS || 2724 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY) 2725 __sock_recv_cmsgs(msg, sk, skb); 2726 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2727 sock_write_timestamp(sk, skb->tstamp); 2728 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2729 sock_write_timestamp(sk, 0); 2730 } 2731 2732 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2733 2734 /** 2735 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2736 * @sk: socket sending this packet 2737 * @tsflags: timestamping flags to use 2738 * @tx_flags: completed with instructions for time stamping 2739 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2740 * 2741 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2742 */ 2743 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2744 __u8 *tx_flags, __u32 *tskey) 2745 { 2746 if (unlikely(tsflags)) { 2747 __sock_tx_timestamp(tsflags, tx_flags); 2748 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2749 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2750 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2751 } 2752 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2753 *tx_flags |= SKBTX_WIFI_STATUS; 2754 } 2755 2756 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2757 __u8 *tx_flags) 2758 { 2759 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2760 } 2761 2762 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2763 { 2764 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2765 &skb_shinfo(skb)->tskey); 2766 } 2767 2768 static inline bool sk_is_inet(const struct sock *sk) 2769 { 2770 int family = READ_ONCE(sk->sk_family); 2771 2772 return family == AF_INET || family == AF_INET6; 2773 } 2774 2775 static inline bool sk_is_tcp(const struct sock *sk) 2776 { 2777 return sk_is_inet(sk) && 2778 sk->sk_type == SOCK_STREAM && 2779 sk->sk_protocol == IPPROTO_TCP; 2780 } 2781 2782 static inline bool sk_is_udp(const struct sock *sk) 2783 { 2784 return sk_is_inet(sk) && 2785 sk->sk_type == SOCK_DGRAM && 2786 sk->sk_protocol == IPPROTO_UDP; 2787 } 2788 2789 static inline bool sk_is_stream_unix(const struct sock *sk) 2790 { 2791 return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM; 2792 } 2793 2794 /** 2795 * sk_eat_skb - Release a skb if it is no longer needed 2796 * @sk: socket to eat this skb from 2797 * @skb: socket buffer to eat 2798 * 2799 * This routine must be called with interrupts disabled or with the socket 2800 * locked so that the sk_buff queue operation is ok. 2801 */ 2802 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2803 { 2804 __skb_unlink(skb, &sk->sk_receive_queue); 2805 __kfree_skb(skb); 2806 } 2807 2808 static inline bool 2809 skb_sk_is_prefetched(struct sk_buff *skb) 2810 { 2811 #ifdef CONFIG_INET 2812 return skb->destructor == sock_pfree; 2813 #else 2814 return false; 2815 #endif /* CONFIG_INET */ 2816 } 2817 2818 /* This helper checks if a socket is a full socket, 2819 * ie _not_ a timewait or request socket. 2820 */ 2821 static inline bool sk_fullsock(const struct sock *sk) 2822 { 2823 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2824 } 2825 2826 static inline bool 2827 sk_is_refcounted(struct sock *sk) 2828 { 2829 /* Only full sockets have sk->sk_flags. */ 2830 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2831 } 2832 2833 /** 2834 * skb_steal_sock - steal a socket from an sk_buff 2835 * @skb: sk_buff to steal the socket from 2836 * @refcounted: is set to true if the socket is reference-counted 2837 * @prefetched: is set to true if the socket was assigned from bpf 2838 */ 2839 static inline struct sock * 2840 skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched) 2841 { 2842 if (skb->sk) { 2843 struct sock *sk = skb->sk; 2844 2845 *refcounted = true; 2846 *prefetched = skb_sk_is_prefetched(skb); 2847 if (*prefetched) 2848 *refcounted = sk_is_refcounted(sk); 2849 skb->destructor = NULL; 2850 skb->sk = NULL; 2851 return sk; 2852 } 2853 *prefetched = false; 2854 *refcounted = false; 2855 return NULL; 2856 } 2857 2858 /* Checks if this SKB belongs to an HW offloaded socket 2859 * and whether any SW fallbacks are required based on dev. 2860 * Check decrypted mark in case skb_orphan() cleared socket. 2861 */ 2862 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2863 struct net_device *dev) 2864 { 2865 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2866 struct sock *sk = skb->sk; 2867 2868 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2869 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2870 #ifdef CONFIG_TLS_DEVICE 2871 } else if (unlikely(skb->decrypted)) { 2872 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2873 kfree_skb(skb); 2874 skb = NULL; 2875 #endif 2876 } 2877 #endif 2878 2879 return skb; 2880 } 2881 2882 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2883 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2884 */ 2885 static inline bool sk_listener(const struct sock *sk) 2886 { 2887 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2888 } 2889 2890 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2891 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2892 int type); 2893 2894 bool sk_ns_capable(const struct sock *sk, 2895 struct user_namespace *user_ns, int cap); 2896 bool sk_capable(const struct sock *sk, int cap); 2897 bool sk_net_capable(const struct sock *sk, int cap); 2898 2899 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2900 2901 /* Take into consideration the size of the struct sk_buff overhead in the 2902 * determination of these values, since that is non-constant across 2903 * platforms. This makes socket queueing behavior and performance 2904 * not depend upon such differences. 2905 */ 2906 #define _SK_MEM_PACKETS 256 2907 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2908 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2909 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2910 2911 extern __u32 sysctl_wmem_max; 2912 extern __u32 sysctl_rmem_max; 2913 2914 extern int sysctl_tstamp_allow_data; 2915 2916 extern __u32 sysctl_wmem_default; 2917 extern __u32 sysctl_rmem_default; 2918 2919 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2920 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2921 2922 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2923 { 2924 /* Does this proto have per netns sysctl_wmem ? */ 2925 if (proto->sysctl_wmem_offset) 2926 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2927 2928 return READ_ONCE(*proto->sysctl_wmem); 2929 } 2930 2931 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2932 { 2933 /* Does this proto have per netns sysctl_rmem ? */ 2934 if (proto->sysctl_rmem_offset) 2935 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2936 2937 return READ_ONCE(*proto->sysctl_rmem); 2938 } 2939 2940 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2941 * Some wifi drivers need to tweak it to get more chunks. 2942 * They can use this helper from their ndo_start_xmit() 2943 */ 2944 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2945 { 2946 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2947 return; 2948 WRITE_ONCE(sk->sk_pacing_shift, val); 2949 } 2950 2951 /* if a socket is bound to a device, check that the given device 2952 * index is either the same or that the socket is bound to an L3 2953 * master device and the given device index is also enslaved to 2954 * that L3 master 2955 */ 2956 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2957 { 2958 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2959 int mdif; 2960 2961 if (!bound_dev_if || bound_dev_if == dif) 2962 return true; 2963 2964 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2965 if (mdif && mdif == bound_dev_if) 2966 return true; 2967 2968 return false; 2969 } 2970 2971 void sock_def_readable(struct sock *sk); 2972 2973 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2974 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2975 int sock_set_timestamping(struct sock *sk, int optname, 2976 struct so_timestamping timestamping); 2977 2978 void sock_enable_timestamps(struct sock *sk); 2979 void sock_no_linger(struct sock *sk); 2980 void sock_set_keepalive(struct sock *sk); 2981 void sock_set_priority(struct sock *sk, u32 priority); 2982 void sock_set_rcvbuf(struct sock *sk, int val); 2983 void sock_set_mark(struct sock *sk, u32 val); 2984 void sock_set_reuseaddr(struct sock *sk); 2985 void sock_set_reuseport(struct sock *sk); 2986 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2987 2988 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2989 2990 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2991 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2992 sockptr_t optval, int optlen, bool old_timeval); 2993 2994 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 2995 void __user *arg, void *karg, size_t size); 2996 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 2997 static inline bool sk_is_readable(struct sock *sk) 2998 { 2999 if (sk->sk_prot->sock_is_readable) 3000 return sk->sk_prot->sock_is_readable(sk); 3001 return false; 3002 } 3003 #endif /* _SOCK_H */ 3004