1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 union { 165 __addrpair skc_addrpair; 166 struct { 167 __be32 skc_daddr; 168 __be32 skc_rcv_saddr; 169 }; 170 }; 171 union { 172 unsigned int skc_hash; 173 __u16 skc_u16hashes[2]; 174 }; 175 /* skc_dport && skc_num must be grouped as well */ 176 union { 177 __portpair skc_portpair; 178 struct { 179 __be16 skc_dport; 180 __u16 skc_num; 181 }; 182 }; 183 184 unsigned short skc_family; 185 volatile unsigned char skc_state; 186 unsigned char skc_reuse:4; 187 unsigned char skc_reuseport:1; 188 unsigned char skc_ipv6only:1; 189 unsigned char skc_net_refcnt:1; 190 int skc_bound_dev_if; 191 union { 192 struct hlist_node skc_bind_node; 193 struct hlist_node skc_portaddr_node; 194 }; 195 struct proto *skc_prot; 196 possible_net_t skc_net; 197 198 #if IS_ENABLED(CONFIG_IPV6) 199 struct in6_addr skc_v6_daddr; 200 struct in6_addr skc_v6_rcv_saddr; 201 #endif 202 203 atomic64_t skc_cookie; 204 205 /* following fields are padding to force 206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 207 * assuming IPV6 is enabled. We use this padding differently 208 * for different kind of 'sockets' 209 */ 210 union { 211 unsigned long skc_flags; 212 struct sock *skc_listener; /* request_sock */ 213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 214 }; 215 /* 216 * fields between dontcopy_begin/dontcopy_end 217 * are not copied in sock_copy() 218 */ 219 /* private: */ 220 int skc_dontcopy_begin[0]; 221 /* public: */ 222 union { 223 struct hlist_node skc_node; 224 struct hlist_nulls_node skc_nulls_node; 225 }; 226 unsigned short skc_tx_queue_mapping; 227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 228 unsigned short skc_rx_queue_mapping; 229 #endif 230 union { 231 int skc_incoming_cpu; 232 u32 skc_rcv_wnd; 233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 234 }; 235 236 refcount_t skc_refcnt; 237 /* private: */ 238 int skc_dontcopy_end[0]; 239 union { 240 u32 skc_rxhash; 241 u32 skc_window_clamp; 242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 243 }; 244 /* public: */ 245 }; 246 247 struct bpf_local_storage; 248 struct sk_filter; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 304 * @sk_busy_poll_budget: napi processing budget when busypolling 305 * @sk_priority: %SO_PRIORITY setting 306 * @sk_type: socket type (%SOCK_STREAM, etc) 307 * @sk_protocol: which protocol this socket belongs in this network family 308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_txrehash: enable TX hash rethink 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 322 * Sockets that can be used under memory reclaim should 323 * set this to false. 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_wait_pending: number of threads blocked on this socket 340 * @sk_state_change: callback to indicate change in the state of the sock 341 * @sk_data_ready: callback to indicate there is data to be processed 342 * @sk_write_space: callback to indicate there is bf sending space available 343 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 344 * @sk_backlog_rcv: callback to process the backlog 345 * @sk_validate_xmit_skb: ptr to an optional validate function 346 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 347 * @sk_reuseport_cb: reuseport group container 348 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 349 * @sk_rcu: used during RCU grace period 350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 353 * @sk_txtime_unused: unused txtime flags 354 * @ns_tracker: tracker for netns reference 355 * @sk_bind2_node: bind node in the bhash2 table 356 */ 357 struct sock { 358 /* 359 * Now struct inet_timewait_sock also uses sock_common, so please just 360 * don't add nothing before this first member (__sk_common) --acme 361 */ 362 struct sock_common __sk_common; 363 #define sk_node __sk_common.skc_node 364 #define sk_nulls_node __sk_common.skc_nulls_node 365 #define sk_refcnt __sk_common.skc_refcnt 366 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 367 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 368 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 369 #endif 370 371 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 372 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 373 #define sk_hash __sk_common.skc_hash 374 #define sk_portpair __sk_common.skc_portpair 375 #define sk_num __sk_common.skc_num 376 #define sk_dport __sk_common.skc_dport 377 #define sk_addrpair __sk_common.skc_addrpair 378 #define sk_daddr __sk_common.skc_daddr 379 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 380 #define sk_family __sk_common.skc_family 381 #define sk_state __sk_common.skc_state 382 #define sk_reuse __sk_common.skc_reuse 383 #define sk_reuseport __sk_common.skc_reuseport 384 #define sk_ipv6only __sk_common.skc_ipv6only 385 #define sk_net_refcnt __sk_common.skc_net_refcnt 386 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 387 #define sk_bind_node __sk_common.skc_bind_node 388 #define sk_prot __sk_common.skc_prot 389 #define sk_net __sk_common.skc_net 390 #define sk_v6_daddr __sk_common.skc_v6_daddr 391 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 392 #define sk_cookie __sk_common.skc_cookie 393 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 394 #define sk_flags __sk_common.skc_flags 395 #define sk_rxhash __sk_common.skc_rxhash 396 397 /* early demux fields */ 398 struct dst_entry __rcu *sk_rx_dst; 399 int sk_rx_dst_ifindex; 400 u32 sk_rx_dst_cookie; 401 402 socket_lock_t sk_lock; 403 atomic_t sk_drops; 404 int sk_rcvlowat; 405 struct sk_buff_head sk_error_queue; 406 struct sk_buff_head sk_receive_queue; 407 /* 408 * The backlog queue is special, it is always used with 409 * the per-socket spinlock held and requires low latency 410 * access. Therefore we special case it's implementation. 411 * Note : rmem_alloc is in this structure to fill a hole 412 * on 64bit arches, not because its logically part of 413 * backlog. 414 */ 415 struct { 416 atomic_t rmem_alloc; 417 int len; 418 struct sk_buff *head; 419 struct sk_buff *tail; 420 } sk_backlog; 421 422 #define sk_rmem_alloc sk_backlog.rmem_alloc 423 424 int sk_forward_alloc; 425 u32 sk_reserved_mem; 426 #ifdef CONFIG_NET_RX_BUSY_POLL 427 unsigned int sk_ll_usec; 428 /* ===== mostly read cache line ===== */ 429 unsigned int sk_napi_id; 430 #endif 431 int sk_rcvbuf; 432 int sk_wait_pending; 433 434 struct sk_filter __rcu *sk_filter; 435 union { 436 struct socket_wq __rcu *sk_wq; 437 /* private: */ 438 struct socket_wq *sk_wq_raw; 439 /* public: */ 440 }; 441 #ifdef CONFIG_XFRM 442 struct xfrm_policy __rcu *sk_policy[2]; 443 #endif 444 445 struct dst_entry __rcu *sk_dst_cache; 446 atomic_t sk_omem_alloc; 447 int sk_sndbuf; 448 449 /* ===== cache line for TX ===== */ 450 int sk_wmem_queued; 451 refcount_t sk_wmem_alloc; 452 unsigned long sk_tsq_flags; 453 union { 454 struct sk_buff *sk_send_head; 455 struct rb_root tcp_rtx_queue; 456 }; 457 struct sk_buff_head sk_write_queue; 458 __s32 sk_peek_off; 459 int sk_write_pending; 460 __u32 sk_dst_pending_confirm; 461 u32 sk_pacing_status; /* see enum sk_pacing */ 462 long sk_sndtimeo; 463 struct timer_list sk_timer; 464 __u32 sk_priority; 465 __u32 sk_mark; 466 unsigned long sk_pacing_rate; /* bytes per second */ 467 unsigned long sk_max_pacing_rate; 468 struct page_frag sk_frag; 469 netdev_features_t sk_route_caps; 470 int sk_gso_type; 471 unsigned int sk_gso_max_size; 472 gfp_t sk_allocation; 473 __u32 sk_txhash; 474 475 /* 476 * Because of non atomicity rules, all 477 * changes are protected by socket lock. 478 */ 479 u8 sk_gso_disabled : 1, 480 sk_kern_sock : 1, 481 sk_no_check_tx : 1, 482 sk_no_check_rx : 1, 483 sk_userlocks : 4; 484 u8 sk_pacing_shift; 485 u16 sk_type; 486 u16 sk_protocol; 487 u16 sk_gso_max_segs; 488 unsigned long sk_lingertime; 489 struct proto *sk_prot_creator; 490 rwlock_t sk_callback_lock; 491 int sk_err, 492 sk_err_soft; 493 u32 sk_ack_backlog; 494 u32 sk_max_ack_backlog; 495 kuid_t sk_uid; 496 u8 sk_txrehash; 497 #ifdef CONFIG_NET_RX_BUSY_POLL 498 u8 sk_prefer_busy_poll; 499 u16 sk_busy_poll_budget; 500 #endif 501 spinlock_t sk_peer_lock; 502 int sk_bind_phc; 503 struct pid *sk_peer_pid; 504 const struct cred *sk_peer_cred; 505 506 long sk_rcvtimeo; 507 ktime_t sk_stamp; 508 #if BITS_PER_LONG==32 509 seqlock_t sk_stamp_seq; 510 #endif 511 atomic_t sk_tskey; 512 atomic_t sk_zckey; 513 u32 sk_tsflags; 514 u8 sk_shutdown; 515 516 u8 sk_clockid; 517 u8 sk_txtime_deadline_mode : 1, 518 sk_txtime_report_errors : 1, 519 sk_txtime_unused : 6; 520 bool sk_use_task_frag; 521 522 struct socket *sk_socket; 523 void *sk_user_data; 524 #ifdef CONFIG_SECURITY 525 void *sk_security; 526 #endif 527 struct sock_cgroup_data sk_cgrp_data; 528 struct mem_cgroup *sk_memcg; 529 void (*sk_state_change)(struct sock *sk); 530 void (*sk_data_ready)(struct sock *sk); 531 void (*sk_write_space)(struct sock *sk); 532 void (*sk_error_report)(struct sock *sk); 533 int (*sk_backlog_rcv)(struct sock *sk, 534 struct sk_buff *skb); 535 #ifdef CONFIG_SOCK_VALIDATE_XMIT 536 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 537 struct net_device *dev, 538 struct sk_buff *skb); 539 #endif 540 void (*sk_destruct)(struct sock *sk); 541 struct sock_reuseport __rcu *sk_reuseport_cb; 542 #ifdef CONFIG_BPF_SYSCALL 543 struct bpf_local_storage __rcu *sk_bpf_storage; 544 #endif 545 struct rcu_head sk_rcu; 546 netns_tracker ns_tracker; 547 struct hlist_node sk_bind2_node; 548 }; 549 550 enum sk_pacing { 551 SK_PACING_NONE = 0, 552 SK_PACING_NEEDED = 1, 553 SK_PACING_FQ = 2, 554 }; 555 556 /* flag bits in sk_user_data 557 * 558 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 559 * not be suitable for copying when cloning the socket. For instance, 560 * it can point to a reference counted object. sk_user_data bottom 561 * bit is set if pointer must not be copied. 562 * 563 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 564 * managed/owned by a BPF reuseport array. This bit should be set 565 * when sk_user_data's sk is added to the bpf's reuseport_array. 566 * 567 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 568 * sk_user_data points to psock type. This bit should be set 569 * when sk_user_data is assigned to a psock object. 570 */ 571 #define SK_USER_DATA_NOCOPY 1UL 572 #define SK_USER_DATA_BPF 2UL 573 #define SK_USER_DATA_PSOCK 4UL 574 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 575 SK_USER_DATA_PSOCK) 576 577 /** 578 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 579 * @sk: socket 580 */ 581 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 582 { 583 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 584 } 585 586 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 587 588 /** 589 * __locked_read_sk_user_data_with_flags - return the pointer 590 * only if argument flags all has been set in sk_user_data. Otherwise 591 * return NULL 592 * 593 * @sk: socket 594 * @flags: flag bits 595 * 596 * The caller must be holding sk->sk_callback_lock. 597 */ 598 static inline void * 599 __locked_read_sk_user_data_with_flags(const struct sock *sk, 600 uintptr_t flags) 601 { 602 uintptr_t sk_user_data = 603 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 604 lockdep_is_held(&sk->sk_callback_lock)); 605 606 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 607 608 if ((sk_user_data & flags) == flags) 609 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 610 return NULL; 611 } 612 613 /** 614 * __rcu_dereference_sk_user_data_with_flags - return the pointer 615 * only if argument flags all has been set in sk_user_data. Otherwise 616 * return NULL 617 * 618 * @sk: socket 619 * @flags: flag bits 620 */ 621 static inline void * 622 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 623 uintptr_t flags) 624 { 625 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 626 627 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 628 629 if ((sk_user_data & flags) == flags) 630 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 631 return NULL; 632 } 633 634 #define rcu_dereference_sk_user_data(sk) \ 635 __rcu_dereference_sk_user_data_with_flags(sk, 0) 636 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 637 ({ \ 638 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 639 __tmp2 = (uintptr_t)(flags); \ 640 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 641 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 642 rcu_assign_pointer(__sk_user_data((sk)), \ 643 __tmp1 | __tmp2); \ 644 }) 645 #define rcu_assign_sk_user_data(sk, ptr) \ 646 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 647 648 static inline 649 struct net *sock_net(const struct sock *sk) 650 { 651 return read_pnet(&sk->sk_net); 652 } 653 654 static inline 655 void sock_net_set(struct sock *sk, struct net *net) 656 { 657 write_pnet(&sk->sk_net, net); 658 } 659 660 /* 661 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 662 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 663 * on a socket means that the socket will reuse everybody else's port 664 * without looking at the other's sk_reuse value. 665 */ 666 667 #define SK_NO_REUSE 0 668 #define SK_CAN_REUSE 1 669 #define SK_FORCE_REUSE 2 670 671 int sk_set_peek_off(struct sock *sk, int val); 672 673 static inline int sk_peek_offset(const struct sock *sk, int flags) 674 { 675 if (unlikely(flags & MSG_PEEK)) { 676 return READ_ONCE(sk->sk_peek_off); 677 } 678 679 return 0; 680 } 681 682 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 683 { 684 s32 off = READ_ONCE(sk->sk_peek_off); 685 686 if (unlikely(off >= 0)) { 687 off = max_t(s32, off - val, 0); 688 WRITE_ONCE(sk->sk_peek_off, off); 689 } 690 } 691 692 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 693 { 694 sk_peek_offset_bwd(sk, -val); 695 } 696 697 /* 698 * Hashed lists helper routines 699 */ 700 static inline struct sock *sk_entry(const struct hlist_node *node) 701 { 702 return hlist_entry(node, struct sock, sk_node); 703 } 704 705 static inline struct sock *__sk_head(const struct hlist_head *head) 706 { 707 return hlist_entry(head->first, struct sock, sk_node); 708 } 709 710 static inline struct sock *sk_head(const struct hlist_head *head) 711 { 712 return hlist_empty(head) ? NULL : __sk_head(head); 713 } 714 715 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 716 { 717 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 718 } 719 720 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 721 { 722 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 723 } 724 725 static inline struct sock *sk_next(const struct sock *sk) 726 { 727 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 728 } 729 730 static inline struct sock *sk_nulls_next(const struct sock *sk) 731 { 732 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 733 hlist_nulls_entry(sk->sk_nulls_node.next, 734 struct sock, sk_nulls_node) : 735 NULL; 736 } 737 738 static inline bool sk_unhashed(const struct sock *sk) 739 { 740 return hlist_unhashed(&sk->sk_node); 741 } 742 743 static inline bool sk_hashed(const struct sock *sk) 744 { 745 return !sk_unhashed(sk); 746 } 747 748 static inline void sk_node_init(struct hlist_node *node) 749 { 750 node->pprev = NULL; 751 } 752 753 static inline void __sk_del_node(struct sock *sk) 754 { 755 __hlist_del(&sk->sk_node); 756 } 757 758 /* NB: equivalent to hlist_del_init_rcu */ 759 static inline bool __sk_del_node_init(struct sock *sk) 760 { 761 if (sk_hashed(sk)) { 762 __sk_del_node(sk); 763 sk_node_init(&sk->sk_node); 764 return true; 765 } 766 return false; 767 } 768 769 /* Grab socket reference count. This operation is valid only 770 when sk is ALREADY grabbed f.e. it is found in hash table 771 or a list and the lookup is made under lock preventing hash table 772 modifications. 773 */ 774 775 static __always_inline void sock_hold(struct sock *sk) 776 { 777 refcount_inc(&sk->sk_refcnt); 778 } 779 780 /* Ungrab socket in the context, which assumes that socket refcnt 781 cannot hit zero, f.e. it is true in context of any socketcall. 782 */ 783 static __always_inline void __sock_put(struct sock *sk) 784 { 785 refcount_dec(&sk->sk_refcnt); 786 } 787 788 static inline bool sk_del_node_init(struct sock *sk) 789 { 790 bool rc = __sk_del_node_init(sk); 791 792 if (rc) { 793 /* paranoid for a while -acme */ 794 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 795 __sock_put(sk); 796 } 797 return rc; 798 } 799 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 800 801 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 802 { 803 if (sk_hashed(sk)) { 804 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 805 return true; 806 } 807 return false; 808 } 809 810 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 811 { 812 bool rc = __sk_nulls_del_node_init_rcu(sk); 813 814 if (rc) { 815 /* paranoid for a while -acme */ 816 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 817 __sock_put(sk); 818 } 819 return rc; 820 } 821 822 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 823 { 824 hlist_add_head(&sk->sk_node, list); 825 } 826 827 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 828 { 829 sock_hold(sk); 830 __sk_add_node(sk, list); 831 } 832 833 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 834 { 835 sock_hold(sk); 836 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 837 sk->sk_family == AF_INET6) 838 hlist_add_tail_rcu(&sk->sk_node, list); 839 else 840 hlist_add_head_rcu(&sk->sk_node, list); 841 } 842 843 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 844 { 845 sock_hold(sk); 846 hlist_add_tail_rcu(&sk->sk_node, list); 847 } 848 849 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 850 { 851 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 852 } 853 854 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 855 { 856 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 857 } 858 859 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 860 { 861 sock_hold(sk); 862 __sk_nulls_add_node_rcu(sk, list); 863 } 864 865 static inline void __sk_del_bind_node(struct sock *sk) 866 { 867 __hlist_del(&sk->sk_bind_node); 868 } 869 870 static inline void sk_add_bind_node(struct sock *sk, 871 struct hlist_head *list) 872 { 873 hlist_add_head(&sk->sk_bind_node, list); 874 } 875 876 static inline void __sk_del_bind2_node(struct sock *sk) 877 { 878 __hlist_del(&sk->sk_bind2_node); 879 } 880 881 static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list) 882 { 883 hlist_add_head(&sk->sk_bind2_node, list); 884 } 885 886 #define sk_for_each(__sk, list) \ 887 hlist_for_each_entry(__sk, list, sk_node) 888 #define sk_for_each_rcu(__sk, list) \ 889 hlist_for_each_entry_rcu(__sk, list, sk_node) 890 #define sk_nulls_for_each(__sk, node, list) \ 891 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 892 #define sk_nulls_for_each_rcu(__sk, node, list) \ 893 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 894 #define sk_for_each_from(__sk) \ 895 hlist_for_each_entry_from(__sk, sk_node) 896 #define sk_nulls_for_each_from(__sk, node) \ 897 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 898 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 899 #define sk_for_each_safe(__sk, tmp, list) \ 900 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 901 #define sk_for_each_bound(__sk, list) \ 902 hlist_for_each_entry(__sk, list, sk_bind_node) 903 #define sk_for_each_bound_bhash2(__sk, list) \ 904 hlist_for_each_entry(__sk, list, sk_bind2_node) 905 906 /** 907 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 908 * @tpos: the type * to use as a loop cursor. 909 * @pos: the &struct hlist_node to use as a loop cursor. 910 * @head: the head for your list. 911 * @offset: offset of hlist_node within the struct. 912 * 913 */ 914 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 915 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 916 pos != NULL && \ 917 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 918 pos = rcu_dereference(hlist_next_rcu(pos))) 919 920 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 921 { 922 /* Careful only use this in a context where these parameters 923 * can not change and must all be valid, such as recvmsg from 924 * userspace. 925 */ 926 return sk->sk_socket->file->f_cred->user_ns; 927 } 928 929 /* Sock flags */ 930 enum sock_flags { 931 SOCK_DEAD, 932 SOCK_DONE, 933 SOCK_URGINLINE, 934 SOCK_KEEPOPEN, 935 SOCK_LINGER, 936 SOCK_DESTROY, 937 SOCK_BROADCAST, 938 SOCK_TIMESTAMP, 939 SOCK_ZAPPED, 940 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 941 SOCK_DBG, /* %SO_DEBUG setting */ 942 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 943 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 944 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 945 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 946 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 947 SOCK_FASYNC, /* fasync() active */ 948 SOCK_RXQ_OVFL, 949 SOCK_ZEROCOPY, /* buffers from userspace */ 950 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 951 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 952 * Will use last 4 bytes of packet sent from 953 * user-space instead. 954 */ 955 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 956 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 957 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 958 SOCK_TXTIME, 959 SOCK_XDP, /* XDP is attached */ 960 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 961 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 962 }; 963 964 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 965 966 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 967 { 968 nsk->sk_flags = osk->sk_flags; 969 } 970 971 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 972 { 973 __set_bit(flag, &sk->sk_flags); 974 } 975 976 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 977 { 978 __clear_bit(flag, &sk->sk_flags); 979 } 980 981 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 982 int valbool) 983 { 984 if (valbool) 985 sock_set_flag(sk, bit); 986 else 987 sock_reset_flag(sk, bit); 988 } 989 990 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 991 { 992 return test_bit(flag, &sk->sk_flags); 993 } 994 995 #ifdef CONFIG_NET 996 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 997 static inline int sk_memalloc_socks(void) 998 { 999 return static_branch_unlikely(&memalloc_socks_key); 1000 } 1001 1002 void __receive_sock(struct file *file); 1003 #else 1004 1005 static inline int sk_memalloc_socks(void) 1006 { 1007 return 0; 1008 } 1009 1010 static inline void __receive_sock(struct file *file) 1011 { } 1012 #endif 1013 1014 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1015 { 1016 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1017 } 1018 1019 static inline void sk_acceptq_removed(struct sock *sk) 1020 { 1021 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1022 } 1023 1024 static inline void sk_acceptq_added(struct sock *sk) 1025 { 1026 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1027 } 1028 1029 /* Note: If you think the test should be: 1030 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1031 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1032 */ 1033 static inline bool sk_acceptq_is_full(const struct sock *sk) 1034 { 1035 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1036 } 1037 1038 /* 1039 * Compute minimal free write space needed to queue new packets. 1040 */ 1041 static inline int sk_stream_min_wspace(const struct sock *sk) 1042 { 1043 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1044 } 1045 1046 static inline int sk_stream_wspace(const struct sock *sk) 1047 { 1048 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1049 } 1050 1051 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1052 { 1053 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1054 } 1055 1056 void sk_stream_write_space(struct sock *sk); 1057 1058 /* OOB backlog add */ 1059 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1060 { 1061 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1062 skb_dst_force(skb); 1063 1064 if (!sk->sk_backlog.tail) 1065 WRITE_ONCE(sk->sk_backlog.head, skb); 1066 else 1067 sk->sk_backlog.tail->next = skb; 1068 1069 WRITE_ONCE(sk->sk_backlog.tail, skb); 1070 skb->next = NULL; 1071 } 1072 1073 /* 1074 * Take into account size of receive queue and backlog queue 1075 * Do not take into account this skb truesize, 1076 * to allow even a single big packet to come. 1077 */ 1078 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1079 { 1080 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1081 1082 return qsize > limit; 1083 } 1084 1085 /* The per-socket spinlock must be held here. */ 1086 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1087 unsigned int limit) 1088 { 1089 if (sk_rcvqueues_full(sk, limit)) 1090 return -ENOBUFS; 1091 1092 /* 1093 * If the skb was allocated from pfmemalloc reserves, only 1094 * allow SOCK_MEMALLOC sockets to use it as this socket is 1095 * helping free memory 1096 */ 1097 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1098 return -ENOMEM; 1099 1100 __sk_add_backlog(sk, skb); 1101 sk->sk_backlog.len += skb->truesize; 1102 return 0; 1103 } 1104 1105 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1106 1107 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1108 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1109 1110 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1111 { 1112 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1113 return __sk_backlog_rcv(sk, skb); 1114 1115 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1116 tcp_v6_do_rcv, 1117 tcp_v4_do_rcv, 1118 sk, skb); 1119 } 1120 1121 static inline void sk_incoming_cpu_update(struct sock *sk) 1122 { 1123 int cpu = raw_smp_processor_id(); 1124 1125 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1126 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1127 } 1128 1129 static inline void sock_rps_record_flow_hash(__u32 hash) 1130 { 1131 #ifdef CONFIG_RPS 1132 struct rps_sock_flow_table *sock_flow_table; 1133 1134 rcu_read_lock(); 1135 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1136 rps_record_sock_flow(sock_flow_table, hash); 1137 rcu_read_unlock(); 1138 #endif 1139 } 1140 1141 static inline void sock_rps_record_flow(const struct sock *sk) 1142 { 1143 #ifdef CONFIG_RPS 1144 if (static_branch_unlikely(&rfs_needed)) { 1145 /* Reading sk->sk_rxhash might incur an expensive cache line 1146 * miss. 1147 * 1148 * TCP_ESTABLISHED does cover almost all states where RFS 1149 * might be useful, and is cheaper [1] than testing : 1150 * IPv4: inet_sk(sk)->inet_daddr 1151 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1152 * OR an additional socket flag 1153 * [1] : sk_state and sk_prot are in the same cache line. 1154 */ 1155 if (sk->sk_state == TCP_ESTABLISHED) { 1156 /* This READ_ONCE() is paired with the WRITE_ONCE() 1157 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash(). 1158 */ 1159 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash)); 1160 } 1161 } 1162 #endif 1163 } 1164 1165 static inline void sock_rps_save_rxhash(struct sock *sk, 1166 const struct sk_buff *skb) 1167 { 1168 #ifdef CONFIG_RPS 1169 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1170 * here, and another one in sock_rps_record_flow(). 1171 */ 1172 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1173 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1174 #endif 1175 } 1176 1177 static inline void sock_rps_reset_rxhash(struct sock *sk) 1178 { 1179 #ifdef CONFIG_RPS 1180 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1181 WRITE_ONCE(sk->sk_rxhash, 0); 1182 #endif 1183 } 1184 1185 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1186 ({ int __rc; \ 1187 __sk->sk_wait_pending++; \ 1188 release_sock(__sk); \ 1189 __rc = __condition; \ 1190 if (!__rc) { \ 1191 *(__timeo) = wait_woken(__wait, \ 1192 TASK_INTERRUPTIBLE, \ 1193 *(__timeo)); \ 1194 } \ 1195 sched_annotate_sleep(); \ 1196 lock_sock(__sk); \ 1197 __sk->sk_wait_pending--; \ 1198 __rc = __condition; \ 1199 __rc; \ 1200 }) 1201 1202 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1203 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1204 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1205 int sk_stream_error(struct sock *sk, int flags, int err); 1206 void sk_stream_kill_queues(struct sock *sk); 1207 void sk_set_memalloc(struct sock *sk); 1208 void sk_clear_memalloc(struct sock *sk); 1209 1210 void __sk_flush_backlog(struct sock *sk); 1211 1212 static inline bool sk_flush_backlog(struct sock *sk) 1213 { 1214 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1215 __sk_flush_backlog(sk); 1216 return true; 1217 } 1218 return false; 1219 } 1220 1221 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1222 1223 struct request_sock_ops; 1224 struct timewait_sock_ops; 1225 struct inet_hashinfo; 1226 struct raw_hashinfo; 1227 struct smc_hashinfo; 1228 struct module; 1229 struct sk_psock; 1230 1231 /* 1232 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1233 * un-modified. Special care is taken when initializing object to zero. 1234 */ 1235 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1236 { 1237 if (offsetof(struct sock, sk_node.next) != 0) 1238 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1239 memset(&sk->sk_node.pprev, 0, 1240 size - offsetof(struct sock, sk_node.pprev)); 1241 } 1242 1243 /* Networking protocol blocks we attach to sockets. 1244 * socket layer -> transport layer interface 1245 */ 1246 struct proto { 1247 void (*close)(struct sock *sk, 1248 long timeout); 1249 int (*pre_connect)(struct sock *sk, 1250 struct sockaddr *uaddr, 1251 int addr_len); 1252 int (*connect)(struct sock *sk, 1253 struct sockaddr *uaddr, 1254 int addr_len); 1255 int (*disconnect)(struct sock *sk, int flags); 1256 1257 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1258 bool kern); 1259 1260 int (*ioctl)(struct sock *sk, int cmd, 1261 int *karg); 1262 int (*init)(struct sock *sk); 1263 void (*destroy)(struct sock *sk); 1264 void (*shutdown)(struct sock *sk, int how); 1265 int (*setsockopt)(struct sock *sk, int level, 1266 int optname, sockptr_t optval, 1267 unsigned int optlen); 1268 int (*getsockopt)(struct sock *sk, int level, 1269 int optname, char __user *optval, 1270 int __user *option); 1271 void (*keepalive)(struct sock *sk, int valbool); 1272 #ifdef CONFIG_COMPAT 1273 int (*compat_ioctl)(struct sock *sk, 1274 unsigned int cmd, unsigned long arg); 1275 #endif 1276 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1277 size_t len); 1278 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1279 size_t len, int flags, int *addr_len); 1280 void (*splice_eof)(struct socket *sock); 1281 int (*bind)(struct sock *sk, 1282 struct sockaddr *addr, int addr_len); 1283 int (*bind_add)(struct sock *sk, 1284 struct sockaddr *addr, int addr_len); 1285 1286 int (*backlog_rcv) (struct sock *sk, 1287 struct sk_buff *skb); 1288 bool (*bpf_bypass_getsockopt)(int level, 1289 int optname); 1290 1291 void (*release_cb)(struct sock *sk); 1292 1293 /* Keeping track of sk's, looking them up, and port selection methods. */ 1294 int (*hash)(struct sock *sk); 1295 void (*unhash)(struct sock *sk); 1296 void (*rehash)(struct sock *sk); 1297 int (*get_port)(struct sock *sk, unsigned short snum); 1298 void (*put_port)(struct sock *sk); 1299 #ifdef CONFIG_BPF_SYSCALL 1300 int (*psock_update_sk_prot)(struct sock *sk, 1301 struct sk_psock *psock, 1302 bool restore); 1303 #endif 1304 1305 /* Keeping track of sockets in use */ 1306 #ifdef CONFIG_PROC_FS 1307 unsigned int inuse_idx; 1308 #endif 1309 1310 #if IS_ENABLED(CONFIG_MPTCP) 1311 int (*forward_alloc_get)(const struct sock *sk); 1312 #endif 1313 1314 bool (*stream_memory_free)(const struct sock *sk, int wake); 1315 bool (*sock_is_readable)(struct sock *sk); 1316 /* Memory pressure */ 1317 void (*enter_memory_pressure)(struct sock *sk); 1318 void (*leave_memory_pressure)(struct sock *sk); 1319 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1320 int __percpu *per_cpu_fw_alloc; 1321 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1322 1323 /* 1324 * Pressure flag: try to collapse. 1325 * Technical note: it is used by multiple contexts non atomically. 1326 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1327 * All the __sk_mem_schedule() is of this nature: accounting 1328 * is strict, actions are advisory and have some latency. 1329 */ 1330 unsigned long *memory_pressure; 1331 long *sysctl_mem; 1332 1333 int *sysctl_wmem; 1334 int *sysctl_rmem; 1335 u32 sysctl_wmem_offset; 1336 u32 sysctl_rmem_offset; 1337 1338 int max_header; 1339 bool no_autobind; 1340 1341 struct kmem_cache *slab; 1342 unsigned int obj_size; 1343 unsigned int ipv6_pinfo_offset; 1344 slab_flags_t slab_flags; 1345 unsigned int useroffset; /* Usercopy region offset */ 1346 unsigned int usersize; /* Usercopy region size */ 1347 1348 unsigned int __percpu *orphan_count; 1349 1350 struct request_sock_ops *rsk_prot; 1351 struct timewait_sock_ops *twsk_prot; 1352 1353 union { 1354 struct inet_hashinfo *hashinfo; 1355 struct udp_table *udp_table; 1356 struct raw_hashinfo *raw_hash; 1357 struct smc_hashinfo *smc_hash; 1358 } h; 1359 1360 struct module *owner; 1361 1362 char name[32]; 1363 1364 struct list_head node; 1365 int (*diag_destroy)(struct sock *sk, int err); 1366 } __randomize_layout; 1367 1368 int proto_register(struct proto *prot, int alloc_slab); 1369 void proto_unregister(struct proto *prot); 1370 int sock_load_diag_module(int family, int protocol); 1371 1372 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1373 1374 static inline int sk_forward_alloc_get(const struct sock *sk) 1375 { 1376 #if IS_ENABLED(CONFIG_MPTCP) 1377 if (sk->sk_prot->forward_alloc_get) 1378 return sk->sk_prot->forward_alloc_get(sk); 1379 #endif 1380 return sk->sk_forward_alloc; 1381 } 1382 1383 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1384 { 1385 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1386 return false; 1387 1388 return sk->sk_prot->stream_memory_free ? 1389 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1390 tcp_stream_memory_free, sk, wake) : true; 1391 } 1392 1393 static inline bool sk_stream_memory_free(const struct sock *sk) 1394 { 1395 return __sk_stream_memory_free(sk, 0); 1396 } 1397 1398 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1399 { 1400 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1401 __sk_stream_memory_free(sk, wake); 1402 } 1403 1404 static inline bool sk_stream_is_writeable(const struct sock *sk) 1405 { 1406 return __sk_stream_is_writeable(sk, 0); 1407 } 1408 1409 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1410 struct cgroup *ancestor) 1411 { 1412 #ifdef CONFIG_SOCK_CGROUP_DATA 1413 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1414 ancestor); 1415 #else 1416 return -ENOTSUPP; 1417 #endif 1418 } 1419 1420 static inline bool sk_has_memory_pressure(const struct sock *sk) 1421 { 1422 return sk->sk_prot->memory_pressure != NULL; 1423 } 1424 1425 static inline bool sk_under_global_memory_pressure(const struct sock *sk) 1426 { 1427 return sk->sk_prot->memory_pressure && 1428 !!READ_ONCE(*sk->sk_prot->memory_pressure); 1429 } 1430 1431 static inline bool sk_under_memory_pressure(const struct sock *sk) 1432 { 1433 if (!sk->sk_prot->memory_pressure) 1434 return false; 1435 1436 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1437 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1438 return true; 1439 1440 return !!READ_ONCE(*sk->sk_prot->memory_pressure); 1441 } 1442 1443 static inline long 1444 proto_memory_allocated(const struct proto *prot) 1445 { 1446 return max(0L, atomic_long_read(prot->memory_allocated)); 1447 } 1448 1449 static inline long 1450 sk_memory_allocated(const struct sock *sk) 1451 { 1452 return proto_memory_allocated(sk->sk_prot); 1453 } 1454 1455 /* 1 MB per cpu, in page units */ 1456 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1457 1458 static inline void 1459 sk_memory_allocated_add(struct sock *sk, int amt) 1460 { 1461 int local_reserve; 1462 1463 preempt_disable(); 1464 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1465 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) { 1466 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1467 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1468 } 1469 preempt_enable(); 1470 } 1471 1472 static inline void 1473 sk_memory_allocated_sub(struct sock *sk, int amt) 1474 { 1475 int local_reserve; 1476 1477 preempt_disable(); 1478 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt); 1479 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) { 1480 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve); 1481 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated); 1482 } 1483 preempt_enable(); 1484 } 1485 1486 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1487 1488 static inline void sk_sockets_allocated_dec(struct sock *sk) 1489 { 1490 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1491 SK_ALLOC_PERCPU_COUNTER_BATCH); 1492 } 1493 1494 static inline void sk_sockets_allocated_inc(struct sock *sk) 1495 { 1496 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1497 SK_ALLOC_PERCPU_COUNTER_BATCH); 1498 } 1499 1500 static inline u64 1501 sk_sockets_allocated_read_positive(struct sock *sk) 1502 { 1503 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1504 } 1505 1506 static inline int 1507 proto_sockets_allocated_sum_positive(struct proto *prot) 1508 { 1509 return percpu_counter_sum_positive(prot->sockets_allocated); 1510 } 1511 1512 static inline bool 1513 proto_memory_pressure(struct proto *prot) 1514 { 1515 if (!prot->memory_pressure) 1516 return false; 1517 return !!READ_ONCE(*prot->memory_pressure); 1518 } 1519 1520 1521 #ifdef CONFIG_PROC_FS 1522 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1523 struct prot_inuse { 1524 int all; 1525 int val[PROTO_INUSE_NR]; 1526 }; 1527 1528 static inline void sock_prot_inuse_add(const struct net *net, 1529 const struct proto *prot, int val) 1530 { 1531 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1532 } 1533 1534 static inline void sock_inuse_add(const struct net *net, int val) 1535 { 1536 this_cpu_add(net->core.prot_inuse->all, val); 1537 } 1538 1539 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1540 int sock_inuse_get(struct net *net); 1541 #else 1542 static inline void sock_prot_inuse_add(const struct net *net, 1543 const struct proto *prot, int val) 1544 { 1545 } 1546 1547 static inline void sock_inuse_add(const struct net *net, int val) 1548 { 1549 } 1550 #endif 1551 1552 1553 /* With per-bucket locks this operation is not-atomic, so that 1554 * this version is not worse. 1555 */ 1556 static inline int __sk_prot_rehash(struct sock *sk) 1557 { 1558 sk->sk_prot->unhash(sk); 1559 return sk->sk_prot->hash(sk); 1560 } 1561 1562 /* About 10 seconds */ 1563 #define SOCK_DESTROY_TIME (10*HZ) 1564 1565 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1566 #define PROT_SOCK 1024 1567 1568 #define SHUTDOWN_MASK 3 1569 #define RCV_SHUTDOWN 1 1570 #define SEND_SHUTDOWN 2 1571 1572 #define SOCK_BINDADDR_LOCK 4 1573 #define SOCK_BINDPORT_LOCK 8 1574 1575 struct socket_alloc { 1576 struct socket socket; 1577 struct inode vfs_inode; 1578 }; 1579 1580 static inline struct socket *SOCKET_I(struct inode *inode) 1581 { 1582 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1583 } 1584 1585 static inline struct inode *SOCK_INODE(struct socket *socket) 1586 { 1587 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1588 } 1589 1590 /* 1591 * Functions for memory accounting 1592 */ 1593 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1594 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1595 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1596 void __sk_mem_reclaim(struct sock *sk, int amount); 1597 1598 #define SK_MEM_SEND 0 1599 #define SK_MEM_RECV 1 1600 1601 /* sysctl_mem values are in pages */ 1602 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1603 { 1604 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1605 } 1606 1607 static inline int sk_mem_pages(int amt) 1608 { 1609 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1610 } 1611 1612 static inline bool sk_has_account(struct sock *sk) 1613 { 1614 /* return true if protocol supports memory accounting */ 1615 return !!sk->sk_prot->memory_allocated; 1616 } 1617 1618 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1619 { 1620 int delta; 1621 1622 if (!sk_has_account(sk)) 1623 return true; 1624 delta = size - sk->sk_forward_alloc; 1625 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1626 } 1627 1628 static inline bool 1629 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1630 { 1631 int delta; 1632 1633 if (!sk_has_account(sk)) 1634 return true; 1635 delta = size - sk->sk_forward_alloc; 1636 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1637 skb_pfmemalloc(skb); 1638 } 1639 1640 static inline int sk_unused_reserved_mem(const struct sock *sk) 1641 { 1642 int unused_mem; 1643 1644 if (likely(!sk->sk_reserved_mem)) 1645 return 0; 1646 1647 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1648 atomic_read(&sk->sk_rmem_alloc); 1649 1650 return unused_mem > 0 ? unused_mem : 0; 1651 } 1652 1653 static inline void sk_mem_reclaim(struct sock *sk) 1654 { 1655 int reclaimable; 1656 1657 if (!sk_has_account(sk)) 1658 return; 1659 1660 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1661 1662 if (reclaimable >= (int)PAGE_SIZE) 1663 __sk_mem_reclaim(sk, reclaimable); 1664 } 1665 1666 static inline void sk_mem_reclaim_final(struct sock *sk) 1667 { 1668 sk->sk_reserved_mem = 0; 1669 sk_mem_reclaim(sk); 1670 } 1671 1672 static inline void sk_mem_charge(struct sock *sk, int size) 1673 { 1674 if (!sk_has_account(sk)) 1675 return; 1676 sk->sk_forward_alloc -= size; 1677 } 1678 1679 static inline void sk_mem_uncharge(struct sock *sk, int size) 1680 { 1681 if (!sk_has_account(sk)) 1682 return; 1683 sk->sk_forward_alloc += size; 1684 sk_mem_reclaim(sk); 1685 } 1686 1687 /* 1688 * Macro so as to not evaluate some arguments when 1689 * lockdep is not enabled. 1690 * 1691 * Mark both the sk_lock and the sk_lock.slock as a 1692 * per-address-family lock class. 1693 */ 1694 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1695 do { \ 1696 sk->sk_lock.owned = 0; \ 1697 init_waitqueue_head(&sk->sk_lock.wq); \ 1698 spin_lock_init(&(sk)->sk_lock.slock); \ 1699 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1700 sizeof((sk)->sk_lock)); \ 1701 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1702 (skey), (sname)); \ 1703 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1704 } while (0) 1705 1706 static inline bool lockdep_sock_is_held(const struct sock *sk) 1707 { 1708 return lockdep_is_held(&sk->sk_lock) || 1709 lockdep_is_held(&sk->sk_lock.slock); 1710 } 1711 1712 void lock_sock_nested(struct sock *sk, int subclass); 1713 1714 static inline void lock_sock(struct sock *sk) 1715 { 1716 lock_sock_nested(sk, 0); 1717 } 1718 1719 void __lock_sock(struct sock *sk); 1720 void __release_sock(struct sock *sk); 1721 void release_sock(struct sock *sk); 1722 1723 /* BH context may only use the following locking interface. */ 1724 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1725 #define bh_lock_sock_nested(__sk) \ 1726 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1727 SINGLE_DEPTH_NESTING) 1728 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1729 1730 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1731 1732 /** 1733 * lock_sock_fast - fast version of lock_sock 1734 * @sk: socket 1735 * 1736 * This version should be used for very small section, where process wont block 1737 * return false if fast path is taken: 1738 * 1739 * sk_lock.slock locked, owned = 0, BH disabled 1740 * 1741 * return true if slow path is taken: 1742 * 1743 * sk_lock.slock unlocked, owned = 1, BH enabled 1744 */ 1745 static inline bool lock_sock_fast(struct sock *sk) 1746 { 1747 /* The sk_lock has mutex_lock() semantics here. */ 1748 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1749 1750 return __lock_sock_fast(sk); 1751 } 1752 1753 /* fast socket lock variant for caller already holding a [different] socket lock */ 1754 static inline bool lock_sock_fast_nested(struct sock *sk) 1755 { 1756 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1757 1758 return __lock_sock_fast(sk); 1759 } 1760 1761 /** 1762 * unlock_sock_fast - complement of lock_sock_fast 1763 * @sk: socket 1764 * @slow: slow mode 1765 * 1766 * fast unlock socket for user context. 1767 * If slow mode is on, we call regular release_sock() 1768 */ 1769 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1770 __releases(&sk->sk_lock.slock) 1771 { 1772 if (slow) { 1773 release_sock(sk); 1774 __release(&sk->sk_lock.slock); 1775 } else { 1776 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1777 spin_unlock_bh(&sk->sk_lock.slock); 1778 } 1779 } 1780 1781 void sockopt_lock_sock(struct sock *sk); 1782 void sockopt_release_sock(struct sock *sk); 1783 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1784 bool sockopt_capable(int cap); 1785 1786 /* Used by processes to "lock" a socket state, so that 1787 * interrupts and bottom half handlers won't change it 1788 * from under us. It essentially blocks any incoming 1789 * packets, so that we won't get any new data or any 1790 * packets that change the state of the socket. 1791 * 1792 * While locked, BH processing will add new packets to 1793 * the backlog queue. This queue is processed by the 1794 * owner of the socket lock right before it is released. 1795 * 1796 * Since ~2.3.5 it is also exclusive sleep lock serializing 1797 * accesses from user process context. 1798 */ 1799 1800 static inline void sock_owned_by_me(const struct sock *sk) 1801 { 1802 #ifdef CONFIG_LOCKDEP 1803 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1804 #endif 1805 } 1806 1807 static inline bool sock_owned_by_user(const struct sock *sk) 1808 { 1809 sock_owned_by_me(sk); 1810 return sk->sk_lock.owned; 1811 } 1812 1813 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1814 { 1815 return sk->sk_lock.owned; 1816 } 1817 1818 static inline void sock_release_ownership(struct sock *sk) 1819 { 1820 if (sock_owned_by_user_nocheck(sk)) { 1821 sk->sk_lock.owned = 0; 1822 1823 /* The sk_lock has mutex_unlock() semantics: */ 1824 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1825 } 1826 } 1827 1828 /* no reclassification while locks are held */ 1829 static inline bool sock_allow_reclassification(const struct sock *csk) 1830 { 1831 struct sock *sk = (struct sock *)csk; 1832 1833 return !sock_owned_by_user_nocheck(sk) && 1834 !spin_is_locked(&sk->sk_lock.slock); 1835 } 1836 1837 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1838 struct proto *prot, int kern); 1839 void sk_free(struct sock *sk); 1840 void sk_destruct(struct sock *sk); 1841 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1842 void sk_free_unlock_clone(struct sock *sk); 1843 1844 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1845 gfp_t priority); 1846 void __sock_wfree(struct sk_buff *skb); 1847 void sock_wfree(struct sk_buff *skb); 1848 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1849 gfp_t priority); 1850 void skb_orphan_partial(struct sk_buff *skb); 1851 void sock_rfree(struct sk_buff *skb); 1852 void sock_efree(struct sk_buff *skb); 1853 #ifdef CONFIG_INET 1854 void sock_edemux(struct sk_buff *skb); 1855 void sock_pfree(struct sk_buff *skb); 1856 #else 1857 #define sock_edemux sock_efree 1858 #endif 1859 1860 int sk_setsockopt(struct sock *sk, int level, int optname, 1861 sockptr_t optval, unsigned int optlen); 1862 int sock_setsockopt(struct socket *sock, int level, int op, 1863 sockptr_t optval, unsigned int optlen); 1864 1865 int sk_getsockopt(struct sock *sk, int level, int optname, 1866 sockptr_t optval, sockptr_t optlen); 1867 int sock_getsockopt(struct socket *sock, int level, int op, 1868 char __user *optval, int __user *optlen); 1869 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1870 bool timeval, bool time32); 1871 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1872 unsigned long data_len, int noblock, 1873 int *errcode, int max_page_order); 1874 1875 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1876 unsigned long size, 1877 int noblock, int *errcode) 1878 { 1879 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1880 } 1881 1882 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1883 void sock_kfree_s(struct sock *sk, void *mem, int size); 1884 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1885 void sk_send_sigurg(struct sock *sk); 1886 1887 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1888 { 1889 if (sk->sk_socket) 1890 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1891 WRITE_ONCE(sk->sk_prot, proto); 1892 } 1893 1894 struct sockcm_cookie { 1895 u64 transmit_time; 1896 u32 mark; 1897 u32 tsflags; 1898 }; 1899 1900 static inline void sockcm_init(struct sockcm_cookie *sockc, 1901 const struct sock *sk) 1902 { 1903 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags }; 1904 } 1905 1906 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1907 struct sockcm_cookie *sockc); 1908 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1909 struct sockcm_cookie *sockc); 1910 1911 /* 1912 * Functions to fill in entries in struct proto_ops when a protocol 1913 * does not implement a particular function. 1914 */ 1915 int sock_no_bind(struct socket *, struct sockaddr *, int); 1916 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1917 int sock_no_socketpair(struct socket *, struct socket *); 1918 int sock_no_accept(struct socket *, struct socket *, int, bool); 1919 int sock_no_getname(struct socket *, struct sockaddr *, int); 1920 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1921 int sock_no_listen(struct socket *, int); 1922 int sock_no_shutdown(struct socket *, int); 1923 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1924 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1925 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1926 int sock_no_mmap(struct file *file, struct socket *sock, 1927 struct vm_area_struct *vma); 1928 1929 /* 1930 * Functions to fill in entries in struct proto_ops when a protocol 1931 * uses the inet style. 1932 */ 1933 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1934 char __user *optval, int __user *optlen); 1935 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1936 int flags); 1937 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1938 sockptr_t optval, unsigned int optlen); 1939 1940 void sk_common_release(struct sock *sk); 1941 1942 /* 1943 * Default socket callbacks and setup code 1944 */ 1945 1946 /* Initialise core socket variables using an explicit uid. */ 1947 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1948 1949 /* Initialise core socket variables. 1950 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1951 */ 1952 void sock_init_data(struct socket *sock, struct sock *sk); 1953 1954 /* 1955 * Socket reference counting postulates. 1956 * 1957 * * Each user of socket SHOULD hold a reference count. 1958 * * Each access point to socket (an hash table bucket, reference from a list, 1959 * running timer, skb in flight MUST hold a reference count. 1960 * * When reference count hits 0, it means it will never increase back. 1961 * * When reference count hits 0, it means that no references from 1962 * outside exist to this socket and current process on current CPU 1963 * is last user and may/should destroy this socket. 1964 * * sk_free is called from any context: process, BH, IRQ. When 1965 * it is called, socket has no references from outside -> sk_free 1966 * may release descendant resources allocated by the socket, but 1967 * to the time when it is called, socket is NOT referenced by any 1968 * hash tables, lists etc. 1969 * * Packets, delivered from outside (from network or from another process) 1970 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1971 * when they sit in queue. Otherwise, packets will leak to hole, when 1972 * socket is looked up by one cpu and unhasing is made by another CPU. 1973 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1974 * (leak to backlog). Packet socket does all the processing inside 1975 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1976 * use separate SMP lock, so that they are prone too. 1977 */ 1978 1979 /* Ungrab socket and destroy it, if it was the last reference. */ 1980 static inline void sock_put(struct sock *sk) 1981 { 1982 if (refcount_dec_and_test(&sk->sk_refcnt)) 1983 sk_free(sk); 1984 } 1985 /* Generic version of sock_put(), dealing with all sockets 1986 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1987 */ 1988 void sock_gen_put(struct sock *sk); 1989 1990 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1991 unsigned int trim_cap, bool refcounted); 1992 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1993 const int nested) 1994 { 1995 return __sk_receive_skb(sk, skb, nested, 1, true); 1996 } 1997 1998 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1999 { 2000 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2001 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2002 return; 2003 sk->sk_tx_queue_mapping = tx_queue; 2004 } 2005 2006 #define NO_QUEUE_MAPPING USHRT_MAX 2007 2008 static inline void sk_tx_queue_clear(struct sock *sk) 2009 { 2010 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING; 2011 } 2012 2013 static inline int sk_tx_queue_get(const struct sock *sk) 2014 { 2015 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING) 2016 return sk->sk_tx_queue_mapping; 2017 2018 return -1; 2019 } 2020 2021 static inline void __sk_rx_queue_set(struct sock *sk, 2022 const struct sk_buff *skb, 2023 bool force_set) 2024 { 2025 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2026 if (skb_rx_queue_recorded(skb)) { 2027 u16 rx_queue = skb_get_rx_queue(skb); 2028 2029 if (force_set || 2030 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2031 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2032 } 2033 #endif 2034 } 2035 2036 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2037 { 2038 __sk_rx_queue_set(sk, skb, true); 2039 } 2040 2041 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2042 { 2043 __sk_rx_queue_set(sk, skb, false); 2044 } 2045 2046 static inline void sk_rx_queue_clear(struct sock *sk) 2047 { 2048 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2049 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2050 #endif 2051 } 2052 2053 static inline int sk_rx_queue_get(const struct sock *sk) 2054 { 2055 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2056 if (sk) { 2057 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2058 2059 if (res != NO_QUEUE_MAPPING) 2060 return res; 2061 } 2062 #endif 2063 2064 return -1; 2065 } 2066 2067 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2068 { 2069 sk->sk_socket = sock; 2070 } 2071 2072 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2073 { 2074 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2075 return &rcu_dereference_raw(sk->sk_wq)->wait; 2076 } 2077 /* Detach socket from process context. 2078 * Announce socket dead, detach it from wait queue and inode. 2079 * Note that parent inode held reference count on this struct sock, 2080 * we do not release it in this function, because protocol 2081 * probably wants some additional cleanups or even continuing 2082 * to work with this socket (TCP). 2083 */ 2084 static inline void sock_orphan(struct sock *sk) 2085 { 2086 write_lock_bh(&sk->sk_callback_lock); 2087 sock_set_flag(sk, SOCK_DEAD); 2088 sk_set_socket(sk, NULL); 2089 sk->sk_wq = NULL; 2090 write_unlock_bh(&sk->sk_callback_lock); 2091 } 2092 2093 static inline void sock_graft(struct sock *sk, struct socket *parent) 2094 { 2095 WARN_ON(parent->sk); 2096 write_lock_bh(&sk->sk_callback_lock); 2097 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2098 parent->sk = sk; 2099 sk_set_socket(sk, parent); 2100 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2101 security_sock_graft(sk, parent); 2102 write_unlock_bh(&sk->sk_callback_lock); 2103 } 2104 2105 kuid_t sock_i_uid(struct sock *sk); 2106 unsigned long __sock_i_ino(struct sock *sk); 2107 unsigned long sock_i_ino(struct sock *sk); 2108 2109 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2110 { 2111 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2112 } 2113 2114 static inline u32 net_tx_rndhash(void) 2115 { 2116 u32 v = get_random_u32(); 2117 2118 return v ?: 1; 2119 } 2120 2121 static inline void sk_set_txhash(struct sock *sk) 2122 { 2123 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2124 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2125 } 2126 2127 static inline bool sk_rethink_txhash(struct sock *sk) 2128 { 2129 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2130 sk_set_txhash(sk); 2131 return true; 2132 } 2133 return false; 2134 } 2135 2136 static inline struct dst_entry * 2137 __sk_dst_get(struct sock *sk) 2138 { 2139 return rcu_dereference_check(sk->sk_dst_cache, 2140 lockdep_sock_is_held(sk)); 2141 } 2142 2143 static inline struct dst_entry * 2144 sk_dst_get(struct sock *sk) 2145 { 2146 struct dst_entry *dst; 2147 2148 rcu_read_lock(); 2149 dst = rcu_dereference(sk->sk_dst_cache); 2150 if (dst && !rcuref_get(&dst->__rcuref)) 2151 dst = NULL; 2152 rcu_read_unlock(); 2153 return dst; 2154 } 2155 2156 static inline void __dst_negative_advice(struct sock *sk) 2157 { 2158 struct dst_entry *ndst, *dst = __sk_dst_get(sk); 2159 2160 if (dst && dst->ops->negative_advice) { 2161 ndst = dst->ops->negative_advice(dst); 2162 2163 if (ndst != dst) { 2164 rcu_assign_pointer(sk->sk_dst_cache, ndst); 2165 sk_tx_queue_clear(sk); 2166 sk->sk_dst_pending_confirm = 0; 2167 } 2168 } 2169 } 2170 2171 static inline void dst_negative_advice(struct sock *sk) 2172 { 2173 sk_rethink_txhash(sk); 2174 __dst_negative_advice(sk); 2175 } 2176 2177 static inline void 2178 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2179 { 2180 struct dst_entry *old_dst; 2181 2182 sk_tx_queue_clear(sk); 2183 sk->sk_dst_pending_confirm = 0; 2184 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2185 lockdep_sock_is_held(sk)); 2186 rcu_assign_pointer(sk->sk_dst_cache, dst); 2187 dst_release(old_dst); 2188 } 2189 2190 static inline void 2191 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2192 { 2193 struct dst_entry *old_dst; 2194 2195 sk_tx_queue_clear(sk); 2196 sk->sk_dst_pending_confirm = 0; 2197 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst); 2198 dst_release(old_dst); 2199 } 2200 2201 static inline void 2202 __sk_dst_reset(struct sock *sk) 2203 { 2204 __sk_dst_set(sk, NULL); 2205 } 2206 2207 static inline void 2208 sk_dst_reset(struct sock *sk) 2209 { 2210 sk_dst_set(sk, NULL); 2211 } 2212 2213 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2214 2215 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2216 2217 static inline void sk_dst_confirm(struct sock *sk) 2218 { 2219 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2220 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2221 } 2222 2223 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2224 { 2225 if (skb_get_dst_pending_confirm(skb)) { 2226 struct sock *sk = skb->sk; 2227 2228 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2229 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2230 neigh_confirm(n); 2231 } 2232 } 2233 2234 bool sk_mc_loop(struct sock *sk); 2235 2236 static inline bool sk_can_gso(const struct sock *sk) 2237 { 2238 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2239 } 2240 2241 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2242 2243 static inline void sk_gso_disable(struct sock *sk) 2244 { 2245 sk->sk_gso_disabled = 1; 2246 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2247 } 2248 2249 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2250 struct iov_iter *from, char *to, 2251 int copy, int offset) 2252 { 2253 if (skb->ip_summed == CHECKSUM_NONE) { 2254 __wsum csum = 0; 2255 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2256 return -EFAULT; 2257 skb->csum = csum_block_add(skb->csum, csum, offset); 2258 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2259 if (!copy_from_iter_full_nocache(to, copy, from)) 2260 return -EFAULT; 2261 } else if (!copy_from_iter_full(to, copy, from)) 2262 return -EFAULT; 2263 2264 return 0; 2265 } 2266 2267 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2268 struct iov_iter *from, int copy) 2269 { 2270 int err, offset = skb->len; 2271 2272 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2273 copy, offset); 2274 if (err) 2275 __skb_trim(skb, offset); 2276 2277 return err; 2278 } 2279 2280 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2281 struct sk_buff *skb, 2282 struct page *page, 2283 int off, int copy) 2284 { 2285 int err; 2286 2287 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2288 copy, skb->len); 2289 if (err) 2290 return err; 2291 2292 skb_len_add(skb, copy); 2293 sk_wmem_queued_add(sk, copy); 2294 sk_mem_charge(sk, copy); 2295 return 0; 2296 } 2297 2298 /** 2299 * sk_wmem_alloc_get - returns write allocations 2300 * @sk: socket 2301 * 2302 * Return: sk_wmem_alloc minus initial offset of one 2303 */ 2304 static inline int sk_wmem_alloc_get(const struct sock *sk) 2305 { 2306 return refcount_read(&sk->sk_wmem_alloc) - 1; 2307 } 2308 2309 /** 2310 * sk_rmem_alloc_get - returns read allocations 2311 * @sk: socket 2312 * 2313 * Return: sk_rmem_alloc 2314 */ 2315 static inline int sk_rmem_alloc_get(const struct sock *sk) 2316 { 2317 return atomic_read(&sk->sk_rmem_alloc); 2318 } 2319 2320 /** 2321 * sk_has_allocations - check if allocations are outstanding 2322 * @sk: socket 2323 * 2324 * Return: true if socket has write or read allocations 2325 */ 2326 static inline bool sk_has_allocations(const struct sock *sk) 2327 { 2328 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2329 } 2330 2331 /** 2332 * skwq_has_sleeper - check if there are any waiting processes 2333 * @wq: struct socket_wq 2334 * 2335 * Return: true if socket_wq has waiting processes 2336 * 2337 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2338 * barrier call. They were added due to the race found within the tcp code. 2339 * 2340 * Consider following tcp code paths:: 2341 * 2342 * CPU1 CPU2 2343 * sys_select receive packet 2344 * ... ... 2345 * __add_wait_queue update tp->rcv_nxt 2346 * ... ... 2347 * tp->rcv_nxt check sock_def_readable 2348 * ... { 2349 * schedule rcu_read_lock(); 2350 * wq = rcu_dereference(sk->sk_wq); 2351 * if (wq && waitqueue_active(&wq->wait)) 2352 * wake_up_interruptible(&wq->wait) 2353 * ... 2354 * } 2355 * 2356 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2357 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2358 * could then endup calling schedule and sleep forever if there are no more 2359 * data on the socket. 2360 * 2361 */ 2362 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2363 { 2364 return wq && wq_has_sleeper(&wq->wait); 2365 } 2366 2367 /** 2368 * sock_poll_wait - place memory barrier behind the poll_wait call. 2369 * @filp: file 2370 * @sock: socket to wait on 2371 * @p: poll_table 2372 * 2373 * See the comments in the wq_has_sleeper function. 2374 */ 2375 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2376 poll_table *p) 2377 { 2378 if (!poll_does_not_wait(p)) { 2379 poll_wait(filp, &sock->wq.wait, p); 2380 /* We need to be sure we are in sync with the 2381 * socket flags modification. 2382 * 2383 * This memory barrier is paired in the wq_has_sleeper. 2384 */ 2385 smp_mb(); 2386 } 2387 } 2388 2389 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2390 { 2391 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2392 u32 txhash = READ_ONCE(sk->sk_txhash); 2393 2394 if (txhash) { 2395 skb->l4_hash = 1; 2396 skb->hash = txhash; 2397 } 2398 } 2399 2400 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2401 2402 /* 2403 * Queue a received datagram if it will fit. Stream and sequenced 2404 * protocols can't normally use this as they need to fit buffers in 2405 * and play with them. 2406 * 2407 * Inlined as it's very short and called for pretty much every 2408 * packet ever received. 2409 */ 2410 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2411 { 2412 skb_orphan(skb); 2413 skb->sk = sk; 2414 skb->destructor = sock_rfree; 2415 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2416 sk_mem_charge(sk, skb->truesize); 2417 } 2418 2419 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2420 { 2421 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2422 skb_orphan(skb); 2423 skb->destructor = sock_efree; 2424 skb->sk = sk; 2425 return true; 2426 } 2427 return false; 2428 } 2429 2430 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2431 { 2432 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2433 if (skb) { 2434 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2435 skb_set_owner_r(skb, sk); 2436 return skb; 2437 } 2438 __kfree_skb(skb); 2439 } 2440 return NULL; 2441 } 2442 2443 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2444 { 2445 if (skb->destructor != sock_wfree) { 2446 skb_orphan(skb); 2447 return; 2448 } 2449 skb->slow_gro = 1; 2450 } 2451 2452 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2453 unsigned long expires); 2454 2455 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2456 2457 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2458 2459 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2460 struct sk_buff *skb, unsigned int flags, 2461 void (*destructor)(struct sock *sk, 2462 struct sk_buff *skb)); 2463 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2464 2465 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2466 enum skb_drop_reason *reason); 2467 2468 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2469 { 2470 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2471 } 2472 2473 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2474 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2475 2476 /* 2477 * Recover an error report and clear atomically 2478 */ 2479 2480 static inline int sock_error(struct sock *sk) 2481 { 2482 int err; 2483 2484 /* Avoid an atomic operation for the common case. 2485 * This is racy since another cpu/thread can change sk_err under us. 2486 */ 2487 if (likely(data_race(!sk->sk_err))) 2488 return 0; 2489 2490 err = xchg(&sk->sk_err, 0); 2491 return -err; 2492 } 2493 2494 void sk_error_report(struct sock *sk); 2495 2496 static inline unsigned long sock_wspace(struct sock *sk) 2497 { 2498 int amt = 0; 2499 2500 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2501 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2502 if (amt < 0) 2503 amt = 0; 2504 } 2505 return amt; 2506 } 2507 2508 /* Note: 2509 * We use sk->sk_wq_raw, from contexts knowing this 2510 * pointer is not NULL and cannot disappear/change. 2511 */ 2512 static inline void sk_set_bit(int nr, struct sock *sk) 2513 { 2514 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2515 !sock_flag(sk, SOCK_FASYNC)) 2516 return; 2517 2518 set_bit(nr, &sk->sk_wq_raw->flags); 2519 } 2520 2521 static inline void sk_clear_bit(int nr, struct sock *sk) 2522 { 2523 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2524 !sock_flag(sk, SOCK_FASYNC)) 2525 return; 2526 2527 clear_bit(nr, &sk->sk_wq_raw->flags); 2528 } 2529 2530 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2531 { 2532 if (sock_flag(sk, SOCK_FASYNC)) { 2533 rcu_read_lock(); 2534 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2535 rcu_read_unlock(); 2536 } 2537 } 2538 2539 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2540 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2541 * Note: for send buffers, TCP works better if we can build two skbs at 2542 * minimum. 2543 */ 2544 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2545 2546 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2547 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2548 2549 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2550 { 2551 u32 val; 2552 2553 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2554 return; 2555 2556 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2557 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2558 2559 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2560 } 2561 2562 /** 2563 * sk_page_frag - return an appropriate page_frag 2564 * @sk: socket 2565 * 2566 * Use the per task page_frag instead of the per socket one for 2567 * optimization when we know that we're in process context and own 2568 * everything that's associated with %current. 2569 * 2570 * Both direct reclaim and page faults can nest inside other 2571 * socket operations and end up recursing into sk_page_frag() 2572 * while it's already in use: explicitly avoid task page_frag 2573 * when users disable sk_use_task_frag. 2574 * 2575 * Return: a per task page_frag if context allows that, 2576 * otherwise a per socket one. 2577 */ 2578 static inline struct page_frag *sk_page_frag(struct sock *sk) 2579 { 2580 if (sk->sk_use_task_frag) 2581 return ¤t->task_frag; 2582 2583 return &sk->sk_frag; 2584 } 2585 2586 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2587 2588 /* 2589 * Default write policy as shown to user space via poll/select/SIGIO 2590 */ 2591 static inline bool sock_writeable(const struct sock *sk) 2592 { 2593 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2594 } 2595 2596 static inline gfp_t gfp_any(void) 2597 { 2598 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2599 } 2600 2601 static inline gfp_t gfp_memcg_charge(void) 2602 { 2603 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2604 } 2605 2606 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2607 { 2608 return noblock ? 0 : sk->sk_rcvtimeo; 2609 } 2610 2611 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2612 { 2613 return noblock ? 0 : sk->sk_sndtimeo; 2614 } 2615 2616 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2617 { 2618 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2619 2620 return v ?: 1; 2621 } 2622 2623 /* Alas, with timeout socket operations are not restartable. 2624 * Compare this to poll(). 2625 */ 2626 static inline int sock_intr_errno(long timeo) 2627 { 2628 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2629 } 2630 2631 struct sock_skb_cb { 2632 u32 dropcount; 2633 }; 2634 2635 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2636 * using skb->cb[] would keep using it directly and utilize its 2637 * alignement guarantee. 2638 */ 2639 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2640 sizeof(struct sock_skb_cb))) 2641 2642 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2643 SOCK_SKB_CB_OFFSET)) 2644 2645 #define sock_skb_cb_check_size(size) \ 2646 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2647 2648 static inline void 2649 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2650 { 2651 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2652 atomic_read(&sk->sk_drops) : 0; 2653 } 2654 2655 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2656 { 2657 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2658 2659 atomic_add(segs, &sk->sk_drops); 2660 } 2661 2662 static inline ktime_t sock_read_timestamp(struct sock *sk) 2663 { 2664 #if BITS_PER_LONG==32 2665 unsigned int seq; 2666 ktime_t kt; 2667 2668 do { 2669 seq = read_seqbegin(&sk->sk_stamp_seq); 2670 kt = sk->sk_stamp; 2671 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2672 2673 return kt; 2674 #else 2675 return READ_ONCE(sk->sk_stamp); 2676 #endif 2677 } 2678 2679 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2680 { 2681 #if BITS_PER_LONG==32 2682 write_seqlock(&sk->sk_stamp_seq); 2683 sk->sk_stamp = kt; 2684 write_sequnlock(&sk->sk_stamp_seq); 2685 #else 2686 WRITE_ONCE(sk->sk_stamp, kt); 2687 #endif 2688 } 2689 2690 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2691 struct sk_buff *skb); 2692 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2693 struct sk_buff *skb); 2694 2695 static inline void 2696 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2697 { 2698 ktime_t kt = skb->tstamp; 2699 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2700 2701 /* 2702 * generate control messages if 2703 * - receive time stamping in software requested 2704 * - software time stamp available and wanted 2705 * - hardware time stamps available and wanted 2706 */ 2707 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2708 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2709 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2710 (hwtstamps->hwtstamp && 2711 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2712 __sock_recv_timestamp(msg, sk, skb); 2713 else 2714 sock_write_timestamp(sk, kt); 2715 2716 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2717 __sock_recv_wifi_status(msg, sk, skb); 2718 } 2719 2720 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2721 struct sk_buff *skb); 2722 2723 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2724 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2725 struct sk_buff *skb) 2726 { 2727 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2728 (1UL << SOCK_RCVTSTAMP) | \ 2729 (1UL << SOCK_RCVMARK)) 2730 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2731 SOF_TIMESTAMPING_RAW_HARDWARE) 2732 2733 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY) 2734 __sock_recv_cmsgs(msg, sk, skb); 2735 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2736 sock_write_timestamp(sk, skb->tstamp); 2737 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2738 sock_write_timestamp(sk, 0); 2739 } 2740 2741 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2742 2743 /** 2744 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2745 * @sk: socket sending this packet 2746 * @tsflags: timestamping flags to use 2747 * @tx_flags: completed with instructions for time stamping 2748 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2749 * 2750 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2751 */ 2752 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2753 __u8 *tx_flags, __u32 *tskey) 2754 { 2755 if (unlikely(tsflags)) { 2756 __sock_tx_timestamp(tsflags, tx_flags); 2757 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2758 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2759 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2760 } 2761 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2762 *tx_flags |= SKBTX_WIFI_STATUS; 2763 } 2764 2765 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2766 __u8 *tx_flags) 2767 { 2768 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2769 } 2770 2771 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2772 { 2773 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2774 &skb_shinfo(skb)->tskey); 2775 } 2776 2777 static inline bool sk_is_tcp(const struct sock *sk) 2778 { 2779 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP; 2780 } 2781 2782 /** 2783 * sk_eat_skb - Release a skb if it is no longer needed 2784 * @sk: socket to eat this skb from 2785 * @skb: socket buffer to eat 2786 * 2787 * This routine must be called with interrupts disabled or with the socket 2788 * locked so that the sk_buff queue operation is ok. 2789 */ 2790 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2791 { 2792 __skb_unlink(skb, &sk->sk_receive_queue); 2793 __kfree_skb(skb); 2794 } 2795 2796 static inline bool 2797 skb_sk_is_prefetched(struct sk_buff *skb) 2798 { 2799 #ifdef CONFIG_INET 2800 return skb->destructor == sock_pfree; 2801 #else 2802 return false; 2803 #endif /* CONFIG_INET */ 2804 } 2805 2806 /* This helper checks if a socket is a full socket, 2807 * ie _not_ a timewait or request socket. 2808 */ 2809 static inline bool sk_fullsock(const struct sock *sk) 2810 { 2811 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2812 } 2813 2814 static inline bool 2815 sk_is_refcounted(struct sock *sk) 2816 { 2817 /* Only full sockets have sk->sk_flags. */ 2818 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2819 } 2820 2821 /** 2822 * skb_steal_sock - steal a socket from an sk_buff 2823 * @skb: sk_buff to steal the socket from 2824 * @refcounted: is set to true if the socket is reference-counted 2825 * @prefetched: is set to true if the socket was assigned from bpf 2826 */ 2827 static inline struct sock * 2828 skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched) 2829 { 2830 if (skb->sk) { 2831 struct sock *sk = skb->sk; 2832 2833 *refcounted = true; 2834 *prefetched = skb_sk_is_prefetched(skb); 2835 if (*prefetched) 2836 *refcounted = sk_is_refcounted(sk); 2837 skb->destructor = NULL; 2838 skb->sk = NULL; 2839 return sk; 2840 } 2841 *prefetched = false; 2842 *refcounted = false; 2843 return NULL; 2844 } 2845 2846 /* Checks if this SKB belongs to an HW offloaded socket 2847 * and whether any SW fallbacks are required based on dev. 2848 * Check decrypted mark in case skb_orphan() cleared socket. 2849 */ 2850 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2851 struct net_device *dev) 2852 { 2853 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2854 struct sock *sk = skb->sk; 2855 2856 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2857 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2858 #ifdef CONFIG_TLS_DEVICE 2859 } else if (unlikely(skb->decrypted)) { 2860 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2861 kfree_skb(skb); 2862 skb = NULL; 2863 #endif 2864 } 2865 #endif 2866 2867 return skb; 2868 } 2869 2870 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2871 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2872 */ 2873 static inline bool sk_listener(const struct sock *sk) 2874 { 2875 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2876 } 2877 2878 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2879 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2880 int type); 2881 2882 bool sk_ns_capable(const struct sock *sk, 2883 struct user_namespace *user_ns, int cap); 2884 bool sk_capable(const struct sock *sk, int cap); 2885 bool sk_net_capable(const struct sock *sk, int cap); 2886 2887 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2888 2889 /* Take into consideration the size of the struct sk_buff overhead in the 2890 * determination of these values, since that is non-constant across 2891 * platforms. This makes socket queueing behavior and performance 2892 * not depend upon such differences. 2893 */ 2894 #define _SK_MEM_PACKETS 256 2895 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2896 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2897 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2898 2899 extern __u32 sysctl_wmem_max; 2900 extern __u32 sysctl_rmem_max; 2901 2902 extern int sysctl_tstamp_allow_data; 2903 extern int sysctl_optmem_max; 2904 2905 extern __u32 sysctl_wmem_default; 2906 extern __u32 sysctl_rmem_default; 2907 2908 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2909 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2910 2911 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2912 { 2913 /* Does this proto have per netns sysctl_wmem ? */ 2914 if (proto->sysctl_wmem_offset) 2915 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2916 2917 return READ_ONCE(*proto->sysctl_wmem); 2918 } 2919 2920 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2921 { 2922 /* Does this proto have per netns sysctl_rmem ? */ 2923 if (proto->sysctl_rmem_offset) 2924 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2925 2926 return READ_ONCE(*proto->sysctl_rmem); 2927 } 2928 2929 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2930 * Some wifi drivers need to tweak it to get more chunks. 2931 * They can use this helper from their ndo_start_xmit() 2932 */ 2933 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2934 { 2935 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2936 return; 2937 WRITE_ONCE(sk->sk_pacing_shift, val); 2938 } 2939 2940 /* if a socket is bound to a device, check that the given device 2941 * index is either the same or that the socket is bound to an L3 2942 * master device and the given device index is also enslaved to 2943 * that L3 master 2944 */ 2945 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2946 { 2947 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2948 int mdif; 2949 2950 if (!bound_dev_if || bound_dev_if == dif) 2951 return true; 2952 2953 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2954 if (mdif && mdif == bound_dev_if) 2955 return true; 2956 2957 return false; 2958 } 2959 2960 void sock_def_readable(struct sock *sk); 2961 2962 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2963 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2964 int sock_set_timestamping(struct sock *sk, int optname, 2965 struct so_timestamping timestamping); 2966 2967 void sock_enable_timestamps(struct sock *sk); 2968 void sock_no_linger(struct sock *sk); 2969 void sock_set_keepalive(struct sock *sk); 2970 void sock_set_priority(struct sock *sk, u32 priority); 2971 void sock_set_rcvbuf(struct sock *sk, int val); 2972 void sock_set_mark(struct sock *sk, u32 val); 2973 void sock_set_reuseaddr(struct sock *sk); 2974 void sock_set_reuseport(struct sock *sk); 2975 void sock_set_sndtimeo(struct sock *sk, s64 secs); 2976 2977 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 2978 2979 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 2980 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 2981 sockptr_t optval, int optlen, bool old_timeval); 2982 2983 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 2984 void __user *arg, void *karg, size_t size); 2985 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 2986 static inline bool sk_is_readable(struct sock *sk) 2987 { 2988 if (sk->sk_prot->sock_is_readable) 2989 return sk->sk_prot->sock_is_readable(sk); 2990 return false; 2991 } 2992 #endif /* _SOCK_H */ 2993