1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 union { 165 __addrpair skc_addrpair; 166 struct { 167 __be32 skc_daddr; 168 __be32 skc_rcv_saddr; 169 }; 170 }; 171 union { 172 unsigned int skc_hash; 173 __u16 skc_u16hashes[2]; 174 }; 175 /* skc_dport && skc_num must be grouped as well */ 176 union { 177 __portpair skc_portpair; 178 struct { 179 __be16 skc_dport; 180 __u16 skc_num; 181 }; 182 }; 183 184 unsigned short skc_family; 185 volatile unsigned char skc_state; 186 unsigned char skc_reuse:4; 187 unsigned char skc_reuseport:1; 188 unsigned char skc_ipv6only:1; 189 unsigned char skc_net_refcnt:1; 190 int skc_bound_dev_if; 191 union { 192 struct hlist_node skc_bind_node; 193 struct hlist_node skc_portaddr_node; 194 }; 195 struct proto *skc_prot; 196 possible_net_t skc_net; 197 198 #if IS_ENABLED(CONFIG_IPV6) 199 struct in6_addr skc_v6_daddr; 200 struct in6_addr skc_v6_rcv_saddr; 201 #endif 202 203 atomic64_t skc_cookie; 204 205 /* following fields are padding to force 206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 207 * assuming IPV6 is enabled. We use this padding differently 208 * for different kind of 'sockets' 209 */ 210 union { 211 unsigned long skc_flags; 212 struct sock *skc_listener; /* request_sock */ 213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 214 }; 215 /* 216 * fields between dontcopy_begin/dontcopy_end 217 * are not copied in sock_copy() 218 */ 219 /* private: */ 220 int skc_dontcopy_begin[0]; 221 /* public: */ 222 union { 223 struct hlist_node skc_node; 224 struct hlist_nulls_node skc_nulls_node; 225 }; 226 unsigned short skc_tx_queue_mapping; 227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 228 unsigned short skc_rx_queue_mapping; 229 #endif 230 union { 231 int skc_incoming_cpu; 232 u32 skc_rcv_wnd; 233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 234 }; 235 236 refcount_t skc_refcnt; 237 /* private: */ 238 int skc_dontcopy_end[0]; 239 union { 240 u32 skc_rxhash; 241 u32 skc_window_clamp; 242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 243 }; 244 /* public: */ 245 }; 246 247 struct bpf_local_storage; 248 struct sk_filter; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 304 * @sk_busy_poll_budget: napi processing budget when busypolling 305 * @sk_priority: %SO_PRIORITY setting 306 * @sk_type: socket type (%SOCK_STREAM, etc) 307 * @sk_protocol: which protocol this socket belongs in this network family 308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_txrehash: enable TX hash rethink 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 322 * Sockets that can be used under memory reclaim should 323 * set this to false. 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_state_change: callback to indicate change in the state of the sock 340 * @sk_data_ready: callback to indicate there is data to be processed 341 * @sk_write_space: callback to indicate there is bf sending space available 342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 343 * @sk_backlog_rcv: callback to process the backlog 344 * @sk_validate_xmit_skb: ptr to an optional validate function 345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 346 * @sk_reuseport_cb: reuseport group container 347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 348 * @sk_rcu: used during RCU grace period 349 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 350 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 351 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 352 * @sk_txtime_unused: unused txtime flags 353 * @ns_tracker: tracker for netns reference 354 * @sk_bind2_node: bind node in the bhash2 table 355 */ 356 struct sock { 357 /* 358 * Now struct inet_timewait_sock also uses sock_common, so please just 359 * don't add nothing before this first member (__sk_common) --acme 360 */ 361 struct sock_common __sk_common; 362 #define sk_node __sk_common.skc_node 363 #define sk_nulls_node __sk_common.skc_nulls_node 364 #define sk_refcnt __sk_common.skc_refcnt 365 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 366 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 367 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 368 #endif 369 370 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 371 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 372 #define sk_hash __sk_common.skc_hash 373 #define sk_portpair __sk_common.skc_portpair 374 #define sk_num __sk_common.skc_num 375 #define sk_dport __sk_common.skc_dport 376 #define sk_addrpair __sk_common.skc_addrpair 377 #define sk_daddr __sk_common.skc_daddr 378 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 379 #define sk_family __sk_common.skc_family 380 #define sk_state __sk_common.skc_state 381 #define sk_reuse __sk_common.skc_reuse 382 #define sk_reuseport __sk_common.skc_reuseport 383 #define sk_ipv6only __sk_common.skc_ipv6only 384 #define sk_net_refcnt __sk_common.skc_net_refcnt 385 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 386 #define sk_bind_node __sk_common.skc_bind_node 387 #define sk_prot __sk_common.skc_prot 388 #define sk_net __sk_common.skc_net 389 #define sk_v6_daddr __sk_common.skc_v6_daddr 390 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 391 #define sk_cookie __sk_common.skc_cookie 392 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 393 #define sk_flags __sk_common.skc_flags 394 #define sk_rxhash __sk_common.skc_rxhash 395 396 /* early demux fields */ 397 struct dst_entry __rcu *sk_rx_dst; 398 int sk_rx_dst_ifindex; 399 u32 sk_rx_dst_cookie; 400 401 socket_lock_t sk_lock; 402 atomic_t sk_drops; 403 int sk_rcvlowat; 404 struct sk_buff_head sk_error_queue; 405 struct sk_buff_head sk_receive_queue; 406 /* 407 * The backlog queue is special, it is always used with 408 * the per-socket spinlock held and requires low latency 409 * access. Therefore we special case it's implementation. 410 * Note : rmem_alloc is in this structure to fill a hole 411 * on 64bit arches, not because its logically part of 412 * backlog. 413 */ 414 struct { 415 atomic_t rmem_alloc; 416 int len; 417 struct sk_buff *head; 418 struct sk_buff *tail; 419 } sk_backlog; 420 421 #define sk_rmem_alloc sk_backlog.rmem_alloc 422 423 int sk_forward_alloc; 424 u32 sk_reserved_mem; 425 #ifdef CONFIG_NET_RX_BUSY_POLL 426 unsigned int sk_ll_usec; 427 /* ===== mostly read cache line ===== */ 428 unsigned int sk_napi_id; 429 #endif 430 int sk_rcvbuf; 431 432 struct sk_filter __rcu *sk_filter; 433 union { 434 struct socket_wq __rcu *sk_wq; 435 /* private: */ 436 struct socket_wq *sk_wq_raw; 437 /* public: */ 438 }; 439 #ifdef CONFIG_XFRM 440 struct xfrm_policy __rcu *sk_policy[2]; 441 #endif 442 443 struct dst_entry __rcu *sk_dst_cache; 444 atomic_t sk_omem_alloc; 445 int sk_sndbuf; 446 447 /* ===== cache line for TX ===== */ 448 int sk_wmem_queued; 449 refcount_t sk_wmem_alloc; 450 unsigned long sk_tsq_flags; 451 union { 452 struct sk_buff *sk_send_head; 453 struct rb_root tcp_rtx_queue; 454 }; 455 struct sk_buff_head sk_write_queue; 456 __s32 sk_peek_off; 457 int sk_write_pending; 458 __u32 sk_dst_pending_confirm; 459 u32 sk_pacing_status; /* see enum sk_pacing */ 460 long sk_sndtimeo; 461 struct timer_list sk_timer; 462 __u32 sk_priority; 463 __u32 sk_mark; 464 unsigned long sk_pacing_rate; /* bytes per second */ 465 unsigned long sk_max_pacing_rate; 466 struct page_frag sk_frag; 467 netdev_features_t sk_route_caps; 468 int sk_gso_type; 469 unsigned int sk_gso_max_size; 470 gfp_t sk_allocation; 471 __u32 sk_txhash; 472 473 /* 474 * Because of non atomicity rules, all 475 * changes are protected by socket lock. 476 */ 477 u8 sk_gso_disabled : 1, 478 sk_kern_sock : 1, 479 sk_no_check_tx : 1, 480 sk_no_check_rx : 1, 481 sk_userlocks : 4; 482 u8 sk_pacing_shift; 483 u16 sk_type; 484 u16 sk_protocol; 485 u16 sk_gso_max_segs; 486 unsigned long sk_lingertime; 487 struct proto *sk_prot_creator; 488 rwlock_t sk_callback_lock; 489 int sk_err, 490 sk_err_soft; 491 u32 sk_ack_backlog; 492 u32 sk_max_ack_backlog; 493 kuid_t sk_uid; 494 u8 sk_txrehash; 495 #ifdef CONFIG_NET_RX_BUSY_POLL 496 u8 sk_prefer_busy_poll; 497 u16 sk_busy_poll_budget; 498 #endif 499 spinlock_t sk_peer_lock; 500 int sk_bind_phc; 501 struct pid *sk_peer_pid; 502 const struct cred *sk_peer_cred; 503 504 long sk_rcvtimeo; 505 ktime_t sk_stamp; 506 #if BITS_PER_LONG==32 507 seqlock_t sk_stamp_seq; 508 #endif 509 atomic_t sk_tskey; 510 atomic_t sk_zckey; 511 u32 sk_tsflags; 512 u8 sk_shutdown; 513 514 u8 sk_clockid; 515 u8 sk_txtime_deadline_mode : 1, 516 sk_txtime_report_errors : 1, 517 sk_txtime_unused : 6; 518 bool sk_use_task_frag; 519 520 struct socket *sk_socket; 521 void *sk_user_data; 522 #ifdef CONFIG_SECURITY 523 void *sk_security; 524 #endif 525 struct sock_cgroup_data sk_cgrp_data; 526 struct mem_cgroup *sk_memcg; 527 void (*sk_state_change)(struct sock *sk); 528 void (*sk_data_ready)(struct sock *sk); 529 void (*sk_write_space)(struct sock *sk); 530 void (*sk_error_report)(struct sock *sk); 531 int (*sk_backlog_rcv)(struct sock *sk, 532 struct sk_buff *skb); 533 #ifdef CONFIG_SOCK_VALIDATE_XMIT 534 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 535 struct net_device *dev, 536 struct sk_buff *skb); 537 #endif 538 void (*sk_destruct)(struct sock *sk); 539 struct sock_reuseport __rcu *sk_reuseport_cb; 540 #ifdef CONFIG_BPF_SYSCALL 541 struct bpf_local_storage __rcu *sk_bpf_storage; 542 #endif 543 struct rcu_head sk_rcu; 544 netns_tracker ns_tracker; 545 struct hlist_node sk_bind2_node; 546 }; 547 548 enum sk_pacing { 549 SK_PACING_NONE = 0, 550 SK_PACING_NEEDED = 1, 551 SK_PACING_FQ = 2, 552 }; 553 554 /* flag bits in sk_user_data 555 * 556 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 557 * not be suitable for copying when cloning the socket. For instance, 558 * it can point to a reference counted object. sk_user_data bottom 559 * bit is set if pointer must not be copied. 560 * 561 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 562 * managed/owned by a BPF reuseport array. This bit should be set 563 * when sk_user_data's sk is added to the bpf's reuseport_array. 564 * 565 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 566 * sk_user_data points to psock type. This bit should be set 567 * when sk_user_data is assigned to a psock object. 568 */ 569 #define SK_USER_DATA_NOCOPY 1UL 570 #define SK_USER_DATA_BPF 2UL 571 #define SK_USER_DATA_PSOCK 4UL 572 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 573 SK_USER_DATA_PSOCK) 574 575 /** 576 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 577 * @sk: socket 578 */ 579 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 580 { 581 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 582 } 583 584 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 585 586 /** 587 * __locked_read_sk_user_data_with_flags - return the pointer 588 * only if argument flags all has been set in sk_user_data. Otherwise 589 * return NULL 590 * 591 * @sk: socket 592 * @flags: flag bits 593 * 594 * The caller must be holding sk->sk_callback_lock. 595 */ 596 static inline void * 597 __locked_read_sk_user_data_with_flags(const struct sock *sk, 598 uintptr_t flags) 599 { 600 uintptr_t sk_user_data = 601 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 602 lockdep_is_held(&sk->sk_callback_lock)); 603 604 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 605 606 if ((sk_user_data & flags) == flags) 607 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 608 return NULL; 609 } 610 611 /** 612 * __rcu_dereference_sk_user_data_with_flags - return the pointer 613 * only if argument flags all has been set in sk_user_data. Otherwise 614 * return NULL 615 * 616 * @sk: socket 617 * @flags: flag bits 618 */ 619 static inline void * 620 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 621 uintptr_t flags) 622 { 623 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 624 625 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 626 627 if ((sk_user_data & flags) == flags) 628 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 629 return NULL; 630 } 631 632 #define rcu_dereference_sk_user_data(sk) \ 633 __rcu_dereference_sk_user_data_with_flags(sk, 0) 634 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 635 ({ \ 636 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 637 __tmp2 = (uintptr_t)(flags); \ 638 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 639 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 640 rcu_assign_pointer(__sk_user_data((sk)), \ 641 __tmp1 | __tmp2); \ 642 }) 643 #define rcu_assign_sk_user_data(sk, ptr) \ 644 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 645 646 static inline 647 struct net *sock_net(const struct sock *sk) 648 { 649 return read_pnet(&sk->sk_net); 650 } 651 652 static inline 653 void sock_net_set(struct sock *sk, struct net *net) 654 { 655 write_pnet(&sk->sk_net, net); 656 } 657 658 /* 659 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 660 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 661 * on a socket means that the socket will reuse everybody else's port 662 * without looking at the other's sk_reuse value. 663 */ 664 665 #define SK_NO_REUSE 0 666 #define SK_CAN_REUSE 1 667 #define SK_FORCE_REUSE 2 668 669 int sk_set_peek_off(struct sock *sk, int val); 670 671 static inline int sk_peek_offset(const struct sock *sk, int flags) 672 { 673 if (unlikely(flags & MSG_PEEK)) { 674 return READ_ONCE(sk->sk_peek_off); 675 } 676 677 return 0; 678 } 679 680 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 681 { 682 s32 off = READ_ONCE(sk->sk_peek_off); 683 684 if (unlikely(off >= 0)) { 685 off = max_t(s32, off - val, 0); 686 WRITE_ONCE(sk->sk_peek_off, off); 687 } 688 } 689 690 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 691 { 692 sk_peek_offset_bwd(sk, -val); 693 } 694 695 /* 696 * Hashed lists helper routines 697 */ 698 static inline struct sock *sk_entry(const struct hlist_node *node) 699 { 700 return hlist_entry(node, struct sock, sk_node); 701 } 702 703 static inline struct sock *__sk_head(const struct hlist_head *head) 704 { 705 return hlist_entry(head->first, struct sock, sk_node); 706 } 707 708 static inline struct sock *sk_head(const struct hlist_head *head) 709 { 710 return hlist_empty(head) ? NULL : __sk_head(head); 711 } 712 713 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 714 { 715 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 716 } 717 718 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 719 { 720 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 721 } 722 723 static inline struct sock *sk_next(const struct sock *sk) 724 { 725 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 726 } 727 728 static inline struct sock *sk_nulls_next(const struct sock *sk) 729 { 730 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 731 hlist_nulls_entry(sk->sk_nulls_node.next, 732 struct sock, sk_nulls_node) : 733 NULL; 734 } 735 736 static inline bool sk_unhashed(const struct sock *sk) 737 { 738 return hlist_unhashed(&sk->sk_node); 739 } 740 741 static inline bool sk_hashed(const struct sock *sk) 742 { 743 return !sk_unhashed(sk); 744 } 745 746 static inline void sk_node_init(struct hlist_node *node) 747 { 748 node->pprev = NULL; 749 } 750 751 static inline void __sk_del_node(struct sock *sk) 752 { 753 __hlist_del(&sk->sk_node); 754 } 755 756 /* NB: equivalent to hlist_del_init_rcu */ 757 static inline bool __sk_del_node_init(struct sock *sk) 758 { 759 if (sk_hashed(sk)) { 760 __sk_del_node(sk); 761 sk_node_init(&sk->sk_node); 762 return true; 763 } 764 return false; 765 } 766 767 /* Grab socket reference count. This operation is valid only 768 when sk is ALREADY grabbed f.e. it is found in hash table 769 or a list and the lookup is made under lock preventing hash table 770 modifications. 771 */ 772 773 static __always_inline void sock_hold(struct sock *sk) 774 { 775 refcount_inc(&sk->sk_refcnt); 776 } 777 778 /* Ungrab socket in the context, which assumes that socket refcnt 779 cannot hit zero, f.e. it is true in context of any socketcall. 780 */ 781 static __always_inline void __sock_put(struct sock *sk) 782 { 783 refcount_dec(&sk->sk_refcnt); 784 } 785 786 static inline bool sk_del_node_init(struct sock *sk) 787 { 788 bool rc = __sk_del_node_init(sk); 789 790 if (rc) { 791 /* paranoid for a while -acme */ 792 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 793 __sock_put(sk); 794 } 795 return rc; 796 } 797 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 798 799 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 800 { 801 if (sk_hashed(sk)) { 802 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 803 return true; 804 } 805 return false; 806 } 807 808 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 809 { 810 bool rc = __sk_nulls_del_node_init_rcu(sk); 811 812 if (rc) { 813 /* paranoid for a while -acme */ 814 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 815 __sock_put(sk); 816 } 817 return rc; 818 } 819 820 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 821 { 822 hlist_add_head(&sk->sk_node, list); 823 } 824 825 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 826 { 827 sock_hold(sk); 828 __sk_add_node(sk, list); 829 } 830 831 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 832 { 833 sock_hold(sk); 834 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 835 sk->sk_family == AF_INET6) 836 hlist_add_tail_rcu(&sk->sk_node, list); 837 else 838 hlist_add_head_rcu(&sk->sk_node, list); 839 } 840 841 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 842 { 843 sock_hold(sk); 844 hlist_add_tail_rcu(&sk->sk_node, list); 845 } 846 847 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 848 { 849 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 850 } 851 852 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 853 { 854 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 855 } 856 857 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 858 { 859 sock_hold(sk); 860 __sk_nulls_add_node_rcu(sk, list); 861 } 862 863 static inline void __sk_del_bind_node(struct sock *sk) 864 { 865 __hlist_del(&sk->sk_bind_node); 866 } 867 868 static inline void sk_add_bind_node(struct sock *sk, 869 struct hlist_head *list) 870 { 871 hlist_add_head(&sk->sk_bind_node, list); 872 } 873 874 static inline void __sk_del_bind2_node(struct sock *sk) 875 { 876 __hlist_del(&sk->sk_bind2_node); 877 } 878 879 static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list) 880 { 881 hlist_add_head(&sk->sk_bind2_node, list); 882 } 883 884 #define sk_for_each(__sk, list) \ 885 hlist_for_each_entry(__sk, list, sk_node) 886 #define sk_for_each_rcu(__sk, list) \ 887 hlist_for_each_entry_rcu(__sk, list, sk_node) 888 #define sk_nulls_for_each(__sk, node, list) \ 889 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 890 #define sk_nulls_for_each_rcu(__sk, node, list) \ 891 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 892 #define sk_for_each_from(__sk) \ 893 hlist_for_each_entry_from(__sk, sk_node) 894 #define sk_nulls_for_each_from(__sk, node) \ 895 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 896 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 897 #define sk_for_each_safe(__sk, tmp, list) \ 898 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 899 #define sk_for_each_bound(__sk, list) \ 900 hlist_for_each_entry(__sk, list, sk_bind_node) 901 #define sk_for_each_bound_bhash2(__sk, list) \ 902 hlist_for_each_entry(__sk, list, sk_bind2_node) 903 904 /** 905 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 906 * @tpos: the type * to use as a loop cursor. 907 * @pos: the &struct hlist_node to use as a loop cursor. 908 * @head: the head for your list. 909 * @offset: offset of hlist_node within the struct. 910 * 911 */ 912 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 913 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 914 pos != NULL && \ 915 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 916 pos = rcu_dereference(hlist_next_rcu(pos))) 917 918 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 919 { 920 /* Careful only use this in a context where these parameters 921 * can not change and must all be valid, such as recvmsg from 922 * userspace. 923 */ 924 return sk->sk_socket->file->f_cred->user_ns; 925 } 926 927 /* Sock flags */ 928 enum sock_flags { 929 SOCK_DEAD, 930 SOCK_DONE, 931 SOCK_URGINLINE, 932 SOCK_KEEPOPEN, 933 SOCK_LINGER, 934 SOCK_DESTROY, 935 SOCK_BROADCAST, 936 SOCK_TIMESTAMP, 937 SOCK_ZAPPED, 938 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 939 SOCK_DBG, /* %SO_DEBUG setting */ 940 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 941 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 942 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 943 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 944 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 945 SOCK_FASYNC, /* fasync() active */ 946 SOCK_RXQ_OVFL, 947 SOCK_ZEROCOPY, /* buffers from userspace */ 948 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 949 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 950 * Will use last 4 bytes of packet sent from 951 * user-space instead. 952 */ 953 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 954 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 955 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 956 SOCK_TXTIME, 957 SOCK_XDP, /* XDP is attached */ 958 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 959 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 960 }; 961 962 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 963 964 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 965 { 966 nsk->sk_flags = osk->sk_flags; 967 } 968 969 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 970 { 971 __set_bit(flag, &sk->sk_flags); 972 } 973 974 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 975 { 976 __clear_bit(flag, &sk->sk_flags); 977 } 978 979 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 980 int valbool) 981 { 982 if (valbool) 983 sock_set_flag(sk, bit); 984 else 985 sock_reset_flag(sk, bit); 986 } 987 988 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 989 { 990 return test_bit(flag, &sk->sk_flags); 991 } 992 993 #ifdef CONFIG_NET 994 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 995 static inline int sk_memalloc_socks(void) 996 { 997 return static_branch_unlikely(&memalloc_socks_key); 998 } 999 1000 void __receive_sock(struct file *file); 1001 #else 1002 1003 static inline int sk_memalloc_socks(void) 1004 { 1005 return 0; 1006 } 1007 1008 static inline void __receive_sock(struct file *file) 1009 { } 1010 #endif 1011 1012 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1013 { 1014 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1015 } 1016 1017 static inline void sk_acceptq_removed(struct sock *sk) 1018 { 1019 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1020 } 1021 1022 static inline void sk_acceptq_added(struct sock *sk) 1023 { 1024 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1025 } 1026 1027 /* Note: If you think the test should be: 1028 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1029 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1030 */ 1031 static inline bool sk_acceptq_is_full(const struct sock *sk) 1032 { 1033 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1034 } 1035 1036 /* 1037 * Compute minimal free write space needed to queue new packets. 1038 */ 1039 static inline int sk_stream_min_wspace(const struct sock *sk) 1040 { 1041 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1042 } 1043 1044 static inline int sk_stream_wspace(const struct sock *sk) 1045 { 1046 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1047 } 1048 1049 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1050 { 1051 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1052 } 1053 1054 void sk_stream_write_space(struct sock *sk); 1055 1056 /* OOB backlog add */ 1057 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1058 { 1059 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1060 skb_dst_force(skb); 1061 1062 if (!sk->sk_backlog.tail) 1063 WRITE_ONCE(sk->sk_backlog.head, skb); 1064 else 1065 sk->sk_backlog.tail->next = skb; 1066 1067 WRITE_ONCE(sk->sk_backlog.tail, skb); 1068 skb->next = NULL; 1069 } 1070 1071 /* 1072 * Take into account size of receive queue and backlog queue 1073 * Do not take into account this skb truesize, 1074 * to allow even a single big packet to come. 1075 */ 1076 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1077 { 1078 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1079 1080 return qsize > limit; 1081 } 1082 1083 /* The per-socket spinlock must be held here. */ 1084 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1085 unsigned int limit) 1086 { 1087 if (sk_rcvqueues_full(sk, limit)) 1088 return -ENOBUFS; 1089 1090 /* 1091 * If the skb was allocated from pfmemalloc reserves, only 1092 * allow SOCK_MEMALLOC sockets to use it as this socket is 1093 * helping free memory 1094 */ 1095 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1096 return -ENOMEM; 1097 1098 __sk_add_backlog(sk, skb); 1099 sk->sk_backlog.len += skb->truesize; 1100 return 0; 1101 } 1102 1103 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1104 1105 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1106 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1107 1108 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1109 { 1110 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1111 return __sk_backlog_rcv(sk, skb); 1112 1113 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1114 tcp_v6_do_rcv, 1115 tcp_v4_do_rcv, 1116 sk, skb); 1117 } 1118 1119 static inline void sk_incoming_cpu_update(struct sock *sk) 1120 { 1121 int cpu = raw_smp_processor_id(); 1122 1123 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1124 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1125 } 1126 1127 static inline void sock_rps_record_flow_hash(__u32 hash) 1128 { 1129 #ifdef CONFIG_RPS 1130 struct rps_sock_flow_table *sock_flow_table; 1131 1132 rcu_read_lock(); 1133 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1134 rps_record_sock_flow(sock_flow_table, hash); 1135 rcu_read_unlock(); 1136 #endif 1137 } 1138 1139 static inline void sock_rps_record_flow(const struct sock *sk) 1140 { 1141 #ifdef CONFIG_RPS 1142 if (static_branch_unlikely(&rfs_needed)) { 1143 /* Reading sk->sk_rxhash might incur an expensive cache line 1144 * miss. 1145 * 1146 * TCP_ESTABLISHED does cover almost all states where RFS 1147 * might be useful, and is cheaper [1] than testing : 1148 * IPv4: inet_sk(sk)->inet_daddr 1149 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1150 * OR an additional socket flag 1151 * [1] : sk_state and sk_prot are in the same cache line. 1152 */ 1153 if (sk->sk_state == TCP_ESTABLISHED) 1154 sock_rps_record_flow_hash(sk->sk_rxhash); 1155 } 1156 #endif 1157 } 1158 1159 static inline void sock_rps_save_rxhash(struct sock *sk, 1160 const struct sk_buff *skb) 1161 { 1162 #ifdef CONFIG_RPS 1163 if (unlikely(sk->sk_rxhash != skb->hash)) 1164 sk->sk_rxhash = skb->hash; 1165 #endif 1166 } 1167 1168 static inline void sock_rps_reset_rxhash(struct sock *sk) 1169 { 1170 #ifdef CONFIG_RPS 1171 sk->sk_rxhash = 0; 1172 #endif 1173 } 1174 1175 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1176 ({ int __rc; \ 1177 release_sock(__sk); \ 1178 __rc = __condition; \ 1179 if (!__rc) { \ 1180 *(__timeo) = wait_woken(__wait, \ 1181 TASK_INTERRUPTIBLE, \ 1182 *(__timeo)); \ 1183 } \ 1184 sched_annotate_sleep(); \ 1185 lock_sock(__sk); \ 1186 __rc = __condition; \ 1187 __rc; \ 1188 }) 1189 1190 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1191 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1192 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1193 int sk_stream_error(struct sock *sk, int flags, int err); 1194 void sk_stream_kill_queues(struct sock *sk); 1195 void sk_set_memalloc(struct sock *sk); 1196 void sk_clear_memalloc(struct sock *sk); 1197 1198 void __sk_flush_backlog(struct sock *sk); 1199 1200 static inline bool sk_flush_backlog(struct sock *sk) 1201 { 1202 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1203 __sk_flush_backlog(sk); 1204 return true; 1205 } 1206 return false; 1207 } 1208 1209 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1210 1211 struct request_sock_ops; 1212 struct timewait_sock_ops; 1213 struct inet_hashinfo; 1214 struct raw_hashinfo; 1215 struct smc_hashinfo; 1216 struct module; 1217 struct sk_psock; 1218 1219 /* 1220 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1221 * un-modified. Special care is taken when initializing object to zero. 1222 */ 1223 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1224 { 1225 if (offsetof(struct sock, sk_node.next) != 0) 1226 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1227 memset(&sk->sk_node.pprev, 0, 1228 size - offsetof(struct sock, sk_node.pprev)); 1229 } 1230 1231 /* Networking protocol blocks we attach to sockets. 1232 * socket layer -> transport layer interface 1233 */ 1234 struct proto { 1235 void (*close)(struct sock *sk, 1236 long timeout); 1237 int (*pre_connect)(struct sock *sk, 1238 struct sockaddr *uaddr, 1239 int addr_len); 1240 int (*connect)(struct sock *sk, 1241 struct sockaddr *uaddr, 1242 int addr_len); 1243 int (*disconnect)(struct sock *sk, int flags); 1244 1245 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1246 bool kern); 1247 1248 int (*ioctl)(struct sock *sk, int cmd, 1249 unsigned long arg); 1250 int (*init)(struct sock *sk); 1251 void (*destroy)(struct sock *sk); 1252 void (*shutdown)(struct sock *sk, int how); 1253 int (*setsockopt)(struct sock *sk, int level, 1254 int optname, sockptr_t optval, 1255 unsigned int optlen); 1256 int (*getsockopt)(struct sock *sk, int level, 1257 int optname, char __user *optval, 1258 int __user *option); 1259 void (*keepalive)(struct sock *sk, int valbool); 1260 #ifdef CONFIG_COMPAT 1261 int (*compat_ioctl)(struct sock *sk, 1262 unsigned int cmd, unsigned long arg); 1263 #endif 1264 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1265 size_t len); 1266 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1267 size_t len, int flags, int *addr_len); 1268 int (*sendpage)(struct sock *sk, struct page *page, 1269 int offset, size_t size, int flags); 1270 int (*bind)(struct sock *sk, 1271 struct sockaddr *addr, int addr_len); 1272 int (*bind_add)(struct sock *sk, 1273 struct sockaddr *addr, int addr_len); 1274 1275 int (*backlog_rcv) (struct sock *sk, 1276 struct sk_buff *skb); 1277 bool (*bpf_bypass_getsockopt)(int level, 1278 int optname); 1279 1280 void (*release_cb)(struct sock *sk); 1281 1282 /* Keeping track of sk's, looking them up, and port selection methods. */ 1283 int (*hash)(struct sock *sk); 1284 void (*unhash)(struct sock *sk); 1285 void (*rehash)(struct sock *sk); 1286 int (*get_port)(struct sock *sk, unsigned short snum); 1287 void (*put_port)(struct sock *sk); 1288 #ifdef CONFIG_BPF_SYSCALL 1289 int (*psock_update_sk_prot)(struct sock *sk, 1290 struct sk_psock *psock, 1291 bool restore); 1292 #endif 1293 1294 /* Keeping track of sockets in use */ 1295 #ifdef CONFIG_PROC_FS 1296 unsigned int inuse_idx; 1297 #endif 1298 1299 #if IS_ENABLED(CONFIG_MPTCP) 1300 int (*forward_alloc_get)(const struct sock *sk); 1301 #endif 1302 1303 bool (*stream_memory_free)(const struct sock *sk, int wake); 1304 bool (*sock_is_readable)(struct sock *sk); 1305 /* Memory pressure */ 1306 void (*enter_memory_pressure)(struct sock *sk); 1307 void (*leave_memory_pressure)(struct sock *sk); 1308 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1309 int __percpu *per_cpu_fw_alloc; 1310 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1311 1312 /* 1313 * Pressure flag: try to collapse. 1314 * Technical note: it is used by multiple contexts non atomically. 1315 * All the __sk_mem_schedule() is of this nature: accounting 1316 * is strict, actions are advisory and have some latency. 1317 */ 1318 unsigned long *memory_pressure; 1319 long *sysctl_mem; 1320 1321 int *sysctl_wmem; 1322 int *sysctl_rmem; 1323 u32 sysctl_wmem_offset; 1324 u32 sysctl_rmem_offset; 1325 1326 int max_header; 1327 bool no_autobind; 1328 1329 struct kmem_cache *slab; 1330 unsigned int obj_size; 1331 slab_flags_t slab_flags; 1332 unsigned int useroffset; /* Usercopy region offset */ 1333 unsigned int usersize; /* Usercopy region size */ 1334 1335 unsigned int __percpu *orphan_count; 1336 1337 struct request_sock_ops *rsk_prot; 1338 struct timewait_sock_ops *twsk_prot; 1339 1340 union { 1341 struct inet_hashinfo *hashinfo; 1342 struct udp_table *udp_table; 1343 struct raw_hashinfo *raw_hash; 1344 struct smc_hashinfo *smc_hash; 1345 } h; 1346 1347 struct module *owner; 1348 1349 char name[32]; 1350 1351 struct list_head node; 1352 #ifdef SOCK_REFCNT_DEBUG 1353 atomic_t socks; 1354 #endif 1355 int (*diag_destroy)(struct sock *sk, int err); 1356 } __randomize_layout; 1357 1358 int proto_register(struct proto *prot, int alloc_slab); 1359 void proto_unregister(struct proto *prot); 1360 int sock_load_diag_module(int family, int protocol); 1361 1362 #ifdef SOCK_REFCNT_DEBUG 1363 static inline void sk_refcnt_debug_inc(struct sock *sk) 1364 { 1365 atomic_inc(&sk->sk_prot->socks); 1366 } 1367 1368 static inline void sk_refcnt_debug_dec(struct sock *sk) 1369 { 1370 atomic_dec(&sk->sk_prot->socks); 1371 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1372 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1373 } 1374 1375 static inline void sk_refcnt_debug_release(const struct sock *sk) 1376 { 1377 if (refcount_read(&sk->sk_refcnt) != 1) 1378 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1379 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1380 } 1381 #else /* SOCK_REFCNT_DEBUG */ 1382 #define sk_refcnt_debug_inc(sk) do { } while (0) 1383 #define sk_refcnt_debug_dec(sk) do { } while (0) 1384 #define sk_refcnt_debug_release(sk) do { } while (0) 1385 #endif /* SOCK_REFCNT_DEBUG */ 1386 1387 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1388 1389 static inline int sk_forward_alloc_get(const struct sock *sk) 1390 { 1391 #if IS_ENABLED(CONFIG_MPTCP) 1392 if (sk->sk_prot->forward_alloc_get) 1393 return sk->sk_prot->forward_alloc_get(sk); 1394 #endif 1395 return sk->sk_forward_alloc; 1396 } 1397 1398 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1399 { 1400 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1401 return false; 1402 1403 return sk->sk_prot->stream_memory_free ? 1404 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1405 tcp_stream_memory_free, sk, wake) : true; 1406 } 1407 1408 static inline bool sk_stream_memory_free(const struct sock *sk) 1409 { 1410 return __sk_stream_memory_free(sk, 0); 1411 } 1412 1413 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1414 { 1415 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1416 __sk_stream_memory_free(sk, wake); 1417 } 1418 1419 static inline bool sk_stream_is_writeable(const struct sock *sk) 1420 { 1421 return __sk_stream_is_writeable(sk, 0); 1422 } 1423 1424 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1425 struct cgroup *ancestor) 1426 { 1427 #ifdef CONFIG_SOCK_CGROUP_DATA 1428 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1429 ancestor); 1430 #else 1431 return -ENOTSUPP; 1432 #endif 1433 } 1434 1435 static inline bool sk_has_memory_pressure(const struct sock *sk) 1436 { 1437 return sk->sk_prot->memory_pressure != NULL; 1438 } 1439 1440 static inline bool sk_under_memory_pressure(const struct sock *sk) 1441 { 1442 if (!sk->sk_prot->memory_pressure) 1443 return false; 1444 1445 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1446 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1447 return true; 1448 1449 return !!*sk->sk_prot->memory_pressure; 1450 } 1451 1452 static inline long 1453 proto_memory_allocated(const struct proto *prot) 1454 { 1455 return max(0L, atomic_long_read(prot->memory_allocated)); 1456 } 1457 1458 static inline long 1459 sk_memory_allocated(const struct sock *sk) 1460 { 1461 return proto_memory_allocated(sk->sk_prot); 1462 } 1463 1464 /* 1 MB per cpu, in page units */ 1465 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1466 1467 static inline void 1468 sk_memory_allocated_add(struct sock *sk, int amt) 1469 { 1470 int local_reserve; 1471 1472 preempt_disable(); 1473 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1474 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) { 1475 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1476 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1477 } 1478 preempt_enable(); 1479 } 1480 1481 static inline void 1482 sk_memory_allocated_sub(struct sock *sk, int amt) 1483 { 1484 int local_reserve; 1485 1486 preempt_disable(); 1487 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1488 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) { 1489 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1490 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1491 } 1492 preempt_enable(); 1493 } 1494 1495 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1496 1497 static inline void sk_sockets_allocated_dec(struct sock *sk) 1498 { 1499 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1500 SK_ALLOC_PERCPU_COUNTER_BATCH); 1501 } 1502 1503 static inline void sk_sockets_allocated_inc(struct sock *sk) 1504 { 1505 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1506 SK_ALLOC_PERCPU_COUNTER_BATCH); 1507 } 1508 1509 static inline u64 1510 sk_sockets_allocated_read_positive(struct sock *sk) 1511 { 1512 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1513 } 1514 1515 static inline int 1516 proto_sockets_allocated_sum_positive(struct proto *prot) 1517 { 1518 return percpu_counter_sum_positive(prot->sockets_allocated); 1519 } 1520 1521 static inline bool 1522 proto_memory_pressure(struct proto *prot) 1523 { 1524 if (!prot->memory_pressure) 1525 return false; 1526 return !!*prot->memory_pressure; 1527 } 1528 1529 1530 #ifdef CONFIG_PROC_FS 1531 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1532 struct prot_inuse { 1533 int all; 1534 int val[PROTO_INUSE_NR]; 1535 }; 1536 1537 static inline void sock_prot_inuse_add(const struct net *net, 1538 const struct proto *prot, int val) 1539 { 1540 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1541 } 1542 1543 static inline void sock_inuse_add(const struct net *net, int val) 1544 { 1545 this_cpu_add(net->core.prot_inuse->all, val); 1546 } 1547 1548 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1549 int sock_inuse_get(struct net *net); 1550 #else 1551 static inline void sock_prot_inuse_add(const struct net *net, 1552 const struct proto *prot, int val) 1553 { 1554 } 1555 1556 static inline void sock_inuse_add(const struct net *net, int val) 1557 { 1558 } 1559 #endif 1560 1561 1562 /* With per-bucket locks this operation is not-atomic, so that 1563 * this version is not worse. 1564 */ 1565 static inline int __sk_prot_rehash(struct sock *sk) 1566 { 1567 sk->sk_prot->unhash(sk); 1568 return sk->sk_prot->hash(sk); 1569 } 1570 1571 /* About 10 seconds */ 1572 #define SOCK_DESTROY_TIME (10*HZ) 1573 1574 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1575 #define PROT_SOCK 1024 1576 1577 #define SHUTDOWN_MASK 3 1578 #define RCV_SHUTDOWN 1 1579 #define SEND_SHUTDOWN 2 1580 1581 #define SOCK_BINDADDR_LOCK 4 1582 #define SOCK_BINDPORT_LOCK 8 1583 1584 struct socket_alloc { 1585 struct socket socket; 1586 struct inode vfs_inode; 1587 }; 1588 1589 static inline struct socket *SOCKET_I(struct inode *inode) 1590 { 1591 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1592 } 1593 1594 static inline struct inode *SOCK_INODE(struct socket *socket) 1595 { 1596 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1597 } 1598 1599 /* 1600 * Functions for memory accounting 1601 */ 1602 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1603 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1604 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1605 void __sk_mem_reclaim(struct sock *sk, int amount); 1606 1607 #define SK_MEM_SEND 0 1608 #define SK_MEM_RECV 1 1609 1610 /* sysctl_mem values are in pages */ 1611 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1612 { 1613 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1614 } 1615 1616 static inline int sk_mem_pages(int amt) 1617 { 1618 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1619 } 1620 1621 static inline bool sk_has_account(struct sock *sk) 1622 { 1623 /* return true if protocol supports memory accounting */ 1624 return !!sk->sk_prot->memory_allocated; 1625 } 1626 1627 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1628 { 1629 int delta; 1630 1631 if (!sk_has_account(sk)) 1632 return true; 1633 delta = size - sk->sk_forward_alloc; 1634 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1635 } 1636 1637 static inline bool 1638 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1639 { 1640 int delta; 1641 1642 if (!sk_has_account(sk)) 1643 return true; 1644 delta = size - sk->sk_forward_alloc; 1645 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1646 skb_pfmemalloc(skb); 1647 } 1648 1649 static inline int sk_unused_reserved_mem(const struct sock *sk) 1650 { 1651 int unused_mem; 1652 1653 if (likely(!sk->sk_reserved_mem)) 1654 return 0; 1655 1656 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1657 atomic_read(&sk->sk_rmem_alloc); 1658 1659 return unused_mem > 0 ? unused_mem : 0; 1660 } 1661 1662 static inline void sk_mem_reclaim(struct sock *sk) 1663 { 1664 int reclaimable; 1665 1666 if (!sk_has_account(sk)) 1667 return; 1668 1669 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1670 1671 if (reclaimable >= (int)PAGE_SIZE) 1672 __sk_mem_reclaim(sk, reclaimable); 1673 } 1674 1675 static inline void sk_mem_reclaim_final(struct sock *sk) 1676 { 1677 sk->sk_reserved_mem = 0; 1678 sk_mem_reclaim(sk); 1679 } 1680 1681 static inline void sk_mem_charge(struct sock *sk, int size) 1682 { 1683 if (!sk_has_account(sk)) 1684 return; 1685 sk->sk_forward_alloc -= size; 1686 } 1687 1688 static inline void sk_mem_uncharge(struct sock *sk, int size) 1689 { 1690 if (!sk_has_account(sk)) 1691 return; 1692 sk->sk_forward_alloc += size; 1693 sk_mem_reclaim(sk); 1694 } 1695 1696 /* 1697 * Macro so as to not evaluate some arguments when 1698 * lockdep is not enabled. 1699 * 1700 * Mark both the sk_lock and the sk_lock.slock as a 1701 * per-address-family lock class. 1702 */ 1703 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1704 do { \ 1705 sk->sk_lock.owned = 0; \ 1706 init_waitqueue_head(&sk->sk_lock.wq); \ 1707 spin_lock_init(&(sk)->sk_lock.slock); \ 1708 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1709 sizeof((sk)->sk_lock)); \ 1710 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1711 (skey), (sname)); \ 1712 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1713 } while (0) 1714 1715 static inline bool lockdep_sock_is_held(const struct sock *sk) 1716 { 1717 return lockdep_is_held(&sk->sk_lock) || 1718 lockdep_is_held(&sk->sk_lock.slock); 1719 } 1720 1721 void lock_sock_nested(struct sock *sk, int subclass); 1722 1723 static inline void lock_sock(struct sock *sk) 1724 { 1725 lock_sock_nested(sk, 0); 1726 } 1727 1728 void __lock_sock(struct sock *sk); 1729 void __release_sock(struct sock *sk); 1730 void release_sock(struct sock *sk); 1731 1732 /* BH context may only use the following locking interface. */ 1733 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1734 #define bh_lock_sock_nested(__sk) \ 1735 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1736 SINGLE_DEPTH_NESTING) 1737 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1738 1739 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1740 1741 /** 1742 * lock_sock_fast - fast version of lock_sock 1743 * @sk: socket 1744 * 1745 * This version should be used for very small section, where process wont block 1746 * return false if fast path is taken: 1747 * 1748 * sk_lock.slock locked, owned = 0, BH disabled 1749 * 1750 * return true if slow path is taken: 1751 * 1752 * sk_lock.slock unlocked, owned = 1, BH enabled 1753 */ 1754 static inline bool lock_sock_fast(struct sock *sk) 1755 { 1756 /* The sk_lock has mutex_lock() semantics here. */ 1757 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1758 1759 return __lock_sock_fast(sk); 1760 } 1761 1762 /* fast socket lock variant for caller already holding a [different] socket lock */ 1763 static inline bool lock_sock_fast_nested(struct sock *sk) 1764 { 1765 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1766 1767 return __lock_sock_fast(sk); 1768 } 1769 1770 /** 1771 * unlock_sock_fast - complement of lock_sock_fast 1772 * @sk: socket 1773 * @slow: slow mode 1774 * 1775 * fast unlock socket for user context. 1776 * If slow mode is on, we call regular release_sock() 1777 */ 1778 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1779 __releases(&sk->sk_lock.slock) 1780 { 1781 if (slow) { 1782 release_sock(sk); 1783 __release(&sk->sk_lock.slock); 1784 } else { 1785 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1786 spin_unlock_bh(&sk->sk_lock.slock); 1787 } 1788 } 1789 1790 void sockopt_lock_sock(struct sock *sk); 1791 void sockopt_release_sock(struct sock *sk); 1792 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1793 bool sockopt_capable(int cap); 1794 1795 /* Used by processes to "lock" a socket state, so that 1796 * interrupts and bottom half handlers won't change it 1797 * from under us. It essentially blocks any incoming 1798 * packets, so that we won't get any new data or any 1799 * packets that change the state of the socket. 1800 * 1801 * While locked, BH processing will add new packets to 1802 * the backlog queue. This queue is processed by the 1803 * owner of the socket lock right before it is released. 1804 * 1805 * Since ~2.3.5 it is also exclusive sleep lock serializing 1806 * accesses from user process context. 1807 */ 1808 1809 static inline void sock_owned_by_me(const struct sock *sk) 1810 { 1811 #ifdef CONFIG_LOCKDEP 1812 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1813 #endif 1814 } 1815 1816 static inline bool sock_owned_by_user(const struct sock *sk) 1817 { 1818 sock_owned_by_me(sk); 1819 return sk->sk_lock.owned; 1820 } 1821 1822 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1823 { 1824 return sk->sk_lock.owned; 1825 } 1826 1827 static inline void sock_release_ownership(struct sock *sk) 1828 { 1829 if (sock_owned_by_user_nocheck(sk)) { 1830 sk->sk_lock.owned = 0; 1831 1832 /* The sk_lock has mutex_unlock() semantics: */ 1833 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1834 } 1835 } 1836 1837 /* no reclassification while locks are held */ 1838 static inline bool sock_allow_reclassification(const struct sock *csk) 1839 { 1840 struct sock *sk = (struct sock *)csk; 1841 1842 return !sock_owned_by_user_nocheck(sk) && 1843 !spin_is_locked(&sk->sk_lock.slock); 1844 } 1845 1846 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1847 struct proto *prot, int kern); 1848 void sk_free(struct sock *sk); 1849 void sk_destruct(struct sock *sk); 1850 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1851 void sk_free_unlock_clone(struct sock *sk); 1852 1853 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1854 gfp_t priority); 1855 void __sock_wfree(struct sk_buff *skb); 1856 void sock_wfree(struct sk_buff *skb); 1857 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1858 gfp_t priority); 1859 void skb_orphan_partial(struct sk_buff *skb); 1860 void sock_rfree(struct sk_buff *skb); 1861 void sock_efree(struct sk_buff *skb); 1862 #ifdef CONFIG_INET 1863 void sock_edemux(struct sk_buff *skb); 1864 void sock_pfree(struct sk_buff *skb); 1865 #else 1866 #define sock_edemux sock_efree 1867 #endif 1868 1869 int sk_setsockopt(struct sock *sk, int level, int optname, 1870 sockptr_t optval, unsigned int optlen); 1871 int sock_setsockopt(struct socket *sock, int level, int op, 1872 sockptr_t optval, unsigned int optlen); 1873 1874 int sk_getsockopt(struct sock *sk, int level, int optname, 1875 sockptr_t optval, sockptr_t optlen); 1876 int sock_getsockopt(struct socket *sock, int level, int op, 1877 char __user *optval, int __user *optlen); 1878 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1879 bool timeval, bool time32); 1880 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1881 unsigned long data_len, int noblock, 1882 int *errcode, int max_page_order); 1883 1884 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1885 unsigned long size, 1886 int noblock, int *errcode) 1887 { 1888 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1889 } 1890 1891 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1892 void sock_kfree_s(struct sock *sk, void *mem, int size); 1893 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1894 void sk_send_sigurg(struct sock *sk); 1895 1896 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1897 { 1898 if (sk->sk_socket) 1899 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1900 WRITE_ONCE(sk->sk_prot, proto); 1901 } 1902 1903 struct sockcm_cookie { 1904 u64 transmit_time; 1905 u32 mark; 1906 u32 tsflags; 1907 }; 1908 1909 static inline void sockcm_init(struct sockcm_cookie *sockc, 1910 const struct sock *sk) 1911 { 1912 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1913 } 1914 1915 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1916 struct sockcm_cookie *sockc); 1917 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1918 struct sockcm_cookie *sockc); 1919 1920 /* 1921 * Functions to fill in entries in struct proto_ops when a protocol 1922 * does not implement a particular function. 1923 */ 1924 int sock_no_bind(struct socket *, struct sockaddr *, int); 1925 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1926 int sock_no_socketpair(struct socket *, struct socket *); 1927 int sock_no_accept(struct socket *, struct socket *, int, bool); 1928 int sock_no_getname(struct socket *, struct sockaddr *, int); 1929 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1930 int sock_no_listen(struct socket *, int); 1931 int sock_no_shutdown(struct socket *, int); 1932 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1933 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1934 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1935 int sock_no_mmap(struct file *file, struct socket *sock, 1936 struct vm_area_struct *vma); 1937 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1938 size_t size, int flags); 1939 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1940 int offset, size_t size, int flags); 1941 1942 /* 1943 * Functions to fill in entries in struct proto_ops when a protocol 1944 * uses the inet style. 1945 */ 1946 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1947 char __user *optval, int __user *optlen); 1948 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1949 int flags); 1950 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1951 sockptr_t optval, unsigned int optlen); 1952 1953 void sk_common_release(struct sock *sk); 1954 1955 /* 1956 * Default socket callbacks and setup code 1957 */ 1958 1959 /* Initialise core socket variables */ 1960 void sock_init_data(struct socket *sock, struct sock *sk); 1961 1962 /* 1963 * Socket reference counting postulates. 1964 * 1965 * * Each user of socket SHOULD hold a reference count. 1966 * * Each access point to socket (an hash table bucket, reference from a list, 1967 * running timer, skb in flight MUST hold a reference count. 1968 * * When reference count hits 0, it means it will never increase back. 1969 * * When reference count hits 0, it means that no references from 1970 * outside exist to this socket and current process on current CPU 1971 * is last user and may/should destroy this socket. 1972 * * sk_free is called from any context: process, BH, IRQ. When 1973 * it is called, socket has no references from outside -> sk_free 1974 * may release descendant resources allocated by the socket, but 1975 * to the time when it is called, socket is NOT referenced by any 1976 * hash tables, lists etc. 1977 * * Packets, delivered from outside (from network or from another process) 1978 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1979 * when they sit in queue. Otherwise, packets will leak to hole, when 1980 * socket is looked up by one cpu and unhasing is made by another CPU. 1981 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1982 * (leak to backlog). Packet socket does all the processing inside 1983 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1984 * use separate SMP lock, so that they are prone too. 1985 */ 1986 1987 /* Ungrab socket and destroy it, if it was the last reference. */ 1988 static inline void sock_put(struct sock *sk) 1989 { 1990 if (refcount_dec_and_test(&sk->sk_refcnt)) 1991 sk_free(sk); 1992 } 1993 /* Generic version of sock_put(), dealing with all sockets 1994 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1995 */ 1996 void sock_gen_put(struct sock *sk); 1997 1998 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1999 unsigned int trim_cap, bool refcounted); 2000 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 2001 const int nested) 2002 { 2003 return __sk_receive_skb(sk, skb, nested, 1, true); 2004 } 2005 2006 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 2007 { 2008 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2009 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2010 return; 2011 sk->sk_tx_queue_mapping = tx_queue; 2012 } 2013 2014 #define NO_QUEUE_MAPPING USHRT_MAX 2015 2016 static inline void sk_tx_queue_clear(struct sock *sk) 2017 { 2018 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 2019 } 2020 2021 static inline int sk_tx_queue_get(const struct sock *sk) 2022 { 2023 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 2024 return sk->sk_tx_queue_mapping; 2025 2026 return -1; 2027 } 2028 2029 static inline void __sk_rx_queue_set(struct sock *sk, 2030 const struct sk_buff *skb, 2031 bool force_set) 2032 { 2033 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2034 if (skb_rx_queue_recorded(skb)) { 2035 u16 rx_queue = skb_get_rx_queue(skb); 2036 2037 if (force_set || 2038 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2039 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2040 } 2041 #endif 2042 } 2043 2044 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2045 { 2046 __sk_rx_queue_set(sk, skb, true); 2047 } 2048 2049 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2050 { 2051 __sk_rx_queue_set(sk, skb, false); 2052 } 2053 2054 static inline void sk_rx_queue_clear(struct sock *sk) 2055 { 2056 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2057 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2058 #endif 2059 } 2060 2061 static inline int sk_rx_queue_get(const struct sock *sk) 2062 { 2063 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2064 if (sk) { 2065 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2066 2067 if (res != NO_QUEUE_MAPPING) 2068 return res; 2069 } 2070 #endif 2071 2072 return -1; 2073 } 2074 2075 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2076 { 2077 sk->sk_socket = sock; 2078 } 2079 2080 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2081 { 2082 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2083 return &rcu_dereference_raw(sk->sk_wq)->wait; 2084 } 2085 /* Detach socket from process context. 2086 * Announce socket dead, detach it from wait queue and inode. 2087 * Note that parent inode held reference count on this struct sock, 2088 * we do not release it in this function, because protocol 2089 * probably wants some additional cleanups or even continuing 2090 * to work with this socket (TCP). 2091 */ 2092 static inline void sock_orphan(struct sock *sk) 2093 { 2094 write_lock_bh(&sk->sk_callback_lock); 2095 sock_set_flag(sk, SOCK_DEAD); 2096 sk_set_socket(sk, NULL); 2097 sk->sk_wq = NULL; 2098 write_unlock_bh(&sk->sk_callback_lock); 2099 } 2100 2101 static inline void sock_graft(struct sock *sk, struct socket *parent) 2102 { 2103 WARN_ON(parent->sk); 2104 write_lock_bh(&sk->sk_callback_lock); 2105 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2106 parent->sk = sk; 2107 sk_set_socket(sk, parent); 2108 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2109 security_sock_graft(sk, parent); 2110 write_unlock_bh(&sk->sk_callback_lock); 2111 } 2112 2113 kuid_t sock_i_uid(struct sock *sk); 2114 unsigned long sock_i_ino(struct sock *sk); 2115 2116 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2117 { 2118 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2119 } 2120 2121 static inline u32 net_tx_rndhash(void) 2122 { 2123 u32 v = get_random_u32(); 2124 2125 return v ?: 1; 2126 } 2127 2128 static inline void sk_set_txhash(struct sock *sk) 2129 { 2130 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2131 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2132 } 2133 2134 static inline bool sk_rethink_txhash(struct sock *sk) 2135 { 2136 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2137 sk_set_txhash(sk); 2138 return true; 2139 } 2140 return false; 2141 } 2142 2143 static inline struct dst_entry * 2144 __sk_dst_get(struct sock *sk) 2145 { 2146 return rcu_dereference_check(sk->sk_dst_cache, 2147 lockdep_sock_is_held(sk)); 2148 } 2149 2150 static inline struct dst_entry * 2151 sk_dst_get(struct sock *sk) 2152 { 2153 struct dst_entry *dst; 2154 2155 rcu_read_lock(); 2156 dst = rcu_dereference(sk->sk_dst_cache); 2157 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2158 dst = NULL; 2159 rcu_read_unlock(); 2160 return dst; 2161 } 2162 2163 static inline void __dst_negative_advice(struct sock *sk) 2164 { 2165 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2166 2167 if (dst && dst->ops->negative_advice) { 2168 ndst = dst->ops->negative_advice(dst); 2169 2170 if (ndst != dst) { 2171 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2172 sk_tx_queue_clear(sk); 2173 sk->sk_dst_pending_confirm = 0; 2174 } 2175 } 2176 } 2177 2178 static inline void dst_negative_advice(struct sock *sk) 2179 { 2180 sk_rethink_txhash(sk); 2181 __dst_negative_advice(sk); 2182 } 2183 2184 static inline void 2185 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2186 { 2187 struct dst_entry *old_dst; 2188 2189 sk_tx_queue_clear(sk); 2190 sk->sk_dst_pending_confirm = 0; 2191 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2192 lockdep_sock_is_held(sk)); 2193 rcu_assign_pointer(sk->sk_dst_cache, dst); 2194 dst_release(old_dst); 2195 } 2196 2197 static inline void 2198 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2199 { 2200 struct dst_entry *old_dst; 2201 2202 sk_tx_queue_clear(sk); 2203 sk->sk_dst_pending_confirm = 0; 2204 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2205 dst_release(old_dst); 2206 } 2207 2208 static inline void 2209 __sk_dst_reset(struct sock *sk) 2210 { 2211 __sk_dst_set(sk, NULL); 2212 } 2213 2214 static inline void 2215 sk_dst_reset(struct sock *sk) 2216 { 2217 sk_dst_set(sk, NULL); 2218 } 2219 2220 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2221 2222 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2223 2224 static inline void sk_dst_confirm(struct sock *sk) 2225 { 2226 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2227 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2228 } 2229 2230 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2231 { 2232 if (skb_get_dst_pending_confirm(skb)) { 2233 struct sock *sk = skb->sk; 2234 2235 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2236 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2237 neigh_confirm(n); 2238 } 2239 } 2240 2241 bool sk_mc_loop(struct sock *sk); 2242 2243 static inline bool sk_can_gso(const struct sock *sk) 2244 { 2245 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2246 } 2247 2248 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2249 2250 static inline void sk_gso_disable(struct sock *sk) 2251 { 2252 sk->sk_gso_disabled = 1; 2253 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2254 } 2255 2256 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2257 struct iov_iter *from, char *to, 2258 int copy, int offset) 2259 { 2260 if (skb->ip_summed == CHECKSUM_NONE) { 2261 __wsum csum = 0; 2262 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2263 return -EFAULT; 2264 skb->csum = csum_block_add(skb->csum, csum, offset); 2265 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2266 if (!copy_from_iter_full_nocache(to, copy, from)) 2267 return -EFAULT; 2268 } else if (!copy_from_iter_full(to, copy, from)) 2269 return -EFAULT; 2270 2271 return 0; 2272 } 2273 2274 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2275 struct iov_iter *from, int copy) 2276 { 2277 int err, offset = skb->len; 2278 2279 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2280 copy, offset); 2281 if (err) 2282 __skb_trim(skb, offset); 2283 2284 return err; 2285 } 2286 2287 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2288 struct sk_buff *skb, 2289 struct page *page, 2290 int off, int copy) 2291 { 2292 int err; 2293 2294 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2295 copy, skb->len); 2296 if (err) 2297 return err; 2298 2299 skb_len_add(skb, copy); 2300 sk_wmem_queued_add(sk, copy); 2301 sk_mem_charge(sk, copy); 2302 return 0; 2303 } 2304 2305 /** 2306 * sk_wmem_alloc_get - returns write allocations 2307 * @sk: socket 2308 * 2309 * Return: sk_wmem_alloc minus initial offset of one 2310 */ 2311 static inline int sk_wmem_alloc_get(const struct sock *sk) 2312 { 2313 return refcount_read(&sk->sk_wmem_alloc) - 1; 2314 } 2315 2316 /** 2317 * sk_rmem_alloc_get - returns read allocations 2318 * @sk: socket 2319 * 2320 * Return: sk_rmem_alloc 2321 */ 2322 static inline int sk_rmem_alloc_get(const struct sock *sk) 2323 { 2324 return atomic_read(&sk->sk_rmem_alloc); 2325 } 2326 2327 /** 2328 * sk_has_allocations - check if allocations are outstanding 2329 * @sk: socket 2330 * 2331 * Return: true if socket has write or read allocations 2332 */ 2333 static inline bool sk_has_allocations(const struct sock *sk) 2334 { 2335 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2336 } 2337 2338 /** 2339 * skwq_has_sleeper - check if there are any waiting processes 2340 * @wq: struct socket_wq 2341 * 2342 * Return: true if socket_wq has waiting processes 2343 * 2344 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2345 * barrier call. They were added due to the race found within the tcp code. 2346 * 2347 * Consider following tcp code paths:: 2348 * 2349 * CPU1 CPU2 2350 * sys_select receive packet 2351 * ... ... 2352 * __add_wait_queue update tp->rcv_nxt 2353 * ... ... 2354 * tp->rcv_nxt check sock_def_readable 2355 * ... { 2356 * schedule rcu_read_lock(); 2357 * wq = rcu_dereference(sk->sk_wq); 2358 * if (wq && waitqueue_active(&wq->wait)) 2359 * wake_up_interruptible(&wq->wait) 2360 * ... 2361 * } 2362 * 2363 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2364 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2365 * could then endup calling schedule and sleep forever if there are no more 2366 * data on the socket. 2367 * 2368 */ 2369 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2370 { 2371 return wq && wq_has_sleeper(&wq->wait); 2372 } 2373 2374 /** 2375 * sock_poll_wait - place memory barrier behind the poll_wait call. 2376 * @filp: file 2377 * @sock: socket to wait on 2378 * @p: poll_table 2379 * 2380 * See the comments in the wq_has_sleeper function. 2381 */ 2382 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2383 poll_table *p) 2384 { 2385 if (!poll_does_not_wait(p)) { 2386 poll_wait(filp, &sock->wq.wait, p); 2387 /* We need to be sure we are in sync with the 2388 * socket flags modification. 2389 * 2390 * This memory barrier is paired in the wq_has_sleeper. 2391 */ 2392 smp_mb(); 2393 } 2394 } 2395 2396 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2397 { 2398 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2399 u32 txhash = READ_ONCE(sk->sk_txhash); 2400 2401 if (txhash) { 2402 skb->l4_hash = 1; 2403 skb->hash = txhash; 2404 } 2405 } 2406 2407 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2408 2409 /* 2410 * Queue a received datagram if it will fit. Stream and sequenced 2411 * protocols can't normally use this as they need to fit buffers in 2412 * and play with them. 2413 * 2414 * Inlined as it's very short and called for pretty much every 2415 * packet ever received. 2416 */ 2417 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2418 { 2419 skb_orphan(skb); 2420 skb->sk = sk; 2421 skb->destructor = sock_rfree; 2422 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2423 sk_mem_charge(sk, skb->truesize); 2424 } 2425 2426 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2427 { 2428 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2429 skb_orphan(skb); 2430 skb->destructor = sock_efree; 2431 skb->sk = sk; 2432 return true; 2433 } 2434 return false; 2435 } 2436 2437 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2438 { 2439 if (skb->destructor != sock_wfree) { 2440 skb_orphan(skb); 2441 return; 2442 } 2443 skb->slow_gro = 1; 2444 } 2445 2446 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2447 unsigned long expires); 2448 2449 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2450 2451 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2452 2453 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2454 struct sk_buff *skb, unsigned int flags, 2455 void (*destructor)(struct sock *sk, 2456 struct sk_buff *skb)); 2457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2458 2459 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2460 enum skb_drop_reason *reason); 2461 2462 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2463 { 2464 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2465 } 2466 2467 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2468 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2469 2470 /* 2471 * Recover an error report and clear atomically 2472 */ 2473 2474 static inline int sock_error(struct sock *sk) 2475 { 2476 int err; 2477 2478 /* Avoid an atomic operation for the common case. 2479 * This is racy since another cpu/thread can change sk_err under us. 2480 */ 2481 if (likely(data_race(!sk->sk_err))) 2482 return 0; 2483 2484 err = xchg(&sk->sk_err, 0); 2485 return -err; 2486 } 2487 2488 void sk_error_report(struct sock *sk); 2489 2490 static inline unsigned long sock_wspace(struct sock *sk) 2491 { 2492 int amt = 0; 2493 2494 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2495 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2496 if (amt < 0) 2497 amt = 0; 2498 } 2499 return amt; 2500 } 2501 2502 /* Note: 2503 * We use sk->sk_wq_raw, from contexts knowing this 2504 * pointer is not NULL and cannot disappear/change. 2505 */ 2506 static inline void sk_set_bit(int nr, struct sock *sk) 2507 { 2508 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2509 !sock_flag(sk, SOCK_FASYNC)) 2510 return; 2511 2512 set_bit(nr, &sk->sk_wq_raw->flags); 2513 } 2514 2515 static inline void sk_clear_bit(int nr, struct sock *sk) 2516 { 2517 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2518 !sock_flag(sk, SOCK_FASYNC)) 2519 return; 2520 2521 clear_bit(nr, &sk->sk_wq_raw->flags); 2522 } 2523 2524 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2525 { 2526 if (sock_flag(sk, SOCK_FASYNC)) { 2527 rcu_read_lock(); 2528 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2529 rcu_read_unlock(); 2530 } 2531 } 2532 2533 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2534 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2535 * Note: for send buffers, TCP works better if we can build two skbs at 2536 * minimum. 2537 */ 2538 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2539 2540 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2541 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2542 2543 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2544 { 2545 u32 val; 2546 2547 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2548 return; 2549 2550 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2551 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2552 2553 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2554 } 2555 2556 /** 2557 * sk_page_frag - return an appropriate page_frag 2558 * @sk: socket 2559 * 2560 * Use the per task page_frag instead of the per socket one for 2561 * optimization when we know that we're in process context and own 2562 * everything that's associated with %current. 2563 * 2564 * Both direct reclaim and page faults can nest inside other 2565 * socket operations and end up recursing into sk_page_frag() 2566 * while it's already in use: explicitly avoid task page_frag 2567 * when users disable sk_use_task_frag. 2568 * 2569 * Return: a per task page_frag if context allows that, 2570 * otherwise a per socket one. 2571 */ 2572 static inline struct page_frag *sk_page_frag(struct sock *sk) 2573 { 2574 if (sk->sk_use_task_frag) 2575 return ¤t->task_frag; 2576 2577 return &sk->sk_frag; 2578 } 2579 2580 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2581 2582 /* 2583 * Default write policy as shown to user space via poll/select/SIGIO 2584 */ 2585 static inline bool sock_writeable(const struct sock *sk) 2586 { 2587 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2588 } 2589 2590 static inline gfp_t gfp_any(void) 2591 { 2592 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2593 } 2594 2595 static inline gfp_t gfp_memcg_charge(void) 2596 { 2597 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2598 } 2599 2600 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2601 { 2602 return noblock ? 0 : sk->sk_rcvtimeo; 2603 } 2604 2605 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2606 { 2607 return noblock ? 0 : sk->sk_sndtimeo; 2608 } 2609 2610 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2611 { 2612 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2613 2614 return v ?: 1; 2615 } 2616 2617 /* Alas, with timeout socket operations are not restartable. 2618 * Compare this to poll(). 2619 */ 2620 static inline int sock_intr_errno(long timeo) 2621 { 2622 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2623 } 2624 2625 struct sock_skb_cb { 2626 u32 dropcount; 2627 }; 2628 2629 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2630 * using skb->cb[] would keep using it directly and utilize its 2631 * alignement guarantee. 2632 */ 2633 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2634 sizeof(struct sock_skb_cb))) 2635 2636 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2637 SOCK_SKB_CB_OFFSET)) 2638 2639 #define sock_skb_cb_check_size(size) \ 2640 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2641 2642 static inline void 2643 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2644 { 2645 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2646 atomic_read(&sk->sk_drops) : 0; 2647 } 2648 2649 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2650 { 2651 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2652 2653 atomic_add(segs, &sk->sk_drops); 2654 } 2655 2656 static inline ktime_t sock_read_timestamp(struct sock *sk) 2657 { 2658 #if BITS_PER_LONG==32 2659 unsigned int seq; 2660 ktime_t kt; 2661 2662 do { 2663 seq = read_seqbegin(&sk->sk_stamp_seq); 2664 kt = sk->sk_stamp; 2665 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2666 2667 return kt; 2668 #else 2669 return READ_ONCE(sk->sk_stamp); 2670 #endif 2671 } 2672 2673 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2674 { 2675 #if BITS_PER_LONG==32 2676 write_seqlock(&sk->sk_stamp_seq); 2677 sk->sk_stamp = kt; 2678 write_sequnlock(&sk->sk_stamp_seq); 2679 #else 2680 WRITE_ONCE(sk->sk_stamp, kt); 2681 #endif 2682 } 2683 2684 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2685 struct sk_buff *skb); 2686 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2687 struct sk_buff *skb); 2688 2689 static inline void 2690 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2691 { 2692 ktime_t kt = skb->tstamp; 2693 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2694 2695 /* 2696 * generate control messages if 2697 * - receive time stamping in software requested 2698 * - software time stamp available and wanted 2699 * - hardware time stamps available and wanted 2700 */ 2701 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2702 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2703 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2704 (hwtstamps->hwtstamp && 2705 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2706 __sock_recv_timestamp(msg, sk, skb); 2707 else 2708 sock_write_timestamp(sk, kt); 2709 2710 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2711 __sock_recv_wifi_status(msg, sk, skb); 2712 } 2713 2714 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2715 struct sk_buff *skb); 2716 2717 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2718 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2719 struct sk_buff *skb) 2720 { 2721 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2722 (1UL << SOCK_RCVTSTAMP) | \ 2723 (1UL << SOCK_RCVMARK)) 2724 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2725 SOF_TIMESTAMPING_RAW_HARDWARE) 2726 2727 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2728 __sock_recv_cmsgs(msg, sk, skb); 2729 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2730 sock_write_timestamp(sk, skb->tstamp); 2731 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2732 sock_write_timestamp(sk, 0); 2733 } 2734 2735 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2736 2737 /** 2738 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2739 * @sk: socket sending this packet 2740 * @tsflags: timestamping flags to use 2741 * @tx_flags: completed with instructions for time stamping 2742 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2743 * 2744 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2745 */ 2746 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2747 __u8 *tx_flags, __u32 *tskey) 2748 { 2749 if (unlikely(tsflags)) { 2750 __sock_tx_timestamp(tsflags, tx_flags); 2751 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2752 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2753 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2754 } 2755 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2756 *tx_flags |= SKBTX_WIFI_STATUS; 2757 } 2758 2759 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2760 __u8 *tx_flags) 2761 { 2762 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2763 } 2764 2765 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2766 { 2767 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2768 &skb_shinfo(skb)->tskey); 2769 } 2770 2771 static inline bool sk_is_tcp(const struct sock *sk) 2772 { 2773 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2774 } 2775 2776 /** 2777 * sk_eat_skb - Release a skb if it is no longer needed 2778 * @sk: socket to eat this skb from 2779 * @skb: socket buffer to eat 2780 * 2781 * This routine must be called with interrupts disabled or with the socket 2782 * locked so that the sk_buff queue operation is ok. 2783 */ 2784 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2785 { 2786 __skb_unlink(skb, &sk->sk_receive_queue); 2787 __kfree_skb(skb); 2788 } 2789 2790 static inline bool 2791 skb_sk_is_prefetched(struct sk_buff *skb) 2792 { 2793 #ifdef CONFIG_INET 2794 return skb->destructor == sock_pfree; 2795 #else 2796 return false; 2797 #endif /* CONFIG_INET */ 2798 } 2799 2800 /* This helper checks if a socket is a full socket, 2801 * ie _not_ a timewait or request socket. 2802 */ 2803 static inline bool sk_fullsock(const struct sock *sk) 2804 { 2805 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2806 } 2807 2808 static inline bool 2809 sk_is_refcounted(struct sock *sk) 2810 { 2811 /* Only full sockets have sk->sk_flags. */ 2812 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2813 } 2814 2815 /** 2816 * skb_steal_sock - steal a socket from an sk_buff 2817 * @skb: sk_buff to steal the socket from 2818 * @refcounted: is set to true if the socket is reference-counted 2819 */ 2820 static inline struct sock * 2821 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2822 { 2823 if (skb->sk) { 2824 struct sock *sk = skb->sk; 2825 2826 *refcounted = true; 2827 if (skb_sk_is_prefetched(skb)) 2828 *refcounted = sk_is_refcounted(sk); 2829 skb->destructor = NULL; 2830 skb->sk = NULL; 2831 return sk; 2832 } 2833 *refcounted = false; 2834 return NULL; 2835 } 2836 2837 /* Checks if this SKB belongs to an HW offloaded socket 2838 * and whether any SW fallbacks are required based on dev. 2839 * Check decrypted mark in case skb_orphan() cleared socket. 2840 */ 2841 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2842 struct net_device *dev) 2843 { 2844 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2845 struct sock *sk = skb->sk; 2846 2847 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2848 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2849 #ifdef CONFIG_TLS_DEVICE 2850 } else if (unlikely(skb->decrypted)) { 2851 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2852 kfree_skb(skb); 2853 skb = NULL; 2854 #endif 2855 } 2856 #endif 2857 2858 return skb; 2859 } 2860 2861 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2862 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2863 */ 2864 static inline bool sk_listener(const struct sock *sk) 2865 { 2866 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2867 } 2868 2869 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2870 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2871 int type); 2872 2873 bool sk_ns_capable(const struct sock *sk, 2874 struct user_namespace *user_ns, int cap); 2875 bool sk_capable(const struct sock *sk, int cap); 2876 bool sk_net_capable(const struct sock *sk, int cap); 2877 2878 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2879 2880 /* Take into consideration the size of the struct sk_buff overhead in the 2881 * determination of these values, since that is non-constant across 2882 * platforms. This makes socket queueing behavior and performance 2883 * not depend upon such differences. 2884 */ 2885 #define _SK_MEM_PACKETS 256 2886 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2887 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2888 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2889 2890 extern __u32 sysctl_wmem_max; 2891 extern __u32 sysctl_rmem_max; 2892 2893 extern int sysctl_tstamp_allow_data; 2894 extern int sysctl_optmem_max; 2895 2896 extern __u32 sysctl_wmem_default; 2897 extern __u32 sysctl_rmem_default; 2898 2899 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2900 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2901 2902 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2903 { 2904 /* Does this proto have per netns sysctl_wmem ? */ 2905 if (proto->sysctl_wmem_offset) 2906 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2907 2908 return READ_ONCE(*proto->sysctl_wmem); 2909 } 2910 2911 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2912 { 2913 /* Does this proto have per netns sysctl_rmem ? */ 2914 if (proto->sysctl_rmem_offset) 2915 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2916 2917 return READ_ONCE(*proto->sysctl_rmem); 2918 } 2919 2920 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2921 * Some wifi drivers need to tweak it to get more chunks. 2922 * They can use this helper from their ndo_start_xmit() 2923 */ 2924 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2925 { 2926 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2927 return; 2928 WRITE_ONCE(sk->sk_pacing_shift, val); 2929 } 2930 2931 /* if a socket is bound to a device, check that the given device 2932 * index is either the same or that the socket is bound to an L3 2933 * master device and the given device index is also enslaved to 2934 * that L3 master 2935 */ 2936 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2937 { 2938 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2939 int mdif; 2940 2941 if (!bound_dev_if || bound_dev_if == dif) 2942 return true; 2943 2944 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2945 if (mdif && mdif == bound_dev_if) 2946 return true; 2947 2948 return false; 2949 } 2950 2951 void sock_def_readable(struct sock *sk); 2952 2953 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2954 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2955 int sock_set_timestamping(struct sock *sk, int optname, 2956 struct so_timestamping timestamping); 2957 2958 void sock_enable_timestamps(struct sock *sk); 2959 void sock_no_linger(struct sock *sk); 2960 void sock_set_keepalive(struct sock *sk); 2961 void sock_set_priority(struct sock *sk, u32 priority); 2962 void sock_set_rcvbuf(struct sock *sk, int val); 2963 void sock_set_mark(struct sock *sk, u32 val); 2964 void sock_set_reuseaddr(struct sock *sk); 2965 void sock_set_reuseport(struct sock *sk); 2966 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2967 2968 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2969 2970 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2971 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2972 sockptr_t optval, int optlen, bool old_timeval); 2973 2974 static inline bool sk_is_readable(struct sock *sk) 2975 { 2976 if (sk->sk_prot->sock_is_readable) 2977 return sk->sk_prot->sock_is_readable(sk); 2978 return false; 2979 } 2980 #endif /* _SOCK_H */ 2981