1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 union { 165 __addrpair skc_addrpair; 166 struct { 167 __be32 skc_daddr; 168 __be32 skc_rcv_saddr; 169 }; 170 }; 171 union { 172 unsigned int skc_hash; 173 __u16 skc_u16hashes[2]; 174 }; 175 /* skc_dport && skc_num must be grouped as well */ 176 union { 177 __portpair skc_portpair; 178 struct { 179 __be16 skc_dport; 180 __u16 skc_num; 181 }; 182 }; 183 184 unsigned short skc_family; 185 volatile unsigned char skc_state; 186 unsigned char skc_reuse:4; 187 unsigned char skc_reuseport:1; 188 unsigned char skc_ipv6only:1; 189 unsigned char skc_net_refcnt:1; 190 int skc_bound_dev_if; 191 union { 192 struct hlist_node skc_bind_node; 193 struct hlist_node skc_portaddr_node; 194 }; 195 struct proto *skc_prot; 196 possible_net_t skc_net; 197 198 #if IS_ENABLED(CONFIG_IPV6) 199 struct in6_addr skc_v6_daddr; 200 struct in6_addr skc_v6_rcv_saddr; 201 #endif 202 203 atomic64_t skc_cookie; 204 205 /* following fields are padding to force 206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 207 * assuming IPV6 is enabled. We use this padding differently 208 * for different kind of 'sockets' 209 */ 210 union { 211 unsigned long skc_flags; 212 struct sock *skc_listener; /* request_sock */ 213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 214 }; 215 /* 216 * fields between dontcopy_begin/dontcopy_end 217 * are not copied in sock_copy() 218 */ 219 /* private: */ 220 int skc_dontcopy_begin[0]; 221 /* public: */ 222 union { 223 struct hlist_node skc_node; 224 struct hlist_nulls_node skc_nulls_node; 225 }; 226 unsigned short skc_tx_queue_mapping; 227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 228 unsigned short skc_rx_queue_mapping; 229 #endif 230 union { 231 int skc_incoming_cpu; 232 u32 skc_rcv_wnd; 233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 234 }; 235 236 refcount_t skc_refcnt; 237 /* private: */ 238 int skc_dontcopy_end[0]; 239 union { 240 u32 skc_rxhash; 241 u32 skc_window_clamp; 242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 243 }; 244 /* public: */ 245 }; 246 247 struct bpf_local_storage; 248 struct sk_filter; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 304 * @sk_busy_poll_budget: napi processing budget when busypolling 305 * @sk_priority: %SO_PRIORITY setting 306 * @sk_type: socket type (%SOCK_STREAM, etc) 307 * @sk_protocol: which protocol this socket belongs in this network family 308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_txrehash: enable TX hash rethink 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 322 * for timestamping 323 * @sk_tskey: counter to disambiguate concurrent tstamp requests 324 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 325 * @sk_socket: Identd and reporting IO signals 326 * @sk_user_data: RPC layer private data 327 * @sk_frag: cached page frag 328 * @sk_peek_off: current peek_offset value 329 * @sk_send_head: front of stuff to transmit 330 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 331 * @sk_security: used by security modules 332 * @sk_mark: generic packet mark 333 * @sk_cgrp_data: cgroup data for this cgroup 334 * @sk_memcg: this socket's memory cgroup association 335 * @sk_write_pending: a write to stream socket waits to start 336 * @sk_state_change: callback to indicate change in the state of the sock 337 * @sk_data_ready: callback to indicate there is data to be processed 338 * @sk_write_space: callback to indicate there is bf sending space available 339 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 340 * @sk_backlog_rcv: callback to process the backlog 341 * @sk_validate_xmit_skb: ptr to an optional validate function 342 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 343 * @sk_reuseport_cb: reuseport group container 344 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 345 * @sk_rcu: used during RCU grace period 346 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 347 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 348 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 349 * @sk_txtime_unused: unused txtime flags 350 * @ns_tracker: tracker for netns reference 351 */ 352 struct sock { 353 /* 354 * Now struct inet_timewait_sock also uses sock_common, so please just 355 * don't add nothing before this first member (__sk_common) --acme 356 */ 357 struct sock_common __sk_common; 358 #define sk_node __sk_common.skc_node 359 #define sk_nulls_node __sk_common.skc_nulls_node 360 #define sk_refcnt __sk_common.skc_refcnt 361 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 362 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 363 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 364 #endif 365 366 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 367 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 368 #define sk_hash __sk_common.skc_hash 369 #define sk_portpair __sk_common.skc_portpair 370 #define sk_num __sk_common.skc_num 371 #define sk_dport __sk_common.skc_dport 372 #define sk_addrpair __sk_common.skc_addrpair 373 #define sk_daddr __sk_common.skc_daddr 374 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 375 #define sk_family __sk_common.skc_family 376 #define sk_state __sk_common.skc_state 377 #define sk_reuse __sk_common.skc_reuse 378 #define sk_reuseport __sk_common.skc_reuseport 379 #define sk_ipv6only __sk_common.skc_ipv6only 380 #define sk_net_refcnt __sk_common.skc_net_refcnt 381 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 382 #define sk_bind_node __sk_common.skc_bind_node 383 #define sk_prot __sk_common.skc_prot 384 #define sk_net __sk_common.skc_net 385 #define sk_v6_daddr __sk_common.skc_v6_daddr 386 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 387 #define sk_cookie __sk_common.skc_cookie 388 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 389 #define sk_flags __sk_common.skc_flags 390 #define sk_rxhash __sk_common.skc_rxhash 391 392 /* early demux fields */ 393 struct dst_entry __rcu *sk_rx_dst; 394 int sk_rx_dst_ifindex; 395 u32 sk_rx_dst_cookie; 396 397 socket_lock_t sk_lock; 398 atomic_t sk_drops; 399 int sk_rcvlowat; 400 struct sk_buff_head sk_error_queue; 401 struct sk_buff_head sk_receive_queue; 402 /* 403 * The backlog queue is special, it is always used with 404 * the per-socket spinlock held and requires low latency 405 * access. Therefore we special case it's implementation. 406 * Note : rmem_alloc is in this structure to fill a hole 407 * on 64bit arches, not because its logically part of 408 * backlog. 409 */ 410 struct { 411 atomic_t rmem_alloc; 412 int len; 413 struct sk_buff *head; 414 struct sk_buff *tail; 415 } sk_backlog; 416 417 #define sk_rmem_alloc sk_backlog.rmem_alloc 418 419 int sk_forward_alloc; 420 u32 sk_reserved_mem; 421 #ifdef CONFIG_NET_RX_BUSY_POLL 422 unsigned int sk_ll_usec; 423 /* ===== mostly read cache line ===== */ 424 unsigned int sk_napi_id; 425 #endif 426 int sk_rcvbuf; 427 428 struct sk_filter __rcu *sk_filter; 429 union { 430 struct socket_wq __rcu *sk_wq; 431 /* private: */ 432 struct socket_wq *sk_wq_raw; 433 /* public: */ 434 }; 435 #ifdef CONFIG_XFRM 436 struct xfrm_policy __rcu *sk_policy[2]; 437 #endif 438 439 struct dst_entry __rcu *sk_dst_cache; 440 atomic_t sk_omem_alloc; 441 int sk_sndbuf; 442 443 /* ===== cache line for TX ===== */ 444 int sk_wmem_queued; 445 refcount_t sk_wmem_alloc; 446 unsigned long sk_tsq_flags; 447 union { 448 struct sk_buff *sk_send_head; 449 struct rb_root tcp_rtx_queue; 450 }; 451 struct sk_buff_head sk_write_queue; 452 __s32 sk_peek_off; 453 int sk_write_pending; 454 __u32 sk_dst_pending_confirm; 455 u32 sk_pacing_status; /* see enum sk_pacing */ 456 long sk_sndtimeo; 457 struct timer_list sk_timer; 458 __u32 sk_priority; 459 __u32 sk_mark; 460 unsigned long sk_pacing_rate; /* bytes per second */ 461 unsigned long sk_max_pacing_rate; 462 struct page_frag sk_frag; 463 netdev_features_t sk_route_caps; 464 int sk_gso_type; 465 unsigned int sk_gso_max_size; 466 gfp_t sk_allocation; 467 __u32 sk_txhash; 468 469 /* 470 * Because of non atomicity rules, all 471 * changes are protected by socket lock. 472 */ 473 u8 sk_gso_disabled : 1, 474 sk_kern_sock : 1, 475 sk_no_check_tx : 1, 476 sk_no_check_rx : 1, 477 sk_userlocks : 4; 478 u8 sk_pacing_shift; 479 u16 sk_type; 480 u16 sk_protocol; 481 u16 sk_gso_max_segs; 482 unsigned long sk_lingertime; 483 struct proto *sk_prot_creator; 484 rwlock_t sk_callback_lock; 485 int sk_err, 486 sk_err_soft; 487 u32 sk_ack_backlog; 488 u32 sk_max_ack_backlog; 489 kuid_t sk_uid; 490 u8 sk_txrehash; 491 #ifdef CONFIG_NET_RX_BUSY_POLL 492 u8 sk_prefer_busy_poll; 493 u16 sk_busy_poll_budget; 494 #endif 495 spinlock_t sk_peer_lock; 496 int sk_bind_phc; 497 struct pid *sk_peer_pid; 498 const struct cred *sk_peer_cred; 499 500 long sk_rcvtimeo; 501 ktime_t sk_stamp; 502 #if BITS_PER_LONG==32 503 seqlock_t sk_stamp_seq; 504 #endif 505 u16 sk_tsflags; 506 u8 sk_shutdown; 507 atomic_t sk_tskey; 508 atomic_t sk_zckey; 509 510 u8 sk_clockid; 511 u8 sk_txtime_deadline_mode : 1, 512 sk_txtime_report_errors : 1, 513 sk_txtime_unused : 6; 514 515 struct socket *sk_socket; 516 void *sk_user_data; 517 #ifdef CONFIG_SECURITY 518 void *sk_security; 519 #endif 520 struct sock_cgroup_data sk_cgrp_data; 521 struct mem_cgroup *sk_memcg; 522 void (*sk_state_change)(struct sock *sk); 523 void (*sk_data_ready)(struct sock *sk); 524 void (*sk_write_space)(struct sock *sk); 525 void (*sk_error_report)(struct sock *sk); 526 int (*sk_backlog_rcv)(struct sock *sk, 527 struct sk_buff *skb); 528 #ifdef CONFIG_SOCK_VALIDATE_XMIT 529 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 530 struct net_device *dev, 531 struct sk_buff *skb); 532 #endif 533 void (*sk_destruct)(struct sock *sk); 534 struct sock_reuseport __rcu *sk_reuseport_cb; 535 #ifdef CONFIG_BPF_SYSCALL 536 struct bpf_local_storage __rcu *sk_bpf_storage; 537 #endif 538 struct rcu_head sk_rcu; 539 netns_tracker ns_tracker; 540 }; 541 542 enum sk_pacing { 543 SK_PACING_NONE = 0, 544 SK_PACING_NEEDED = 1, 545 SK_PACING_FQ = 2, 546 }; 547 548 /* flag bits in sk_user_data 549 * 550 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 551 * not be suitable for copying when cloning the socket. For instance, 552 * it can point to a reference counted object. sk_user_data bottom 553 * bit is set if pointer must not be copied. 554 * 555 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 556 * managed/owned by a BPF reuseport array. This bit should be set 557 * when sk_user_data's sk is added to the bpf's reuseport_array. 558 * 559 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 560 * sk_user_data points to psock type. This bit should be set 561 * when sk_user_data is assigned to a psock object. 562 */ 563 #define SK_USER_DATA_NOCOPY 1UL 564 #define SK_USER_DATA_BPF 2UL 565 #define SK_USER_DATA_PSOCK 4UL 566 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 567 SK_USER_DATA_PSOCK) 568 569 /** 570 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 571 * @sk: socket 572 */ 573 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 574 { 575 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 576 } 577 578 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 579 580 /** 581 * __locked_read_sk_user_data_with_flags - return the pointer 582 * only if argument flags all has been set in sk_user_data. Otherwise 583 * return NULL 584 * 585 * @sk: socket 586 * @flags: flag bits 587 * 588 * The caller must be holding sk->sk_callback_lock. 589 */ 590 static inline void * 591 __locked_read_sk_user_data_with_flags(const struct sock *sk, 592 uintptr_t flags) 593 { 594 uintptr_t sk_user_data = 595 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 596 lockdep_is_held(&sk->sk_callback_lock)); 597 598 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 599 600 if ((sk_user_data & flags) == flags) 601 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 602 return NULL; 603 } 604 605 /** 606 * __rcu_dereference_sk_user_data_with_flags - return the pointer 607 * only if argument flags all has been set in sk_user_data. Otherwise 608 * return NULL 609 * 610 * @sk: socket 611 * @flags: flag bits 612 */ 613 static inline void * 614 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 615 uintptr_t flags) 616 { 617 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 618 619 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 620 621 if ((sk_user_data & flags) == flags) 622 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 623 return NULL; 624 } 625 626 #define rcu_dereference_sk_user_data(sk) \ 627 __rcu_dereference_sk_user_data_with_flags(sk, 0) 628 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 629 ({ \ 630 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 631 __tmp2 = (uintptr_t)(flags); \ 632 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 633 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 634 rcu_assign_pointer(__sk_user_data((sk)), \ 635 __tmp1 | __tmp2); \ 636 }) 637 #define rcu_assign_sk_user_data(sk, ptr) \ 638 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 639 640 static inline 641 struct net *sock_net(const struct sock *sk) 642 { 643 return read_pnet(&sk->sk_net); 644 } 645 646 static inline 647 void sock_net_set(struct sock *sk, struct net *net) 648 { 649 write_pnet(&sk->sk_net, net); 650 } 651 652 /* 653 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 654 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 655 * on a socket means that the socket will reuse everybody else's port 656 * without looking at the other's sk_reuse value. 657 */ 658 659 #define SK_NO_REUSE 0 660 #define SK_CAN_REUSE 1 661 #define SK_FORCE_REUSE 2 662 663 int sk_set_peek_off(struct sock *sk, int val); 664 665 static inline int sk_peek_offset(const struct sock *sk, int flags) 666 { 667 if (unlikely(flags & MSG_PEEK)) { 668 return READ_ONCE(sk->sk_peek_off); 669 } 670 671 return 0; 672 } 673 674 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 675 { 676 s32 off = READ_ONCE(sk->sk_peek_off); 677 678 if (unlikely(off >= 0)) { 679 off = max_t(s32, off - val, 0); 680 WRITE_ONCE(sk->sk_peek_off, off); 681 } 682 } 683 684 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 685 { 686 sk_peek_offset_bwd(sk, -val); 687 } 688 689 /* 690 * Hashed lists helper routines 691 */ 692 static inline struct sock *sk_entry(const struct hlist_node *node) 693 { 694 return hlist_entry(node, struct sock, sk_node); 695 } 696 697 static inline struct sock *__sk_head(const struct hlist_head *head) 698 { 699 return hlist_entry(head->first, struct sock, sk_node); 700 } 701 702 static inline struct sock *sk_head(const struct hlist_head *head) 703 { 704 return hlist_empty(head) ? NULL : __sk_head(head); 705 } 706 707 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 708 { 709 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 710 } 711 712 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 713 { 714 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 715 } 716 717 static inline struct sock *sk_next(const struct sock *sk) 718 { 719 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 720 } 721 722 static inline struct sock *sk_nulls_next(const struct sock *sk) 723 { 724 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 725 hlist_nulls_entry(sk->sk_nulls_node.next, 726 struct sock, sk_nulls_node) : 727 NULL; 728 } 729 730 static inline bool sk_unhashed(const struct sock *sk) 731 { 732 return hlist_unhashed(&sk->sk_node); 733 } 734 735 static inline bool sk_hashed(const struct sock *sk) 736 { 737 return !sk_unhashed(sk); 738 } 739 740 static inline void sk_node_init(struct hlist_node *node) 741 { 742 node->pprev = NULL; 743 } 744 745 static inline void sk_nulls_node_init(struct hlist_nulls_node *node) 746 { 747 node->pprev = NULL; 748 } 749 750 static inline void __sk_del_node(struct sock *sk) 751 { 752 __hlist_del(&sk->sk_node); 753 } 754 755 /* NB: equivalent to hlist_del_init_rcu */ 756 static inline bool __sk_del_node_init(struct sock *sk) 757 { 758 if (sk_hashed(sk)) { 759 __sk_del_node(sk); 760 sk_node_init(&sk->sk_node); 761 return true; 762 } 763 return false; 764 } 765 766 /* Grab socket reference count. This operation is valid only 767 when sk is ALREADY grabbed f.e. it is found in hash table 768 or a list and the lookup is made under lock preventing hash table 769 modifications. 770 */ 771 772 static __always_inline void sock_hold(struct sock *sk) 773 { 774 refcount_inc(&sk->sk_refcnt); 775 } 776 777 /* Ungrab socket in the context, which assumes that socket refcnt 778 cannot hit zero, f.e. it is true in context of any socketcall. 779 */ 780 static __always_inline void __sock_put(struct sock *sk) 781 { 782 refcount_dec(&sk->sk_refcnt); 783 } 784 785 static inline bool sk_del_node_init(struct sock *sk) 786 { 787 bool rc = __sk_del_node_init(sk); 788 789 if (rc) { 790 /* paranoid for a while -acme */ 791 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 792 __sock_put(sk); 793 } 794 return rc; 795 } 796 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 797 798 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 799 { 800 if (sk_hashed(sk)) { 801 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 802 return true; 803 } 804 return false; 805 } 806 807 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 808 { 809 bool rc = __sk_nulls_del_node_init_rcu(sk); 810 811 if (rc) { 812 /* paranoid for a while -acme */ 813 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 814 __sock_put(sk); 815 } 816 return rc; 817 } 818 819 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 820 { 821 hlist_add_head(&sk->sk_node, list); 822 } 823 824 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 825 { 826 sock_hold(sk); 827 __sk_add_node(sk, list); 828 } 829 830 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 831 { 832 sock_hold(sk); 833 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 834 sk->sk_family == AF_INET6) 835 hlist_add_tail_rcu(&sk->sk_node, list); 836 else 837 hlist_add_head_rcu(&sk->sk_node, list); 838 } 839 840 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 841 { 842 sock_hold(sk); 843 hlist_add_tail_rcu(&sk->sk_node, list); 844 } 845 846 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 847 { 848 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 849 } 850 851 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 852 { 853 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 854 } 855 856 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 857 { 858 sock_hold(sk); 859 __sk_nulls_add_node_rcu(sk, list); 860 } 861 862 static inline void __sk_del_bind_node(struct sock *sk) 863 { 864 __hlist_del(&sk->sk_bind_node); 865 } 866 867 static inline void sk_add_bind_node(struct sock *sk, 868 struct hlist_head *list) 869 { 870 hlist_add_head(&sk->sk_bind_node, list); 871 } 872 873 #define sk_for_each(__sk, list) \ 874 hlist_for_each_entry(__sk, list, sk_node) 875 #define sk_for_each_rcu(__sk, list) \ 876 hlist_for_each_entry_rcu(__sk, list, sk_node) 877 #define sk_nulls_for_each(__sk, node, list) \ 878 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 879 #define sk_nulls_for_each_rcu(__sk, node, list) \ 880 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 881 #define sk_for_each_from(__sk) \ 882 hlist_for_each_entry_from(__sk, sk_node) 883 #define sk_nulls_for_each_from(__sk, node) \ 884 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 885 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 886 #define sk_for_each_safe(__sk, tmp, list) \ 887 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 888 #define sk_for_each_bound(__sk, list) \ 889 hlist_for_each_entry(__sk, list, sk_bind_node) 890 891 /** 892 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 893 * @tpos: the type * to use as a loop cursor. 894 * @pos: the &struct hlist_node to use as a loop cursor. 895 * @head: the head for your list. 896 * @offset: offset of hlist_node within the struct. 897 * 898 */ 899 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 900 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 901 pos != NULL && \ 902 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 903 pos = rcu_dereference(hlist_next_rcu(pos))) 904 905 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 906 { 907 /* Careful only use this in a context where these parameters 908 * can not change and must all be valid, such as recvmsg from 909 * userspace. 910 */ 911 return sk->sk_socket->file->f_cred->user_ns; 912 } 913 914 /* Sock flags */ 915 enum sock_flags { 916 SOCK_DEAD, 917 SOCK_DONE, 918 SOCK_URGINLINE, 919 SOCK_KEEPOPEN, 920 SOCK_LINGER, 921 SOCK_DESTROY, 922 SOCK_BROADCAST, 923 SOCK_TIMESTAMP, 924 SOCK_ZAPPED, 925 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 926 SOCK_DBG, /* %SO_DEBUG setting */ 927 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 928 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 929 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 930 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 931 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 932 SOCK_FASYNC, /* fasync() active */ 933 SOCK_RXQ_OVFL, 934 SOCK_ZEROCOPY, /* buffers from userspace */ 935 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 936 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 937 * Will use last 4 bytes of packet sent from 938 * user-space instead. 939 */ 940 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 941 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 942 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 943 SOCK_TXTIME, 944 SOCK_XDP, /* XDP is attached */ 945 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 946 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 947 }; 948 949 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 950 951 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 952 { 953 nsk->sk_flags = osk->sk_flags; 954 } 955 956 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 957 { 958 __set_bit(flag, &sk->sk_flags); 959 } 960 961 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 962 { 963 __clear_bit(flag, &sk->sk_flags); 964 } 965 966 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 967 int valbool) 968 { 969 if (valbool) 970 sock_set_flag(sk, bit); 971 else 972 sock_reset_flag(sk, bit); 973 } 974 975 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 976 { 977 return test_bit(flag, &sk->sk_flags); 978 } 979 980 #ifdef CONFIG_NET 981 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 982 static inline int sk_memalloc_socks(void) 983 { 984 return static_branch_unlikely(&memalloc_socks_key); 985 } 986 987 void __receive_sock(struct file *file); 988 #else 989 990 static inline int sk_memalloc_socks(void) 991 { 992 return 0; 993 } 994 995 static inline void __receive_sock(struct file *file) 996 { } 997 #endif 998 999 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1000 { 1001 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1002 } 1003 1004 static inline void sk_acceptq_removed(struct sock *sk) 1005 { 1006 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1007 } 1008 1009 static inline void sk_acceptq_added(struct sock *sk) 1010 { 1011 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1012 } 1013 1014 /* Note: If you think the test should be: 1015 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1016 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1017 */ 1018 static inline bool sk_acceptq_is_full(const struct sock *sk) 1019 { 1020 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1021 } 1022 1023 /* 1024 * Compute minimal free write space needed to queue new packets. 1025 */ 1026 static inline int sk_stream_min_wspace(const struct sock *sk) 1027 { 1028 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1029 } 1030 1031 static inline int sk_stream_wspace(const struct sock *sk) 1032 { 1033 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1034 } 1035 1036 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1037 { 1038 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1039 } 1040 1041 void sk_stream_write_space(struct sock *sk); 1042 1043 /* OOB backlog add */ 1044 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1045 { 1046 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1047 skb_dst_force(skb); 1048 1049 if (!sk->sk_backlog.tail) 1050 WRITE_ONCE(sk->sk_backlog.head, skb); 1051 else 1052 sk->sk_backlog.tail->next = skb; 1053 1054 WRITE_ONCE(sk->sk_backlog.tail, skb); 1055 skb->next = NULL; 1056 } 1057 1058 /* 1059 * Take into account size of receive queue and backlog queue 1060 * Do not take into account this skb truesize, 1061 * to allow even a single big packet to come. 1062 */ 1063 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1064 { 1065 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1066 1067 return qsize > limit; 1068 } 1069 1070 /* The per-socket spinlock must be held here. */ 1071 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1072 unsigned int limit) 1073 { 1074 if (sk_rcvqueues_full(sk, limit)) 1075 return -ENOBUFS; 1076 1077 /* 1078 * If the skb was allocated from pfmemalloc reserves, only 1079 * allow SOCK_MEMALLOC sockets to use it as this socket is 1080 * helping free memory 1081 */ 1082 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1083 return -ENOMEM; 1084 1085 __sk_add_backlog(sk, skb); 1086 sk->sk_backlog.len += skb->truesize; 1087 return 0; 1088 } 1089 1090 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1091 1092 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1093 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1094 1095 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1096 { 1097 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1098 return __sk_backlog_rcv(sk, skb); 1099 1100 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1101 tcp_v6_do_rcv, 1102 tcp_v4_do_rcv, 1103 sk, skb); 1104 } 1105 1106 static inline void sk_incoming_cpu_update(struct sock *sk) 1107 { 1108 int cpu = raw_smp_processor_id(); 1109 1110 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1111 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1112 } 1113 1114 static inline void sock_rps_record_flow_hash(__u32 hash) 1115 { 1116 #ifdef CONFIG_RPS 1117 struct rps_sock_flow_table *sock_flow_table; 1118 1119 rcu_read_lock(); 1120 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1121 rps_record_sock_flow(sock_flow_table, hash); 1122 rcu_read_unlock(); 1123 #endif 1124 } 1125 1126 static inline void sock_rps_record_flow(const struct sock *sk) 1127 { 1128 #ifdef CONFIG_RPS 1129 if (static_branch_unlikely(&rfs_needed)) { 1130 /* Reading sk->sk_rxhash might incur an expensive cache line 1131 * miss. 1132 * 1133 * TCP_ESTABLISHED does cover almost all states where RFS 1134 * might be useful, and is cheaper [1] than testing : 1135 * IPv4: inet_sk(sk)->inet_daddr 1136 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1137 * OR an additional socket flag 1138 * [1] : sk_state and sk_prot are in the same cache line. 1139 */ 1140 if (sk->sk_state == TCP_ESTABLISHED) 1141 sock_rps_record_flow_hash(sk->sk_rxhash); 1142 } 1143 #endif 1144 } 1145 1146 static inline void sock_rps_save_rxhash(struct sock *sk, 1147 const struct sk_buff *skb) 1148 { 1149 #ifdef CONFIG_RPS 1150 if (unlikely(sk->sk_rxhash != skb->hash)) 1151 sk->sk_rxhash = skb->hash; 1152 #endif 1153 } 1154 1155 static inline void sock_rps_reset_rxhash(struct sock *sk) 1156 { 1157 #ifdef CONFIG_RPS 1158 sk->sk_rxhash = 0; 1159 #endif 1160 } 1161 1162 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1163 ({ int __rc; \ 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 = __condition; \ 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 unsigned long arg); 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 int (*sendpage)(struct sock *sk, struct page *page, 1256 int offset, size_t size, int flags); 1257 int (*bind)(struct sock *sk, 1258 struct sockaddr *addr, int addr_len); 1259 int (*bind_add)(struct sock *sk, 1260 struct sockaddr *addr, int addr_len); 1261 1262 int (*backlog_rcv) (struct sock *sk, 1263 struct sk_buff *skb); 1264 bool (*bpf_bypass_getsockopt)(int level, 1265 int optname); 1266 1267 void (*release_cb)(struct sock *sk); 1268 1269 /* Keeping track of sk's, looking them up, and port selection methods. */ 1270 int (*hash)(struct sock *sk); 1271 void (*unhash)(struct sock *sk); 1272 void (*rehash)(struct sock *sk); 1273 int (*get_port)(struct sock *sk, unsigned short snum); 1274 void (*put_port)(struct sock *sk); 1275 #ifdef CONFIG_BPF_SYSCALL 1276 int (*psock_update_sk_prot)(struct sock *sk, 1277 struct sk_psock *psock, 1278 bool restore); 1279 #endif 1280 1281 /* Keeping track of sockets in use */ 1282 #ifdef CONFIG_PROC_FS 1283 unsigned int inuse_idx; 1284 #endif 1285 1286 #if IS_ENABLED(CONFIG_MPTCP) 1287 int (*forward_alloc_get)(const struct sock *sk); 1288 #endif 1289 1290 bool (*stream_memory_free)(const struct sock *sk, int wake); 1291 bool (*sock_is_readable)(struct sock *sk); 1292 /* Memory pressure */ 1293 void (*enter_memory_pressure)(struct sock *sk); 1294 void (*leave_memory_pressure)(struct sock *sk); 1295 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1296 int __percpu *per_cpu_fw_alloc; 1297 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1298 1299 /* 1300 * Pressure flag: try to collapse. 1301 * Technical note: it is used by multiple contexts non atomically. 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 slab_flags_t slab_flags; 1319 unsigned int useroffset; /* Usercopy region offset */ 1320 unsigned int usersize; /* Usercopy region size */ 1321 1322 unsigned int __percpu *orphan_count; 1323 1324 struct request_sock_ops *rsk_prot; 1325 struct timewait_sock_ops *twsk_prot; 1326 1327 union { 1328 struct inet_hashinfo *hashinfo; 1329 struct udp_table *udp_table; 1330 struct raw_hashinfo *raw_hash; 1331 struct smc_hashinfo *smc_hash; 1332 } h; 1333 1334 struct module *owner; 1335 1336 char name[32]; 1337 1338 struct list_head node; 1339 #ifdef SOCK_REFCNT_DEBUG 1340 atomic_t socks; 1341 #endif 1342 int (*diag_destroy)(struct sock *sk, int err); 1343 } __randomize_layout; 1344 1345 int proto_register(struct proto *prot, int alloc_slab); 1346 void proto_unregister(struct proto *prot); 1347 int sock_load_diag_module(int family, int protocol); 1348 1349 #ifdef SOCK_REFCNT_DEBUG 1350 static inline void sk_refcnt_debug_inc(struct sock *sk) 1351 { 1352 atomic_inc(&sk->sk_prot->socks); 1353 } 1354 1355 static inline void sk_refcnt_debug_dec(struct sock *sk) 1356 { 1357 atomic_dec(&sk->sk_prot->socks); 1358 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1359 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1360 } 1361 1362 static inline void sk_refcnt_debug_release(const struct sock *sk) 1363 { 1364 if (refcount_read(&sk->sk_refcnt) != 1) 1365 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1366 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1367 } 1368 #else /* SOCK_REFCNT_DEBUG */ 1369 #define sk_refcnt_debug_inc(sk) do { } while (0) 1370 #define sk_refcnt_debug_dec(sk) do { } while (0) 1371 #define sk_refcnt_debug_release(sk) do { } while (0) 1372 #endif /* SOCK_REFCNT_DEBUG */ 1373 1374 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1375 1376 static inline int sk_forward_alloc_get(const struct sock *sk) 1377 { 1378 #if IS_ENABLED(CONFIG_MPTCP) 1379 if (sk->sk_prot->forward_alloc_get) 1380 return sk->sk_prot->forward_alloc_get(sk); 1381 #endif 1382 return sk->sk_forward_alloc; 1383 } 1384 1385 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1386 { 1387 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1388 return false; 1389 1390 return sk->sk_prot->stream_memory_free ? 1391 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1392 tcp_stream_memory_free, sk, wake) : true; 1393 } 1394 1395 static inline bool sk_stream_memory_free(const struct sock *sk) 1396 { 1397 return __sk_stream_memory_free(sk, 0); 1398 } 1399 1400 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1401 { 1402 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1403 __sk_stream_memory_free(sk, wake); 1404 } 1405 1406 static inline bool sk_stream_is_writeable(const struct sock *sk) 1407 { 1408 return __sk_stream_is_writeable(sk, 0); 1409 } 1410 1411 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1412 struct cgroup *ancestor) 1413 { 1414 #ifdef CONFIG_SOCK_CGROUP_DATA 1415 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1416 ancestor); 1417 #else 1418 return -ENOTSUPP; 1419 #endif 1420 } 1421 1422 static inline bool sk_has_memory_pressure(const struct sock *sk) 1423 { 1424 return sk->sk_prot->memory_pressure != NULL; 1425 } 1426 1427 static inline bool sk_under_memory_pressure(const struct sock *sk) 1428 { 1429 if (!sk->sk_prot->memory_pressure) 1430 return false; 1431 1432 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1433 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1434 return true; 1435 1436 return !!*sk->sk_prot->memory_pressure; 1437 } 1438 1439 static inline long 1440 proto_memory_allocated(const struct proto *prot) 1441 { 1442 return max(0L, atomic_long_read(prot->memory_allocated)); 1443 } 1444 1445 static inline long 1446 sk_memory_allocated(const struct sock *sk) 1447 { 1448 return proto_memory_allocated(sk->sk_prot); 1449 } 1450 1451 /* 1 MB per cpu, in page units */ 1452 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1453 1454 static inline void 1455 sk_memory_allocated_add(struct sock *sk, int amt) 1456 { 1457 int local_reserve; 1458 1459 preempt_disable(); 1460 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1461 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) { 1462 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1463 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1464 } 1465 preempt_enable(); 1466 } 1467 1468 static inline void 1469 sk_memory_allocated_sub(struct sock *sk, int amt) 1470 { 1471 int local_reserve; 1472 1473 preempt_disable(); 1474 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1475 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) { 1476 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1477 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1478 } 1479 preempt_enable(); 1480 } 1481 1482 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1483 1484 static inline void sk_sockets_allocated_dec(struct sock *sk) 1485 { 1486 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1487 SK_ALLOC_PERCPU_COUNTER_BATCH); 1488 } 1489 1490 static inline void sk_sockets_allocated_inc(struct sock *sk) 1491 { 1492 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1493 SK_ALLOC_PERCPU_COUNTER_BATCH); 1494 } 1495 1496 static inline u64 1497 sk_sockets_allocated_read_positive(struct sock *sk) 1498 { 1499 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1500 } 1501 1502 static inline int 1503 proto_sockets_allocated_sum_positive(struct proto *prot) 1504 { 1505 return percpu_counter_sum_positive(prot->sockets_allocated); 1506 } 1507 1508 static inline bool 1509 proto_memory_pressure(struct proto *prot) 1510 { 1511 if (!prot->memory_pressure) 1512 return false; 1513 return !!*prot->memory_pressure; 1514 } 1515 1516 1517 #ifdef CONFIG_PROC_FS 1518 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1519 struct prot_inuse { 1520 int all; 1521 int val[PROTO_INUSE_NR]; 1522 }; 1523 1524 static inline void sock_prot_inuse_add(const struct net *net, 1525 const struct proto *prot, int val) 1526 { 1527 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1528 } 1529 1530 static inline void sock_inuse_add(const struct net *net, int val) 1531 { 1532 this_cpu_add(net->core.prot_inuse->all, val); 1533 } 1534 1535 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1536 int sock_inuse_get(struct net *net); 1537 #else 1538 static inline void sock_prot_inuse_add(const struct net *net, 1539 const struct proto *prot, int val) 1540 { 1541 } 1542 1543 static inline void sock_inuse_add(const struct net *net, int val) 1544 { 1545 } 1546 #endif 1547 1548 1549 /* With per-bucket locks this operation is not-atomic, so that 1550 * this version is not worse. 1551 */ 1552 static inline int __sk_prot_rehash(struct sock *sk) 1553 { 1554 sk->sk_prot->unhash(sk); 1555 return sk->sk_prot->hash(sk); 1556 } 1557 1558 /* About 10 seconds */ 1559 #define SOCK_DESTROY_TIME (10*HZ) 1560 1561 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1562 #define PROT_SOCK 1024 1563 1564 #define SHUTDOWN_MASK 3 1565 #define RCV_SHUTDOWN 1 1566 #define SEND_SHUTDOWN 2 1567 1568 #define SOCK_BINDADDR_LOCK 4 1569 #define SOCK_BINDPORT_LOCK 8 1570 1571 struct socket_alloc { 1572 struct socket socket; 1573 struct inode vfs_inode; 1574 }; 1575 1576 static inline struct socket *SOCKET_I(struct inode *inode) 1577 { 1578 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1579 } 1580 1581 static inline struct inode *SOCK_INODE(struct socket *socket) 1582 { 1583 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1584 } 1585 1586 /* 1587 * Functions for memory accounting 1588 */ 1589 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1590 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1591 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1592 void __sk_mem_reclaim(struct sock *sk, int amount); 1593 1594 #define SK_MEM_SEND 0 1595 #define SK_MEM_RECV 1 1596 1597 /* sysctl_mem values are in pages */ 1598 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1599 { 1600 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1601 } 1602 1603 static inline int sk_mem_pages(int amt) 1604 { 1605 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1606 } 1607 1608 static inline bool sk_has_account(struct sock *sk) 1609 { 1610 /* return true if protocol supports memory accounting */ 1611 return !!sk->sk_prot->memory_allocated; 1612 } 1613 1614 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1615 { 1616 int delta; 1617 1618 if (!sk_has_account(sk)) 1619 return true; 1620 delta = size - sk->sk_forward_alloc; 1621 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1622 } 1623 1624 static inline bool 1625 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1626 { 1627 int delta; 1628 1629 if (!sk_has_account(sk)) 1630 return true; 1631 delta = size - sk->sk_forward_alloc; 1632 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1633 skb_pfmemalloc(skb); 1634 } 1635 1636 static inline int sk_unused_reserved_mem(const struct sock *sk) 1637 { 1638 int unused_mem; 1639 1640 if (likely(!sk->sk_reserved_mem)) 1641 return 0; 1642 1643 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1644 atomic_read(&sk->sk_rmem_alloc); 1645 1646 return unused_mem > 0 ? unused_mem : 0; 1647 } 1648 1649 static inline void sk_mem_reclaim(struct sock *sk) 1650 { 1651 int reclaimable; 1652 1653 if (!sk_has_account(sk)) 1654 return; 1655 1656 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1657 1658 if (reclaimable >= (int)PAGE_SIZE) 1659 __sk_mem_reclaim(sk, reclaimable); 1660 } 1661 1662 static inline void sk_mem_reclaim_final(struct sock *sk) 1663 { 1664 sk->sk_reserved_mem = 0; 1665 sk_mem_reclaim(sk); 1666 } 1667 1668 static inline void sk_mem_charge(struct sock *sk, int size) 1669 { 1670 if (!sk_has_account(sk)) 1671 return; 1672 sk->sk_forward_alloc -= size; 1673 } 1674 1675 static inline void sk_mem_uncharge(struct sock *sk, int size) 1676 { 1677 if (!sk_has_account(sk)) 1678 return; 1679 sk->sk_forward_alloc += size; 1680 sk_mem_reclaim(sk); 1681 } 1682 1683 /* 1684 * Macro so as to not evaluate some arguments when 1685 * lockdep is not enabled. 1686 * 1687 * Mark both the sk_lock and the sk_lock.slock as a 1688 * per-address-family lock class. 1689 */ 1690 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1691 do { \ 1692 sk->sk_lock.owned = 0; \ 1693 init_waitqueue_head(&sk->sk_lock.wq); \ 1694 spin_lock_init(&(sk)->sk_lock.slock); \ 1695 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1696 sizeof((sk)->sk_lock)); \ 1697 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1698 (skey), (sname)); \ 1699 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1700 } while (0) 1701 1702 static inline bool lockdep_sock_is_held(const struct sock *sk) 1703 { 1704 return lockdep_is_held(&sk->sk_lock) || 1705 lockdep_is_held(&sk->sk_lock.slock); 1706 } 1707 1708 void lock_sock_nested(struct sock *sk, int subclass); 1709 1710 static inline void lock_sock(struct sock *sk) 1711 { 1712 lock_sock_nested(sk, 0); 1713 } 1714 1715 void __lock_sock(struct sock *sk); 1716 void __release_sock(struct sock *sk); 1717 void release_sock(struct sock *sk); 1718 1719 /* BH context may only use the following locking interface. */ 1720 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1721 #define bh_lock_sock_nested(__sk) \ 1722 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1723 SINGLE_DEPTH_NESTING) 1724 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1725 1726 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1727 1728 /** 1729 * lock_sock_fast - fast version of lock_sock 1730 * @sk: socket 1731 * 1732 * This version should be used for very small section, where process wont block 1733 * return false if fast path is taken: 1734 * 1735 * sk_lock.slock locked, owned = 0, BH disabled 1736 * 1737 * return true if slow path is taken: 1738 * 1739 * sk_lock.slock unlocked, owned = 1, BH enabled 1740 */ 1741 static inline bool lock_sock_fast(struct sock *sk) 1742 { 1743 /* The sk_lock has mutex_lock() semantics here. */ 1744 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1745 1746 return __lock_sock_fast(sk); 1747 } 1748 1749 /* fast socket lock variant for caller already holding a [different] socket lock */ 1750 static inline bool lock_sock_fast_nested(struct sock *sk) 1751 { 1752 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1753 1754 return __lock_sock_fast(sk); 1755 } 1756 1757 /** 1758 * unlock_sock_fast - complement of lock_sock_fast 1759 * @sk: socket 1760 * @slow: slow mode 1761 * 1762 * fast unlock socket for user context. 1763 * If slow mode is on, we call regular release_sock() 1764 */ 1765 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1766 __releases(&sk->sk_lock.slock) 1767 { 1768 if (slow) { 1769 release_sock(sk); 1770 __release(&sk->sk_lock.slock); 1771 } else { 1772 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1773 spin_unlock_bh(&sk->sk_lock.slock); 1774 } 1775 } 1776 1777 /* Used by processes to "lock" a socket state, so that 1778 * interrupts and bottom half handlers won't change it 1779 * from under us. It essentially blocks any incoming 1780 * packets, so that we won't get any new data or any 1781 * packets that change the state of the socket. 1782 * 1783 * While locked, BH processing will add new packets to 1784 * the backlog queue. This queue is processed by the 1785 * owner of the socket lock right before it is released. 1786 * 1787 * Since ~2.3.5 it is also exclusive sleep lock serializing 1788 * accesses from user process context. 1789 */ 1790 1791 static inline void sock_owned_by_me(const struct sock *sk) 1792 { 1793 #ifdef CONFIG_LOCKDEP 1794 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1795 #endif 1796 } 1797 1798 static inline bool sock_owned_by_user(const struct sock *sk) 1799 { 1800 sock_owned_by_me(sk); 1801 return sk->sk_lock.owned; 1802 } 1803 1804 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1805 { 1806 return sk->sk_lock.owned; 1807 } 1808 1809 static inline void sock_release_ownership(struct sock *sk) 1810 { 1811 if (sock_owned_by_user_nocheck(sk)) { 1812 sk->sk_lock.owned = 0; 1813 1814 /* The sk_lock has mutex_unlock() semantics: */ 1815 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1816 } 1817 } 1818 1819 /* no reclassification while locks are held */ 1820 static inline bool sock_allow_reclassification(const struct sock *csk) 1821 { 1822 struct sock *sk = (struct sock *)csk; 1823 1824 return !sock_owned_by_user_nocheck(sk) && 1825 !spin_is_locked(&sk->sk_lock.slock); 1826 } 1827 1828 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1829 struct proto *prot, int kern); 1830 void sk_free(struct sock *sk); 1831 void sk_destruct(struct sock *sk); 1832 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1833 void sk_free_unlock_clone(struct sock *sk); 1834 1835 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1836 gfp_t priority); 1837 void __sock_wfree(struct sk_buff *skb); 1838 void sock_wfree(struct sk_buff *skb); 1839 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1840 gfp_t priority); 1841 void skb_orphan_partial(struct sk_buff *skb); 1842 void sock_rfree(struct sk_buff *skb); 1843 void sock_efree(struct sk_buff *skb); 1844 #ifdef CONFIG_INET 1845 void sock_edemux(struct sk_buff *skb); 1846 void sock_pfree(struct sk_buff *skb); 1847 #else 1848 #define sock_edemux sock_efree 1849 #endif 1850 1851 int sock_setsockopt(struct socket *sock, int level, int op, 1852 sockptr_t optval, unsigned int optlen); 1853 1854 int sock_getsockopt(struct socket *sock, int level, int op, 1855 char __user *optval, int __user *optlen); 1856 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1857 bool timeval, bool time32); 1858 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1859 unsigned long data_len, int noblock, 1860 int *errcode, int max_page_order); 1861 1862 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1863 unsigned long size, 1864 int noblock, int *errcode) 1865 { 1866 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1867 } 1868 1869 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1870 void sock_kfree_s(struct sock *sk, void *mem, int size); 1871 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1872 void sk_send_sigurg(struct sock *sk); 1873 1874 struct sockcm_cookie { 1875 u64 transmit_time; 1876 u32 mark; 1877 u16 tsflags; 1878 }; 1879 1880 static inline void sockcm_init(struct sockcm_cookie *sockc, 1881 const struct sock *sk) 1882 { 1883 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1884 } 1885 1886 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1887 struct sockcm_cookie *sockc); 1888 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1889 struct sockcm_cookie *sockc); 1890 1891 /* 1892 * Functions to fill in entries in struct proto_ops when a protocol 1893 * does not implement a particular function. 1894 */ 1895 int sock_no_bind(struct socket *, struct sockaddr *, int); 1896 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1897 int sock_no_socketpair(struct socket *, struct socket *); 1898 int sock_no_accept(struct socket *, struct socket *, int, bool); 1899 int sock_no_getname(struct socket *, struct sockaddr *, int); 1900 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1901 int sock_no_listen(struct socket *, int); 1902 int sock_no_shutdown(struct socket *, int); 1903 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1904 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1905 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1906 int sock_no_mmap(struct file *file, struct socket *sock, 1907 struct vm_area_struct *vma); 1908 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1909 size_t size, int flags); 1910 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1911 int offset, size_t size, int flags); 1912 1913 /* 1914 * Functions to fill in entries in struct proto_ops when a protocol 1915 * uses the inet style. 1916 */ 1917 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1918 char __user *optval, int __user *optlen); 1919 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1920 int flags); 1921 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1922 sockptr_t optval, unsigned int optlen); 1923 1924 void sk_common_release(struct sock *sk); 1925 1926 /* 1927 * Default socket callbacks and setup code 1928 */ 1929 1930 /* Initialise core socket variables */ 1931 void sock_init_data(struct socket *sock, struct sock *sk); 1932 1933 /* 1934 * Socket reference counting postulates. 1935 * 1936 * * Each user of socket SHOULD hold a reference count. 1937 * * Each access point to socket (an hash table bucket, reference from a list, 1938 * running timer, skb in flight MUST hold a reference count. 1939 * * When reference count hits 0, it means it will never increase back. 1940 * * When reference count hits 0, it means that no references from 1941 * outside exist to this socket and current process on current CPU 1942 * is last user and may/should destroy this socket. 1943 * * sk_free is called from any context: process, BH, IRQ. When 1944 * it is called, socket has no references from outside -> sk_free 1945 * may release descendant resources allocated by the socket, but 1946 * to the time when it is called, socket is NOT referenced by any 1947 * hash tables, lists etc. 1948 * * Packets, delivered from outside (from network or from another process) 1949 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1950 * when they sit in queue. Otherwise, packets will leak to hole, when 1951 * socket is looked up by one cpu and unhasing is made by another CPU. 1952 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1953 * (leak to backlog). Packet socket does all the processing inside 1954 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1955 * use separate SMP lock, so that they are prone too. 1956 */ 1957 1958 /* Ungrab socket and destroy it, if it was the last reference. */ 1959 static inline void sock_put(struct sock *sk) 1960 { 1961 if (refcount_dec_and_test(&sk->sk_refcnt)) 1962 sk_free(sk); 1963 } 1964 /* Generic version of sock_put(), dealing with all sockets 1965 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1966 */ 1967 void sock_gen_put(struct sock *sk); 1968 1969 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1970 unsigned int trim_cap, bool refcounted); 1971 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1972 const int nested) 1973 { 1974 return __sk_receive_skb(sk, skb, nested, 1, true); 1975 } 1976 1977 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1978 { 1979 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1980 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1981 return; 1982 sk->sk_tx_queue_mapping = tx_queue; 1983 } 1984 1985 #define NO_QUEUE_MAPPING USHRT_MAX 1986 1987 static inline void sk_tx_queue_clear(struct sock *sk) 1988 { 1989 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 1990 } 1991 1992 static inline int sk_tx_queue_get(const struct sock *sk) 1993 { 1994 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 1995 return sk->sk_tx_queue_mapping; 1996 1997 return -1; 1998 } 1999 2000 static inline void __sk_rx_queue_set(struct sock *sk, 2001 const struct sk_buff *skb, 2002 bool force_set) 2003 { 2004 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2005 if (skb_rx_queue_recorded(skb)) { 2006 u16 rx_queue = skb_get_rx_queue(skb); 2007 2008 if (force_set || 2009 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2010 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2011 } 2012 #endif 2013 } 2014 2015 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2016 { 2017 __sk_rx_queue_set(sk, skb, true); 2018 } 2019 2020 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2021 { 2022 __sk_rx_queue_set(sk, skb, false); 2023 } 2024 2025 static inline void sk_rx_queue_clear(struct sock *sk) 2026 { 2027 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2028 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2029 #endif 2030 } 2031 2032 static inline int sk_rx_queue_get(const struct sock *sk) 2033 { 2034 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2035 if (sk) { 2036 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2037 2038 if (res != NO_QUEUE_MAPPING) 2039 return res; 2040 } 2041 #endif 2042 2043 return -1; 2044 } 2045 2046 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2047 { 2048 sk->sk_socket = sock; 2049 } 2050 2051 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2052 { 2053 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2054 return &rcu_dereference_raw(sk->sk_wq)->wait; 2055 } 2056 /* Detach socket from process context. 2057 * Announce socket dead, detach it from wait queue and inode. 2058 * Note that parent inode held reference count on this struct sock, 2059 * we do not release it in this function, because protocol 2060 * probably wants some additional cleanups or even continuing 2061 * to work with this socket (TCP). 2062 */ 2063 static inline void sock_orphan(struct sock *sk) 2064 { 2065 write_lock_bh(&sk->sk_callback_lock); 2066 sock_set_flag(sk, SOCK_DEAD); 2067 sk_set_socket(sk, NULL); 2068 sk->sk_wq = NULL; 2069 write_unlock_bh(&sk->sk_callback_lock); 2070 } 2071 2072 static inline void sock_graft(struct sock *sk, struct socket *parent) 2073 { 2074 WARN_ON(parent->sk); 2075 write_lock_bh(&sk->sk_callback_lock); 2076 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2077 parent->sk = sk; 2078 sk_set_socket(sk, parent); 2079 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2080 security_sock_graft(sk, parent); 2081 write_unlock_bh(&sk->sk_callback_lock); 2082 } 2083 2084 kuid_t sock_i_uid(struct sock *sk); 2085 unsigned long sock_i_ino(struct sock *sk); 2086 2087 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2088 { 2089 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2090 } 2091 2092 static inline u32 net_tx_rndhash(void) 2093 { 2094 u32 v = prandom_u32(); 2095 2096 return v ?: 1; 2097 } 2098 2099 static inline void sk_set_txhash(struct sock *sk) 2100 { 2101 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2102 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2103 } 2104 2105 static inline bool sk_rethink_txhash(struct sock *sk) 2106 { 2107 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2108 sk_set_txhash(sk); 2109 return true; 2110 } 2111 return false; 2112 } 2113 2114 static inline struct dst_entry * 2115 __sk_dst_get(struct sock *sk) 2116 { 2117 return rcu_dereference_check(sk->sk_dst_cache, 2118 lockdep_sock_is_held(sk)); 2119 } 2120 2121 static inline struct dst_entry * 2122 sk_dst_get(struct sock *sk) 2123 { 2124 struct dst_entry *dst; 2125 2126 rcu_read_lock(); 2127 dst = rcu_dereference(sk->sk_dst_cache); 2128 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2129 dst = NULL; 2130 rcu_read_unlock(); 2131 return dst; 2132 } 2133 2134 static inline void __dst_negative_advice(struct sock *sk) 2135 { 2136 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2137 2138 if (dst && dst->ops->negative_advice) { 2139 ndst = dst->ops->negative_advice(dst); 2140 2141 if (ndst != dst) { 2142 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2143 sk_tx_queue_clear(sk); 2144 sk->sk_dst_pending_confirm = 0; 2145 } 2146 } 2147 } 2148 2149 static inline void dst_negative_advice(struct sock *sk) 2150 { 2151 sk_rethink_txhash(sk); 2152 __dst_negative_advice(sk); 2153 } 2154 2155 static inline void 2156 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2157 { 2158 struct dst_entry *old_dst; 2159 2160 sk_tx_queue_clear(sk); 2161 sk->sk_dst_pending_confirm = 0; 2162 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2163 lockdep_sock_is_held(sk)); 2164 rcu_assign_pointer(sk->sk_dst_cache, dst); 2165 dst_release(old_dst); 2166 } 2167 2168 static inline void 2169 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2170 { 2171 struct dst_entry *old_dst; 2172 2173 sk_tx_queue_clear(sk); 2174 sk->sk_dst_pending_confirm = 0; 2175 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2176 dst_release(old_dst); 2177 } 2178 2179 static inline void 2180 __sk_dst_reset(struct sock *sk) 2181 { 2182 __sk_dst_set(sk, NULL); 2183 } 2184 2185 static inline void 2186 sk_dst_reset(struct sock *sk) 2187 { 2188 sk_dst_set(sk, NULL); 2189 } 2190 2191 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2192 2193 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2194 2195 static inline void sk_dst_confirm(struct sock *sk) 2196 { 2197 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2198 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2199 } 2200 2201 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2202 { 2203 if (skb_get_dst_pending_confirm(skb)) { 2204 struct sock *sk = skb->sk; 2205 2206 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2207 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2208 neigh_confirm(n); 2209 } 2210 } 2211 2212 bool sk_mc_loop(struct sock *sk); 2213 2214 static inline bool sk_can_gso(const struct sock *sk) 2215 { 2216 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2217 } 2218 2219 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2220 2221 static inline void sk_gso_disable(struct sock *sk) 2222 { 2223 sk->sk_gso_disabled = 1; 2224 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2225 } 2226 2227 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2228 struct iov_iter *from, char *to, 2229 int copy, int offset) 2230 { 2231 if (skb->ip_summed == CHECKSUM_NONE) { 2232 __wsum csum = 0; 2233 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2234 return -EFAULT; 2235 skb->csum = csum_block_add(skb->csum, csum, offset); 2236 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2237 if (!copy_from_iter_full_nocache(to, copy, from)) 2238 return -EFAULT; 2239 } else if (!copy_from_iter_full(to, copy, from)) 2240 return -EFAULT; 2241 2242 return 0; 2243 } 2244 2245 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2246 struct iov_iter *from, int copy) 2247 { 2248 int err, offset = skb->len; 2249 2250 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2251 copy, offset); 2252 if (err) 2253 __skb_trim(skb, offset); 2254 2255 return err; 2256 } 2257 2258 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2259 struct sk_buff *skb, 2260 struct page *page, 2261 int off, int copy) 2262 { 2263 int err; 2264 2265 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2266 copy, skb->len); 2267 if (err) 2268 return err; 2269 2270 skb_len_add(skb, copy); 2271 sk_wmem_queued_add(sk, copy); 2272 sk_mem_charge(sk, copy); 2273 return 0; 2274 } 2275 2276 /** 2277 * sk_wmem_alloc_get - returns write allocations 2278 * @sk: socket 2279 * 2280 * Return: sk_wmem_alloc minus initial offset of one 2281 */ 2282 static inline int sk_wmem_alloc_get(const struct sock *sk) 2283 { 2284 return refcount_read(&sk->sk_wmem_alloc) - 1; 2285 } 2286 2287 /** 2288 * sk_rmem_alloc_get - returns read allocations 2289 * @sk: socket 2290 * 2291 * Return: sk_rmem_alloc 2292 */ 2293 static inline int sk_rmem_alloc_get(const struct sock *sk) 2294 { 2295 return atomic_read(&sk->sk_rmem_alloc); 2296 } 2297 2298 /** 2299 * sk_has_allocations - check if allocations are outstanding 2300 * @sk: socket 2301 * 2302 * Return: true if socket has write or read allocations 2303 */ 2304 static inline bool sk_has_allocations(const struct sock *sk) 2305 { 2306 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2307 } 2308 2309 /** 2310 * skwq_has_sleeper - check if there are any waiting processes 2311 * @wq: struct socket_wq 2312 * 2313 * Return: true if socket_wq has waiting processes 2314 * 2315 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2316 * barrier call. They were added due to the race found within the tcp code. 2317 * 2318 * Consider following tcp code paths:: 2319 * 2320 * CPU1 CPU2 2321 * sys_select receive packet 2322 * ... ... 2323 * __add_wait_queue update tp->rcv_nxt 2324 * ... ... 2325 * tp->rcv_nxt check sock_def_readable 2326 * ... { 2327 * schedule rcu_read_lock(); 2328 * wq = rcu_dereference(sk->sk_wq); 2329 * if (wq && waitqueue_active(&wq->wait)) 2330 * wake_up_interruptible(&wq->wait) 2331 * ... 2332 * } 2333 * 2334 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2335 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2336 * could then endup calling schedule and sleep forever if there are no more 2337 * data on the socket. 2338 * 2339 */ 2340 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2341 { 2342 return wq && wq_has_sleeper(&wq->wait); 2343 } 2344 2345 /** 2346 * sock_poll_wait - place memory barrier behind the poll_wait call. 2347 * @filp: file 2348 * @sock: socket to wait on 2349 * @p: poll_table 2350 * 2351 * See the comments in the wq_has_sleeper function. 2352 */ 2353 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2354 poll_table *p) 2355 { 2356 if (!poll_does_not_wait(p)) { 2357 poll_wait(filp, &sock->wq.wait, p); 2358 /* We need to be sure we are in sync with the 2359 * socket flags modification. 2360 * 2361 * This memory barrier is paired in the wq_has_sleeper. 2362 */ 2363 smp_mb(); 2364 } 2365 } 2366 2367 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2368 { 2369 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2370 u32 txhash = READ_ONCE(sk->sk_txhash); 2371 2372 if (txhash) { 2373 skb->l4_hash = 1; 2374 skb->hash = txhash; 2375 } 2376 } 2377 2378 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2379 2380 /* 2381 * Queue a received datagram if it will fit. Stream and sequenced 2382 * protocols can't normally use this as they need to fit buffers in 2383 * and play with them. 2384 * 2385 * Inlined as it's very short and called for pretty much every 2386 * packet ever received. 2387 */ 2388 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2389 { 2390 skb_orphan(skb); 2391 skb->sk = sk; 2392 skb->destructor = sock_rfree; 2393 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2394 sk_mem_charge(sk, skb->truesize); 2395 } 2396 2397 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2398 { 2399 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2400 skb_orphan(skb); 2401 skb->destructor = sock_efree; 2402 skb->sk = sk; 2403 return true; 2404 } 2405 return false; 2406 } 2407 2408 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2409 { 2410 if (skb->destructor != sock_wfree) { 2411 skb_orphan(skb); 2412 return; 2413 } 2414 skb->slow_gro = 1; 2415 } 2416 2417 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2418 unsigned long expires); 2419 2420 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2421 2422 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2423 2424 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2425 struct sk_buff *skb, unsigned int flags, 2426 void (*destructor)(struct sock *sk, 2427 struct sk_buff *skb)); 2428 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2429 2430 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2431 enum skb_drop_reason *reason); 2432 2433 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2434 { 2435 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2436 } 2437 2438 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2439 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2440 2441 /* 2442 * Recover an error report and clear atomically 2443 */ 2444 2445 static inline int sock_error(struct sock *sk) 2446 { 2447 int err; 2448 2449 /* Avoid an atomic operation for the common case. 2450 * This is racy since another cpu/thread can change sk_err under us. 2451 */ 2452 if (likely(data_race(!sk->sk_err))) 2453 return 0; 2454 2455 err = xchg(&sk->sk_err, 0); 2456 return -err; 2457 } 2458 2459 void sk_error_report(struct sock *sk); 2460 2461 static inline unsigned long sock_wspace(struct sock *sk) 2462 { 2463 int amt = 0; 2464 2465 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2466 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2467 if (amt < 0) 2468 amt = 0; 2469 } 2470 return amt; 2471 } 2472 2473 /* Note: 2474 * We use sk->sk_wq_raw, from contexts knowing this 2475 * pointer is not NULL and cannot disappear/change. 2476 */ 2477 static inline void sk_set_bit(int nr, struct sock *sk) 2478 { 2479 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2480 !sock_flag(sk, SOCK_FASYNC)) 2481 return; 2482 2483 set_bit(nr, &sk->sk_wq_raw->flags); 2484 } 2485 2486 static inline void sk_clear_bit(int nr, struct sock *sk) 2487 { 2488 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2489 !sock_flag(sk, SOCK_FASYNC)) 2490 return; 2491 2492 clear_bit(nr, &sk->sk_wq_raw->flags); 2493 } 2494 2495 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2496 { 2497 if (sock_flag(sk, SOCK_FASYNC)) { 2498 rcu_read_lock(); 2499 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2500 rcu_read_unlock(); 2501 } 2502 } 2503 2504 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2505 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2506 * Note: for send buffers, TCP works better if we can build two skbs at 2507 * minimum. 2508 */ 2509 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2510 2511 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2512 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2513 2514 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2515 { 2516 u32 val; 2517 2518 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2519 return; 2520 2521 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2522 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2523 2524 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2525 } 2526 2527 /** 2528 * sk_page_frag - return an appropriate page_frag 2529 * @sk: socket 2530 * 2531 * Use the per task page_frag instead of the per socket one for 2532 * optimization when we know that we're in process context and own 2533 * everything that's associated with %current. 2534 * 2535 * Both direct reclaim and page faults can nest inside other 2536 * socket operations and end up recursing into sk_page_frag() 2537 * while it's already in use: explicitly avoid task page_frag 2538 * usage if the caller is potentially doing any of them. 2539 * This assumes that page fault handlers use the GFP_NOFS flags. 2540 * 2541 * Return: a per task page_frag if context allows that, 2542 * otherwise a per socket one. 2543 */ 2544 static inline struct page_frag *sk_page_frag(struct sock *sk) 2545 { 2546 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) == 2547 (__GFP_DIRECT_RECLAIM | __GFP_FS)) 2548 return ¤t->task_frag; 2549 2550 return &sk->sk_frag; 2551 } 2552 2553 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2554 2555 /* 2556 * Default write policy as shown to user space via poll/select/SIGIO 2557 */ 2558 static inline bool sock_writeable(const struct sock *sk) 2559 { 2560 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2561 } 2562 2563 static inline gfp_t gfp_any(void) 2564 { 2565 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2566 } 2567 2568 static inline gfp_t gfp_memcg_charge(void) 2569 { 2570 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2571 } 2572 2573 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2574 { 2575 return noblock ? 0 : sk->sk_rcvtimeo; 2576 } 2577 2578 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2579 { 2580 return noblock ? 0 : sk->sk_sndtimeo; 2581 } 2582 2583 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2584 { 2585 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2586 2587 return v ?: 1; 2588 } 2589 2590 /* Alas, with timeout socket operations are not restartable. 2591 * Compare this to poll(). 2592 */ 2593 static inline int sock_intr_errno(long timeo) 2594 { 2595 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2596 } 2597 2598 struct sock_skb_cb { 2599 u32 dropcount; 2600 }; 2601 2602 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2603 * using skb->cb[] would keep using it directly and utilize its 2604 * alignement guarantee. 2605 */ 2606 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2607 sizeof(struct sock_skb_cb))) 2608 2609 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2610 SOCK_SKB_CB_OFFSET)) 2611 2612 #define sock_skb_cb_check_size(size) \ 2613 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2614 2615 static inline void 2616 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2617 { 2618 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2619 atomic_read(&sk->sk_drops) : 0; 2620 } 2621 2622 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2623 { 2624 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2625 2626 atomic_add(segs, &sk->sk_drops); 2627 } 2628 2629 static inline ktime_t sock_read_timestamp(struct sock *sk) 2630 { 2631 #if BITS_PER_LONG==32 2632 unsigned int seq; 2633 ktime_t kt; 2634 2635 do { 2636 seq = read_seqbegin(&sk->sk_stamp_seq); 2637 kt = sk->sk_stamp; 2638 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2639 2640 return kt; 2641 #else 2642 return READ_ONCE(sk->sk_stamp); 2643 #endif 2644 } 2645 2646 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2647 { 2648 #if BITS_PER_LONG==32 2649 write_seqlock(&sk->sk_stamp_seq); 2650 sk->sk_stamp = kt; 2651 write_sequnlock(&sk->sk_stamp_seq); 2652 #else 2653 WRITE_ONCE(sk->sk_stamp, kt); 2654 #endif 2655 } 2656 2657 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2658 struct sk_buff *skb); 2659 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2660 struct sk_buff *skb); 2661 2662 static inline void 2663 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2664 { 2665 ktime_t kt = skb->tstamp; 2666 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2667 2668 /* 2669 * generate control messages if 2670 * - receive time stamping in software requested 2671 * - software time stamp available and wanted 2672 * - hardware time stamps available and wanted 2673 */ 2674 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2675 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2676 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2677 (hwtstamps->hwtstamp && 2678 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2679 __sock_recv_timestamp(msg, sk, skb); 2680 else 2681 sock_write_timestamp(sk, kt); 2682 2683 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2684 __sock_recv_wifi_status(msg, sk, skb); 2685 } 2686 2687 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2688 struct sk_buff *skb); 2689 2690 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2691 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2692 struct sk_buff *skb) 2693 { 2694 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2695 (1UL << SOCK_RCVTSTAMP) | \ 2696 (1UL << SOCK_RCVMARK)) 2697 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2698 SOF_TIMESTAMPING_RAW_HARDWARE) 2699 2700 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2701 __sock_recv_cmsgs(msg, sk, skb); 2702 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2703 sock_write_timestamp(sk, skb->tstamp); 2704 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2705 sock_write_timestamp(sk, 0); 2706 } 2707 2708 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2709 2710 /** 2711 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2712 * @sk: socket sending this packet 2713 * @tsflags: timestamping flags to use 2714 * @tx_flags: completed with instructions for time stamping 2715 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2716 * 2717 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2718 */ 2719 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2720 __u8 *tx_flags, __u32 *tskey) 2721 { 2722 if (unlikely(tsflags)) { 2723 __sock_tx_timestamp(tsflags, tx_flags); 2724 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2725 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2726 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2727 } 2728 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2729 *tx_flags |= SKBTX_WIFI_STATUS; 2730 } 2731 2732 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2733 __u8 *tx_flags) 2734 { 2735 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2736 } 2737 2738 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2739 { 2740 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2741 &skb_shinfo(skb)->tskey); 2742 } 2743 2744 static inline bool sk_is_tcp(const struct sock *sk) 2745 { 2746 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2747 } 2748 2749 /** 2750 * sk_eat_skb - Release a skb if it is no longer needed 2751 * @sk: socket to eat this skb from 2752 * @skb: socket buffer to eat 2753 * 2754 * This routine must be called with interrupts disabled or with the socket 2755 * locked so that the sk_buff queue operation is ok. 2756 */ 2757 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2758 { 2759 __skb_unlink(skb, &sk->sk_receive_queue); 2760 __kfree_skb(skb); 2761 } 2762 2763 static inline bool 2764 skb_sk_is_prefetched(struct sk_buff *skb) 2765 { 2766 #ifdef CONFIG_INET 2767 return skb->destructor == sock_pfree; 2768 #else 2769 return false; 2770 #endif /* CONFIG_INET */ 2771 } 2772 2773 /* This helper checks if a socket is a full socket, 2774 * ie _not_ a timewait or request socket. 2775 */ 2776 static inline bool sk_fullsock(const struct sock *sk) 2777 { 2778 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2779 } 2780 2781 static inline bool 2782 sk_is_refcounted(struct sock *sk) 2783 { 2784 /* Only full sockets have sk->sk_flags. */ 2785 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2786 } 2787 2788 /** 2789 * skb_steal_sock - steal a socket from an sk_buff 2790 * @skb: sk_buff to steal the socket from 2791 * @refcounted: is set to true if the socket is reference-counted 2792 */ 2793 static inline struct sock * 2794 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2795 { 2796 if (skb->sk) { 2797 struct sock *sk = skb->sk; 2798 2799 *refcounted = true; 2800 if (skb_sk_is_prefetched(skb)) 2801 *refcounted = sk_is_refcounted(sk); 2802 skb->destructor = NULL; 2803 skb->sk = NULL; 2804 return sk; 2805 } 2806 *refcounted = false; 2807 return NULL; 2808 } 2809 2810 /* Checks if this SKB belongs to an HW offloaded socket 2811 * and whether any SW fallbacks are required based on dev. 2812 * Check decrypted mark in case skb_orphan() cleared socket. 2813 */ 2814 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2815 struct net_device *dev) 2816 { 2817 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2818 struct sock *sk = skb->sk; 2819 2820 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2821 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2822 #ifdef CONFIG_TLS_DEVICE 2823 } else if (unlikely(skb->decrypted)) { 2824 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2825 kfree_skb(skb); 2826 skb = NULL; 2827 #endif 2828 } 2829 #endif 2830 2831 return skb; 2832 } 2833 2834 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2835 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2836 */ 2837 static inline bool sk_listener(const struct sock *sk) 2838 { 2839 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2840 } 2841 2842 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2843 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2844 int type); 2845 2846 bool sk_ns_capable(const struct sock *sk, 2847 struct user_namespace *user_ns, int cap); 2848 bool sk_capable(const struct sock *sk, int cap); 2849 bool sk_net_capable(const struct sock *sk, int cap); 2850 2851 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2852 2853 /* Take into consideration the size of the struct sk_buff overhead in the 2854 * determination of these values, since that is non-constant across 2855 * platforms. This makes socket queueing behavior and performance 2856 * not depend upon such differences. 2857 */ 2858 #define _SK_MEM_PACKETS 256 2859 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2860 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2861 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2862 2863 extern __u32 sysctl_wmem_max; 2864 extern __u32 sysctl_rmem_max; 2865 2866 extern int sysctl_tstamp_allow_data; 2867 extern int sysctl_optmem_max; 2868 2869 extern __u32 sysctl_wmem_default; 2870 extern __u32 sysctl_rmem_default; 2871 2872 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2873 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2874 2875 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2876 { 2877 /* Does this proto have per netns sysctl_wmem ? */ 2878 if (proto->sysctl_wmem_offset) 2879 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2880 2881 return READ_ONCE(*proto->sysctl_wmem); 2882 } 2883 2884 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2885 { 2886 /* Does this proto have per netns sysctl_rmem ? */ 2887 if (proto->sysctl_rmem_offset) 2888 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2889 2890 return READ_ONCE(*proto->sysctl_rmem); 2891 } 2892 2893 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2894 * Some wifi drivers need to tweak it to get more chunks. 2895 * They can use this helper from their ndo_start_xmit() 2896 */ 2897 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2898 { 2899 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2900 return; 2901 WRITE_ONCE(sk->sk_pacing_shift, val); 2902 } 2903 2904 /* if a socket is bound to a device, check that the given device 2905 * index is either the same or that the socket is bound to an L3 2906 * master device and the given device index is also enslaved to 2907 * that L3 master 2908 */ 2909 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2910 { 2911 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2912 int mdif; 2913 2914 if (!bound_dev_if || bound_dev_if == dif) 2915 return true; 2916 2917 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2918 if (mdif && mdif == bound_dev_if) 2919 return true; 2920 2921 return false; 2922 } 2923 2924 void sock_def_readable(struct sock *sk); 2925 2926 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2927 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2928 int sock_set_timestamping(struct sock *sk, int optname, 2929 struct so_timestamping timestamping); 2930 2931 void sock_enable_timestamps(struct sock *sk); 2932 void sock_no_linger(struct sock *sk); 2933 void sock_set_keepalive(struct sock *sk); 2934 void sock_set_priority(struct sock *sk, u32 priority); 2935 void sock_set_rcvbuf(struct sock *sk, int val); 2936 void sock_set_mark(struct sock *sk, u32 val); 2937 void sock_set_reuseaddr(struct sock *sk); 2938 void sock_set_reuseport(struct sock *sk); 2939 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2940 2941 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2942 2943 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2944 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2945 sockptr_t optval, int optlen, bool old_timeval); 2946 2947 static inline bool sk_is_readable(struct sock *sk) 2948 { 2949 if (sk->sk_prot->sock_is_readable) 2950 return sk->sk_prot->sock_is_readable(sk); 2951 return false; 2952 } 2953 #endif /* _SOCK_H */ 2954