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