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