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