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_nulls_node_init(struct hlist_nulls_node *node) 748 { 749 node->pprev = NULL; 750 } 751 752 static inline void __sk_del_node(struct sock *sk) 753 { 754 __hlist_del(&sk->sk_node); 755 } 756 757 /* NB: equivalent to hlist_del_init_rcu */ 758 static inline bool __sk_del_node_init(struct sock *sk) 759 { 760 if (sk_hashed(sk)) { 761 __sk_del_node(sk); 762 sk_node_init(&sk->sk_node); 763 return true; 764 } 765 return false; 766 } 767 768 /* Grab socket reference count. This operation is valid only 769 when sk is ALREADY grabbed f.e. it is found in hash table 770 or a list and the lookup is made under lock preventing hash table 771 modifications. 772 */ 773 774 static __always_inline void sock_hold(struct sock *sk) 775 { 776 refcount_inc(&sk->sk_refcnt); 777 } 778 779 /* Ungrab socket in the context, which assumes that socket refcnt 780 cannot hit zero, f.e. it is true in context of any socketcall. 781 */ 782 static __always_inline void __sock_put(struct sock *sk) 783 { 784 refcount_dec(&sk->sk_refcnt); 785 } 786 787 static inline bool sk_del_node_init(struct sock *sk) 788 { 789 bool rc = __sk_del_node_init(sk); 790 791 if (rc) { 792 /* paranoid for a while -acme */ 793 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 794 __sock_put(sk); 795 } 796 return rc; 797 } 798 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 799 800 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 801 { 802 if (sk_hashed(sk)) { 803 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 804 return true; 805 } 806 return false; 807 } 808 809 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 810 { 811 bool rc = __sk_nulls_del_node_init_rcu(sk); 812 813 if (rc) { 814 /* paranoid for a while -acme */ 815 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 816 __sock_put(sk); 817 } 818 return rc; 819 } 820 821 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 822 { 823 hlist_add_head(&sk->sk_node, list); 824 } 825 826 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 827 { 828 sock_hold(sk); 829 __sk_add_node(sk, list); 830 } 831 832 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 833 { 834 sock_hold(sk); 835 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 836 sk->sk_family == AF_INET6) 837 hlist_add_tail_rcu(&sk->sk_node, list); 838 else 839 hlist_add_head_rcu(&sk->sk_node, list); 840 } 841 842 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 843 { 844 sock_hold(sk); 845 hlist_add_tail_rcu(&sk->sk_node, list); 846 } 847 848 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 849 { 850 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 851 } 852 853 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 854 { 855 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 856 } 857 858 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 859 { 860 sock_hold(sk); 861 __sk_nulls_add_node_rcu(sk, list); 862 } 863 864 static inline void __sk_del_bind_node(struct sock *sk) 865 { 866 __hlist_del(&sk->sk_bind_node); 867 } 868 869 static inline void sk_add_bind_node(struct sock *sk, 870 struct hlist_head *list) 871 { 872 hlist_add_head(&sk->sk_bind_node, list); 873 } 874 875 static inline void __sk_del_bind2_node(struct sock *sk) 876 { 877 __hlist_del(&sk->sk_bind2_node); 878 } 879 880 static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list) 881 { 882 hlist_add_head(&sk->sk_bind2_node, list); 883 } 884 885 #define sk_for_each(__sk, list) \ 886 hlist_for_each_entry(__sk, list, sk_node) 887 #define sk_for_each_rcu(__sk, list) \ 888 hlist_for_each_entry_rcu(__sk, list, sk_node) 889 #define sk_nulls_for_each(__sk, node, list) \ 890 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 891 #define sk_nulls_for_each_rcu(__sk, node, list) \ 892 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 893 #define sk_for_each_from(__sk) \ 894 hlist_for_each_entry_from(__sk, sk_node) 895 #define sk_nulls_for_each_from(__sk, node) \ 896 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 897 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 898 #define sk_for_each_safe(__sk, tmp, list) \ 899 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 900 #define sk_for_each_bound(__sk, list) \ 901 hlist_for_each_entry(__sk, list, sk_bind_node) 902 #define sk_for_each_bound_bhash2(__sk, list) \ 903 hlist_for_each_entry(__sk, list, sk_bind2_node) 904 905 /** 906 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 907 * @tpos: the type * to use as a loop cursor. 908 * @pos: the &struct hlist_node to use as a loop cursor. 909 * @head: the head for your list. 910 * @offset: offset of hlist_node within the struct. 911 * 912 */ 913 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 914 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 915 pos != NULL && \ 916 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 917 pos = rcu_dereference(hlist_next_rcu(pos))) 918 919 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 920 { 921 /* Careful only use this in a context where these parameters 922 * can not change and must all be valid, such as recvmsg from 923 * userspace. 924 */ 925 return sk->sk_socket->file->f_cred->user_ns; 926 } 927 928 /* Sock flags */ 929 enum sock_flags { 930 SOCK_DEAD, 931 SOCK_DONE, 932 SOCK_URGINLINE, 933 SOCK_KEEPOPEN, 934 SOCK_LINGER, 935 SOCK_DESTROY, 936 SOCK_BROADCAST, 937 SOCK_TIMESTAMP, 938 SOCK_ZAPPED, 939 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 940 SOCK_DBG, /* %SO_DEBUG setting */ 941 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 942 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 943 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 944 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 945 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 946 SOCK_FASYNC, /* fasync() active */ 947 SOCK_RXQ_OVFL, 948 SOCK_ZEROCOPY, /* buffers from userspace */ 949 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 950 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 951 * Will use last 4 bytes of packet sent from 952 * user-space instead. 953 */ 954 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 955 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 956 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 957 SOCK_TXTIME, 958 SOCK_XDP, /* XDP is attached */ 959 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 960 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 961 }; 962 963 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 964 965 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 966 { 967 nsk->sk_flags = osk->sk_flags; 968 } 969 970 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 971 { 972 __set_bit(flag, &sk->sk_flags); 973 } 974 975 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 976 { 977 __clear_bit(flag, &sk->sk_flags); 978 } 979 980 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 981 int valbool) 982 { 983 if (valbool) 984 sock_set_flag(sk, bit); 985 else 986 sock_reset_flag(sk, bit); 987 } 988 989 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 990 { 991 return test_bit(flag, &sk->sk_flags); 992 } 993 994 #ifdef CONFIG_NET 995 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 996 static inline int sk_memalloc_socks(void) 997 { 998 return static_branch_unlikely(&memalloc_socks_key); 999 } 1000 1001 void __receive_sock(struct file *file); 1002 #else 1003 1004 static inline int sk_memalloc_socks(void) 1005 { 1006 return 0; 1007 } 1008 1009 static inline void __receive_sock(struct file *file) 1010 { } 1011 #endif 1012 1013 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1014 { 1015 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1016 } 1017 1018 static inline void sk_acceptq_removed(struct sock *sk) 1019 { 1020 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1021 } 1022 1023 static inline void sk_acceptq_added(struct sock *sk) 1024 { 1025 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1026 } 1027 1028 /* Note: If you think the test should be: 1029 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1030 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1031 */ 1032 static inline bool sk_acceptq_is_full(const struct sock *sk) 1033 { 1034 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1035 } 1036 1037 /* 1038 * Compute minimal free write space needed to queue new packets. 1039 */ 1040 static inline int sk_stream_min_wspace(const struct sock *sk) 1041 { 1042 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1043 } 1044 1045 static inline int sk_stream_wspace(const struct sock *sk) 1046 { 1047 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1048 } 1049 1050 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1051 { 1052 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1053 } 1054 1055 void sk_stream_write_space(struct sock *sk); 1056 1057 /* OOB backlog add */ 1058 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1059 { 1060 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1061 skb_dst_force(skb); 1062 1063 if (!sk->sk_backlog.tail) 1064 WRITE_ONCE(sk->sk_backlog.head, skb); 1065 else 1066 sk->sk_backlog.tail->next = skb; 1067 1068 WRITE_ONCE(sk->sk_backlog.tail, skb); 1069 skb->next = NULL; 1070 } 1071 1072 /* 1073 * Take into account size of receive queue and backlog queue 1074 * Do not take into account this skb truesize, 1075 * to allow even a single big packet to come. 1076 */ 1077 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1078 { 1079 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1080 1081 return qsize > limit; 1082 } 1083 1084 /* The per-socket spinlock must be held here. */ 1085 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1086 unsigned int limit) 1087 { 1088 if (sk_rcvqueues_full(sk, limit)) 1089 return -ENOBUFS; 1090 1091 /* 1092 * If the skb was allocated from pfmemalloc reserves, only 1093 * allow SOCK_MEMALLOC sockets to use it as this socket is 1094 * helping free memory 1095 */ 1096 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1097 return -ENOMEM; 1098 1099 __sk_add_backlog(sk, skb); 1100 sk->sk_backlog.len += skb->truesize; 1101 return 0; 1102 } 1103 1104 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1105 1106 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1107 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1108 1109 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1110 { 1111 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1112 return __sk_backlog_rcv(sk, skb); 1113 1114 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1115 tcp_v6_do_rcv, 1116 tcp_v4_do_rcv, 1117 sk, skb); 1118 } 1119 1120 static inline void sk_incoming_cpu_update(struct sock *sk) 1121 { 1122 int cpu = raw_smp_processor_id(); 1123 1124 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1125 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1126 } 1127 1128 static inline void sock_rps_record_flow_hash(__u32 hash) 1129 { 1130 #ifdef CONFIG_RPS 1131 struct rps_sock_flow_table *sock_flow_table; 1132 1133 rcu_read_lock(); 1134 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1135 rps_record_sock_flow(sock_flow_table, hash); 1136 rcu_read_unlock(); 1137 #endif 1138 } 1139 1140 static inline void sock_rps_record_flow(const struct sock *sk) 1141 { 1142 #ifdef CONFIG_RPS 1143 if (static_branch_unlikely(&rfs_needed)) { 1144 /* Reading sk->sk_rxhash might incur an expensive cache line 1145 * miss. 1146 * 1147 * TCP_ESTABLISHED does cover almost all states where RFS 1148 * might be useful, and is cheaper [1] than testing : 1149 * IPv4: inet_sk(sk)->inet_daddr 1150 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1151 * OR an additional socket flag 1152 * [1] : sk_state and sk_prot are in the same cache line. 1153 */ 1154 if (sk->sk_state == TCP_ESTABLISHED) 1155 sock_rps_record_flow_hash(sk->sk_rxhash); 1156 } 1157 #endif 1158 } 1159 1160 static inline void sock_rps_save_rxhash(struct sock *sk, 1161 const struct sk_buff *skb) 1162 { 1163 #ifdef CONFIG_RPS 1164 if (unlikely(sk->sk_rxhash != skb->hash)) 1165 sk->sk_rxhash = skb->hash; 1166 #endif 1167 } 1168 1169 static inline void sock_rps_reset_rxhash(struct sock *sk) 1170 { 1171 #ifdef CONFIG_RPS 1172 sk->sk_rxhash = 0; 1173 #endif 1174 } 1175 1176 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1177 ({ int __rc; \ 1178 release_sock(__sk); \ 1179 __rc = __condition; \ 1180 if (!__rc) { \ 1181 *(__timeo) = wait_woken(__wait, \ 1182 TASK_INTERRUPTIBLE, \ 1183 *(__timeo)); \ 1184 } \ 1185 sched_annotate_sleep(); \ 1186 lock_sock(__sk); \ 1187 __rc = __condition; \ 1188 __rc; \ 1189 }) 1190 1191 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1192 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1193 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1194 int sk_stream_error(struct sock *sk, int flags, int err); 1195 void sk_stream_kill_queues(struct sock *sk); 1196 void sk_set_memalloc(struct sock *sk); 1197 void sk_clear_memalloc(struct sock *sk); 1198 1199 void __sk_flush_backlog(struct sock *sk); 1200 1201 static inline bool sk_flush_backlog(struct sock *sk) 1202 { 1203 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1204 __sk_flush_backlog(sk); 1205 return true; 1206 } 1207 return false; 1208 } 1209 1210 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1211 1212 struct request_sock_ops; 1213 struct timewait_sock_ops; 1214 struct inet_hashinfo; 1215 struct raw_hashinfo; 1216 struct smc_hashinfo; 1217 struct module; 1218 struct sk_psock; 1219 1220 /* 1221 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1222 * un-modified. Special care is taken when initializing object to zero. 1223 */ 1224 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1225 { 1226 if (offsetof(struct sock, sk_node.next) != 0) 1227 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1228 memset(&sk->sk_node.pprev, 0, 1229 size - offsetof(struct sock, sk_node.pprev)); 1230 } 1231 1232 /* Networking protocol blocks we attach to sockets. 1233 * socket layer -> transport layer interface 1234 */ 1235 struct proto { 1236 void (*close)(struct sock *sk, 1237 long timeout); 1238 int (*pre_connect)(struct sock *sk, 1239 struct sockaddr *uaddr, 1240 int addr_len); 1241 int (*connect)(struct sock *sk, 1242 struct sockaddr *uaddr, 1243 int addr_len); 1244 int (*disconnect)(struct sock *sk, int flags); 1245 1246 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1247 bool kern); 1248 1249 int (*ioctl)(struct sock *sk, int cmd, 1250 unsigned long arg); 1251 int (*init)(struct sock *sk); 1252 void (*destroy)(struct sock *sk); 1253 void (*shutdown)(struct sock *sk, int how); 1254 int (*setsockopt)(struct sock *sk, int level, 1255 int optname, sockptr_t optval, 1256 unsigned int optlen); 1257 int (*getsockopt)(struct sock *sk, int level, 1258 int optname, char __user *optval, 1259 int __user *option); 1260 void (*keepalive)(struct sock *sk, int valbool); 1261 #ifdef CONFIG_COMPAT 1262 int (*compat_ioctl)(struct sock *sk, 1263 unsigned int cmd, unsigned long arg); 1264 #endif 1265 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1266 size_t len); 1267 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1268 size_t len, int flags, int *addr_len); 1269 int (*sendpage)(struct sock *sk, struct page *page, 1270 int offset, size_t size, int flags); 1271 int (*bind)(struct sock *sk, 1272 struct sockaddr *addr, int addr_len); 1273 int (*bind_add)(struct sock *sk, 1274 struct sockaddr *addr, int addr_len); 1275 1276 int (*backlog_rcv) (struct sock *sk, 1277 struct sk_buff *skb); 1278 bool (*bpf_bypass_getsockopt)(int level, 1279 int optname); 1280 1281 void (*release_cb)(struct sock *sk); 1282 1283 /* Keeping track of sk's, looking them up, and port selection methods. */ 1284 int (*hash)(struct sock *sk); 1285 void (*unhash)(struct sock *sk); 1286 void (*rehash)(struct sock *sk); 1287 int (*get_port)(struct sock *sk, unsigned short snum); 1288 void (*put_port)(struct sock *sk); 1289 #ifdef CONFIG_BPF_SYSCALL 1290 int (*psock_update_sk_prot)(struct sock *sk, 1291 struct sk_psock *psock, 1292 bool restore); 1293 #endif 1294 1295 /* Keeping track of sockets in use */ 1296 #ifdef CONFIG_PROC_FS 1297 unsigned int inuse_idx; 1298 #endif 1299 1300 #if IS_ENABLED(CONFIG_MPTCP) 1301 int (*forward_alloc_get)(const struct sock *sk); 1302 #endif 1303 1304 bool (*stream_memory_free)(const struct sock *sk, int wake); 1305 bool (*sock_is_readable)(struct sock *sk); 1306 /* Memory pressure */ 1307 void (*enter_memory_pressure)(struct sock *sk); 1308 void (*leave_memory_pressure)(struct sock *sk); 1309 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1310 int __percpu *per_cpu_fw_alloc; 1311 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1312 1313 /* 1314 * Pressure flag: try to collapse. 1315 * Technical note: it is used by multiple contexts non atomically. 1316 * All the __sk_mem_schedule() is of this nature: accounting 1317 * is strict, actions are advisory and have some latency. 1318 */ 1319 unsigned long *memory_pressure; 1320 long *sysctl_mem; 1321 1322 int *sysctl_wmem; 1323 int *sysctl_rmem; 1324 u32 sysctl_wmem_offset; 1325 u32 sysctl_rmem_offset; 1326 1327 int max_header; 1328 bool no_autobind; 1329 1330 struct kmem_cache *slab; 1331 unsigned int obj_size; 1332 slab_flags_t slab_flags; 1333 unsigned int useroffset; /* Usercopy region offset */ 1334 unsigned int usersize; /* Usercopy region size */ 1335 1336 unsigned int __percpu *orphan_count; 1337 1338 struct request_sock_ops *rsk_prot; 1339 struct timewait_sock_ops *twsk_prot; 1340 1341 union { 1342 struct inet_hashinfo *hashinfo; 1343 struct udp_table *udp_table; 1344 struct raw_hashinfo *raw_hash; 1345 struct smc_hashinfo *smc_hash; 1346 } h; 1347 1348 struct module *owner; 1349 1350 char name[32]; 1351 1352 struct list_head node; 1353 #ifdef SOCK_REFCNT_DEBUG 1354 atomic_t socks; 1355 #endif 1356 int (*diag_destroy)(struct sock *sk, int err); 1357 } __randomize_layout; 1358 1359 int proto_register(struct proto *prot, int alloc_slab); 1360 void proto_unregister(struct proto *prot); 1361 int sock_load_diag_module(int family, int protocol); 1362 1363 #ifdef SOCK_REFCNT_DEBUG 1364 static inline void sk_refcnt_debug_inc(struct sock *sk) 1365 { 1366 atomic_inc(&sk->sk_prot->socks); 1367 } 1368 1369 static inline void sk_refcnt_debug_dec(struct sock *sk) 1370 { 1371 atomic_dec(&sk->sk_prot->socks); 1372 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n", 1373 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks)); 1374 } 1375 1376 static inline void sk_refcnt_debug_release(const struct sock *sk) 1377 { 1378 if (refcount_read(&sk->sk_refcnt) != 1) 1379 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n", 1380 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt)); 1381 } 1382 #else /* SOCK_REFCNT_DEBUG */ 1383 #define sk_refcnt_debug_inc(sk) do { } while (0) 1384 #define sk_refcnt_debug_dec(sk) do { } while (0) 1385 #define sk_refcnt_debug_release(sk) do { } while (0) 1386 #endif /* SOCK_REFCNT_DEBUG */ 1387 1388 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1389 1390 static inline int sk_forward_alloc_get(const struct sock *sk) 1391 { 1392 #if IS_ENABLED(CONFIG_MPTCP) 1393 if (sk->sk_prot->forward_alloc_get) 1394 return sk->sk_prot->forward_alloc_get(sk); 1395 #endif 1396 return sk->sk_forward_alloc; 1397 } 1398 1399 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1400 { 1401 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1402 return false; 1403 1404 return sk->sk_prot->stream_memory_free ? 1405 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1406 tcp_stream_memory_free, sk, wake) : true; 1407 } 1408 1409 static inline bool sk_stream_memory_free(const struct sock *sk) 1410 { 1411 return __sk_stream_memory_free(sk, 0); 1412 } 1413 1414 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1415 { 1416 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1417 __sk_stream_memory_free(sk, wake); 1418 } 1419 1420 static inline bool sk_stream_is_writeable(const struct sock *sk) 1421 { 1422 return __sk_stream_is_writeable(sk, 0); 1423 } 1424 1425 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1426 struct cgroup *ancestor) 1427 { 1428 #ifdef CONFIG_SOCK_CGROUP_DATA 1429 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1430 ancestor); 1431 #else 1432 return -ENOTSUPP; 1433 #endif 1434 } 1435 1436 static inline bool sk_has_memory_pressure(const struct sock *sk) 1437 { 1438 return sk->sk_prot->memory_pressure != NULL; 1439 } 1440 1441 static inline bool sk_under_memory_pressure(const struct sock *sk) 1442 { 1443 if (!sk->sk_prot->memory_pressure) 1444 return false; 1445 1446 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1447 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1448 return true; 1449 1450 return !!*sk->sk_prot->memory_pressure; 1451 } 1452 1453 static inline long 1454 proto_memory_allocated(const struct proto *prot) 1455 { 1456 return max(0L, atomic_long_read(prot->memory_allocated)); 1457 } 1458 1459 static inline long 1460 sk_memory_allocated(const struct sock *sk) 1461 { 1462 return proto_memory_allocated(sk->sk_prot); 1463 } 1464 1465 /* 1 MB per cpu, in page units */ 1466 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1467 1468 static inline void 1469 sk_memory_allocated_add(struct sock *sk, int amt) 1470 { 1471 int local_reserve; 1472 1473 preempt_disable(); 1474 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1475 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) { 1476 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1477 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1478 } 1479 preempt_enable(); 1480 } 1481 1482 static inline void 1483 sk_memory_allocated_sub(struct sock *sk, int amt) 1484 { 1485 int local_reserve; 1486 1487 preempt_disable(); 1488 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1489 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) { 1490 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1491 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1492 } 1493 preempt_enable(); 1494 } 1495 1496 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1497 1498 static inline void sk_sockets_allocated_dec(struct sock *sk) 1499 { 1500 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1501 SK_ALLOC_PERCPU_COUNTER_BATCH); 1502 } 1503 1504 static inline void sk_sockets_allocated_inc(struct sock *sk) 1505 { 1506 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1507 SK_ALLOC_PERCPU_COUNTER_BATCH); 1508 } 1509 1510 static inline u64 1511 sk_sockets_allocated_read_positive(struct sock *sk) 1512 { 1513 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1514 } 1515 1516 static inline int 1517 proto_sockets_allocated_sum_positive(struct proto *prot) 1518 { 1519 return percpu_counter_sum_positive(prot->sockets_allocated); 1520 } 1521 1522 static inline bool 1523 proto_memory_pressure(struct proto *prot) 1524 { 1525 if (!prot->memory_pressure) 1526 return false; 1527 return !!*prot->memory_pressure; 1528 } 1529 1530 1531 #ifdef CONFIG_PROC_FS 1532 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1533 struct prot_inuse { 1534 int all; 1535 int val[PROTO_INUSE_NR]; 1536 }; 1537 1538 static inline void sock_prot_inuse_add(const struct net *net, 1539 const struct proto *prot, int val) 1540 { 1541 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1542 } 1543 1544 static inline void sock_inuse_add(const struct net *net, int val) 1545 { 1546 this_cpu_add(net->core.prot_inuse->all, val); 1547 } 1548 1549 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1550 int sock_inuse_get(struct net *net); 1551 #else 1552 static inline void sock_prot_inuse_add(const struct net *net, 1553 const struct proto *prot, int val) 1554 { 1555 } 1556 1557 static inline void sock_inuse_add(const struct net *net, int val) 1558 { 1559 } 1560 #endif 1561 1562 1563 /* With per-bucket locks this operation is not-atomic, so that 1564 * this version is not worse. 1565 */ 1566 static inline int __sk_prot_rehash(struct sock *sk) 1567 { 1568 sk->sk_prot->unhash(sk); 1569 return sk->sk_prot->hash(sk); 1570 } 1571 1572 /* About 10 seconds */ 1573 #define SOCK_DESTROY_TIME (10*HZ) 1574 1575 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1576 #define PROT_SOCK 1024 1577 1578 #define SHUTDOWN_MASK 3 1579 #define RCV_SHUTDOWN 1 1580 #define SEND_SHUTDOWN 2 1581 1582 #define SOCK_BINDADDR_LOCK 4 1583 #define SOCK_BINDPORT_LOCK 8 1584 1585 struct socket_alloc { 1586 struct socket socket; 1587 struct inode vfs_inode; 1588 }; 1589 1590 static inline struct socket *SOCKET_I(struct inode *inode) 1591 { 1592 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1593 } 1594 1595 static inline struct inode *SOCK_INODE(struct socket *socket) 1596 { 1597 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1598 } 1599 1600 /* 1601 * Functions for memory accounting 1602 */ 1603 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1604 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1605 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1606 void __sk_mem_reclaim(struct sock *sk, int amount); 1607 1608 #define SK_MEM_SEND 0 1609 #define SK_MEM_RECV 1 1610 1611 /* sysctl_mem values are in pages */ 1612 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1613 { 1614 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1615 } 1616 1617 static inline int sk_mem_pages(int amt) 1618 { 1619 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1620 } 1621 1622 static inline bool sk_has_account(struct sock *sk) 1623 { 1624 /* return true if protocol supports memory accounting */ 1625 return !!sk->sk_prot->memory_allocated; 1626 } 1627 1628 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1629 { 1630 int delta; 1631 1632 if (!sk_has_account(sk)) 1633 return true; 1634 delta = size - sk->sk_forward_alloc; 1635 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1636 } 1637 1638 static inline bool 1639 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1640 { 1641 int delta; 1642 1643 if (!sk_has_account(sk)) 1644 return true; 1645 delta = size - sk->sk_forward_alloc; 1646 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1647 skb_pfmemalloc(skb); 1648 } 1649 1650 static inline int sk_unused_reserved_mem(const struct sock *sk) 1651 { 1652 int unused_mem; 1653 1654 if (likely(!sk->sk_reserved_mem)) 1655 return 0; 1656 1657 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1658 atomic_read(&sk->sk_rmem_alloc); 1659 1660 return unused_mem > 0 ? unused_mem : 0; 1661 } 1662 1663 static inline void sk_mem_reclaim(struct sock *sk) 1664 { 1665 int reclaimable; 1666 1667 if (!sk_has_account(sk)) 1668 return; 1669 1670 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1671 1672 if (reclaimable >= (int)PAGE_SIZE) 1673 __sk_mem_reclaim(sk, reclaimable); 1674 } 1675 1676 static inline void sk_mem_reclaim_final(struct sock *sk) 1677 { 1678 sk->sk_reserved_mem = 0; 1679 sk_mem_reclaim(sk); 1680 } 1681 1682 static inline void sk_mem_charge(struct sock *sk, int size) 1683 { 1684 if (!sk_has_account(sk)) 1685 return; 1686 sk->sk_forward_alloc -= size; 1687 } 1688 1689 static inline void sk_mem_uncharge(struct sock *sk, int size) 1690 { 1691 if (!sk_has_account(sk)) 1692 return; 1693 sk->sk_forward_alloc += size; 1694 sk_mem_reclaim(sk); 1695 } 1696 1697 /* 1698 * Macro so as to not evaluate some arguments when 1699 * lockdep is not enabled. 1700 * 1701 * Mark both the sk_lock and the sk_lock.slock as a 1702 * per-address-family lock class. 1703 */ 1704 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1705 do { \ 1706 sk->sk_lock.owned = 0; \ 1707 init_waitqueue_head(&sk->sk_lock.wq); \ 1708 spin_lock_init(&(sk)->sk_lock.slock); \ 1709 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1710 sizeof((sk)->sk_lock)); \ 1711 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1712 (skey), (sname)); \ 1713 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1714 } while (0) 1715 1716 static inline bool lockdep_sock_is_held(const struct sock *sk) 1717 { 1718 return lockdep_is_held(&sk->sk_lock) || 1719 lockdep_is_held(&sk->sk_lock.slock); 1720 } 1721 1722 void lock_sock_nested(struct sock *sk, int subclass); 1723 1724 static inline void lock_sock(struct sock *sk) 1725 { 1726 lock_sock_nested(sk, 0); 1727 } 1728 1729 void __lock_sock(struct sock *sk); 1730 void __release_sock(struct sock *sk); 1731 void release_sock(struct sock *sk); 1732 1733 /* BH context may only use the following locking interface. */ 1734 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1735 #define bh_lock_sock_nested(__sk) \ 1736 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1737 SINGLE_DEPTH_NESTING) 1738 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1739 1740 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1741 1742 /** 1743 * lock_sock_fast - fast version of lock_sock 1744 * @sk: socket 1745 * 1746 * This version should be used for very small section, where process wont block 1747 * return false if fast path is taken: 1748 * 1749 * sk_lock.slock locked, owned = 0, BH disabled 1750 * 1751 * return true if slow path is taken: 1752 * 1753 * sk_lock.slock unlocked, owned = 1, BH enabled 1754 */ 1755 static inline bool lock_sock_fast(struct sock *sk) 1756 { 1757 /* The sk_lock has mutex_lock() semantics here. */ 1758 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1759 1760 return __lock_sock_fast(sk); 1761 } 1762 1763 /* fast socket lock variant for caller already holding a [different] socket lock */ 1764 static inline bool lock_sock_fast_nested(struct sock *sk) 1765 { 1766 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1767 1768 return __lock_sock_fast(sk); 1769 } 1770 1771 /** 1772 * unlock_sock_fast - complement of lock_sock_fast 1773 * @sk: socket 1774 * @slow: slow mode 1775 * 1776 * fast unlock socket for user context. 1777 * If slow mode is on, we call regular release_sock() 1778 */ 1779 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1780 __releases(&sk->sk_lock.slock) 1781 { 1782 if (slow) { 1783 release_sock(sk); 1784 __release(&sk->sk_lock.slock); 1785 } else { 1786 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1787 spin_unlock_bh(&sk->sk_lock.slock); 1788 } 1789 } 1790 1791 void sockopt_lock_sock(struct sock *sk); 1792 void sockopt_release_sock(struct sock *sk); 1793 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1794 bool sockopt_capable(int cap); 1795 1796 /* Used by processes to "lock" a socket state, so that 1797 * interrupts and bottom half handlers won't change it 1798 * from under us. It essentially blocks any incoming 1799 * packets, so that we won't get any new data or any 1800 * packets that change the state of the socket. 1801 * 1802 * While locked, BH processing will add new packets to 1803 * the backlog queue. This queue is processed by the 1804 * owner of the socket lock right before it is released. 1805 * 1806 * Since ~2.3.5 it is also exclusive sleep lock serializing 1807 * accesses from user process context. 1808 */ 1809 1810 static inline void sock_owned_by_me(const struct sock *sk) 1811 { 1812 #ifdef CONFIG_LOCKDEP 1813 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1814 #endif 1815 } 1816 1817 static inline bool sock_owned_by_user(const struct sock *sk) 1818 { 1819 sock_owned_by_me(sk); 1820 return sk->sk_lock.owned; 1821 } 1822 1823 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1824 { 1825 return sk->sk_lock.owned; 1826 } 1827 1828 static inline void sock_release_ownership(struct sock *sk) 1829 { 1830 if (sock_owned_by_user_nocheck(sk)) { 1831 sk->sk_lock.owned = 0; 1832 1833 /* The sk_lock has mutex_unlock() semantics: */ 1834 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1835 } 1836 } 1837 1838 /* no reclassification while locks are held */ 1839 static inline bool sock_allow_reclassification(const struct sock *csk) 1840 { 1841 struct sock *sk = (struct sock *)csk; 1842 1843 return !sock_owned_by_user_nocheck(sk) && 1844 !spin_is_locked(&sk->sk_lock.slock); 1845 } 1846 1847 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1848 struct proto *prot, int kern); 1849 void sk_free(struct sock *sk); 1850 void sk_destruct(struct sock *sk); 1851 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1852 void sk_free_unlock_clone(struct sock *sk); 1853 1854 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1855 gfp_t priority); 1856 void __sock_wfree(struct sk_buff *skb); 1857 void sock_wfree(struct sk_buff *skb); 1858 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1859 gfp_t priority); 1860 void skb_orphan_partial(struct sk_buff *skb); 1861 void sock_rfree(struct sk_buff *skb); 1862 void sock_efree(struct sk_buff *skb); 1863 #ifdef CONFIG_INET 1864 void sock_edemux(struct sk_buff *skb); 1865 void sock_pfree(struct sk_buff *skb); 1866 #else 1867 #define sock_edemux sock_efree 1868 #endif 1869 1870 int sk_setsockopt(struct sock *sk, int level, int optname, 1871 sockptr_t optval, unsigned int optlen); 1872 int sock_setsockopt(struct socket *sock, int level, int op, 1873 sockptr_t optval, unsigned int optlen); 1874 1875 int sk_getsockopt(struct sock *sk, int level, int optname, 1876 sockptr_t optval, sockptr_t optlen); 1877 int sock_getsockopt(struct socket *sock, int level, int op, 1878 char __user *optval, int __user *optlen); 1879 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1880 bool timeval, bool time32); 1881 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1882 unsigned long data_len, int noblock, 1883 int *errcode, int max_page_order); 1884 1885 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1886 unsigned long size, 1887 int noblock, int *errcode) 1888 { 1889 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1890 } 1891 1892 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1893 void sock_kfree_s(struct sock *sk, void *mem, int size); 1894 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1895 void sk_send_sigurg(struct sock *sk); 1896 1897 struct sockcm_cookie { 1898 u64 transmit_time; 1899 u32 mark; 1900 u16 tsflags; 1901 }; 1902 1903 static inline void sockcm_init(struct sockcm_cookie *sockc, 1904 const struct sock *sk) 1905 { 1906 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1907 } 1908 1909 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg, 1910 struct sockcm_cookie *sockc); 1911 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1912 struct sockcm_cookie *sockc); 1913 1914 /* 1915 * Functions to fill in entries in struct proto_ops when a protocol 1916 * does not implement a particular function. 1917 */ 1918 int sock_no_bind(struct socket *, struct sockaddr *, int); 1919 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1920 int sock_no_socketpair(struct socket *, struct socket *); 1921 int sock_no_accept(struct socket *, struct socket *, int, bool); 1922 int sock_no_getname(struct socket *, struct sockaddr *, int); 1923 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1924 int sock_no_listen(struct socket *, int); 1925 int sock_no_shutdown(struct socket *, int); 1926 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1927 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1928 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1929 int sock_no_mmap(struct file *file, struct socket *sock, 1930 struct vm_area_struct *vma); 1931 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, 1932 size_t size, int flags); 1933 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page, 1934 int offset, size_t size, int flags); 1935 1936 /* 1937 * Functions to fill in entries in struct proto_ops when a protocol 1938 * uses the inet style. 1939 */ 1940 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1941 char __user *optval, int __user *optlen); 1942 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1943 int flags); 1944 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1945 sockptr_t optval, unsigned int optlen); 1946 1947 void sk_common_release(struct sock *sk); 1948 1949 /* 1950 * Default socket callbacks and setup code 1951 */ 1952 1953 /* Initialise core socket variables */ 1954 void sock_init_data(struct socket *sock, struct sock *sk); 1955 1956 /* 1957 * Socket reference counting postulates. 1958 * 1959 * * Each user of socket SHOULD hold a reference count. 1960 * * Each access point to socket (an hash table bucket, reference from a list, 1961 * running timer, skb in flight MUST hold a reference count. 1962 * * When reference count hits 0, it means it will never increase back. 1963 * * When reference count hits 0, it means that no references from 1964 * outside exist to this socket and current process on current CPU 1965 * is last user and may/should destroy this socket. 1966 * * sk_free is called from any context: process, BH, IRQ. When 1967 * it is called, socket has no references from outside -> sk_free 1968 * may release descendant resources allocated by the socket, but 1969 * to the time when it is called, socket is NOT referenced by any 1970 * hash tables, lists etc. 1971 * * Packets, delivered from outside (from network or from another process) 1972 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1973 * when they sit in queue. Otherwise, packets will leak to hole, when 1974 * socket is looked up by one cpu and unhasing is made by another CPU. 1975 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1976 * (leak to backlog). Packet socket does all the processing inside 1977 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1978 * use separate SMP lock, so that they are prone too. 1979 */ 1980 1981 /* Ungrab socket and destroy it, if it was the last reference. */ 1982 static inline void sock_put(struct sock *sk) 1983 { 1984 if (refcount_dec_and_test(&sk->sk_refcnt)) 1985 sk_free(sk); 1986 } 1987 /* Generic version of sock_put(), dealing with all sockets 1988 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1989 */ 1990 void sock_gen_put(struct sock *sk); 1991 1992 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1993 unsigned int trim_cap, bool refcounted); 1994 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1995 const int nested) 1996 { 1997 return __sk_receive_skb(sk, skb, nested, 1, true); 1998 } 1999 2000 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 2001 { 2002 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2003 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2004 return; 2005 sk->sk_tx_queue_mapping = tx_queue; 2006 } 2007 2008 #define NO_QUEUE_MAPPING USHRT_MAX 2009 2010 static inline void sk_tx_queue_clear(struct sock *sk) 2011 { 2012 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 2013 } 2014 2015 static inline int sk_tx_queue_get(const struct sock *sk) 2016 { 2017 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 2018 return sk->sk_tx_queue_mapping; 2019 2020 return -1; 2021 } 2022 2023 static inline void __sk_rx_queue_set(struct sock *sk, 2024 const struct sk_buff *skb, 2025 bool force_set) 2026 { 2027 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2028 if (skb_rx_queue_recorded(skb)) { 2029 u16 rx_queue = skb_get_rx_queue(skb); 2030 2031 if (force_set || 2032 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2033 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2034 } 2035 #endif 2036 } 2037 2038 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2039 { 2040 __sk_rx_queue_set(sk, skb, true); 2041 } 2042 2043 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2044 { 2045 __sk_rx_queue_set(sk, skb, false); 2046 } 2047 2048 static inline void sk_rx_queue_clear(struct sock *sk) 2049 { 2050 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2051 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2052 #endif 2053 } 2054 2055 static inline int sk_rx_queue_get(const struct sock *sk) 2056 { 2057 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2058 if (sk) { 2059 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2060 2061 if (res != NO_QUEUE_MAPPING) 2062 return res; 2063 } 2064 #endif 2065 2066 return -1; 2067 } 2068 2069 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2070 { 2071 sk->sk_socket = sock; 2072 } 2073 2074 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2075 { 2076 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2077 return &rcu_dereference_raw(sk->sk_wq)->wait; 2078 } 2079 /* Detach socket from process context. 2080 * Announce socket dead, detach it from wait queue and inode. 2081 * Note that parent inode held reference count on this struct sock, 2082 * we do not release it in this function, because protocol 2083 * probably wants some additional cleanups or even continuing 2084 * to work with this socket (TCP). 2085 */ 2086 static inline void sock_orphan(struct sock *sk) 2087 { 2088 write_lock_bh(&sk->sk_callback_lock); 2089 sock_set_flag(sk, SOCK_DEAD); 2090 sk_set_socket(sk, NULL); 2091 sk->sk_wq = NULL; 2092 write_unlock_bh(&sk->sk_callback_lock); 2093 } 2094 2095 static inline void sock_graft(struct sock *sk, struct socket *parent) 2096 { 2097 WARN_ON(parent->sk); 2098 write_lock_bh(&sk->sk_callback_lock); 2099 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2100 parent->sk = sk; 2101 sk_set_socket(sk, parent); 2102 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2103 security_sock_graft(sk, parent); 2104 write_unlock_bh(&sk->sk_callback_lock); 2105 } 2106 2107 kuid_t sock_i_uid(struct sock *sk); 2108 unsigned long sock_i_ino(struct sock *sk); 2109 2110 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2111 { 2112 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2113 } 2114 2115 static inline u32 net_tx_rndhash(void) 2116 { 2117 u32 v = prandom_u32(); 2118 2119 return v ?: 1; 2120 } 2121 2122 static inline void sk_set_txhash(struct sock *sk) 2123 { 2124 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2125 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2126 } 2127 2128 static inline bool sk_rethink_txhash(struct sock *sk) 2129 { 2130 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2131 sk_set_txhash(sk); 2132 return true; 2133 } 2134 return false; 2135 } 2136 2137 static inline struct dst_entry * 2138 __sk_dst_get(struct sock *sk) 2139 { 2140 return rcu_dereference_check(sk->sk_dst_cache, 2141 lockdep_sock_is_held(sk)); 2142 } 2143 2144 static inline struct dst_entry * 2145 sk_dst_get(struct sock *sk) 2146 { 2147 struct dst_entry *dst; 2148 2149 rcu_read_lock(); 2150 dst = rcu_dereference(sk->sk_dst_cache); 2151 if (dst && !atomic_inc_not_zero(&dst->__refcnt)) 2152 dst = NULL; 2153 rcu_read_unlock(); 2154 return dst; 2155 } 2156 2157 static inline void __dst_negative_advice(struct sock *sk) 2158 { 2159 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2160 2161 if (dst && dst->ops->negative_advice) { 2162 ndst = dst->ops->negative_advice(dst); 2163 2164 if (ndst != dst) { 2165 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2166 sk_tx_queue_clear(sk); 2167 sk->sk_dst_pending_confirm = 0; 2168 } 2169 } 2170 } 2171 2172 static inline void dst_negative_advice(struct sock *sk) 2173 { 2174 sk_rethink_txhash(sk); 2175 __dst_negative_advice(sk); 2176 } 2177 2178 static inline void 2179 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2180 { 2181 struct dst_entry *old_dst; 2182 2183 sk_tx_queue_clear(sk); 2184 sk->sk_dst_pending_confirm = 0; 2185 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2186 lockdep_sock_is_held(sk)); 2187 rcu_assign_pointer(sk->sk_dst_cache, dst); 2188 dst_release(old_dst); 2189 } 2190 2191 static inline void 2192 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2193 { 2194 struct dst_entry *old_dst; 2195 2196 sk_tx_queue_clear(sk); 2197 sk->sk_dst_pending_confirm = 0; 2198 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2199 dst_release(old_dst); 2200 } 2201 2202 static inline void 2203 __sk_dst_reset(struct sock *sk) 2204 { 2205 __sk_dst_set(sk, NULL); 2206 } 2207 2208 static inline void 2209 sk_dst_reset(struct sock *sk) 2210 { 2211 sk_dst_set(sk, NULL); 2212 } 2213 2214 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2215 2216 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2217 2218 static inline void sk_dst_confirm(struct sock *sk) 2219 { 2220 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2221 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2222 } 2223 2224 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2225 { 2226 if (skb_get_dst_pending_confirm(skb)) { 2227 struct sock *sk = skb->sk; 2228 2229 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2230 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2231 neigh_confirm(n); 2232 } 2233 } 2234 2235 bool sk_mc_loop(struct sock *sk); 2236 2237 static inline bool sk_can_gso(const struct sock *sk) 2238 { 2239 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2240 } 2241 2242 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2243 2244 static inline void sk_gso_disable(struct sock *sk) 2245 { 2246 sk->sk_gso_disabled = 1; 2247 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2248 } 2249 2250 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2251 struct iov_iter *from, char *to, 2252 int copy, int offset) 2253 { 2254 if (skb->ip_summed == CHECKSUM_NONE) { 2255 __wsum csum = 0; 2256 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2257 return -EFAULT; 2258 skb->csum = csum_block_add(skb->csum, csum, offset); 2259 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2260 if (!copy_from_iter_full_nocache(to, copy, from)) 2261 return -EFAULT; 2262 } else if (!copy_from_iter_full(to, copy, from)) 2263 return -EFAULT; 2264 2265 return 0; 2266 } 2267 2268 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2269 struct iov_iter *from, int copy) 2270 { 2271 int err, offset = skb->len; 2272 2273 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2274 copy, offset); 2275 if (err) 2276 __skb_trim(skb, offset); 2277 2278 return err; 2279 } 2280 2281 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2282 struct sk_buff *skb, 2283 struct page *page, 2284 int off, int copy) 2285 { 2286 int err; 2287 2288 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2289 copy, skb->len); 2290 if (err) 2291 return err; 2292 2293 skb_len_add(skb, copy); 2294 sk_wmem_queued_add(sk, copy); 2295 sk_mem_charge(sk, copy); 2296 return 0; 2297 } 2298 2299 /** 2300 * sk_wmem_alloc_get - returns write allocations 2301 * @sk: socket 2302 * 2303 * Return: sk_wmem_alloc minus initial offset of one 2304 */ 2305 static inline int sk_wmem_alloc_get(const struct sock *sk) 2306 { 2307 return refcount_read(&sk->sk_wmem_alloc) - 1; 2308 } 2309 2310 /** 2311 * sk_rmem_alloc_get - returns read allocations 2312 * @sk: socket 2313 * 2314 * Return: sk_rmem_alloc 2315 */ 2316 static inline int sk_rmem_alloc_get(const struct sock *sk) 2317 { 2318 return atomic_read(&sk->sk_rmem_alloc); 2319 } 2320 2321 /** 2322 * sk_has_allocations - check if allocations are outstanding 2323 * @sk: socket 2324 * 2325 * Return: true if socket has write or read allocations 2326 */ 2327 static inline bool sk_has_allocations(const struct sock *sk) 2328 { 2329 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2330 } 2331 2332 /** 2333 * skwq_has_sleeper - check if there are any waiting processes 2334 * @wq: struct socket_wq 2335 * 2336 * Return: true if socket_wq has waiting processes 2337 * 2338 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2339 * barrier call. They were added due to the race found within the tcp code. 2340 * 2341 * Consider following tcp code paths:: 2342 * 2343 * CPU1 CPU2 2344 * sys_select receive packet 2345 * ... ... 2346 * __add_wait_queue update tp->rcv_nxt 2347 * ... ... 2348 * tp->rcv_nxt check sock_def_readable 2349 * ... { 2350 * schedule rcu_read_lock(); 2351 * wq = rcu_dereference(sk->sk_wq); 2352 * if (wq && waitqueue_active(&wq->wait)) 2353 * wake_up_interruptible(&wq->wait) 2354 * ... 2355 * } 2356 * 2357 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2358 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2359 * could then endup calling schedule and sleep forever if there are no more 2360 * data on the socket. 2361 * 2362 */ 2363 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2364 { 2365 return wq && wq_has_sleeper(&wq->wait); 2366 } 2367 2368 /** 2369 * sock_poll_wait - place memory barrier behind the poll_wait call. 2370 * @filp: file 2371 * @sock: socket to wait on 2372 * @p: poll_table 2373 * 2374 * See the comments in the wq_has_sleeper function. 2375 */ 2376 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2377 poll_table *p) 2378 { 2379 if (!poll_does_not_wait(p)) { 2380 poll_wait(filp, &sock->wq.wait, p); 2381 /* We need to be sure we are in sync with the 2382 * socket flags modification. 2383 * 2384 * This memory barrier is paired in the wq_has_sleeper. 2385 */ 2386 smp_mb(); 2387 } 2388 } 2389 2390 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2391 { 2392 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2393 u32 txhash = READ_ONCE(sk->sk_txhash); 2394 2395 if (txhash) { 2396 skb->l4_hash = 1; 2397 skb->hash = txhash; 2398 } 2399 } 2400 2401 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2402 2403 /* 2404 * Queue a received datagram if it will fit. Stream and sequenced 2405 * protocols can't normally use this as they need to fit buffers in 2406 * and play with them. 2407 * 2408 * Inlined as it's very short and called for pretty much every 2409 * packet ever received. 2410 */ 2411 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2412 { 2413 skb_orphan(skb); 2414 skb->sk = sk; 2415 skb->destructor = sock_rfree; 2416 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2417 sk_mem_charge(sk, skb->truesize); 2418 } 2419 2420 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2421 { 2422 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2423 skb_orphan(skb); 2424 skb->destructor = sock_efree; 2425 skb->sk = sk; 2426 return true; 2427 } 2428 return false; 2429 } 2430 2431 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2432 { 2433 if (skb->destructor != sock_wfree) { 2434 skb_orphan(skb); 2435 return; 2436 } 2437 skb->slow_gro = 1; 2438 } 2439 2440 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2441 unsigned long expires); 2442 2443 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2444 2445 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2446 2447 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2448 struct sk_buff *skb, unsigned int flags, 2449 void (*destructor)(struct sock *sk, 2450 struct sk_buff *skb)); 2451 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2452 2453 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2454 enum skb_drop_reason *reason); 2455 2456 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2457 { 2458 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2459 } 2460 2461 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2462 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2463 2464 /* 2465 * Recover an error report and clear atomically 2466 */ 2467 2468 static inline int sock_error(struct sock *sk) 2469 { 2470 int err; 2471 2472 /* Avoid an atomic operation for the common case. 2473 * This is racy since another cpu/thread can change sk_err under us. 2474 */ 2475 if (likely(data_race(!sk->sk_err))) 2476 return 0; 2477 2478 err = xchg(&sk->sk_err, 0); 2479 return -err; 2480 } 2481 2482 void sk_error_report(struct sock *sk); 2483 2484 static inline unsigned long sock_wspace(struct sock *sk) 2485 { 2486 int amt = 0; 2487 2488 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2489 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2490 if (amt < 0) 2491 amt = 0; 2492 } 2493 return amt; 2494 } 2495 2496 /* Note: 2497 * We use sk->sk_wq_raw, from contexts knowing this 2498 * pointer is not NULL and cannot disappear/change. 2499 */ 2500 static inline void sk_set_bit(int nr, struct sock *sk) 2501 { 2502 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2503 !sock_flag(sk, SOCK_FASYNC)) 2504 return; 2505 2506 set_bit(nr, &sk->sk_wq_raw->flags); 2507 } 2508 2509 static inline void sk_clear_bit(int nr, struct sock *sk) 2510 { 2511 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2512 !sock_flag(sk, SOCK_FASYNC)) 2513 return; 2514 2515 clear_bit(nr, &sk->sk_wq_raw->flags); 2516 } 2517 2518 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2519 { 2520 if (sock_flag(sk, SOCK_FASYNC)) { 2521 rcu_read_lock(); 2522 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2523 rcu_read_unlock(); 2524 } 2525 } 2526 2527 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2528 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2529 * Note: for send buffers, TCP works better if we can build two skbs at 2530 * minimum. 2531 */ 2532 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2533 2534 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2535 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2536 2537 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2538 { 2539 u32 val; 2540 2541 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2542 return; 2543 2544 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2545 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2546 2547 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2548 } 2549 2550 /** 2551 * sk_page_frag - return an appropriate page_frag 2552 * @sk: socket 2553 * 2554 * Use the per task page_frag instead of the per socket one for 2555 * optimization when we know that we're in process context and own 2556 * everything that's associated with %current. 2557 * 2558 * Both direct reclaim and page faults can nest inside other 2559 * socket operations and end up recursing into sk_page_frag() 2560 * while it's already in use: explicitly avoid task page_frag 2561 * usage if the caller is potentially doing any of them. 2562 * This assumes that page fault handlers use the GFP_NOFS flags. 2563 * 2564 * Return: a per task page_frag if context allows that, 2565 * otherwise a per socket one. 2566 */ 2567 static inline struct page_frag *sk_page_frag(struct sock *sk) 2568 { 2569 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) == 2570 (__GFP_DIRECT_RECLAIM | __GFP_FS)) 2571 return ¤t->task_frag; 2572 2573 return &sk->sk_frag; 2574 } 2575 2576 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2577 2578 /* 2579 * Default write policy as shown to user space via poll/select/SIGIO 2580 */ 2581 static inline bool sock_writeable(const struct sock *sk) 2582 { 2583 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2584 } 2585 2586 static inline gfp_t gfp_any(void) 2587 { 2588 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2589 } 2590 2591 static inline gfp_t gfp_memcg_charge(void) 2592 { 2593 return in_softirq() ? GFP_NOWAIT : GFP_KERNEL; 2594 } 2595 2596 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2597 { 2598 return noblock ? 0 : sk->sk_rcvtimeo; 2599 } 2600 2601 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2602 { 2603 return noblock ? 0 : sk->sk_sndtimeo; 2604 } 2605 2606 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2607 { 2608 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2609 2610 return v ?: 1; 2611 } 2612 2613 /* Alas, with timeout socket operations are not restartable. 2614 * Compare this to poll(). 2615 */ 2616 static inline int sock_intr_errno(long timeo) 2617 { 2618 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2619 } 2620 2621 struct sock_skb_cb { 2622 u32 dropcount; 2623 }; 2624 2625 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2626 * using skb->cb[] would keep using it directly and utilize its 2627 * alignement guarantee. 2628 */ 2629 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2630 sizeof(struct sock_skb_cb))) 2631 2632 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2633 SOCK_SKB_CB_OFFSET)) 2634 2635 #define sock_skb_cb_check_size(size) \ 2636 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2637 2638 static inline void 2639 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2640 { 2641 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2642 atomic_read(&sk->sk_drops) : 0; 2643 } 2644 2645 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2646 { 2647 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2648 2649 atomic_add(segs, &sk->sk_drops); 2650 } 2651 2652 static inline ktime_t sock_read_timestamp(struct sock *sk) 2653 { 2654 #if BITS_PER_LONG==32 2655 unsigned int seq; 2656 ktime_t kt; 2657 2658 do { 2659 seq = read_seqbegin(&sk->sk_stamp_seq); 2660 kt = sk->sk_stamp; 2661 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2662 2663 return kt; 2664 #else 2665 return READ_ONCE(sk->sk_stamp); 2666 #endif 2667 } 2668 2669 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2670 { 2671 #if BITS_PER_LONG==32 2672 write_seqlock(&sk->sk_stamp_seq); 2673 sk->sk_stamp = kt; 2674 write_sequnlock(&sk->sk_stamp_seq); 2675 #else 2676 WRITE_ONCE(sk->sk_stamp, kt); 2677 #endif 2678 } 2679 2680 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2681 struct sk_buff *skb); 2682 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2683 struct sk_buff *skb); 2684 2685 static inline void 2686 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2687 { 2688 ktime_t kt = skb->tstamp; 2689 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2690 2691 /* 2692 * generate control messages if 2693 * - receive time stamping in software requested 2694 * - software time stamp available and wanted 2695 * - hardware time stamps available and wanted 2696 */ 2697 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2698 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2699 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2700 (hwtstamps->hwtstamp && 2701 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2702 __sock_recv_timestamp(msg, sk, skb); 2703 else 2704 sock_write_timestamp(sk, kt); 2705 2706 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid) 2707 __sock_recv_wifi_status(msg, sk, skb); 2708 } 2709 2710 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2711 struct sk_buff *skb); 2712 2713 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2714 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2715 struct sk_buff *skb) 2716 { 2717 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2718 (1UL << SOCK_RCVTSTAMP) | \ 2719 (1UL << SOCK_RCVMARK)) 2720 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2721 SOF_TIMESTAMPING_RAW_HARDWARE) 2722 2723 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2724 __sock_recv_cmsgs(msg, sk, skb); 2725 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2726 sock_write_timestamp(sk, skb->tstamp); 2727 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP)) 2728 sock_write_timestamp(sk, 0); 2729 } 2730 2731 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2732 2733 /** 2734 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2735 * @sk: socket sending this packet 2736 * @tsflags: timestamping flags to use 2737 * @tx_flags: completed with instructions for time stamping 2738 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2739 * 2740 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2741 */ 2742 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2743 __u8 *tx_flags, __u32 *tskey) 2744 { 2745 if (unlikely(tsflags)) { 2746 __sock_tx_timestamp(tsflags, tx_flags); 2747 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2748 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2749 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2750 } 2751 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2752 *tx_flags |= SKBTX_WIFI_STATUS; 2753 } 2754 2755 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2756 __u8 *tx_flags) 2757 { 2758 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2759 } 2760 2761 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2762 { 2763 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2764 &skb_shinfo(skb)->tskey); 2765 } 2766 2767 static inline bool sk_is_tcp(const struct sock *sk) 2768 { 2769 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2770 } 2771 2772 /** 2773 * sk_eat_skb - Release a skb if it is no longer needed 2774 * @sk: socket to eat this skb from 2775 * @skb: socket buffer to eat 2776 * 2777 * This routine must be called with interrupts disabled or with the socket 2778 * locked so that the sk_buff queue operation is ok. 2779 */ 2780 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2781 { 2782 __skb_unlink(skb, &sk->sk_receive_queue); 2783 __kfree_skb(skb); 2784 } 2785 2786 static inline bool 2787 skb_sk_is_prefetched(struct sk_buff *skb) 2788 { 2789 #ifdef CONFIG_INET 2790 return skb->destructor == sock_pfree; 2791 #else 2792 return false; 2793 #endif /* CONFIG_INET */ 2794 } 2795 2796 /* This helper checks if a socket is a full socket, 2797 * ie _not_ a timewait or request socket. 2798 */ 2799 static inline bool sk_fullsock(const struct sock *sk) 2800 { 2801 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2802 } 2803 2804 static inline bool 2805 sk_is_refcounted(struct sock *sk) 2806 { 2807 /* Only full sockets have sk->sk_flags. */ 2808 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2809 } 2810 2811 /** 2812 * skb_steal_sock - steal a socket from an sk_buff 2813 * @skb: sk_buff to steal the socket from 2814 * @refcounted: is set to true if the socket is reference-counted 2815 */ 2816 static inline struct sock * 2817 skb_steal_sock(struct sk_buff *skb, bool *refcounted) 2818 { 2819 if (skb->sk) { 2820 struct sock *sk = skb->sk; 2821 2822 *refcounted = true; 2823 if (skb_sk_is_prefetched(skb)) 2824 *refcounted = sk_is_refcounted(sk); 2825 skb->destructor = NULL; 2826 skb->sk = NULL; 2827 return sk; 2828 } 2829 *refcounted = false; 2830 return NULL; 2831 } 2832 2833 /* Checks if this SKB belongs to an HW offloaded socket 2834 * and whether any SW fallbacks are required based on dev. 2835 * Check decrypted mark in case skb_orphan() cleared socket. 2836 */ 2837 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2838 struct net_device *dev) 2839 { 2840 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2841 struct sock *sk = skb->sk; 2842 2843 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2844 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2845 #ifdef CONFIG_TLS_DEVICE 2846 } else if (unlikely(skb->decrypted)) { 2847 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2848 kfree_skb(skb); 2849 skb = NULL; 2850 #endif 2851 } 2852 #endif 2853 2854 return skb; 2855 } 2856 2857 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2858 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2859 */ 2860 static inline bool sk_listener(const struct sock *sk) 2861 { 2862 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2863 } 2864 2865 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2866 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2867 int type); 2868 2869 bool sk_ns_capable(const struct sock *sk, 2870 struct user_namespace *user_ns, int cap); 2871 bool sk_capable(const struct sock *sk, int cap); 2872 bool sk_net_capable(const struct sock *sk, int cap); 2873 2874 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2875 2876 /* Take into consideration the size of the struct sk_buff overhead in the 2877 * determination of these values, since that is non-constant across 2878 * platforms. This makes socket queueing behavior and performance 2879 * not depend upon such differences. 2880 */ 2881 #define _SK_MEM_PACKETS 256 2882 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2883 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2884 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2885 2886 extern __u32 sysctl_wmem_max; 2887 extern __u32 sysctl_rmem_max; 2888 2889 extern int sysctl_tstamp_allow_data; 2890 extern int sysctl_optmem_max; 2891 2892 extern __u32 sysctl_wmem_default; 2893 extern __u32 sysctl_rmem_default; 2894 2895 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2896 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2897 2898 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2899 { 2900 /* Does this proto have per netns sysctl_wmem ? */ 2901 if (proto->sysctl_wmem_offset) 2902 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2903 2904 return READ_ONCE(*proto->sysctl_wmem); 2905 } 2906 2907 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2908 { 2909 /* Does this proto have per netns sysctl_rmem ? */ 2910 if (proto->sysctl_rmem_offset) 2911 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2912 2913 return READ_ONCE(*proto->sysctl_rmem); 2914 } 2915 2916 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2917 * Some wifi drivers need to tweak it to get more chunks. 2918 * They can use this helper from their ndo_start_xmit() 2919 */ 2920 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2921 { 2922 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2923 return; 2924 WRITE_ONCE(sk->sk_pacing_shift, val); 2925 } 2926 2927 /* if a socket is bound to a device, check that the given device 2928 * index is either the same or that the socket is bound to an L3 2929 * master device and the given device index is also enslaved to 2930 * that L3 master 2931 */ 2932 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2933 { 2934 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2935 int mdif; 2936 2937 if (!bound_dev_if || bound_dev_if == dif) 2938 return true; 2939 2940 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2941 if (mdif && mdif == bound_dev_if) 2942 return true; 2943 2944 return false; 2945 } 2946 2947 void sock_def_readable(struct sock *sk); 2948 2949 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2950 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2951 int sock_set_timestamping(struct sock *sk, int optname, 2952 struct so_timestamping timestamping); 2953 2954 void sock_enable_timestamps(struct sock *sk); 2955 void sock_no_linger(struct sock *sk); 2956 void sock_set_keepalive(struct sock *sk); 2957 void sock_set_priority(struct sock *sk, u32 priority); 2958 void sock_set_rcvbuf(struct sock *sk, int val); 2959 void sock_set_mark(struct sock *sk, u32 val); 2960 void sock_set_reuseaddr(struct sock *sk); 2961 void sock_set_reuseport(struct sock *sk); 2962 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2963 2964 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2965 2966 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2967 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2968 sockptr_t optval, int optlen, bool old_timeval); 2969 2970 static inline bool sk_is_readable(struct sock *sk) 2971 { 2972 if (sk->sk_prot->sock_is_readable) 2973 return sk->sk_prot->sock_is_readable(sk); 2974 return false; 2975 } 2976 #endif /* _SOCK_H */ 2977