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 struct socket_drop_counters { 106 atomic_t drops0 ____cacheline_aligned_in_smp; 107 atomic_t drops1 ____cacheline_aligned_in_smp; 108 }; 109 110 /** 111 * struct sock_common - minimal network layer representation of sockets 112 * @skc_daddr: Foreign IPv4 addr 113 * @skc_rcv_saddr: Bound local IPv4 addr 114 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 115 * @skc_hash: hash value used with various protocol lookup tables 116 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 117 * @skc_dport: placeholder for inet_dport/tw_dport 118 * @skc_num: placeholder for inet_num/tw_num 119 * @skc_portpair: __u32 union of @skc_dport & @skc_num 120 * @skc_family: network address family 121 * @skc_state: Connection state 122 * @skc_reuse: %SO_REUSEADDR setting 123 * @skc_reuseport: %SO_REUSEPORT setting 124 * @skc_ipv6only: socket is IPV6 only 125 * @skc_net_refcnt: socket is using net ref counting 126 * @skc_bound_dev_if: bound device index if != 0 127 * @skc_bind_node: bind hash linkage for various protocol lookup tables 128 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 129 * @skc_prot: protocol handlers inside a network family 130 * @skc_net: reference to the network namespace of this socket 131 * @skc_v6_daddr: IPV6 destination address 132 * @skc_v6_rcv_saddr: IPV6 source address 133 * @skc_cookie: socket's cookie value 134 * @skc_node: main hash linkage for various protocol lookup tables 135 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 136 * @skc_tx_queue_mapping: tx queue number for this connection 137 * @skc_rx_queue_mapping: rx queue number for this connection 138 * @skc_flags: place holder for sk_flags 139 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 140 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 141 * @skc_listener: connection request listener socket (aka rsk_listener) 142 * [union with @skc_flags] 143 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 144 * [union with @skc_flags] 145 * @skc_incoming_cpu: record/match cpu processing incoming packets 146 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 147 * [union with @skc_incoming_cpu] 148 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 149 * [union with @skc_incoming_cpu] 150 * @skc_refcnt: reference count 151 * 152 * This is the minimal network layer representation of sockets, the header 153 * for struct sock and struct inet_timewait_sock. 154 */ 155 struct sock_common { 156 union { 157 __addrpair skc_addrpair; 158 struct { 159 __be32 skc_daddr; 160 __be32 skc_rcv_saddr; 161 }; 162 }; 163 union { 164 unsigned int skc_hash; 165 __u16 skc_u16hashes[2]; 166 }; 167 /* skc_dport && skc_num must be grouped as well */ 168 union { 169 __portpair skc_portpair; 170 struct { 171 __be16 skc_dport; 172 __u16 skc_num; 173 }; 174 }; 175 176 unsigned short skc_family; 177 volatile unsigned char skc_state; 178 unsigned char skc_reuse:4; 179 unsigned char skc_reuseport:1; 180 unsigned char skc_ipv6only:1; 181 unsigned char skc_net_refcnt:1; 182 int skc_bound_dev_if; 183 union { 184 struct hlist_node skc_bind_node; 185 struct hlist_node skc_portaddr_node; 186 }; 187 struct proto *skc_prot; 188 possible_net_t skc_net; 189 190 #if IS_ENABLED(CONFIG_IPV6) 191 struct in6_addr skc_v6_daddr; 192 struct in6_addr skc_v6_rcv_saddr; 193 #endif 194 195 atomic64_t skc_cookie; 196 197 /* following fields are padding to force 198 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 199 * assuming IPV6 is enabled. We use this padding differently 200 * for different kind of 'sockets' 201 */ 202 union { 203 unsigned long skc_flags; 204 struct sock *skc_listener; /* request_sock */ 205 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 206 }; 207 /* 208 * fields between dontcopy_begin/dontcopy_end 209 * are not copied in sock_copy() 210 */ 211 /* private: */ 212 int skc_dontcopy_begin[0]; 213 /* public: */ 214 union { 215 struct hlist_node skc_node; 216 struct hlist_nulls_node skc_nulls_node; 217 }; 218 unsigned short skc_tx_queue_mapping; 219 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 220 unsigned short skc_rx_queue_mapping; 221 #endif 222 union { 223 int skc_incoming_cpu; 224 u32 skc_rcv_wnd; 225 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 226 }; 227 228 refcount_t skc_refcnt; 229 /* private: */ 230 int skc_dontcopy_end[0]; 231 union { 232 u32 skc_rxhash; 233 u32 skc_window_clamp; 234 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 235 }; 236 /* public: */ 237 }; 238 239 struct bpf_local_storage; 240 struct sk_filter; 241 242 /** 243 * struct sock - network layer representation of sockets 244 * @__sk_common: shared layout with inet_timewait_sock 245 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 246 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 247 * @sk_lock: synchronizer 248 * @sk_kern_sock: True if sock is using kernel lock classes 249 * @sk_rcvbuf: size of receive buffer in bytes 250 * @sk_wq: sock wait queue and async head 251 * @sk_rx_dst: receive input route used by early demux 252 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 253 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 254 * @sk_dst_cache: destination cache 255 * @sk_dst_pending_confirm: need to confirm neighbour 256 * @sk_policy: flow policy 257 * @sk_receive_queue: incoming packets 258 * @sk_wmem_alloc: transmit queue bytes committed 259 * @sk_tsq_flags: TCP Small Queues flags 260 * @sk_write_queue: Packet sending queue 261 * @sk_omem_alloc: "o" is "option" or "other" 262 * @sk_wmem_queued: persistent queue size 263 * @sk_forward_alloc: space allocated forward 264 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 265 * @sk_napi_id: id of the last napi context to receive data for sk 266 * @sk_ll_usec: usecs to busypoll when there is no data 267 * @sk_allocation: allocation mode 268 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 269 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 270 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 271 * @sk_sndbuf: size of send buffer in bytes 272 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 273 * @sk_no_check_rx: allow zero checksum in RX packets 274 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 275 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 276 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 277 * @sk_gso_max_size: Maximum GSO segment size to build 278 * @sk_gso_max_segs: Maximum number of GSO segments 279 * @sk_pacing_shift: scaling factor for TCP Small Queues 280 * @sk_lingertime: %SO_LINGER l_linger setting 281 * @sk_backlog: always used with the per-socket spinlock held 282 * @sk_callback_lock: used with the callbacks in the end of this struct 283 * @sk_error_queue: rarely used 284 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 285 * IPV6_ADDRFORM for instance) 286 * @sk_err: last error 287 * @sk_err_soft: errors that don't cause failure but are the cause of a 288 * persistent failure not just 'timed out' 289 * @sk_drops: raw/udp drops counter 290 * @sk_drop_counters: optional pointer to socket_drop_counters 291 * @sk_ack_backlog: current listen backlog 292 * @sk_max_ack_backlog: listen backlog set in listen() 293 * @sk_uid: user id of owner 294 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 295 * @sk_busy_poll_budget: napi processing budget when busypolling 296 * @sk_priority: %SO_PRIORITY setting 297 * @sk_type: socket type (%SOCK_STREAM, etc) 298 * @sk_protocol: which protocol this socket belongs in this network family 299 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 300 * @sk_peer_pid: &struct pid for this socket's peer 301 * @sk_peer_cred: %SO_PEERCRED setting 302 * @sk_rcvlowat: %SO_RCVLOWAT setting 303 * @sk_rcvtimeo: %SO_RCVTIMEO setting 304 * @sk_sndtimeo: %SO_SNDTIMEO setting 305 * @sk_txhash: computed flow hash for use on transmit 306 * @sk_txrehash: enable TX hash rethink 307 * @sk_filter: socket filtering instructions 308 * @sk_timer: sock cleanup timer 309 * @sk_stamp: time stamp of last packet received 310 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 311 * @sk_tsflags: SO_TIMESTAMPING flags 312 * @sk_bpf_cb_flags: used in bpf_setsockopt() 313 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 314 * Sockets that can be used under memory reclaim should 315 * set this to false. 316 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 317 * for timestamping 318 * @sk_tskey: counter to disambiguate concurrent tstamp requests 319 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 320 * @sk_socket: Identd and reporting IO signals 321 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 322 * @sk_frag: cached page frag 323 * @sk_peek_off: current peek_offset value 324 * @sk_send_head: front of stuff to transmit 325 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 326 * @sk_security: used by security modules 327 * @sk_mark: generic packet mark 328 * @sk_cgrp_data: cgroup data for this cgroup 329 * @sk_memcg: this socket's memory cgroup association 330 * @sk_write_pending: a write to stream socket waits to start 331 * @sk_disconnects: number of disconnect operations performed on this sock 332 * @sk_state_change: callback to indicate change in the state of the sock 333 * @sk_data_ready: callback to indicate there is data to be processed 334 * @sk_write_space: callback to indicate there is bf sending space available 335 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 336 * @sk_backlog_rcv: callback to process the backlog 337 * @sk_validate_xmit_skb: ptr to an optional validate function 338 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 339 * @sk_reuseport_cb: reuseport group container 340 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 341 * @sk_rcu: used during RCU grace period 342 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 343 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 344 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 345 * @sk_txtime_unused: unused txtime flags 346 * @sk_scm_recv_flags: all flags used by scm_recv() 347 * @sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS 348 * @sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY 349 * @sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD 350 * @sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS 351 * @sk_scm_unused: unused flags for scm_recv() 352 * @ns_tracker: tracker for netns reference 353 * @sk_user_frags: xarray of pages the user is holding a reference on. 354 * @sk_owner: reference to the real owner of the socket that calls 355 * sock_lock_init_class_and_name(). 356 */ 357 struct sock { 358 /* 359 * Now struct inet_timewait_sock also uses sock_common, so please just 360 * don't add nothing before this first member (__sk_common) --acme 361 */ 362 struct sock_common __sk_common; 363 #define sk_node __sk_common.skc_node 364 #define sk_nulls_node __sk_common.skc_nulls_node 365 #define sk_refcnt __sk_common.skc_refcnt 366 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 367 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 368 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 369 #endif 370 371 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 372 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 373 #define sk_hash __sk_common.skc_hash 374 #define sk_portpair __sk_common.skc_portpair 375 #define sk_num __sk_common.skc_num 376 #define sk_dport __sk_common.skc_dport 377 #define sk_addrpair __sk_common.skc_addrpair 378 #define sk_daddr __sk_common.skc_daddr 379 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 380 #define sk_family __sk_common.skc_family 381 #define sk_state __sk_common.skc_state 382 #define sk_reuse __sk_common.skc_reuse 383 #define sk_reuseport __sk_common.skc_reuseport 384 #define sk_ipv6only __sk_common.skc_ipv6only 385 #define sk_net_refcnt __sk_common.skc_net_refcnt 386 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 387 #define sk_bind_node __sk_common.skc_bind_node 388 #define sk_prot __sk_common.skc_prot 389 #define sk_net __sk_common.skc_net 390 #define sk_v6_daddr __sk_common.skc_v6_daddr 391 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 392 #define sk_cookie __sk_common.skc_cookie 393 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 394 #define sk_flags __sk_common.skc_flags 395 #define sk_rxhash __sk_common.skc_rxhash 396 397 __cacheline_group_begin(sock_write_rx); 398 399 atomic_t sk_drops; 400 __s32 sk_peek_off; 401 struct sk_buff_head sk_error_queue; 402 struct sk_buff_head sk_receive_queue; 403 /* 404 * The backlog queue is special, it is always used with 405 * the per-socket spinlock held and requires low latency 406 * access. Therefore we special case it's implementation. 407 * Note : rmem_alloc is in this structure to fill a hole 408 * on 64bit arches, not because its logically part of 409 * backlog. 410 */ 411 struct { 412 atomic_t rmem_alloc; 413 int len; 414 struct sk_buff *head; 415 struct sk_buff *tail; 416 } sk_backlog; 417 #define sk_rmem_alloc sk_backlog.rmem_alloc 418 419 __cacheline_group_end(sock_write_rx); 420 421 __cacheline_group_begin(sock_read_rx); 422 /* early demux fields */ 423 struct dst_entry __rcu *sk_rx_dst; 424 int sk_rx_dst_ifindex; 425 u32 sk_rx_dst_cookie; 426 427 #ifdef CONFIG_NET_RX_BUSY_POLL 428 unsigned int sk_ll_usec; 429 unsigned int sk_napi_id; 430 u16 sk_busy_poll_budget; 431 u8 sk_prefer_busy_poll; 432 #endif 433 u8 sk_userlocks; 434 int sk_rcvbuf; 435 436 struct sk_filter __rcu *sk_filter; 437 union { 438 struct socket_wq __rcu *sk_wq; 439 /* private: */ 440 struct socket_wq *sk_wq_raw; 441 /* public: */ 442 }; 443 444 void (*sk_data_ready)(struct sock *sk); 445 long sk_rcvtimeo; 446 int sk_rcvlowat; 447 __cacheline_group_end(sock_read_rx); 448 449 __cacheline_group_begin(sock_read_rxtx); 450 int sk_err; 451 struct socket *sk_socket; 452 #ifdef CONFIG_MEMCG 453 struct mem_cgroup *sk_memcg; 454 #endif 455 #ifdef CONFIG_XFRM 456 struct xfrm_policy __rcu *sk_policy[2]; 457 #endif 458 struct socket_drop_counters *sk_drop_counters; 459 __cacheline_group_end(sock_read_rxtx); 460 461 __cacheline_group_begin(sock_write_rxtx); 462 socket_lock_t sk_lock; 463 u32 sk_reserved_mem; 464 int sk_forward_alloc; 465 u32 sk_tsflags; 466 __cacheline_group_end(sock_write_rxtx); 467 468 __cacheline_group_begin(sock_write_tx); 469 int sk_write_pending; 470 atomic_t sk_omem_alloc; 471 int sk_sndbuf; 472 473 int sk_wmem_queued; 474 refcount_t sk_wmem_alloc; 475 unsigned long sk_tsq_flags; 476 union { 477 struct sk_buff *sk_send_head; 478 struct rb_root tcp_rtx_queue; 479 }; 480 struct sk_buff_head sk_write_queue; 481 u32 sk_dst_pending_confirm; 482 u32 sk_pacing_status; /* see enum sk_pacing */ 483 struct page_frag sk_frag; 484 struct timer_list sk_timer; 485 486 unsigned long sk_pacing_rate; /* bytes per second */ 487 atomic_t sk_zckey; 488 atomic_t sk_tskey; 489 __cacheline_group_end(sock_write_tx); 490 491 __cacheline_group_begin(sock_read_tx); 492 unsigned long sk_max_pacing_rate; 493 long sk_sndtimeo; 494 u32 sk_priority; 495 u32 sk_mark; 496 struct dst_entry __rcu *sk_dst_cache; 497 netdev_features_t sk_route_caps; 498 #ifdef CONFIG_SOCK_VALIDATE_XMIT 499 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 500 struct net_device *dev, 501 struct sk_buff *skb); 502 #endif 503 u16 sk_gso_type; 504 u16 sk_gso_max_segs; 505 unsigned int sk_gso_max_size; 506 gfp_t sk_allocation; 507 u32 sk_txhash; 508 u8 sk_pacing_shift; 509 bool sk_use_task_frag; 510 __cacheline_group_end(sock_read_tx); 511 512 /* 513 * Because of non atomicity rules, all 514 * changes are protected by socket lock. 515 */ 516 u8 sk_gso_disabled : 1, 517 sk_kern_sock : 1, 518 sk_no_check_tx : 1, 519 sk_no_check_rx : 1; 520 u8 sk_shutdown; 521 u16 sk_type; 522 u16 sk_protocol; 523 unsigned long sk_lingertime; 524 struct proto *sk_prot_creator; 525 rwlock_t sk_callback_lock; 526 int sk_err_soft; 527 u32 sk_ack_backlog; 528 u32 sk_max_ack_backlog; 529 kuid_t sk_uid; 530 spinlock_t sk_peer_lock; 531 int sk_bind_phc; 532 struct pid *sk_peer_pid; 533 const struct cred *sk_peer_cred; 534 535 ktime_t sk_stamp; 536 #if BITS_PER_LONG==32 537 seqlock_t sk_stamp_seq; 538 #endif 539 int sk_disconnects; 540 541 union { 542 u8 sk_txrehash; 543 u8 sk_scm_recv_flags; 544 struct { 545 u8 sk_scm_credentials : 1, 546 sk_scm_security : 1, 547 sk_scm_pidfd : 1, 548 sk_scm_rights : 1, 549 sk_scm_unused : 4; 550 }; 551 }; 552 u8 sk_clockid; 553 u8 sk_txtime_deadline_mode : 1, 554 sk_txtime_report_errors : 1, 555 sk_txtime_unused : 6; 556 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG)) 557 u8 sk_bpf_cb_flags; 558 559 void *sk_user_data; 560 #ifdef CONFIG_SECURITY 561 void *sk_security; 562 #endif 563 struct sock_cgroup_data sk_cgrp_data; 564 void (*sk_state_change)(struct sock *sk); 565 void (*sk_write_space)(struct sock *sk); 566 void (*sk_error_report)(struct sock *sk); 567 int (*sk_backlog_rcv)(struct sock *sk, 568 struct sk_buff *skb); 569 void (*sk_destruct)(struct sock *sk); 570 struct sock_reuseport __rcu *sk_reuseport_cb; 571 #ifdef CONFIG_BPF_SYSCALL 572 struct bpf_local_storage __rcu *sk_bpf_storage; 573 #endif 574 struct rcu_head sk_rcu; 575 netns_tracker ns_tracker; 576 struct xarray sk_user_frags; 577 578 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 579 struct module *sk_owner; 580 #endif 581 }; 582 583 struct sock_bh_locked { 584 struct sock *sock; 585 local_lock_t bh_lock; 586 }; 587 588 enum sk_pacing { 589 SK_PACING_NONE = 0, 590 SK_PACING_NEEDED = 1, 591 SK_PACING_FQ = 2, 592 }; 593 594 /* flag bits in sk_user_data 595 * 596 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 597 * not be suitable for copying when cloning the socket. For instance, 598 * it can point to a reference counted object. sk_user_data bottom 599 * bit is set if pointer must not be copied. 600 * 601 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 602 * managed/owned by a BPF reuseport array. This bit should be set 603 * when sk_user_data's sk is added to the bpf's reuseport_array. 604 * 605 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 606 * sk_user_data points to psock type. This bit should be set 607 * when sk_user_data is assigned to a psock object. 608 */ 609 #define SK_USER_DATA_NOCOPY 1UL 610 #define SK_USER_DATA_BPF 2UL 611 #define SK_USER_DATA_PSOCK 4UL 612 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 613 SK_USER_DATA_PSOCK) 614 615 /** 616 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 617 * @sk: socket 618 */ 619 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 620 { 621 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 622 } 623 624 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 625 626 /** 627 * __locked_read_sk_user_data_with_flags - return the pointer 628 * only if argument flags all has been set in sk_user_data. Otherwise 629 * return NULL 630 * 631 * @sk: socket 632 * @flags: flag bits 633 * 634 * The caller must be holding sk->sk_callback_lock. 635 */ 636 static inline void * 637 __locked_read_sk_user_data_with_flags(const struct sock *sk, 638 uintptr_t flags) 639 { 640 uintptr_t sk_user_data = 641 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 642 lockdep_is_held(&sk->sk_callback_lock)); 643 644 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 645 646 if ((sk_user_data & flags) == flags) 647 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 648 return NULL; 649 } 650 651 /** 652 * __rcu_dereference_sk_user_data_with_flags - return the pointer 653 * only if argument flags all has been set in sk_user_data. Otherwise 654 * return NULL 655 * 656 * @sk: socket 657 * @flags: flag bits 658 */ 659 static inline void * 660 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 661 uintptr_t flags) 662 { 663 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 664 665 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 666 667 if ((sk_user_data & flags) == flags) 668 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 669 return NULL; 670 } 671 672 #define rcu_dereference_sk_user_data(sk) \ 673 __rcu_dereference_sk_user_data_with_flags(sk, 0) 674 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 675 ({ \ 676 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 677 __tmp2 = (uintptr_t)(flags); \ 678 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 679 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 680 rcu_assign_pointer(__sk_user_data((sk)), \ 681 __tmp1 | __tmp2); \ 682 }) 683 #define rcu_assign_sk_user_data(sk, ptr) \ 684 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 685 686 static inline 687 struct net *sock_net(const struct sock *sk) 688 { 689 return read_pnet(&sk->sk_net); 690 } 691 692 static inline 693 void sock_net_set(struct sock *sk, struct net *net) 694 { 695 write_pnet(&sk->sk_net, net); 696 } 697 698 /* 699 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 700 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 701 * on a socket means that the socket will reuse everybody else's port 702 * without looking at the other's sk_reuse value. 703 */ 704 705 #define SK_NO_REUSE 0 706 #define SK_CAN_REUSE 1 707 #define SK_FORCE_REUSE 2 708 709 int sk_set_peek_off(struct sock *sk, int val); 710 711 static inline int sk_peek_offset(const struct sock *sk, int flags) 712 { 713 if (unlikely(flags & MSG_PEEK)) { 714 return READ_ONCE(sk->sk_peek_off); 715 } 716 717 return 0; 718 } 719 720 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 721 { 722 s32 off = READ_ONCE(sk->sk_peek_off); 723 724 if (unlikely(off >= 0)) { 725 off = max_t(s32, off - val, 0); 726 WRITE_ONCE(sk->sk_peek_off, off); 727 } 728 } 729 730 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 731 { 732 sk_peek_offset_bwd(sk, -val); 733 } 734 735 /* 736 * Hashed lists helper routines 737 */ 738 static inline struct sock *sk_entry(const struct hlist_node *node) 739 { 740 return hlist_entry(node, struct sock, sk_node); 741 } 742 743 static inline struct sock *__sk_head(const struct hlist_head *head) 744 { 745 return hlist_entry(head->first, struct sock, sk_node); 746 } 747 748 static inline struct sock *sk_head(const struct hlist_head *head) 749 { 750 return hlist_empty(head) ? NULL : __sk_head(head); 751 } 752 753 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 754 { 755 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 756 } 757 758 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 759 { 760 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 761 } 762 763 static inline struct sock *sk_next(const struct sock *sk) 764 { 765 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 766 } 767 768 static inline struct sock *sk_nulls_next(const struct sock *sk) 769 { 770 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 771 hlist_nulls_entry(sk->sk_nulls_node.next, 772 struct sock, sk_nulls_node) : 773 NULL; 774 } 775 776 static inline bool sk_unhashed(const struct sock *sk) 777 { 778 return hlist_unhashed(&sk->sk_node); 779 } 780 781 static inline bool sk_hashed(const struct sock *sk) 782 { 783 return !sk_unhashed(sk); 784 } 785 786 static inline void sk_node_init(struct hlist_node *node) 787 { 788 node->pprev = NULL; 789 } 790 791 static inline void __sk_del_node(struct sock *sk) 792 { 793 __hlist_del(&sk->sk_node); 794 } 795 796 /* NB: equivalent to hlist_del_init_rcu */ 797 static inline bool __sk_del_node_init(struct sock *sk) 798 { 799 if (sk_hashed(sk)) { 800 __sk_del_node(sk); 801 sk_node_init(&sk->sk_node); 802 return true; 803 } 804 return false; 805 } 806 807 /* Grab socket reference count. This operation is valid only 808 when sk is ALREADY grabbed f.e. it is found in hash table 809 or a list and the lookup is made under lock preventing hash table 810 modifications. 811 */ 812 813 static __always_inline void sock_hold(struct sock *sk) 814 { 815 refcount_inc(&sk->sk_refcnt); 816 } 817 818 /* Ungrab socket in the context, which assumes that socket refcnt 819 cannot hit zero, f.e. it is true in context of any socketcall. 820 */ 821 static __always_inline void __sock_put(struct sock *sk) 822 { 823 refcount_dec(&sk->sk_refcnt); 824 } 825 826 static inline bool sk_del_node_init(struct sock *sk) 827 { 828 bool rc = __sk_del_node_init(sk); 829 830 if (rc) { 831 /* paranoid for a while -acme */ 832 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 833 __sock_put(sk); 834 } 835 return rc; 836 } 837 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 838 839 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 840 { 841 if (sk_hashed(sk)) { 842 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 843 return true; 844 } 845 return false; 846 } 847 848 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 849 { 850 bool rc = __sk_nulls_del_node_init_rcu(sk); 851 852 if (rc) { 853 /* paranoid for a while -acme */ 854 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 855 __sock_put(sk); 856 } 857 return rc; 858 } 859 860 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 861 { 862 hlist_add_head(&sk->sk_node, list); 863 } 864 865 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 866 { 867 sock_hold(sk); 868 __sk_add_node(sk, list); 869 } 870 871 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 872 { 873 sock_hold(sk); 874 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 875 sk->sk_family == AF_INET6) 876 hlist_add_tail_rcu(&sk->sk_node, list); 877 else 878 hlist_add_head_rcu(&sk->sk_node, list); 879 } 880 881 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 882 { 883 sock_hold(sk); 884 hlist_add_tail_rcu(&sk->sk_node, list); 885 } 886 887 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 888 { 889 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 890 } 891 892 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 893 { 894 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 895 } 896 897 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 898 { 899 sock_hold(sk); 900 __sk_nulls_add_node_rcu(sk, list); 901 } 902 903 static inline void __sk_del_bind_node(struct sock *sk) 904 { 905 __hlist_del(&sk->sk_bind_node); 906 } 907 908 static inline void sk_add_bind_node(struct sock *sk, 909 struct hlist_head *list) 910 { 911 hlist_add_head(&sk->sk_bind_node, list); 912 } 913 914 #define sk_for_each(__sk, list) \ 915 hlist_for_each_entry(__sk, list, sk_node) 916 #define sk_for_each_rcu(__sk, list) \ 917 hlist_for_each_entry_rcu(__sk, list, sk_node) 918 #define sk_nulls_for_each(__sk, node, list) \ 919 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 920 #define sk_nulls_for_each_rcu(__sk, node, list) \ 921 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 922 #define sk_for_each_from(__sk) \ 923 hlist_for_each_entry_from(__sk, sk_node) 924 #define sk_nulls_for_each_from(__sk, node) \ 925 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 926 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 927 #define sk_for_each_safe(__sk, tmp, list) \ 928 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 929 #define sk_for_each_bound(__sk, list) \ 930 hlist_for_each_entry(__sk, list, sk_bind_node) 931 #define sk_for_each_bound_safe(__sk, tmp, list) \ 932 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node) 933 934 /** 935 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 936 * @tpos: the type * to use as a loop cursor. 937 * @pos: the &struct hlist_node to use as a loop cursor. 938 * @head: the head for your list. 939 * @offset: offset of hlist_node within the struct. 940 * 941 */ 942 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 943 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 944 pos != NULL && \ 945 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 946 pos = rcu_dereference(hlist_next_rcu(pos))) 947 948 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 949 { 950 /* Careful only use this in a context where these parameters 951 * can not change and must all be valid, such as recvmsg from 952 * userspace. 953 */ 954 return sk->sk_socket->file->f_cred->user_ns; 955 } 956 957 /* Sock flags */ 958 enum sock_flags { 959 SOCK_DEAD, 960 SOCK_DONE, 961 SOCK_URGINLINE, 962 SOCK_KEEPOPEN, 963 SOCK_LINGER, 964 SOCK_DESTROY, 965 SOCK_BROADCAST, 966 SOCK_TIMESTAMP, 967 SOCK_ZAPPED, 968 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 969 SOCK_DBG, /* %SO_DEBUG setting */ 970 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 971 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 972 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 973 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 974 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 975 SOCK_FASYNC, /* fasync() active */ 976 SOCK_RXQ_OVFL, 977 SOCK_ZEROCOPY, /* buffers from userspace */ 978 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 979 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 980 * Will use last 4 bytes of packet sent from 981 * user-space instead. 982 */ 983 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 984 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 985 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 986 SOCK_TXTIME, 987 SOCK_XDP, /* XDP is attached */ 988 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 989 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 990 SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */ 991 SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */ 992 }; 993 994 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 995 /* 996 * The highest bit of sk_tsflags is reserved for kernel-internal 997 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that 998 * SOF_TIMESTAMPING* values do not reach this reserved area 999 */ 1000 #define SOCKCM_FLAG_TS_OPT_ID BIT(31) 1001 1002 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 1003 { 1004 nsk->sk_flags = osk->sk_flags; 1005 } 1006 1007 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 1008 { 1009 __set_bit(flag, &sk->sk_flags); 1010 } 1011 1012 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 1013 { 1014 __clear_bit(flag, &sk->sk_flags); 1015 } 1016 1017 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 1018 int valbool) 1019 { 1020 if (valbool) 1021 sock_set_flag(sk, bit); 1022 else 1023 sock_reset_flag(sk, bit); 1024 } 1025 1026 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 1027 { 1028 return test_bit(flag, &sk->sk_flags); 1029 } 1030 1031 #ifdef CONFIG_NET 1032 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 1033 static inline int sk_memalloc_socks(void) 1034 { 1035 return static_branch_unlikely(&memalloc_socks_key); 1036 } 1037 1038 void __receive_sock(struct file *file); 1039 #else 1040 1041 static inline int sk_memalloc_socks(void) 1042 { 1043 return 0; 1044 } 1045 1046 static inline void __receive_sock(struct file *file) 1047 { } 1048 #endif 1049 1050 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1051 { 1052 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1053 } 1054 1055 static inline void sk_acceptq_removed(struct sock *sk) 1056 { 1057 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1058 } 1059 1060 static inline void sk_acceptq_added(struct sock *sk) 1061 { 1062 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1063 } 1064 1065 /* Note: If you think the test should be: 1066 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1067 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1068 */ 1069 static inline bool sk_acceptq_is_full(const struct sock *sk) 1070 { 1071 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1072 } 1073 1074 /* 1075 * Compute minimal free write space needed to queue new packets. 1076 */ 1077 static inline int sk_stream_min_wspace(const struct sock *sk) 1078 { 1079 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1080 } 1081 1082 static inline int sk_stream_wspace(const struct sock *sk) 1083 { 1084 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1085 } 1086 1087 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1088 { 1089 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1090 } 1091 1092 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1093 { 1094 /* Paired with lockless reads of sk->sk_forward_alloc */ 1095 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1096 } 1097 1098 void sk_stream_write_space(struct sock *sk); 1099 1100 /* OOB backlog add */ 1101 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1102 { 1103 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1104 skb_dst_force(skb); 1105 1106 if (!sk->sk_backlog.tail) 1107 WRITE_ONCE(sk->sk_backlog.head, skb); 1108 else 1109 sk->sk_backlog.tail->next = skb; 1110 1111 WRITE_ONCE(sk->sk_backlog.tail, skb); 1112 skb->next = NULL; 1113 } 1114 1115 /* 1116 * Take into account size of receive queue and backlog queue 1117 * Do not take into account this skb truesize, 1118 * to allow even a single big packet to come. 1119 */ 1120 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1121 { 1122 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1123 1124 return qsize > limit; 1125 } 1126 1127 /* The per-socket spinlock must be held here. */ 1128 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1129 unsigned int limit) 1130 { 1131 if (sk_rcvqueues_full(sk, limit)) 1132 return -ENOBUFS; 1133 1134 /* 1135 * If the skb was allocated from pfmemalloc reserves, only 1136 * allow SOCK_MEMALLOC sockets to use it as this socket is 1137 * helping free memory 1138 */ 1139 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1140 return -ENOMEM; 1141 1142 __sk_add_backlog(sk, skb); 1143 sk->sk_backlog.len += skb->truesize; 1144 return 0; 1145 } 1146 1147 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1148 1149 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1150 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1151 1152 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1153 { 1154 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1155 return __sk_backlog_rcv(sk, skb); 1156 1157 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1158 tcp_v6_do_rcv, 1159 tcp_v4_do_rcv, 1160 sk, skb); 1161 } 1162 1163 static inline void sk_incoming_cpu_update(struct sock *sk) 1164 { 1165 int cpu = raw_smp_processor_id(); 1166 1167 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1168 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1169 } 1170 1171 1172 static inline void sock_rps_save_rxhash(struct sock *sk, 1173 const struct sk_buff *skb) 1174 { 1175 #ifdef CONFIG_RPS 1176 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1177 * here, and another one in sock_rps_record_flow(). 1178 */ 1179 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1180 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1181 #endif 1182 } 1183 1184 static inline void sock_rps_reset_rxhash(struct sock *sk) 1185 { 1186 #ifdef CONFIG_RPS 1187 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1188 WRITE_ONCE(sk->sk_rxhash, 0); 1189 #endif 1190 } 1191 1192 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1193 ({ int __rc, __dis = __sk->sk_disconnects; \ 1194 release_sock(__sk); \ 1195 __rc = __condition; \ 1196 if (!__rc) { \ 1197 *(__timeo) = wait_woken(__wait, \ 1198 TASK_INTERRUPTIBLE, \ 1199 *(__timeo)); \ 1200 } \ 1201 sched_annotate_sleep(); \ 1202 lock_sock(__sk); \ 1203 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1204 __rc; \ 1205 }) 1206 1207 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1208 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1209 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1210 int sk_stream_error(struct sock *sk, int flags, int err); 1211 void sk_stream_kill_queues(struct sock *sk); 1212 void sk_set_memalloc(struct sock *sk); 1213 void sk_clear_memalloc(struct sock *sk); 1214 1215 void __sk_flush_backlog(struct sock *sk); 1216 1217 static inline bool sk_flush_backlog(struct sock *sk) 1218 { 1219 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1220 __sk_flush_backlog(sk); 1221 return true; 1222 } 1223 return false; 1224 } 1225 1226 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1227 1228 struct request_sock_ops; 1229 struct timewait_sock_ops; 1230 struct inet_hashinfo; 1231 struct raw_hashinfo; 1232 struct smc_hashinfo; 1233 struct module; 1234 struct sk_psock; 1235 1236 /* 1237 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1238 * un-modified. Special care is taken when initializing object to zero. 1239 */ 1240 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1241 { 1242 if (offsetof(struct sock, sk_node.next) != 0) 1243 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1244 memset(&sk->sk_node.pprev, 0, 1245 size - offsetof(struct sock, sk_node.pprev)); 1246 } 1247 1248 struct proto_accept_arg { 1249 int flags; 1250 int err; 1251 int is_empty; 1252 bool kern; 1253 }; 1254 1255 /* Networking protocol blocks we attach to sockets. 1256 * socket layer -> transport layer interface 1257 */ 1258 struct proto { 1259 void (*close)(struct sock *sk, 1260 long timeout); 1261 int (*pre_connect)(struct sock *sk, 1262 struct sockaddr *uaddr, 1263 int addr_len); 1264 int (*connect)(struct sock *sk, 1265 struct sockaddr *uaddr, 1266 int addr_len); 1267 int (*disconnect)(struct sock *sk, int flags); 1268 1269 struct sock * (*accept)(struct sock *sk, 1270 struct proto_accept_arg *arg); 1271 1272 int (*ioctl)(struct sock *sk, int cmd, 1273 int *karg); 1274 int (*init)(struct sock *sk); 1275 void (*destroy)(struct sock *sk); 1276 void (*shutdown)(struct sock *sk, int how); 1277 int (*setsockopt)(struct sock *sk, int level, 1278 int optname, sockptr_t optval, 1279 unsigned int optlen); 1280 int (*getsockopt)(struct sock *sk, int level, 1281 int optname, char __user *optval, 1282 int __user *option); 1283 void (*keepalive)(struct sock *sk, int valbool); 1284 #ifdef CONFIG_COMPAT 1285 int (*compat_ioctl)(struct sock *sk, 1286 unsigned int cmd, unsigned long arg); 1287 #endif 1288 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1289 size_t len); 1290 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1291 size_t len, int flags, int *addr_len); 1292 void (*splice_eof)(struct socket *sock); 1293 int (*bind)(struct sock *sk, 1294 struct sockaddr *addr, int addr_len); 1295 int (*bind_add)(struct sock *sk, 1296 struct sockaddr *addr, int addr_len); 1297 1298 int (*backlog_rcv) (struct sock *sk, 1299 struct sk_buff *skb); 1300 bool (*bpf_bypass_getsockopt)(int level, 1301 int optname); 1302 1303 void (*release_cb)(struct sock *sk); 1304 1305 /* Keeping track of sk's, looking them up, and port selection methods. */ 1306 int (*hash)(struct sock *sk); 1307 void (*unhash)(struct sock *sk); 1308 void (*rehash)(struct sock *sk); 1309 int (*get_port)(struct sock *sk, unsigned short snum); 1310 void (*put_port)(struct sock *sk); 1311 #ifdef CONFIG_BPF_SYSCALL 1312 int (*psock_update_sk_prot)(struct sock *sk, 1313 struct sk_psock *psock, 1314 bool restore); 1315 #endif 1316 1317 /* Keeping track of sockets in use */ 1318 #ifdef CONFIG_PROC_FS 1319 unsigned int inuse_idx; 1320 #endif 1321 1322 bool (*stream_memory_free)(const struct sock *sk, int wake); 1323 bool (*sock_is_readable)(struct sock *sk); 1324 /* Memory pressure */ 1325 void (*enter_memory_pressure)(struct sock *sk); 1326 void (*leave_memory_pressure)(struct sock *sk); 1327 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1328 int __percpu *per_cpu_fw_alloc; 1329 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1330 1331 /* 1332 * Pressure flag: try to collapse. 1333 * Technical note: it is used by multiple contexts non atomically. 1334 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1335 * All the __sk_mem_schedule() is of this nature: accounting 1336 * is strict, actions are advisory and have some latency. 1337 */ 1338 unsigned long *memory_pressure; 1339 long *sysctl_mem; 1340 1341 int *sysctl_wmem; 1342 int *sysctl_rmem; 1343 u32 sysctl_wmem_offset; 1344 u32 sysctl_rmem_offset; 1345 1346 int max_header; 1347 bool no_autobind; 1348 1349 struct kmem_cache *slab; 1350 unsigned int obj_size; 1351 unsigned int ipv6_pinfo_offset; 1352 slab_flags_t slab_flags; 1353 unsigned int useroffset; /* Usercopy region offset */ 1354 unsigned int usersize; /* Usercopy region size */ 1355 1356 unsigned int __percpu *orphan_count; 1357 1358 struct request_sock_ops *rsk_prot; 1359 struct timewait_sock_ops *twsk_prot; 1360 1361 union { 1362 struct inet_hashinfo *hashinfo; 1363 struct udp_table *udp_table; 1364 struct raw_hashinfo *raw_hash; 1365 struct smc_hashinfo *smc_hash; 1366 } h; 1367 1368 struct module *owner; 1369 1370 char name[32]; 1371 1372 struct list_head node; 1373 int (*diag_destroy)(struct sock *sk, int err); 1374 } __randomize_layout; 1375 1376 int proto_register(struct proto *prot, int alloc_slab); 1377 void proto_unregister(struct proto *prot); 1378 int sock_load_diag_module(int family, int protocol); 1379 1380 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1381 1382 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1383 { 1384 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1385 return false; 1386 1387 return sk->sk_prot->stream_memory_free ? 1388 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1389 tcp_stream_memory_free, sk, wake) : true; 1390 } 1391 1392 static inline bool sk_stream_memory_free(const struct sock *sk) 1393 { 1394 return __sk_stream_memory_free(sk, 0); 1395 } 1396 1397 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1398 { 1399 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1400 __sk_stream_memory_free(sk, wake); 1401 } 1402 1403 static inline bool sk_stream_is_writeable(const struct sock *sk) 1404 { 1405 return __sk_stream_is_writeable(sk, 0); 1406 } 1407 1408 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1409 struct cgroup *ancestor) 1410 { 1411 #ifdef CONFIG_SOCK_CGROUP_DATA 1412 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1413 ancestor); 1414 #else 1415 return -ENOTSUPP; 1416 #endif 1417 } 1418 1419 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1420 1421 static inline void sk_sockets_allocated_dec(struct sock *sk) 1422 { 1423 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1424 SK_ALLOC_PERCPU_COUNTER_BATCH); 1425 } 1426 1427 static inline void sk_sockets_allocated_inc(struct sock *sk) 1428 { 1429 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1430 SK_ALLOC_PERCPU_COUNTER_BATCH); 1431 } 1432 1433 static inline u64 1434 sk_sockets_allocated_read_positive(struct sock *sk) 1435 { 1436 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1437 } 1438 1439 static inline int 1440 proto_sockets_allocated_sum_positive(struct proto *prot) 1441 { 1442 return percpu_counter_sum_positive(prot->sockets_allocated); 1443 } 1444 1445 #ifdef CONFIG_PROC_FS 1446 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1447 struct prot_inuse { 1448 int all; 1449 int val[PROTO_INUSE_NR]; 1450 }; 1451 1452 static inline void sock_prot_inuse_add(const struct net *net, 1453 const struct proto *prot, int val) 1454 { 1455 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1456 } 1457 1458 static inline void sock_inuse_add(const struct net *net, int val) 1459 { 1460 this_cpu_add(net->core.prot_inuse->all, val); 1461 } 1462 1463 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1464 int sock_inuse_get(struct net *net); 1465 #else 1466 static inline void sock_prot_inuse_add(const struct net *net, 1467 const struct proto *prot, int val) 1468 { 1469 } 1470 1471 static inline void sock_inuse_add(const struct net *net, int val) 1472 { 1473 } 1474 #endif 1475 1476 1477 /* With per-bucket locks this operation is not-atomic, so that 1478 * this version is not worse. 1479 */ 1480 static inline int __sk_prot_rehash(struct sock *sk) 1481 { 1482 sk->sk_prot->unhash(sk); 1483 return sk->sk_prot->hash(sk); 1484 } 1485 1486 /* About 10 seconds */ 1487 #define SOCK_DESTROY_TIME (10*HZ) 1488 1489 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1490 #define PROT_SOCK 1024 1491 1492 #define SHUTDOWN_MASK 3 1493 #define RCV_SHUTDOWN 1 1494 #define SEND_SHUTDOWN 2 1495 1496 #define SOCK_BINDADDR_LOCK 4 1497 #define SOCK_BINDPORT_LOCK 8 1498 1499 struct socket_alloc { 1500 struct socket socket; 1501 struct inode vfs_inode; 1502 }; 1503 1504 static inline struct socket *SOCKET_I(struct inode *inode) 1505 { 1506 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1507 } 1508 1509 static inline struct inode *SOCK_INODE(struct socket *socket) 1510 { 1511 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1512 } 1513 1514 /* 1515 * Functions for memory accounting 1516 */ 1517 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1518 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1519 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1520 void __sk_mem_reclaim(struct sock *sk, int amount); 1521 1522 #define SK_MEM_SEND 0 1523 #define SK_MEM_RECV 1 1524 1525 /* sysctl_mem values are in pages */ 1526 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1527 { 1528 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1529 } 1530 1531 static inline int sk_mem_pages(int amt) 1532 { 1533 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1534 } 1535 1536 static inline bool sk_has_account(struct sock *sk) 1537 { 1538 /* return true if protocol supports memory accounting */ 1539 return !!sk->sk_prot->memory_allocated; 1540 } 1541 1542 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1543 { 1544 int delta; 1545 1546 if (!sk_has_account(sk)) 1547 return true; 1548 delta = size - sk->sk_forward_alloc; 1549 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1550 } 1551 1552 static inline bool 1553 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc) 1554 { 1555 int delta; 1556 1557 if (!sk_has_account(sk)) 1558 return true; 1559 delta = size - sk->sk_forward_alloc; 1560 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1561 pfmemalloc; 1562 } 1563 1564 static inline bool 1565 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size) 1566 { 1567 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb)); 1568 } 1569 1570 static inline int sk_unused_reserved_mem(const struct sock *sk) 1571 { 1572 int unused_mem; 1573 1574 if (likely(!sk->sk_reserved_mem)) 1575 return 0; 1576 1577 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1578 atomic_read(&sk->sk_rmem_alloc); 1579 1580 return unused_mem > 0 ? unused_mem : 0; 1581 } 1582 1583 static inline void sk_mem_reclaim(struct sock *sk) 1584 { 1585 int reclaimable; 1586 1587 if (!sk_has_account(sk)) 1588 return; 1589 1590 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1591 1592 if (reclaimable >= (int)PAGE_SIZE) 1593 __sk_mem_reclaim(sk, reclaimable); 1594 } 1595 1596 static inline void sk_mem_reclaim_final(struct sock *sk) 1597 { 1598 sk->sk_reserved_mem = 0; 1599 sk_mem_reclaim(sk); 1600 } 1601 1602 static inline void sk_mem_charge(struct sock *sk, int size) 1603 { 1604 if (!sk_has_account(sk)) 1605 return; 1606 sk_forward_alloc_add(sk, -size); 1607 } 1608 1609 static inline void sk_mem_uncharge(struct sock *sk, int size) 1610 { 1611 if (!sk_has_account(sk)) 1612 return; 1613 sk_forward_alloc_add(sk, size); 1614 sk_mem_reclaim(sk); 1615 } 1616 1617 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 1618 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1619 { 1620 __module_get(owner); 1621 sk->sk_owner = owner; 1622 } 1623 1624 static inline void sk_owner_clear(struct sock *sk) 1625 { 1626 sk->sk_owner = NULL; 1627 } 1628 1629 static inline void sk_owner_put(struct sock *sk) 1630 { 1631 module_put(sk->sk_owner); 1632 } 1633 #else 1634 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1635 { 1636 } 1637 1638 static inline void sk_owner_clear(struct sock *sk) 1639 { 1640 } 1641 1642 static inline void sk_owner_put(struct sock *sk) 1643 { 1644 } 1645 #endif 1646 /* 1647 * Macro so as to not evaluate some arguments when 1648 * lockdep is not enabled. 1649 * 1650 * Mark both the sk_lock and the sk_lock.slock as a 1651 * per-address-family lock class. 1652 */ 1653 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1654 do { \ 1655 sk_owner_set(sk, THIS_MODULE); \ 1656 sk->sk_lock.owned = 0; \ 1657 init_waitqueue_head(&sk->sk_lock.wq); \ 1658 spin_lock_init(&(sk)->sk_lock.slock); \ 1659 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1660 sizeof((sk)->sk_lock)); \ 1661 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1662 (skey), (sname)); \ 1663 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1664 } while (0) 1665 1666 static inline bool lockdep_sock_is_held(const struct sock *sk) 1667 { 1668 return lockdep_is_held(&sk->sk_lock) || 1669 lockdep_is_held(&sk->sk_lock.slock); 1670 } 1671 1672 void lock_sock_nested(struct sock *sk, int subclass); 1673 1674 static inline void lock_sock(struct sock *sk) 1675 { 1676 lock_sock_nested(sk, 0); 1677 } 1678 1679 void __lock_sock(struct sock *sk); 1680 void __release_sock(struct sock *sk); 1681 void release_sock(struct sock *sk); 1682 1683 /* BH context may only use the following locking interface. */ 1684 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1685 #define bh_lock_sock_nested(__sk) \ 1686 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1687 SINGLE_DEPTH_NESTING) 1688 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1689 1690 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1691 1692 /** 1693 * lock_sock_fast - fast version of lock_sock 1694 * @sk: socket 1695 * 1696 * This version should be used for very small section, where process won't block 1697 * return false if fast path is taken: 1698 * 1699 * sk_lock.slock locked, owned = 0, BH disabled 1700 * 1701 * return true if slow path is taken: 1702 * 1703 * sk_lock.slock unlocked, owned = 1, BH enabled 1704 */ 1705 static inline bool lock_sock_fast(struct sock *sk) 1706 { 1707 /* The sk_lock has mutex_lock() semantics here. */ 1708 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1709 1710 return __lock_sock_fast(sk); 1711 } 1712 1713 /* fast socket lock variant for caller already holding a [different] socket lock */ 1714 static inline bool lock_sock_fast_nested(struct sock *sk) 1715 { 1716 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1717 1718 return __lock_sock_fast(sk); 1719 } 1720 1721 /** 1722 * unlock_sock_fast - complement of lock_sock_fast 1723 * @sk: socket 1724 * @slow: slow mode 1725 * 1726 * fast unlock socket for user context. 1727 * If slow mode is on, we call regular release_sock() 1728 */ 1729 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1730 __releases(&sk->sk_lock.slock) 1731 { 1732 if (slow) { 1733 release_sock(sk); 1734 __release(&sk->sk_lock.slock); 1735 } else { 1736 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1737 spin_unlock_bh(&sk->sk_lock.slock); 1738 } 1739 } 1740 1741 void sockopt_lock_sock(struct sock *sk); 1742 void sockopt_release_sock(struct sock *sk); 1743 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1744 bool sockopt_capable(int cap); 1745 1746 /* Used by processes to "lock" a socket state, so that 1747 * interrupts and bottom half handlers won't change it 1748 * from under us. It essentially blocks any incoming 1749 * packets, so that we won't get any new data or any 1750 * packets that change the state of the socket. 1751 * 1752 * While locked, BH processing will add new packets to 1753 * the backlog queue. This queue is processed by the 1754 * owner of the socket lock right before it is released. 1755 * 1756 * Since ~2.3.5 it is also exclusive sleep lock serializing 1757 * accesses from user process context. 1758 */ 1759 1760 static inline void sock_owned_by_me(const struct sock *sk) 1761 { 1762 #ifdef CONFIG_LOCKDEP 1763 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1764 #endif 1765 } 1766 1767 static inline void sock_not_owned_by_me(const struct sock *sk) 1768 { 1769 #ifdef CONFIG_LOCKDEP 1770 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1771 #endif 1772 } 1773 1774 static inline bool sock_owned_by_user(const struct sock *sk) 1775 { 1776 sock_owned_by_me(sk); 1777 return sk->sk_lock.owned; 1778 } 1779 1780 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1781 { 1782 return sk->sk_lock.owned; 1783 } 1784 1785 static inline void sock_release_ownership(struct sock *sk) 1786 { 1787 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1788 sk->sk_lock.owned = 0; 1789 1790 /* The sk_lock has mutex_unlock() semantics: */ 1791 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1792 } 1793 1794 /* no reclassification while locks are held */ 1795 static inline bool sock_allow_reclassification(const struct sock *csk) 1796 { 1797 struct sock *sk = (struct sock *)csk; 1798 1799 return !sock_owned_by_user_nocheck(sk) && 1800 !spin_is_locked(&sk->sk_lock.slock); 1801 } 1802 1803 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1804 struct proto *prot, int kern); 1805 void sk_free(struct sock *sk); 1806 void sk_net_refcnt_upgrade(struct sock *sk); 1807 void sk_destruct(struct sock *sk); 1808 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1809 1810 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1811 gfp_t priority); 1812 void __sock_wfree(struct sk_buff *skb); 1813 void sock_wfree(struct sk_buff *skb); 1814 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1815 gfp_t priority); 1816 void skb_orphan_partial(struct sk_buff *skb); 1817 void sock_rfree(struct sk_buff *skb); 1818 void sock_efree(struct sk_buff *skb); 1819 #ifdef CONFIG_INET 1820 void sock_edemux(struct sk_buff *skb); 1821 void sock_pfree(struct sk_buff *skb); 1822 1823 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk) 1824 { 1825 skb_orphan(skb); 1826 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 1827 skb->sk = sk; 1828 skb->destructor = sock_edemux; 1829 } 1830 } 1831 #else 1832 #define sock_edemux sock_efree 1833 #endif 1834 1835 int sk_setsockopt(struct sock *sk, int level, int optname, 1836 sockptr_t optval, unsigned int optlen); 1837 int sock_setsockopt(struct socket *sock, int level, int op, 1838 sockptr_t optval, unsigned int optlen); 1839 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1840 int optname, sockptr_t optval, int optlen); 1841 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1842 int optname, sockptr_t optval, sockptr_t optlen); 1843 1844 int sk_getsockopt(struct sock *sk, int level, int optname, 1845 sockptr_t optval, sockptr_t optlen); 1846 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1847 bool timeval, bool time32); 1848 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1849 unsigned long data_len, int noblock, 1850 int *errcode, int max_page_order); 1851 1852 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1853 unsigned long size, 1854 int noblock, int *errcode) 1855 { 1856 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1857 } 1858 1859 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1860 void *sock_kmemdup(struct sock *sk, const void *src, 1861 int size, gfp_t priority); 1862 void sock_kfree_s(struct sock *sk, void *mem, int size); 1863 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1864 void sk_send_sigurg(struct sock *sk); 1865 1866 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1867 { 1868 if (sk->sk_socket) 1869 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1870 WRITE_ONCE(sk->sk_prot, proto); 1871 } 1872 1873 struct sockcm_cookie { 1874 u64 transmit_time; 1875 u32 mark; 1876 u32 tsflags; 1877 u32 ts_opt_id; 1878 u32 priority; 1879 u32 dmabuf_id; 1880 }; 1881 1882 static inline void sockcm_init(struct sockcm_cookie *sockc, 1883 const struct sock *sk) 1884 { 1885 *sockc = (struct sockcm_cookie) { 1886 .mark = READ_ONCE(sk->sk_mark), 1887 .tsflags = READ_ONCE(sk->sk_tsflags), 1888 .priority = READ_ONCE(sk->sk_priority), 1889 }; 1890 } 1891 1892 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1893 struct sockcm_cookie *sockc); 1894 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1895 struct sockcm_cookie *sockc); 1896 1897 /* 1898 * Functions to fill in entries in struct proto_ops when a protocol 1899 * does not implement a particular function. 1900 */ 1901 int sock_no_bind(struct socket *, struct sockaddr *, int); 1902 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1903 int sock_no_socketpair(struct socket *, struct socket *); 1904 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *); 1905 int sock_no_getname(struct socket *, struct sockaddr *, int); 1906 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1907 int sock_no_listen(struct socket *, int); 1908 int sock_no_shutdown(struct socket *, int); 1909 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1910 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1911 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1912 int sock_no_mmap(struct file *file, struct socket *sock, 1913 struct vm_area_struct *vma); 1914 1915 /* 1916 * Functions to fill in entries in struct proto_ops when a protocol 1917 * uses the inet style. 1918 */ 1919 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1920 char __user *optval, int __user *optlen); 1921 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1922 int flags); 1923 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1924 sockptr_t optval, unsigned int optlen); 1925 1926 void sk_common_release(struct sock *sk); 1927 1928 /* 1929 * Default socket callbacks and setup code 1930 */ 1931 1932 /* Initialise core socket variables using an explicit uid. */ 1933 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1934 1935 /* Initialise core socket variables. 1936 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1937 */ 1938 void sock_init_data(struct socket *sock, struct sock *sk); 1939 1940 /* 1941 * Socket reference counting postulates. 1942 * 1943 * * Each user of socket SHOULD hold a reference count. 1944 * * Each access point to socket (an hash table bucket, reference from a list, 1945 * running timer, skb in flight MUST hold a reference count. 1946 * * When reference count hits 0, it means it will never increase back. 1947 * * When reference count hits 0, it means that no references from 1948 * outside exist to this socket and current process on current CPU 1949 * is last user and may/should destroy this socket. 1950 * * sk_free is called from any context: process, BH, IRQ. When 1951 * it is called, socket has no references from outside -> sk_free 1952 * may release descendant resources allocated by the socket, but 1953 * to the time when it is called, socket is NOT referenced by any 1954 * hash tables, lists etc. 1955 * * Packets, delivered from outside (from network or from another process) 1956 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1957 * when they sit in queue. Otherwise, packets will leak to hole, when 1958 * socket is looked up by one cpu and unhasing is made by another CPU. 1959 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1960 * (leak to backlog). Packet socket does all the processing inside 1961 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1962 * use separate SMP lock, so that they are prone too. 1963 */ 1964 1965 /* Ungrab socket and destroy it, if it was the last reference. */ 1966 static inline void sock_put(struct sock *sk) 1967 { 1968 if (refcount_dec_and_test(&sk->sk_refcnt)) 1969 sk_free(sk); 1970 } 1971 /* Generic version of sock_put(), dealing with all sockets 1972 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1973 */ 1974 void sock_gen_put(struct sock *sk); 1975 1976 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1977 unsigned int trim_cap, bool refcounted); 1978 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1979 const int nested) 1980 { 1981 return __sk_receive_skb(sk, skb, nested, 1, true); 1982 } 1983 1984 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1985 { 1986 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1987 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1988 return; 1989 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1990 * other WRITE_ONCE() because socket lock might be not held. 1991 */ 1992 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 1993 } 1994 1995 #define NO_QUEUE_MAPPING USHRT_MAX 1996 1997 static inline void sk_tx_queue_clear(struct sock *sk) 1998 { 1999 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2000 * other WRITE_ONCE() because socket lock might be not held. 2001 */ 2002 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 2003 } 2004 2005 static inline int sk_tx_queue_get(const struct sock *sk) 2006 { 2007 if (sk) { 2008 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 2009 * and sk_tx_queue_set(). 2010 */ 2011 int val = READ_ONCE(sk->sk_tx_queue_mapping); 2012 2013 if (val != NO_QUEUE_MAPPING) 2014 return val; 2015 } 2016 return -1; 2017 } 2018 2019 static inline void __sk_rx_queue_set(struct sock *sk, 2020 const struct sk_buff *skb, 2021 bool force_set) 2022 { 2023 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2024 if (skb_rx_queue_recorded(skb)) { 2025 u16 rx_queue = skb_get_rx_queue(skb); 2026 2027 if (force_set || 2028 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2029 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2030 } 2031 #endif 2032 } 2033 2034 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2035 { 2036 __sk_rx_queue_set(sk, skb, true); 2037 } 2038 2039 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2040 { 2041 __sk_rx_queue_set(sk, skb, false); 2042 } 2043 2044 static inline void sk_rx_queue_clear(struct sock *sk) 2045 { 2046 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2047 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2048 #endif 2049 } 2050 2051 static inline int sk_rx_queue_get(const struct sock *sk) 2052 { 2053 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2054 if (sk) { 2055 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2056 2057 if (res != NO_QUEUE_MAPPING) 2058 return res; 2059 } 2060 #endif 2061 2062 return -1; 2063 } 2064 2065 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2066 { 2067 sk->sk_socket = sock; 2068 } 2069 2070 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2071 { 2072 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2073 return &rcu_dereference_raw(sk->sk_wq)->wait; 2074 } 2075 /* Detach socket from process context. 2076 * Announce socket dead, detach it from wait queue and inode. 2077 * Note that parent inode held reference count on this struct sock, 2078 * we do not release it in this function, because protocol 2079 * probably wants some additional cleanups or even continuing 2080 * to work with this socket (TCP). 2081 */ 2082 static inline void sock_orphan(struct sock *sk) 2083 { 2084 write_lock_bh(&sk->sk_callback_lock); 2085 sock_set_flag(sk, SOCK_DEAD); 2086 sk_set_socket(sk, NULL); 2087 sk->sk_wq = NULL; 2088 /* Note: sk_uid is unchanged. */ 2089 write_unlock_bh(&sk->sk_callback_lock); 2090 } 2091 2092 static inline void sock_graft(struct sock *sk, struct socket *parent) 2093 { 2094 WARN_ON(parent->sk); 2095 write_lock_bh(&sk->sk_callback_lock); 2096 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2097 parent->sk = sk; 2098 sk_set_socket(sk, parent); 2099 WRITE_ONCE(sk->sk_uid, SOCK_INODE(parent)->i_uid); 2100 security_sock_graft(sk, parent); 2101 write_unlock_bh(&sk->sk_callback_lock); 2102 } 2103 2104 static inline kuid_t sk_uid(const struct sock *sk) 2105 { 2106 /* Paired with WRITE_ONCE() in sockfs_setattr() */ 2107 return READ_ONCE(sk->sk_uid); 2108 } 2109 2110 unsigned long __sock_i_ino(struct sock *sk); 2111 unsigned long sock_i_ino(struct sock *sk); 2112 2113 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2114 { 2115 return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0); 2116 } 2117 2118 static inline u32 net_tx_rndhash(void) 2119 { 2120 u32 v = get_random_u32(); 2121 2122 return v ?: 1; 2123 } 2124 2125 static inline void sk_set_txhash(struct sock *sk) 2126 { 2127 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2128 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2129 } 2130 2131 static inline bool sk_rethink_txhash(struct sock *sk) 2132 { 2133 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2134 sk_set_txhash(sk); 2135 return true; 2136 } 2137 return false; 2138 } 2139 2140 static inline struct dst_entry * 2141 __sk_dst_get(const struct sock *sk) 2142 { 2143 return rcu_dereference_check(sk->sk_dst_cache, 2144 lockdep_sock_is_held(sk)); 2145 } 2146 2147 static inline struct dst_entry * 2148 sk_dst_get(const struct sock *sk) 2149 { 2150 struct dst_entry *dst; 2151 2152 rcu_read_lock(); 2153 dst = rcu_dereference(sk->sk_dst_cache); 2154 if (dst && !rcuref_get(&dst->__rcuref)) 2155 dst = NULL; 2156 rcu_read_unlock(); 2157 return dst; 2158 } 2159 2160 static inline void __dst_negative_advice(struct sock *sk) 2161 { 2162 struct dst_entry *dst = __sk_dst_get(sk); 2163 2164 if (dst && dst->ops->negative_advice) 2165 dst->ops->negative_advice(sk, dst); 2166 } 2167 2168 static inline void dst_negative_advice(struct sock *sk) 2169 { 2170 sk_rethink_txhash(sk); 2171 __dst_negative_advice(sk); 2172 } 2173 2174 static inline void 2175 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2176 { 2177 struct dst_entry *old_dst; 2178 2179 sk_tx_queue_clear(sk); 2180 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2181 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2182 lockdep_sock_is_held(sk)); 2183 rcu_assign_pointer(sk->sk_dst_cache, dst); 2184 dst_release(old_dst); 2185 } 2186 2187 static inline void 2188 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2189 { 2190 struct dst_entry *old_dst; 2191 2192 sk_tx_queue_clear(sk); 2193 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2194 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst))); 2195 dst_release(old_dst); 2196 } 2197 2198 static inline void 2199 __sk_dst_reset(struct sock *sk) 2200 { 2201 __sk_dst_set(sk, NULL); 2202 } 2203 2204 static inline void 2205 sk_dst_reset(struct sock *sk) 2206 { 2207 sk_dst_set(sk, NULL); 2208 } 2209 2210 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2211 2212 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2213 2214 static inline void sk_dst_confirm(struct sock *sk) 2215 { 2216 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2217 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2218 } 2219 2220 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2221 { 2222 if (skb_get_dst_pending_confirm(skb)) { 2223 struct sock *sk = skb->sk; 2224 2225 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2226 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2227 neigh_confirm(n); 2228 } 2229 } 2230 2231 bool sk_mc_loop(const struct sock *sk); 2232 2233 static inline bool sk_can_gso(const struct sock *sk) 2234 { 2235 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2236 } 2237 2238 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2239 2240 static inline void sk_gso_disable(struct sock *sk) 2241 { 2242 sk->sk_gso_disabled = 1; 2243 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2244 } 2245 2246 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2247 struct iov_iter *from, char *to, 2248 int copy, int offset) 2249 { 2250 if (skb->ip_summed == CHECKSUM_NONE) { 2251 __wsum csum = 0; 2252 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2253 return -EFAULT; 2254 skb->csum = csum_block_add(skb->csum, csum, offset); 2255 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2256 if (!copy_from_iter_full_nocache(to, copy, from)) 2257 return -EFAULT; 2258 } else if (!copy_from_iter_full(to, copy, from)) 2259 return -EFAULT; 2260 2261 return 0; 2262 } 2263 2264 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2265 struct iov_iter *from, int copy) 2266 { 2267 int err, offset = skb->len; 2268 2269 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2270 copy, offset); 2271 if (err) 2272 __skb_trim(skb, offset); 2273 2274 return err; 2275 } 2276 2277 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2278 struct sk_buff *skb, 2279 struct page *page, 2280 int off, int copy) 2281 { 2282 int err; 2283 2284 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2285 copy, skb->len); 2286 if (err) 2287 return err; 2288 2289 skb_len_add(skb, copy); 2290 sk_wmem_queued_add(sk, copy); 2291 sk_mem_charge(sk, copy); 2292 return 0; 2293 } 2294 2295 /** 2296 * sk_wmem_alloc_get - returns write allocations 2297 * @sk: socket 2298 * 2299 * Return: sk_wmem_alloc minus initial offset of one 2300 */ 2301 static inline int sk_wmem_alloc_get(const struct sock *sk) 2302 { 2303 return refcount_read(&sk->sk_wmem_alloc) - 1; 2304 } 2305 2306 /** 2307 * sk_rmem_alloc_get - returns read allocations 2308 * @sk: socket 2309 * 2310 * Return: sk_rmem_alloc 2311 */ 2312 static inline int sk_rmem_alloc_get(const struct sock *sk) 2313 { 2314 return atomic_read(&sk->sk_rmem_alloc); 2315 } 2316 2317 /** 2318 * sk_has_allocations - check if allocations are outstanding 2319 * @sk: socket 2320 * 2321 * Return: true if socket has write or read allocations 2322 */ 2323 static inline bool sk_has_allocations(const struct sock *sk) 2324 { 2325 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2326 } 2327 2328 /** 2329 * skwq_has_sleeper - check if there are any waiting processes 2330 * @wq: struct socket_wq 2331 * 2332 * Return: true if socket_wq has waiting processes 2333 * 2334 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2335 * barrier call. They were added due to the race found within the tcp code. 2336 * 2337 * Consider following tcp code paths:: 2338 * 2339 * CPU1 CPU2 2340 * sys_select receive packet 2341 * ... ... 2342 * __add_wait_queue update tp->rcv_nxt 2343 * ... ... 2344 * tp->rcv_nxt check sock_def_readable 2345 * ... { 2346 * schedule rcu_read_lock(); 2347 * wq = rcu_dereference(sk->sk_wq); 2348 * if (wq && waitqueue_active(&wq->wait)) 2349 * wake_up_interruptible(&wq->wait) 2350 * ... 2351 * } 2352 * 2353 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2354 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2355 * could then endup calling schedule and sleep forever if there are no more 2356 * data on the socket. 2357 * 2358 */ 2359 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2360 { 2361 return wq && wq_has_sleeper(&wq->wait); 2362 } 2363 2364 /** 2365 * sock_poll_wait - wrapper for the poll_wait call. 2366 * @filp: file 2367 * @sock: socket to wait on 2368 * @p: poll_table 2369 * 2370 * See the comments in the wq_has_sleeper function. 2371 */ 2372 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2373 poll_table *p) 2374 { 2375 /* Provides a barrier we need to be sure we are in sync 2376 * with the socket flags modification. 2377 * 2378 * This memory barrier is paired in the wq_has_sleeper. 2379 */ 2380 poll_wait(filp, &sock->wq.wait, p); 2381 } 2382 2383 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2384 { 2385 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2386 u32 txhash = READ_ONCE(sk->sk_txhash); 2387 2388 if (txhash) { 2389 skb->l4_hash = 1; 2390 skb->hash = txhash; 2391 } 2392 } 2393 2394 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2395 2396 /* 2397 * Queue a received datagram if it will fit. Stream and sequenced 2398 * protocols can't normally use this as they need to fit buffers in 2399 * and play with them. 2400 * 2401 * Inlined as it's very short and called for pretty much every 2402 * packet ever received. 2403 */ 2404 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2405 { 2406 skb_orphan(skb); 2407 skb->sk = sk; 2408 skb->destructor = sock_rfree; 2409 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2410 sk_mem_charge(sk, skb->truesize); 2411 } 2412 2413 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2414 { 2415 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2416 skb_orphan(skb); 2417 skb->destructor = sock_efree; 2418 skb->sk = sk; 2419 return true; 2420 } 2421 return false; 2422 } 2423 2424 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2425 { 2426 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2427 if (skb) { 2428 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2429 skb_set_owner_r(skb, sk); 2430 return skb; 2431 } 2432 __kfree_skb(skb); 2433 } 2434 return NULL; 2435 } 2436 2437 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2438 { 2439 if (skb->destructor != sock_wfree) { 2440 skb_orphan(skb); 2441 return; 2442 } 2443 skb->slow_gro = 1; 2444 } 2445 2446 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2447 unsigned long expires); 2448 2449 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2450 2451 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2452 2453 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2454 struct sk_buff *skb, unsigned int flags, 2455 void (*destructor)(struct sock *sk, 2456 struct sk_buff *skb)); 2457 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2458 2459 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2460 enum skb_drop_reason *reason); 2461 2462 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2463 { 2464 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2465 } 2466 2467 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2468 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2469 2470 /* 2471 * Recover an error report and clear atomically 2472 */ 2473 2474 static inline int sock_error(struct sock *sk) 2475 { 2476 int err; 2477 2478 /* Avoid an atomic operation for the common case. 2479 * This is racy since another cpu/thread can change sk_err under us. 2480 */ 2481 if (likely(data_race(!sk->sk_err))) 2482 return 0; 2483 2484 err = xchg(&sk->sk_err, 0); 2485 return -err; 2486 } 2487 2488 void sk_error_report(struct sock *sk); 2489 2490 static inline unsigned long sock_wspace(struct sock *sk) 2491 { 2492 int amt = 0; 2493 2494 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2495 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2496 if (amt < 0) 2497 amt = 0; 2498 } 2499 return amt; 2500 } 2501 2502 /* Note: 2503 * We use sk->sk_wq_raw, from contexts knowing this 2504 * pointer is not NULL and cannot disappear/change. 2505 */ 2506 static inline void sk_set_bit(int nr, struct sock *sk) 2507 { 2508 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2509 !sock_flag(sk, SOCK_FASYNC)) 2510 return; 2511 2512 set_bit(nr, &sk->sk_wq_raw->flags); 2513 } 2514 2515 static inline void sk_clear_bit(int nr, struct sock *sk) 2516 { 2517 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2518 !sock_flag(sk, SOCK_FASYNC)) 2519 return; 2520 2521 clear_bit(nr, &sk->sk_wq_raw->flags); 2522 } 2523 2524 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2525 { 2526 if (sock_flag(sk, SOCK_FASYNC)) { 2527 rcu_read_lock(); 2528 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2529 rcu_read_unlock(); 2530 } 2531 } 2532 2533 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band) 2534 { 2535 if (unlikely(sock_flag(sk, SOCK_FASYNC))) 2536 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2537 } 2538 2539 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2540 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2541 * Note: for send buffers, TCP works better if we can build two skbs at 2542 * minimum. 2543 */ 2544 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2545 2546 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2547 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2548 2549 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2550 { 2551 u32 val; 2552 2553 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2554 return; 2555 2556 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2557 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2558 2559 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2560 } 2561 2562 /** 2563 * sk_page_frag - return an appropriate page_frag 2564 * @sk: socket 2565 * 2566 * Use the per task page_frag instead of the per socket one for 2567 * optimization when we know that we're in process context and own 2568 * everything that's associated with %current. 2569 * 2570 * Both direct reclaim and page faults can nest inside other 2571 * socket operations and end up recursing into sk_page_frag() 2572 * while it's already in use: explicitly avoid task page_frag 2573 * when users disable sk_use_task_frag. 2574 * 2575 * Return: a per task page_frag if context allows that, 2576 * otherwise a per socket one. 2577 */ 2578 static inline struct page_frag *sk_page_frag(struct sock *sk) 2579 { 2580 if (sk->sk_use_task_frag) 2581 return ¤t->task_frag; 2582 2583 return &sk->sk_frag; 2584 } 2585 2586 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2587 2588 /* 2589 * Default write policy as shown to user space via poll/select/SIGIO 2590 */ 2591 static inline bool sock_writeable(const struct sock *sk) 2592 { 2593 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2594 } 2595 2596 static inline gfp_t gfp_any(void) 2597 { 2598 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2599 } 2600 2601 static inline gfp_t gfp_memcg_charge(void) 2602 { 2603 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2604 } 2605 2606 #ifdef CONFIG_MEMCG 2607 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk) 2608 { 2609 return sk->sk_memcg; 2610 } 2611 2612 static inline bool mem_cgroup_sk_enabled(const struct sock *sk) 2613 { 2614 return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk); 2615 } 2616 2617 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk) 2618 { 2619 struct mem_cgroup *memcg = mem_cgroup_from_sk(sk); 2620 2621 #ifdef CONFIG_MEMCG_V1 2622 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 2623 return !!memcg->tcpmem_pressure; 2624 #endif /* CONFIG_MEMCG_V1 */ 2625 2626 do { 2627 if (time_before64(get_jiffies_64(), mem_cgroup_get_socket_pressure(memcg))) 2628 return true; 2629 } while ((memcg = parent_mem_cgroup(memcg))); 2630 2631 return false; 2632 } 2633 #else 2634 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk) 2635 { 2636 return NULL; 2637 } 2638 2639 static inline bool mem_cgroup_sk_enabled(const struct sock *sk) 2640 { 2641 return false; 2642 } 2643 2644 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk) 2645 { 2646 return false; 2647 } 2648 #endif 2649 2650 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2651 { 2652 return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo); 2653 } 2654 2655 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2656 { 2657 return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo); 2658 } 2659 2660 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2661 { 2662 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2663 2664 return v ?: 1; 2665 } 2666 2667 /* Alas, with timeout socket operations are not restartable. 2668 * Compare this to poll(). 2669 */ 2670 static inline int sock_intr_errno(long timeo) 2671 { 2672 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2673 } 2674 2675 struct sock_skb_cb { 2676 u32 dropcount; 2677 }; 2678 2679 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2680 * using skb->cb[] would keep using it directly and utilize its 2681 * alignment guarantee. 2682 */ 2683 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \ 2684 sizeof(struct sock_skb_cb)) 2685 2686 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2687 SOCK_SKB_CB_OFFSET)) 2688 2689 #define sock_skb_cb_check_size(size) \ 2690 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2691 2692 static inline void sk_drops_add(struct sock *sk, int segs) 2693 { 2694 struct socket_drop_counters *sdc = sk->sk_drop_counters; 2695 2696 if (sdc) { 2697 int n = numa_node_id() % 2; 2698 2699 if (n) 2700 atomic_add(segs, &sdc->drops1); 2701 else 2702 atomic_add(segs, &sdc->drops0); 2703 } else { 2704 atomic_add(segs, &sk->sk_drops); 2705 } 2706 } 2707 2708 static inline void sk_drops_inc(struct sock *sk) 2709 { 2710 sk_drops_add(sk, 1); 2711 } 2712 2713 static inline int sk_drops_read(const struct sock *sk) 2714 { 2715 const struct socket_drop_counters *sdc = sk->sk_drop_counters; 2716 2717 if (sdc) { 2718 DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops)); 2719 return atomic_read(&sdc->drops0) + atomic_read(&sdc->drops1); 2720 } 2721 return atomic_read(&sk->sk_drops); 2722 } 2723 2724 static inline void sk_drops_reset(struct sock *sk) 2725 { 2726 struct socket_drop_counters *sdc = sk->sk_drop_counters; 2727 2728 if (sdc) { 2729 atomic_set(&sdc->drops0, 0); 2730 atomic_set(&sdc->drops1, 0); 2731 } 2732 atomic_set(&sk->sk_drops, 0); 2733 } 2734 2735 static inline void 2736 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2737 { 2738 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2739 sk_drops_read(sk) : 0; 2740 } 2741 2742 static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb) 2743 { 2744 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2745 2746 sk_drops_add(sk, segs); 2747 } 2748 2749 static inline ktime_t sock_read_timestamp(struct sock *sk) 2750 { 2751 #if BITS_PER_LONG==32 2752 unsigned int seq; 2753 ktime_t kt; 2754 2755 do { 2756 seq = read_seqbegin(&sk->sk_stamp_seq); 2757 kt = sk->sk_stamp; 2758 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2759 2760 return kt; 2761 #else 2762 return READ_ONCE(sk->sk_stamp); 2763 #endif 2764 } 2765 2766 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2767 { 2768 #if BITS_PER_LONG==32 2769 write_seqlock(&sk->sk_stamp_seq); 2770 sk->sk_stamp = kt; 2771 write_sequnlock(&sk->sk_stamp_seq); 2772 #else 2773 WRITE_ONCE(sk->sk_stamp, kt); 2774 #endif 2775 } 2776 2777 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2778 struct sk_buff *skb); 2779 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2780 struct sk_buff *skb); 2781 2782 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk); 2783 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk, 2784 struct timespec64 *ts); 2785 2786 static inline void 2787 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2788 { 2789 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2790 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2791 ktime_t kt = skb->tstamp; 2792 /* 2793 * generate control messages if 2794 * - receive time stamping in software requested 2795 * - software time stamp available and wanted 2796 * - hardware time stamps available and wanted 2797 */ 2798 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2799 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2800 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2801 (hwtstamps->hwtstamp && 2802 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2803 __sock_recv_timestamp(msg, sk, skb); 2804 else 2805 sock_write_timestamp(sk, kt); 2806 2807 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2808 __sock_recv_wifi_status(msg, sk, skb); 2809 } 2810 2811 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2812 struct sk_buff *skb); 2813 2814 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2815 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2816 struct sk_buff *skb) 2817 { 2818 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2819 (1UL << SOCK_RCVTSTAMP) | \ 2820 (1UL << SOCK_RCVMARK) | \ 2821 (1UL << SOCK_RCVPRIORITY) | \ 2822 (1UL << SOCK_TIMESTAMPING_ANY)) 2823 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2824 SOF_TIMESTAMPING_RAW_HARDWARE) 2825 2826 if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS) 2827 __sock_recv_cmsgs(msg, sk, skb); 2828 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2829 sock_write_timestamp(sk, skb->tstamp); 2830 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2831 sock_write_timestamp(sk, 0); 2832 } 2833 2834 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags); 2835 2836 /** 2837 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2838 * @sk: socket sending this packet 2839 * @sockc: pointer to socket cmsg cookie to get timestamping info 2840 * @tx_flags: completed with instructions for time stamping 2841 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2842 * 2843 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2844 */ 2845 static inline void _sock_tx_timestamp(struct sock *sk, 2846 const struct sockcm_cookie *sockc, 2847 __u8 *tx_flags, __u32 *tskey) 2848 { 2849 __u32 tsflags = sockc->tsflags; 2850 2851 if (unlikely(tsflags)) { 2852 __sock_tx_timestamp(tsflags, tx_flags); 2853 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2854 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) { 2855 if (tsflags & SOCKCM_FLAG_TS_OPT_ID) 2856 *tskey = sockc->ts_opt_id; 2857 else 2858 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2859 } 2860 } 2861 } 2862 2863 static inline void sock_tx_timestamp(struct sock *sk, 2864 const struct sockcm_cookie *sockc, 2865 __u8 *tx_flags) 2866 { 2867 _sock_tx_timestamp(sk, sockc, tx_flags, NULL); 2868 } 2869 2870 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, 2871 const struct sockcm_cookie *sockc) 2872 { 2873 _sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags, 2874 &skb_shinfo(skb)->tskey); 2875 } 2876 2877 static inline bool sk_is_inet(const struct sock *sk) 2878 { 2879 int family = READ_ONCE(sk->sk_family); 2880 2881 return family == AF_INET || family == AF_INET6; 2882 } 2883 2884 static inline bool sk_is_tcp(const struct sock *sk) 2885 { 2886 return sk_is_inet(sk) && 2887 sk->sk_type == SOCK_STREAM && 2888 sk->sk_protocol == IPPROTO_TCP; 2889 } 2890 2891 static inline bool sk_is_udp(const struct sock *sk) 2892 { 2893 return sk_is_inet(sk) && 2894 sk->sk_type == SOCK_DGRAM && 2895 sk->sk_protocol == IPPROTO_UDP; 2896 } 2897 2898 static inline bool sk_is_unix(const struct sock *sk) 2899 { 2900 return sk->sk_family == AF_UNIX; 2901 } 2902 2903 static inline bool sk_is_stream_unix(const struct sock *sk) 2904 { 2905 return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM; 2906 } 2907 2908 static inline bool sk_is_vsock(const struct sock *sk) 2909 { 2910 return sk->sk_family == AF_VSOCK; 2911 } 2912 2913 static inline bool sk_may_scm_recv(const struct sock *sk) 2914 { 2915 return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) || 2916 sk->sk_family == AF_NETLINK || 2917 (IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH); 2918 } 2919 2920 /** 2921 * sk_eat_skb - Release a skb if it is no longer needed 2922 * @sk: socket to eat this skb from 2923 * @skb: socket buffer to eat 2924 * 2925 * This routine must be called with interrupts disabled or with the socket 2926 * locked so that the sk_buff queue operation is ok. 2927 */ 2928 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2929 { 2930 __skb_unlink(skb, &sk->sk_receive_queue); 2931 __kfree_skb(skb); 2932 } 2933 2934 static inline bool 2935 skb_sk_is_prefetched(struct sk_buff *skb) 2936 { 2937 #ifdef CONFIG_INET 2938 return skb->destructor == sock_pfree; 2939 #else 2940 return false; 2941 #endif /* CONFIG_INET */ 2942 } 2943 2944 /* This helper checks if a socket is a full socket, 2945 * ie _not_ a timewait or request socket. 2946 */ 2947 static inline bool sk_fullsock(const struct sock *sk) 2948 { 2949 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2950 } 2951 2952 static inline bool 2953 sk_is_refcounted(struct sock *sk) 2954 { 2955 /* Only full sockets have sk->sk_flags. */ 2956 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2957 } 2958 2959 static inline bool 2960 sk_requests_wifi_status(struct sock *sk) 2961 { 2962 return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS); 2963 } 2964 2965 /* Checks if this SKB belongs to an HW offloaded socket 2966 * and whether any SW fallbacks are required based on dev. 2967 * Check decrypted mark in case skb_orphan() cleared socket. 2968 */ 2969 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2970 struct net_device *dev) 2971 { 2972 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2973 struct sock *sk = skb->sk; 2974 2975 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2976 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2977 } else if (unlikely(skb_is_decrypted(skb))) { 2978 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2979 kfree_skb(skb); 2980 skb = NULL; 2981 } 2982 #endif 2983 2984 return skb; 2985 } 2986 2987 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2988 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2989 */ 2990 static inline bool sk_listener(const struct sock *sk) 2991 { 2992 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2993 } 2994 2995 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT 2996 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE) 2997 * TCP RST and ACK can be attached to TIME_WAIT. 2998 */ 2999 static inline bool sk_listener_or_tw(const struct sock *sk) 3000 { 3001 return (1 << READ_ONCE(sk->sk_state)) & 3002 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT); 3003 } 3004 3005 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 3006 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 3007 int type); 3008 3009 bool sk_ns_capable(const struct sock *sk, 3010 struct user_namespace *user_ns, int cap); 3011 bool sk_capable(const struct sock *sk, int cap); 3012 bool sk_net_capable(const struct sock *sk, int cap); 3013 3014 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 3015 3016 /* Take into consideration the size of the struct sk_buff overhead in the 3017 * determination of these values, since that is non-constant across 3018 * platforms. This makes socket queueing behavior and performance 3019 * not depend upon such differences. 3020 */ 3021 #define _SK_MEM_PACKETS 256 3022 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 3023 #define SK_WMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 3024 #define SK_RMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 3025 3026 extern __u32 sysctl_wmem_max; 3027 extern __u32 sysctl_rmem_max; 3028 3029 extern __u32 sysctl_wmem_default; 3030 extern __u32 sysctl_rmem_default; 3031 3032 #define SKB_FRAG_PAGE_ORDER get_order(32768) 3033 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3034 3035 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 3036 { 3037 /* Does this proto have per netns sysctl_wmem ? */ 3038 if (proto->sysctl_wmem_offset) 3039 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 3040 3041 return READ_ONCE(*proto->sysctl_wmem); 3042 } 3043 3044 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 3045 { 3046 /* Does this proto have per netns sysctl_rmem ? */ 3047 if (proto->sysctl_rmem_offset) 3048 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 3049 3050 return READ_ONCE(*proto->sysctl_rmem); 3051 } 3052 3053 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 3054 * Some wifi drivers need to tweak it to get more chunks. 3055 * They can use this helper from their ndo_start_xmit() 3056 */ 3057 static inline void sk_pacing_shift_update(struct sock *sk, int val) 3058 { 3059 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 3060 return; 3061 WRITE_ONCE(sk->sk_pacing_shift, val); 3062 } 3063 3064 /* if a socket is bound to a device, check that the given device 3065 * index is either the same or that the socket is bound to an L3 3066 * master device and the given device index is also enslaved to 3067 * that L3 master 3068 */ 3069 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 3070 { 3071 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 3072 int mdif; 3073 3074 if (!bound_dev_if || bound_dev_if == dif) 3075 return true; 3076 3077 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 3078 if (mdif && mdif == bound_dev_if) 3079 return true; 3080 3081 return false; 3082 } 3083 3084 void sock_def_readable(struct sock *sk); 3085 3086 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 3087 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 3088 int sock_set_timestamping(struct sock *sk, int optname, 3089 struct so_timestamping timestamping); 3090 3091 #if defined(CONFIG_CGROUP_BPF) 3092 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op); 3093 #else 3094 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 3095 { 3096 } 3097 #endif 3098 void sock_no_linger(struct sock *sk); 3099 void sock_set_keepalive(struct sock *sk); 3100 void sock_set_priority(struct sock *sk, u32 priority); 3101 void sock_set_rcvbuf(struct sock *sk, int val); 3102 void sock_set_mark(struct sock *sk, u32 val); 3103 void sock_set_reuseaddr(struct sock *sk); 3104 void sock_set_reuseport(struct sock *sk); 3105 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3106 3107 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 3108 3109 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3110 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3111 sockptr_t optval, int optlen, bool old_timeval); 3112 3113 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3114 void __user *arg, void *karg, size_t size); 3115 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3116 static inline bool sk_is_readable(struct sock *sk) 3117 { 3118 const struct proto *prot = READ_ONCE(sk->sk_prot); 3119 3120 if (prot->sock_is_readable) 3121 return prot->sock_is_readable(sk); 3122 3123 return false; 3124 } 3125 #endif /* _SOCK_H */ 3126