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 union { 85 struct slock_owned { 86 int owned; 87 spinlock_t slock; 88 }; 89 long combined; 90 }; 91 wait_queue_head_t wq; 92 /* 93 * We express the mutex-alike socket_lock semantics 94 * to the lock validator by explicitly managing 95 * the slock as a lock variant (in addition to 96 * the slock itself): 97 */ 98 #ifdef CONFIG_DEBUG_LOCK_ALLOC 99 struct lockdep_map dep_map; 100 #endif 101 } socket_lock_t; 102 103 struct sock; 104 struct proto; 105 struct net; 106 107 typedef __u32 __bitwise __portpair; 108 typedef __u64 __bitwise __addrpair; 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_bypass_prot_mem: bypass the per-protocol memory accounting for skb 127 * @skc_bound_dev_if: bound device index if != 0 128 * @skc_bind_node: bind hash linkage for various protocol lookup tables 129 * @skc_portaddr_node: second hash linkage for UDP 130 * @skc_prot: protocol handlers inside a network family 131 * @skc_net: reference to the network namespace of this socket 132 * @skc_v6_daddr: IPV6 destination address 133 * @skc_v6_rcv_saddr: IPV6 source address 134 * @skc_cookie: socket's cookie value 135 * @skc_node: main hash linkage for various protocol lookup tables 136 * @skc_nulls_node: main hash linkage for TCP 137 * @skc_tx_queue_mapping: tx queue number for this connection 138 * @skc_rx_queue_mapping: rx queue number for this connection 139 * @skc_flags: place holder for sk_flags 140 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 141 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 142 * @skc_listener: connection request listener socket (aka rsk_listener) 143 * [union with @skc_flags] 144 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 145 * [union with @skc_flags] 146 * @skc_incoming_cpu: record/match cpu processing incoming packets 147 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 148 * [union with @skc_incoming_cpu] 149 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 150 * [union with @skc_incoming_cpu] 151 * @skc_refcnt: reference count 152 * 153 * This is the minimal network layer representation of sockets, the header 154 * for struct sock and struct inet_timewait_sock. 155 */ 156 struct sock_common { 157 union { 158 __addrpair skc_addrpair; 159 struct { 160 __be32 skc_daddr; 161 __be32 skc_rcv_saddr; 162 }; 163 }; 164 union { 165 unsigned int skc_hash; 166 __u16 skc_u16hashes[2]; 167 }; 168 /* skc_dport && skc_num must be grouped as well */ 169 union { 170 __portpair skc_portpair; 171 struct { 172 __be16 skc_dport; 173 __u16 skc_num; 174 }; 175 }; 176 177 unsigned short skc_family; 178 volatile unsigned char skc_state; 179 unsigned char skc_reuse:4; 180 unsigned char skc_reuseport:1; 181 unsigned char skc_ipv6only:1; 182 unsigned char skc_net_refcnt:1; 183 unsigned char skc_bypass_prot_mem:1; 184 int skc_bound_dev_if; 185 union { 186 struct hlist_node skc_bind_node; 187 struct hlist_node skc_portaddr_node; 188 }; 189 struct proto *skc_prot; 190 possible_net_t skc_net; 191 192 #if IS_ENABLED(CONFIG_IPV6) 193 struct in6_addr skc_v6_daddr; 194 struct in6_addr skc_v6_rcv_saddr; 195 #endif 196 197 atomic64_t skc_cookie; 198 199 /* following fields are padding to force 200 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 201 * assuming IPV6 is enabled. We use this padding differently 202 * for different kind of 'sockets' 203 */ 204 union { 205 unsigned long skc_flags; 206 struct sock *skc_listener; /* request_sock */ 207 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 208 }; 209 /* 210 * fields between dontcopy_begin/dontcopy_end 211 * are not copied in sock_copy() 212 */ 213 /* private: */ 214 int skc_dontcopy_begin[0]; 215 /* public: */ 216 union { 217 struct hlist_node skc_node; 218 struct hlist_nulls_node skc_nulls_node; 219 }; 220 unsigned short skc_tx_queue_mapping; 221 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 222 unsigned short skc_rx_queue_mapping; 223 #endif 224 union { 225 int skc_incoming_cpu; 226 u32 skc_rcv_wnd; 227 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 228 }; 229 230 refcount_t skc_refcnt; 231 /* private: */ 232 int skc_dontcopy_end[0]; 233 union { 234 u32 skc_rxhash; 235 u32 skc_window_clamp; 236 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 237 }; 238 /* public: */ 239 }; 240 241 struct bpf_local_storage; 242 struct sk_filter; 243 244 /** 245 * struct sock - network layer representation of sockets 246 * @__sk_common: shared layout with inet_timewait_sock 247 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 248 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 249 * @sk_lock: synchronizer 250 * @sk_kern_sock: True if sock is using kernel lock classes 251 * @sk_rcvbuf: size of receive buffer in bytes 252 * @sk_wq: sock wait queue and async head 253 * @sk_rx_dst: receive input route used by early demux 254 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 255 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 256 * @sk_dst_cache: destination cache 257 * @sk_dst_pending_confirm: need to confirm neighbour 258 * @sk_policy: flow policy 259 * @psp_assoc: PSP association, if socket is PSP-secured 260 * @sk_receive_queue: incoming packets 261 * @sk_wmem_alloc: transmit queue bytes committed 262 * @sk_tsq_flags: TCP Small Queues flags 263 * @sk_write_queue: Packet sending queue 264 * @sk_omem_alloc: "o" is "option" or "other" 265 * @sk_wmem_queued: persistent queue size 266 * @sk_forward_alloc: space allocated forward 267 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 268 * @sk_napi_id: id of the last napi context to receive data for sk 269 * @sk_ll_usec: usecs to busypoll when there is no data 270 * @sk_allocation: allocation mode 271 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 272 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 273 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 274 * @sk_sndbuf: size of send buffer in bytes 275 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 276 * @sk_no_check_rx: allow zero checksum in RX packets 277 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 278 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 279 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 280 * @sk_gso_max_size: Maximum GSO segment size to build 281 * @sk_gso_max_segs: Maximum number of GSO segments 282 * @sk_pacing_shift: scaling factor for TCP Small Queues 283 * @sk_lingertime: %SO_LINGER l_linger setting 284 * @sk_backlog: always used with the per-socket spinlock held 285 * @sk_callback_lock: used with the callbacks in the end of this struct 286 * @sk_error_queue: rarely used 287 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 288 * IPV6_ADDRFORM for instance) 289 * @sk_err: last error 290 * @sk_err_soft: errors that don't cause failure but are the cause of a 291 * persistent failure not just 'timed out' 292 * @sk_drops: raw/udp drops counter 293 * @sk_drop_counters: optional pointer to numa_drop_counters 294 * @sk_ack_backlog: current listen backlog 295 * @sk_max_ack_backlog: listen backlog set in listen() 296 * @sk_uid: user id of owner 297 * @sk_ino: inode number (zero if orphaned) 298 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 299 * @sk_busy_poll_budget: napi processing budget when busypolling 300 * @sk_priority: %SO_PRIORITY setting 301 * @sk_type: socket type (%SOCK_STREAM, etc) 302 * @sk_protocol: which protocol this socket belongs in this network family 303 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 304 * @sk_peer_pid: &struct pid for this socket's peer, by pid type 305 * @sk_peer_cred: %SO_PEERCRED setting 306 * @sk_rcvlowat: %SO_RCVLOWAT setting 307 * @sk_rcvtimeo: %SO_RCVTIMEO setting 308 * @sk_sndtimeo: %SO_SNDTIMEO setting 309 * @sk_txhash: computed flow hash for use on transmit 310 * @sk_txrehash: enable TX hash rethink 311 * @sk_filter: socket filtering instructions 312 * @sk_timer: sock cleanup timer 313 * @tcp_retransmit_timer: tcp retransmit timer 314 * @mptcp_retransmit_timer: mptcp retransmit timer 315 * @sk_stamp: time stamp of last packet received 316 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 317 * @sk_tsflags: SO_TIMESTAMPING flags 318 * @sk_bpf_cb_flags: used in bpf_setsockopt() 319 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 320 * Sockets that can be used under memory reclaim should 321 * set this to false. 322 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 323 * for timestamping 324 * @sk_tskey: counter to disambiguate concurrent tstamp requests 325 * @sk_tx_queue_mapping_jiffies: time in jiffies of last @sk_tx_queue_mapping refresh. 326 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 327 * @sk_socket: Identd and reporting IO signals 328 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 329 * @sk_frag: cached page frag 330 * @sk_peek_off: current peek_offset value 331 * @sk_send_head: front of stuff to transmit 332 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 333 * @sk_security: used by security modules 334 * @sk_mark: generic packet mark 335 * @sk_cgrp_data: cgroup data for this cgroup 336 * @sk_memcg: this socket's memory cgroup association 337 * @sk_write_pending: a write to stream socket waits to start 338 * @sk_disconnects: number of disconnect operations performed on this sock 339 * @sk_state_change: callback to indicate change in the state of the sock 340 * @sk_data_ready: callback to indicate there is data to be processed 341 * @sk_write_space: callback to indicate there is bf sending space available 342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 343 * @sk_backlog_rcv: callback to process the backlog 344 * @sk_validate_xmit_skb: ptr to an optional validate function 345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 346 * @sk_reuseport_cb: reuseport group container 347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 348 * @sk_rcu: used during RCU grace period 349 * @sk_freeptr: used for SLAB_TYPESAFE_BY_RCU managed sockets 350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 353 * @sk_txtime_unused: unused txtime flags 354 * @sk_scm_recv_flags: all flags used by scm_recv() 355 * @sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS 356 * @sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY 357 * @sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD 358 * @sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS 359 * @sk_scm_pidfd_thread: flagged by SO_PASSPIDFD_THREAD to recv a thread SCM_PIDFD 360 * @sk_scm_unused: unused flags for scm_recv() 361 * @ns_tracker: tracker for netns reference 362 * @sk_user_frags: xarray of pages the user is holding a reference on. 363 * @sk_owner: reference to the real owner of the socket that calls 364 * sock_lock_init_class_and_name(). 365 */ 366 struct sock { 367 /* 368 * Now struct inet_timewait_sock also uses sock_common, so please just 369 * don't add nothing before this first member (__sk_common) --acme 370 */ 371 struct sock_common __sk_common; 372 #define sk_node __sk_common.skc_node 373 #define sk_nulls_node __sk_common.skc_nulls_node 374 #define sk_refcnt __sk_common.skc_refcnt 375 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 376 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 377 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 378 #endif 379 380 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 381 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 382 #define sk_hash __sk_common.skc_hash 383 #define sk_portpair __sk_common.skc_portpair 384 #define sk_num __sk_common.skc_num 385 #define sk_dport __sk_common.skc_dport 386 #define sk_addrpair __sk_common.skc_addrpair 387 #define sk_daddr __sk_common.skc_daddr 388 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 389 #define sk_family __sk_common.skc_family 390 #define sk_state __sk_common.skc_state 391 #define sk_reuse __sk_common.skc_reuse 392 #define sk_reuseport __sk_common.skc_reuseport 393 #define sk_ipv6only __sk_common.skc_ipv6only 394 #define sk_net_refcnt __sk_common.skc_net_refcnt 395 #define sk_bypass_prot_mem __sk_common.skc_bypass_prot_mem 396 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 397 #define sk_bind_node __sk_common.skc_bind_node 398 #define sk_prot __sk_common.skc_prot 399 #define sk_net __sk_common.skc_net 400 #define sk_v6_daddr __sk_common.skc_v6_daddr 401 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 402 #define sk_cookie __sk_common.skc_cookie 403 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 404 #define sk_flags __sk_common.skc_flags 405 #define sk_rxhash __sk_common.skc_rxhash 406 407 __cacheline_group_begin(sock_write_rx); 408 409 atomic_t sk_drops; 410 __s32 sk_peek_off; 411 struct sk_buff_head sk_error_queue; 412 struct sk_buff_head sk_receive_queue; 413 /* 414 * The backlog queue is special, it is always used with 415 * the per-socket spinlock held and requires low latency 416 * access. Therefore we special case it's implementation. 417 * Note : rmem_alloc is in this structure to fill a hole 418 * on 64bit arches, not because its logically part of 419 * backlog. 420 */ 421 struct { 422 atomic_t rmem_alloc; 423 int len; 424 struct sk_buff *head; 425 struct sk_buff *tail; 426 } sk_backlog; 427 #define sk_rmem_alloc sk_backlog.rmem_alloc 428 429 __cacheline_group_end(sock_write_rx); 430 431 __cacheline_group_begin(sock_read_rx); 432 /* early demux fields */ 433 struct dst_entry __rcu *sk_rx_dst; 434 int sk_rx_dst_ifindex; 435 u32 sk_rx_dst_cookie; 436 437 #ifdef CONFIG_NET_RX_BUSY_POLL 438 unsigned int sk_ll_usec; 439 unsigned int sk_napi_id; 440 u16 sk_busy_poll_budget; 441 u8 sk_prefer_busy_poll; 442 #endif 443 u8 sk_userlocks; 444 int sk_rcvbuf; 445 446 struct sk_filter __rcu *sk_filter; 447 union { 448 struct socket_wq __rcu *sk_wq; 449 /* private: */ 450 struct socket_wq *sk_wq_raw; 451 /* public: */ 452 }; 453 454 void (*sk_data_ready)(struct sock *sk); 455 long sk_rcvtimeo; 456 int sk_rcvlowat; 457 __cacheline_group_end(sock_read_rx); 458 459 __cacheline_group_begin(sock_read_rxtx); 460 int sk_err; 461 struct socket *sk_socket; 462 #ifdef CONFIG_MEMCG 463 struct mem_cgroup *sk_memcg; 464 #endif 465 #ifdef CONFIG_XFRM 466 struct xfrm_policy __rcu *sk_policy[2]; 467 #endif 468 #if IS_ENABLED(CONFIG_INET_PSP) 469 struct psp_assoc __rcu *psp_assoc; 470 #endif 471 __cacheline_group_end(sock_read_rxtx); 472 473 __cacheline_group_begin(sock_write_rxtx); 474 socket_lock_t sk_lock; 475 u32 sk_reserved_mem; 476 int sk_forward_alloc; 477 u32 sk_tsflags; 478 __cacheline_group_end(sock_write_rxtx); 479 480 __cacheline_group_begin(sock_write_tx); 481 int sk_write_pending; 482 atomic_t sk_omem_alloc; 483 int sk_err_soft; 484 485 int sk_wmem_queued; 486 refcount_t sk_wmem_alloc; 487 unsigned long sk_tsq_flags; 488 union { 489 struct sk_buff *sk_send_head; 490 struct rb_root tcp_rtx_queue; 491 }; 492 struct sk_buff_head sk_write_queue; 493 struct page_frag sk_frag; 494 union { 495 struct timer_list sk_timer; 496 struct timer_list tcp_retransmit_timer; 497 struct timer_list mptcp_retransmit_timer; 498 }; 499 unsigned long sk_pacing_rate; /* bytes per second */ 500 atomic_t sk_zckey; 501 atomic_t sk_tskey; 502 unsigned long sk_tx_queue_mapping_jiffies; 503 __cacheline_group_end(sock_write_tx); 504 505 __cacheline_group_begin(sock_read_tx); 506 u32 sk_dst_pending_confirm; 507 u32 sk_pacing_status; /* see enum sk_pacing */ 508 unsigned long sk_max_pacing_rate; 509 long sk_sndtimeo; 510 u32 sk_priority; 511 u32 sk_mark; 512 kuid_t sk_uid; 513 u16 sk_protocol; 514 u16 sk_type; 515 struct dst_entry __rcu *sk_dst_cache; 516 netdev_features_t sk_route_caps; 517 #ifdef CONFIG_SOCK_VALIDATE_XMIT 518 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 519 struct net_device *dev, 520 struct sk_buff *skb); 521 #endif 522 u16 sk_gso_type; 523 u16 sk_gso_max_segs; 524 unsigned int sk_gso_max_size; 525 gfp_t sk_allocation; 526 u32 sk_txhash; 527 int sk_sndbuf; 528 u8 sk_pacing_shift; 529 bool sk_use_task_frag; 530 __cacheline_group_end(sock_read_tx); 531 532 /* 533 * Because of non atomicity rules, all 534 * changes are protected by socket lock. 535 */ 536 u8 sk_gso_disabled : 1, 537 sk_kern_sock : 1, 538 sk_no_check_tx : 1, 539 sk_no_check_rx : 1; 540 u8 sk_shutdown; 541 unsigned long sk_lingertime; 542 struct proto *sk_prot_creator; 543 rwlock_t sk_callback_lock; 544 u32 sk_ack_backlog; 545 u32 sk_max_ack_backlog; 546 u64 sk_ino; 547 spinlock_t sk_peer_lock; 548 int sk_bind_phc; 549 DECLARE_PIDS(sk_peer_pid, PIDTYPE_TGID); 550 const struct cred *sk_peer_cred; 551 552 ktime_t sk_stamp; 553 #if BITS_PER_LONG==32 554 seqlock_t sk_stamp_seq; 555 #endif 556 int sk_disconnects; 557 558 union { 559 u8 sk_txrehash; 560 u8 sk_scm_recv_flags; 561 struct { 562 u8 sk_scm_credentials : 1, 563 sk_scm_security : 1, 564 sk_scm_pidfd : 1, 565 sk_scm_rights : 1, 566 sk_scm_pidfd_thread : 1, 567 sk_scm_unused : 3; 568 }; 569 }; 570 u8 sk_clockid; 571 u8 sk_txtime_deadline_mode : 1, 572 sk_txtime_report_errors : 1, 573 sk_txtime_unused : 6; 574 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG)) 575 u8 sk_bpf_cb_flags; 576 577 void *sk_user_data; 578 #ifdef CONFIG_SECURITY 579 void *sk_security; 580 #endif 581 struct sock_cgroup_data sk_cgrp_data; 582 void (*sk_state_change)(struct sock *sk); 583 void (*sk_write_space)(struct sock *sk); 584 void (*sk_error_report)(struct sock *sk); 585 int (*sk_backlog_rcv)(struct sock *sk, 586 struct sk_buff *skb); 587 void (*sk_destruct)(struct sock *sk); 588 struct sock_reuseport __rcu *sk_reuseport_cb; 589 #ifdef CONFIG_BPF_SYSCALL 590 struct bpf_local_storage __rcu *sk_bpf_storage; 591 #endif 592 struct numa_drop_counters *sk_drop_counters; 593 /* sockets using SLAB_TYPESAFE_BY_RCU can use sk_freeptr. 594 * By the time kfree() is called, sk_rcu can not be in 595 * use and can be mangled. 596 */ 597 union { 598 struct rcu_head sk_rcu; 599 freeptr_t sk_freeptr; 600 }; 601 netns_tracker ns_tracker; 602 struct xarray sk_user_frags; 603 604 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 605 struct module *sk_owner; 606 #endif 607 }; 608 609 struct sock_bh_locked { 610 struct sock *sock; 611 local_lock_t bh_lock; 612 }; 613 614 enum sk_pacing { 615 SK_PACING_NONE = 0, 616 SK_PACING_NEEDED = 1, 617 SK_PACING_FQ = 2, 618 }; 619 620 /* flag bits in sk_user_data 621 * 622 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 623 * not be suitable for copying when cloning the socket. For instance, 624 * it can point to a reference counted object. sk_user_data bottom 625 * bit is set if pointer must not be copied. 626 * 627 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 628 * managed/owned by a BPF reuseport array. This bit should be set 629 * when sk_user_data's sk is added to the bpf's reuseport_array. 630 * 631 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 632 * sk_user_data points to psock type. This bit should be set 633 * when sk_user_data is assigned to a psock object. 634 */ 635 #define SK_USER_DATA_NOCOPY 1UL 636 #define SK_USER_DATA_BPF 2UL 637 #define SK_USER_DATA_PSOCK 4UL 638 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 639 SK_USER_DATA_PSOCK) 640 641 /** 642 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 643 * @sk: socket 644 */ 645 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 646 { 647 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 648 } 649 650 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 651 652 /** 653 * __locked_read_sk_user_data_with_flags - return the pointer 654 * only if argument flags all has been set in sk_user_data. Otherwise 655 * return NULL 656 * 657 * @sk: socket 658 * @flags: flag bits 659 * 660 * The caller must be holding sk->sk_callback_lock. 661 */ 662 static inline void * 663 __locked_read_sk_user_data_with_flags(const struct sock *sk, 664 uintptr_t flags) 665 { 666 uintptr_t sk_user_data = 667 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 668 lockdep_is_held(&sk->sk_callback_lock)); 669 670 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 671 672 if ((sk_user_data & flags) == flags) 673 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 674 return NULL; 675 } 676 677 /** 678 * __rcu_dereference_sk_user_data_with_flags - return the pointer 679 * only if argument flags all has been set in sk_user_data. Otherwise 680 * return NULL 681 * 682 * @sk: socket 683 * @flags: flag bits 684 */ 685 static inline void * 686 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 687 uintptr_t flags) 688 { 689 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 690 691 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 692 693 if ((sk_user_data & flags) == flags) 694 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 695 return NULL; 696 } 697 698 #define rcu_dereference_sk_user_data(sk) \ 699 __rcu_dereference_sk_user_data_with_flags(sk, 0) 700 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 701 ({ \ 702 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 703 __tmp2 = (uintptr_t)(flags); \ 704 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 705 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 706 rcu_assign_pointer(__sk_user_data((sk)), \ 707 __tmp1 | __tmp2); \ 708 }) 709 #define rcu_assign_sk_user_data(sk, ptr) \ 710 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 711 712 static inline 713 struct net *sock_net(const struct sock *sk) 714 { 715 return read_pnet(&sk->sk_net); 716 } 717 718 static inline 719 void sock_net_set(struct sock *sk, struct net *net) 720 { 721 write_pnet(&sk->sk_net, net); 722 } 723 724 /* 725 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 726 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 727 * on a socket means that the socket will reuse everybody else's port 728 * without looking at the other's sk_reuse value. 729 */ 730 731 #define SK_NO_REUSE 0 732 #define SK_CAN_REUSE 1 733 #define SK_FORCE_REUSE 2 734 735 int sk_set_peek_off(struct sock *sk, int val); 736 737 static inline int sk_peek_offset(const struct sock *sk, int flags) 738 { 739 if (unlikely(flags & MSG_PEEK)) { 740 return READ_ONCE(sk->sk_peek_off); 741 } 742 743 return 0; 744 } 745 746 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 747 { 748 s32 off = READ_ONCE(sk->sk_peek_off); 749 750 if (unlikely(off >= 0)) { 751 off = max_t(s32, off - val, 0); 752 WRITE_ONCE(sk->sk_peek_off, off); 753 } 754 } 755 756 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 757 { 758 sk_peek_offset_bwd(sk, -val); 759 } 760 761 /* 762 * Hashed lists helper routines 763 */ 764 static inline struct sock *sk_entry(const struct hlist_node *node) 765 { 766 return hlist_entry(node, struct sock, sk_node); 767 } 768 769 static inline struct sock *__sk_head(const struct hlist_head *head) 770 { 771 return hlist_entry(head->first, struct sock, sk_node); 772 } 773 774 static inline struct sock *sk_head(const struct hlist_head *head) 775 { 776 return hlist_empty(head) ? NULL : __sk_head(head); 777 } 778 779 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 780 { 781 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 782 } 783 784 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 785 { 786 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 787 } 788 789 static inline struct sock *sk_next(const struct sock *sk) 790 { 791 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 792 } 793 794 static inline struct sock *sk_nulls_next(const struct sock *sk) 795 { 796 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 797 hlist_nulls_entry(sk->sk_nulls_node.next, 798 struct sock, sk_nulls_node) : 799 NULL; 800 } 801 802 static inline bool sk_unhashed(const struct sock *sk) 803 { 804 return hlist_unhashed(&sk->sk_node); 805 } 806 807 static inline bool sk_hashed(const struct sock *sk) 808 { 809 return !sk_unhashed(sk); 810 } 811 812 static inline void sk_node_init(struct hlist_node *node) 813 { 814 node->pprev = NULL; 815 } 816 817 static inline void __sk_del_node(struct sock *sk) 818 { 819 __hlist_del(&sk->sk_node); 820 } 821 822 /* NB: equivalent to hlist_del_init_rcu */ 823 static inline bool __sk_del_node_init(struct sock *sk) 824 { 825 if (sk_hashed(sk)) { 826 __sk_del_node(sk); 827 sk_node_init(&sk->sk_node); 828 return true; 829 } 830 return false; 831 } 832 833 /* Grab socket reference count. This operation is valid only 834 when sk is ALREADY grabbed f.e. it is found in hash table 835 or a list and the lookup is made under lock preventing hash table 836 modifications. 837 */ 838 839 static __always_inline void sock_hold(struct sock *sk) 840 { 841 refcount_inc(&sk->sk_refcnt); 842 } 843 844 /* Ungrab socket in the context, which assumes that socket refcnt 845 cannot hit zero, f.e. it is true in context of any socketcall. 846 */ 847 static __always_inline void __sock_put(struct sock *sk) 848 { 849 refcount_dec(&sk->sk_refcnt); 850 } 851 852 static inline bool sk_del_node_init(struct sock *sk) 853 { 854 bool rc = __sk_del_node_init(sk); 855 856 if (rc) 857 __sock_put(sk); 858 859 return rc; 860 } 861 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 862 863 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 864 { 865 if (sk_hashed(sk)) { 866 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 867 return true; 868 } 869 return false; 870 } 871 872 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 873 { 874 bool rc = __sk_nulls_del_node_init_rcu(sk); 875 876 if (rc) 877 __sock_put(sk); 878 879 return rc; 880 } 881 882 static inline bool sk_nulls_replace_node_init_rcu(struct sock *old, 883 struct sock *new) 884 { 885 if (sk_hashed(old)) { 886 hlist_nulls_replace_init_rcu(&old->sk_nulls_node, 887 &new->sk_nulls_node); 888 __sock_put(old); 889 return true; 890 } 891 892 return false; 893 } 894 895 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 896 { 897 hlist_add_head(&sk->sk_node, list); 898 } 899 900 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 901 { 902 sock_hold(sk); 903 __sk_add_node(sk, list); 904 } 905 906 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 907 { 908 sock_hold(sk); 909 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 910 sk->sk_family == AF_INET6) 911 hlist_add_tail_rcu(&sk->sk_node, list); 912 else 913 hlist_add_head_rcu(&sk->sk_node, list); 914 } 915 916 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 917 { 918 sock_hold(sk); 919 hlist_add_tail_rcu(&sk->sk_node, list); 920 } 921 922 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 923 { 924 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 925 } 926 927 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 928 { 929 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 930 } 931 932 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 933 { 934 sock_hold(sk); 935 __sk_nulls_add_node_rcu(sk, list); 936 } 937 938 static inline void __sk_del_bind_node(struct sock *sk) 939 { 940 __hlist_del(&sk->sk_bind_node); 941 } 942 943 static inline void sk_add_bind_node(struct sock *sk, 944 struct hlist_head *list) 945 { 946 hlist_add_head(&sk->sk_bind_node, list); 947 } 948 949 #define sk_for_each(__sk, list) \ 950 hlist_for_each_entry(__sk, list, sk_node) 951 #define sk_for_each_rcu(__sk, list) \ 952 hlist_for_each_entry_rcu(__sk, list, sk_node) 953 #define sk_nulls_for_each(__sk, node, list) \ 954 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 955 #define sk_nulls_for_each_rcu(__sk, node, list) \ 956 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 957 #define sk_for_each_from(__sk) \ 958 hlist_for_each_entry_from(__sk, sk_node) 959 #define sk_nulls_for_each_from(__sk, node) \ 960 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 961 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 962 #define sk_for_each_safe(__sk, tmp, list) \ 963 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 964 #define sk_for_each_bound(__sk, list) \ 965 hlist_for_each_entry(__sk, list, sk_bind_node) 966 #define sk_for_each_bound_safe(__sk, tmp, list) \ 967 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node) 968 969 /** 970 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 971 * @tpos: the type * to use as a loop cursor. 972 * @pos: the &struct hlist_node to use as a loop cursor. 973 * @head: the head for your list. 974 * @offset: offset of hlist_node within the struct. 975 * 976 */ 977 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 978 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 979 pos != NULL && \ 980 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 981 pos = rcu_dereference(hlist_next_rcu(pos))) 982 983 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 984 { 985 /* Careful only use this in a context where these parameters 986 * can not change and must all be valid, such as recvmsg from 987 * userspace. 988 */ 989 return sk->sk_socket->file->f_cred->user_ns; 990 } 991 992 /* Sock flags */ 993 enum sock_flags { 994 SOCK_DEAD, 995 SOCK_DONE, 996 SOCK_URGINLINE, 997 SOCK_KEEPOPEN, 998 SOCK_LINGER, 999 SOCK_DESTROY, 1000 SOCK_BROADCAST, 1001 SOCK_TIMESTAMP, 1002 SOCK_ZAPPED, 1003 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 1004 SOCK_DBG, /* %SO_DEBUG setting */ 1005 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 1006 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 1007 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 1008 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 1009 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 1010 SOCK_FASYNC, /* fasync() active */ 1011 SOCK_RXQ_OVFL, 1012 SOCK_ZEROCOPY, /* buffers from userspace */ 1013 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 1014 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 1015 * Will use last 4 bytes of packet sent from 1016 * user-space instead. 1017 */ 1018 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 1019 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 1020 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 1021 SOCK_TXTIME, 1022 SOCK_XDP, /* XDP is attached */ 1023 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 1024 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 1025 SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */ 1026 SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */ 1027 }; 1028 1029 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 1030 /* 1031 * The highest bit of sk_tsflags is reserved for kernel-internal 1032 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that 1033 * SOF_TIMESTAMPING* values do not reach this reserved area 1034 */ 1035 #define SOCKCM_FLAG_TS_OPT_ID BIT(31) 1036 1037 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 1038 { 1039 nsk->sk_flags = osk->sk_flags; 1040 } 1041 1042 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 1043 { 1044 __set_bit(flag, &sk->sk_flags); 1045 } 1046 1047 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 1048 { 1049 __clear_bit(flag, &sk->sk_flags); 1050 } 1051 1052 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 1053 int valbool) 1054 { 1055 if (valbool) 1056 sock_set_flag(sk, bit); 1057 else 1058 sock_reset_flag(sk, bit); 1059 } 1060 1061 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 1062 { 1063 return test_bit(flag, &sk->sk_flags); 1064 } 1065 1066 #ifdef CONFIG_NET 1067 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 1068 static inline int sk_memalloc_socks(void) 1069 { 1070 return static_branch_unlikely(&memalloc_socks_key); 1071 } 1072 1073 void __receive_sock(struct file *file); 1074 #else 1075 1076 static inline int sk_memalloc_socks(void) 1077 { 1078 return 0; 1079 } 1080 1081 static inline void __receive_sock(struct file *file) 1082 { } 1083 #endif 1084 1085 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1086 { 1087 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1088 } 1089 1090 static inline void sk_acceptq_removed(struct sock *sk) 1091 { 1092 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1093 } 1094 1095 static inline void sk_acceptq_added(struct sock *sk) 1096 { 1097 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1098 } 1099 1100 /* Note: If you think the test should be: 1101 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1102 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1103 */ 1104 static inline bool sk_acceptq_is_full(const struct sock *sk) 1105 { 1106 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1107 } 1108 1109 /* 1110 * Compute minimal free write space needed to queue new packets. 1111 */ 1112 static inline int sk_stream_min_wspace(const struct sock *sk) 1113 { 1114 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1115 } 1116 1117 static inline int sk_stream_wspace(const struct sock *sk) 1118 { 1119 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1120 } 1121 1122 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1123 { 1124 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1125 } 1126 1127 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1128 { 1129 /* Paired with lockless reads of sk->sk_forward_alloc */ 1130 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1131 } 1132 1133 void sk_stream_write_space(struct sock *sk); 1134 1135 /* OOB backlog add */ 1136 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1137 { 1138 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1139 skb_dst_force(skb); 1140 1141 if (!sk->sk_backlog.tail) 1142 WRITE_ONCE(sk->sk_backlog.head, skb); 1143 else 1144 sk->sk_backlog.tail->next = skb; 1145 1146 WRITE_ONCE(sk->sk_backlog.tail, skb); 1147 skb->next = NULL; 1148 } 1149 1150 /* 1151 * Take into account size of receive queue and backlog queue 1152 * Do not take into account this skb truesize, 1153 * to allow even a single big packet to come. 1154 */ 1155 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1156 { 1157 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1158 1159 return qsize > limit; 1160 } 1161 1162 /* The per-socket spinlock must be held here. */ 1163 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1164 unsigned int limit) 1165 { 1166 if (sk_rcvqueues_full(sk, limit)) 1167 return -ENOBUFS; 1168 1169 /* 1170 * If the skb was allocated from pfmemalloc reserves, only 1171 * allow SOCK_MEMALLOC sockets to use it as this socket is 1172 * helping free memory 1173 */ 1174 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1175 return -ENOMEM; 1176 1177 __sk_add_backlog(sk, skb); 1178 sk->sk_backlog.len += skb->truesize; 1179 return 0; 1180 } 1181 1182 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1183 1184 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1185 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1186 1187 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1188 { 1189 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1190 return __sk_backlog_rcv(sk, skb); 1191 1192 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1193 tcp_v6_do_rcv, 1194 tcp_v4_do_rcv, 1195 sk, skb); 1196 } 1197 1198 static inline void sk_incoming_cpu_update(struct sock *sk) 1199 { 1200 int cpu = raw_smp_processor_id(); 1201 1202 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1203 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1204 } 1205 1206 1207 static inline void sock_rps_save_rxhash(struct sock *sk, 1208 const struct sk_buff *skb) 1209 { 1210 #ifdef CONFIG_RPS 1211 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1212 * here, and another one in sock_rps_record_flow(). 1213 */ 1214 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1215 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1216 #endif 1217 } 1218 1219 static inline void sock_rps_reset_rxhash(struct sock *sk) 1220 { 1221 #ifdef CONFIG_RPS 1222 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1223 WRITE_ONCE(sk->sk_rxhash, 0); 1224 #endif 1225 } 1226 1227 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1228 ({ int __rc, __dis = __sk->sk_disconnects; \ 1229 release_sock(__sk); \ 1230 __rc = __condition; \ 1231 if (!__rc) { \ 1232 *(__timeo) = wait_woken(__wait, \ 1233 TASK_INTERRUPTIBLE, \ 1234 *(__timeo)); \ 1235 } \ 1236 sched_annotate_sleep(); \ 1237 lock_sock(__sk); \ 1238 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1239 __rc; \ 1240 }) 1241 1242 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1243 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1244 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1245 int sk_stream_error(struct sock *sk, int flags, int err); 1246 void sk_stream_kill_queues(struct sock *sk); 1247 void sk_set_memalloc(struct sock *sk); 1248 void sk_clear_memalloc(struct sock *sk); 1249 1250 void __sk_flush_backlog(struct sock *sk); 1251 1252 static inline bool sk_flush_backlog(struct sock *sk) 1253 { 1254 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1255 __sk_flush_backlog(sk); 1256 return true; 1257 } 1258 return false; 1259 } 1260 1261 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1262 1263 struct request_sock_ops; 1264 struct timewait_sock_ops; 1265 struct inet_hashinfo; 1266 struct raw_hashinfo; 1267 struct smc_hashinfo; 1268 struct module; 1269 struct sk_psock; 1270 1271 /* 1272 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1273 * un-modified. Special care is taken when initializing object to zero. 1274 */ 1275 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1276 { 1277 if (offsetof(struct sock, sk_node.next) != 0) 1278 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1279 memset(&sk->sk_node.pprev, 0, 1280 size - offsetof(struct sock, sk_node.pprev)); 1281 } 1282 1283 struct proto_accept_arg { 1284 int flags; 1285 int err; 1286 int is_empty; 1287 bool kern; 1288 }; 1289 1290 /* Networking protocol blocks we attach to sockets. 1291 * socket layer -> transport layer interface 1292 */ 1293 struct proto { 1294 void (*close)(struct sock *sk, 1295 long timeout); 1296 int (*pre_connect)(struct sock *sk, 1297 struct sockaddr_unsized *uaddr, 1298 int addr_len); 1299 int (*connect)(struct sock *sk, 1300 struct sockaddr_unsized *uaddr, 1301 int addr_len); 1302 int (*disconnect)(struct sock *sk, int flags); 1303 1304 struct sock * (*accept)(struct sock *sk, 1305 struct proto_accept_arg *arg); 1306 1307 int (*ioctl)(struct sock *sk, int cmd, 1308 int *karg); 1309 int (*init)(struct sock *sk); 1310 void (*destroy)(struct sock *sk); 1311 void (*shutdown)(struct sock *sk, int how); 1312 int (*setsockopt)(struct sock *sk, int level, 1313 int optname, sockptr_t optval, 1314 unsigned int optlen); 1315 int (*getsockopt)(struct sock *sk, int level, 1316 int optname, char __user *optval, 1317 int __user *option); 1318 void (*keepalive)(struct sock *sk, int valbool); 1319 #ifdef CONFIG_COMPAT 1320 int (*compat_ioctl)(struct sock *sk, 1321 unsigned int cmd, unsigned long arg); 1322 #endif 1323 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1324 size_t len); 1325 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1326 size_t len, int flags); 1327 void (*splice_eof)(struct socket *sock); 1328 int (*bind)(struct sock *sk, 1329 struct sockaddr_unsized *addr, int addr_len); 1330 int (*bind_add)(struct sock *sk, 1331 struct sockaddr_unsized *addr, int addr_len); 1332 1333 int (*backlog_rcv) (struct sock *sk, 1334 struct sk_buff *skb); 1335 bool (*bpf_bypass_getsockopt)(int level, 1336 int optname); 1337 1338 void (*release_cb)(struct sock *sk); 1339 1340 /* Keeping track of sk's, looking them up, and port selection methods. */ 1341 int (*hash)(struct sock *sk); 1342 void (*unhash)(struct sock *sk); 1343 void (*rehash)(struct sock *sk); 1344 int (*get_port)(struct sock *sk, unsigned short snum); 1345 void (*put_port)(struct sock *sk); 1346 #ifdef CONFIG_BPF_SYSCALL 1347 int (*psock_update_sk_prot)(struct sock *sk, 1348 struct sk_psock *psock, 1349 bool restore); 1350 #endif 1351 1352 /* Keeping track of sockets in use */ 1353 #ifdef CONFIG_PROC_FS 1354 unsigned int inuse_idx; 1355 #endif 1356 1357 bool (*stream_memory_free)(const struct sock *sk, int wake); 1358 bool (*sock_is_readable)(struct sock *sk); 1359 /* Memory pressure */ 1360 void (*enter_memory_pressure)(struct sock *sk); 1361 void (*leave_memory_pressure)(struct sock *sk); 1362 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1363 int __percpu *per_cpu_fw_alloc; 1364 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1365 1366 /* 1367 * Pressure flag: try to collapse. 1368 * Technical note: it is used by multiple contexts non atomically. 1369 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1370 * All the __sk_mem_schedule() is of this nature: accounting 1371 * is strict, actions are advisory and have some latency. 1372 */ 1373 unsigned long *memory_pressure; 1374 long *sysctl_mem; 1375 1376 int *sysctl_wmem; 1377 int *sysctl_rmem; 1378 u32 sysctl_wmem_offset; 1379 u32 sysctl_rmem_offset; 1380 1381 int max_header; 1382 bool no_autobind; 1383 1384 struct kmem_cache *slab; 1385 unsigned int obj_size; 1386 unsigned int freeptr_offset; 1387 unsigned int ipv6_pinfo_offset; 1388 slab_flags_t slab_flags; 1389 unsigned int useroffset; /* Usercopy region offset */ 1390 unsigned int usersize; /* Usercopy region size */ 1391 1392 struct request_sock_ops *rsk_prot; 1393 struct timewait_sock_ops *twsk_prot; 1394 1395 union { 1396 struct inet_hashinfo *hashinfo; 1397 struct raw_hashinfo *raw_hash; 1398 struct smc_hashinfo *smc_hash; 1399 } h; 1400 1401 struct module *owner; 1402 1403 char name[32]; 1404 1405 struct list_head node; 1406 int (*diag_destroy)(struct sock *sk, int err); 1407 } __randomize_layout; 1408 1409 int proto_register(struct proto *prot, int alloc_slab); 1410 void proto_unregister(struct proto *prot); 1411 int sock_load_diag_module(int family, int protocol); 1412 1413 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1414 1415 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1416 { 1417 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1418 return false; 1419 1420 return sk->sk_prot->stream_memory_free ? 1421 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1422 tcp_stream_memory_free, sk, wake) : true; 1423 } 1424 1425 static inline bool sk_stream_memory_free(const struct sock *sk) 1426 { 1427 return __sk_stream_memory_free(sk, 0); 1428 } 1429 1430 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1431 { 1432 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1433 __sk_stream_memory_free(sk, wake); 1434 } 1435 1436 static inline bool sk_stream_is_writeable(const struct sock *sk) 1437 { 1438 return __sk_stream_is_writeable(sk, 0); 1439 } 1440 1441 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1442 struct cgroup *ancestor) 1443 { 1444 #ifdef CONFIG_SOCK_CGROUP_DATA 1445 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1446 ancestor); 1447 #else 1448 return -ENOTSUPP; 1449 #endif 1450 } 1451 1452 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1453 1454 static inline void sk_sockets_allocated_dec(struct sock *sk) 1455 { 1456 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1457 SK_ALLOC_PERCPU_COUNTER_BATCH); 1458 } 1459 1460 static inline void sk_sockets_allocated_inc(struct sock *sk) 1461 { 1462 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1463 SK_ALLOC_PERCPU_COUNTER_BATCH); 1464 } 1465 1466 static inline u64 1467 sk_sockets_allocated_read_positive(struct sock *sk) 1468 { 1469 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1470 } 1471 1472 static inline int 1473 proto_sockets_allocated_sum_positive(struct proto *prot) 1474 { 1475 return percpu_counter_sum_positive(prot->sockets_allocated); 1476 } 1477 1478 #ifdef CONFIG_PROC_FS 1479 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1480 struct prot_inuse { 1481 int all; 1482 int val[PROTO_INUSE_NR]; 1483 }; 1484 1485 static inline void sock_prot_inuse_add(const struct net *net, 1486 const struct proto *prot, int val) 1487 { 1488 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1489 } 1490 1491 static inline void sock_inuse_add(const struct net *net, int val) 1492 { 1493 this_cpu_add(net->core.prot_inuse->all, val); 1494 } 1495 1496 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1497 int sock_inuse_get(struct net *net); 1498 #else 1499 static inline void sock_prot_inuse_add(const struct net *net, 1500 const struct proto *prot, int val) 1501 { 1502 } 1503 1504 static inline void sock_inuse_add(const struct net *net, int val) 1505 { 1506 } 1507 #endif 1508 1509 1510 /* With per-bucket locks this operation is not-atomic, so that 1511 * this version is not worse. 1512 */ 1513 static inline int __sk_prot_rehash(struct sock *sk) 1514 { 1515 sk->sk_prot->unhash(sk); 1516 return sk->sk_prot->hash(sk); 1517 } 1518 1519 /* About 10 seconds */ 1520 #define SOCK_DESTROY_TIME (10*HZ) 1521 1522 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1523 #define PROT_SOCK 1024 1524 1525 #define SHUTDOWN_MASK 3 1526 #define RCV_SHUTDOWN 1 1527 #define SEND_SHUTDOWN 2 1528 1529 #define SOCK_BINDADDR_LOCK 4 1530 #define SOCK_BINDPORT_LOCK 8 1531 /** 1532 * define SOCK_CONNECT_BIND - &sock->sk_userlocks flag for auto-bind at connect() time 1533 */ 1534 #define SOCK_CONNECT_BIND 16 1535 1536 struct socket_alloc { 1537 struct socket socket; 1538 struct inode vfs_inode; 1539 }; 1540 1541 static inline struct socket *SOCKET_I(struct inode *inode) 1542 { 1543 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1544 } 1545 1546 static inline struct inode *SOCK_INODE(struct socket *socket) 1547 { 1548 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1549 } 1550 1551 /* 1552 * Functions for memory accounting 1553 */ 1554 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1555 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1556 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1557 void __sk_mem_reclaim(struct sock *sk, int amount); 1558 1559 #define SK_MEM_SEND 0 1560 #define SK_MEM_RECV 1 1561 1562 /* sysctl_mem values are in pages */ 1563 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1564 { 1565 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1566 } 1567 1568 static inline int sk_mem_pages(int amt) 1569 { 1570 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1571 } 1572 1573 static inline bool sk_has_account(struct sock *sk) 1574 { 1575 /* return true if protocol supports memory accounting */ 1576 return !!sk->sk_prot->memory_allocated; 1577 } 1578 1579 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1580 { 1581 int delta; 1582 1583 if (!sk_has_account(sk)) 1584 return true; 1585 delta = size - sk->sk_forward_alloc; 1586 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1587 } 1588 1589 static inline bool 1590 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc) 1591 { 1592 int delta; 1593 1594 if (!sk_has_account(sk)) 1595 return true; 1596 delta = size - sk->sk_forward_alloc; 1597 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1598 pfmemalloc; 1599 } 1600 1601 static inline bool 1602 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size) 1603 { 1604 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb)); 1605 } 1606 1607 static inline int sk_unused_reserved_mem(const struct sock *sk) 1608 { 1609 int unused_mem; 1610 1611 if (likely(!sk->sk_reserved_mem)) 1612 return 0; 1613 1614 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1615 atomic_read(&sk->sk_rmem_alloc); 1616 1617 return unused_mem > 0 ? unused_mem : 0; 1618 } 1619 1620 static inline void sk_mem_reclaim(struct sock *sk) 1621 { 1622 int reclaimable; 1623 1624 if (!sk_has_account(sk)) 1625 return; 1626 1627 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1628 1629 if (reclaimable >= (int)PAGE_SIZE) 1630 __sk_mem_reclaim(sk, reclaimable); 1631 } 1632 1633 static inline void sk_mem_reclaim_final(struct sock *sk) 1634 { 1635 sk->sk_reserved_mem = 0; 1636 sk_mem_reclaim(sk); 1637 } 1638 1639 static inline void sk_mem_charge(struct sock *sk, int size) 1640 { 1641 if (!sk_has_account(sk)) 1642 return; 1643 sk_forward_alloc_add(sk, -size); 1644 } 1645 1646 static inline void sk_mem_uncharge(struct sock *sk, int size) 1647 { 1648 if (!sk_has_account(sk)) 1649 return; 1650 sk_forward_alloc_add(sk, size); 1651 sk_mem_reclaim(sk); 1652 } 1653 1654 void __sk_charge(struct sock *sk, gfp_t gfp); 1655 1656 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 1657 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1658 { 1659 __module_get(owner); 1660 sk->sk_owner = owner; 1661 } 1662 1663 static inline void sk_owner_clear(struct sock *sk) 1664 { 1665 sk->sk_owner = NULL; 1666 } 1667 1668 static inline void sk_owner_put(struct sock *sk) 1669 { 1670 module_put(sk->sk_owner); 1671 } 1672 #else 1673 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1674 { 1675 } 1676 1677 static inline void sk_owner_clear(struct sock *sk) 1678 { 1679 } 1680 1681 static inline void sk_owner_put(struct sock *sk) 1682 { 1683 } 1684 #endif 1685 /* 1686 * Macro so as to not evaluate some arguments when 1687 * lockdep is not enabled. 1688 * 1689 * Mark both the sk_lock and the sk_lock.slock as a 1690 * per-address-family lock class. 1691 */ 1692 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1693 do { \ 1694 sk_owner_set(sk, THIS_MODULE); \ 1695 sk->sk_lock.owned = 0; \ 1696 init_waitqueue_head(&sk->sk_lock.wq); \ 1697 spin_lock_init(&(sk)->sk_lock.slock); \ 1698 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1699 sizeof((sk)->sk_lock)); \ 1700 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1701 (skey), (sname)); \ 1702 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1703 } while (0) 1704 1705 static inline bool lockdep_sock_is_held(const struct sock *sk) 1706 { 1707 return lockdep_is_held(&sk->sk_lock) || 1708 lockdep_is_held(&sk->sk_lock.slock); 1709 } 1710 1711 void lock_sock_nested(struct sock *sk, int subclass); 1712 1713 static inline void lock_sock(struct sock *sk) 1714 { 1715 lock_sock_nested(sk, 0); 1716 } 1717 1718 void __release_sock(struct sock *sk); 1719 void release_sock(struct sock *sk); 1720 1721 /* BH context may only use the following locking interface. */ 1722 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1723 #define bh_lock_sock_nested(__sk) \ 1724 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1725 SINGLE_DEPTH_NESTING) 1726 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1727 1728 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1729 1730 /** 1731 * lock_sock_fast - fast version of lock_sock 1732 * @sk: socket 1733 * 1734 * This version should be used for very small section, where process won't block 1735 * return false if fast path is taken: 1736 * 1737 * sk_lock.slock locked, owned = 0, BH disabled 1738 * 1739 * return true if slow path is taken: 1740 * 1741 * sk_lock.slock unlocked, owned = 1, BH enabled 1742 */ 1743 static inline bool lock_sock_fast(struct sock *sk) 1744 { 1745 /* The sk_lock has mutex_lock() semantics here. */ 1746 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1747 1748 return __lock_sock_fast(sk); 1749 } 1750 1751 /* fast socket lock variant for caller already holding a [different] socket lock */ 1752 static inline bool lock_sock_fast_nested(struct sock *sk) 1753 { 1754 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1755 1756 return __lock_sock_fast(sk); 1757 } 1758 1759 /** 1760 * unlock_sock_fast - complement of lock_sock_fast 1761 * @sk: socket 1762 * @slow: slow mode 1763 * 1764 * fast unlock socket for user context. 1765 * If slow mode is on, we call regular release_sock() 1766 */ 1767 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1768 __releases(&sk->sk_lock.slock) 1769 { 1770 if (slow) { 1771 release_sock(sk); 1772 __release(&sk->sk_lock.slock); 1773 } else { 1774 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1775 spin_unlock_bh(&sk->sk_lock.slock); 1776 } 1777 } 1778 1779 void sockopt_lock_sock(struct sock *sk); 1780 void sockopt_release_sock(struct sock *sk); 1781 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1782 bool sockopt_capable(int cap); 1783 1784 /* Used by processes to "lock" a socket state, so that 1785 * interrupts and bottom half handlers won't change it 1786 * from under us. It essentially blocks any incoming 1787 * packets, so that we won't get any new data or any 1788 * packets that change the state of the socket. 1789 * 1790 * While locked, BH processing will add new packets to 1791 * the backlog queue. This queue is processed by the 1792 * owner of the socket lock right before it is released. 1793 * 1794 * Since ~2.3.5 it is also exclusive sleep lock serializing 1795 * accesses from user process context. 1796 */ 1797 1798 static inline void sock_owned_by_me(const struct sock *sk) 1799 { 1800 #ifdef CONFIG_LOCKDEP 1801 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1802 #endif 1803 } 1804 1805 static inline void sock_not_owned_by_me(const struct sock *sk) 1806 { 1807 #ifdef CONFIG_LOCKDEP 1808 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1809 #endif 1810 } 1811 1812 static inline bool sock_owned_by_user(const struct sock *sk) 1813 { 1814 sock_owned_by_me(sk); 1815 return sk->sk_lock.owned; 1816 } 1817 1818 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1819 { 1820 return sk->sk_lock.owned; 1821 } 1822 1823 static inline void sock_release_ownership(struct sock *sk) 1824 { 1825 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1826 sk->sk_lock.owned = 0; 1827 1828 /* The sk_lock has mutex_unlock() semantics: */ 1829 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1830 } 1831 1832 /* no reclassification while locks are held */ 1833 static inline bool sock_allow_reclassification(const struct sock *csk) 1834 { 1835 struct sock *sk = (struct sock *)csk; 1836 1837 return !sock_owned_by_user_nocheck(sk) && 1838 !spin_is_locked(&sk->sk_lock.slock); 1839 } 1840 1841 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1842 struct proto *prot, int kern); 1843 void sk_free(struct sock *sk); 1844 void sk_net_refcnt_upgrade(struct sock *sk); 1845 void sk_destruct(struct sock *sk); 1846 struct sock *sk_clone(const struct sock *sk, const gfp_t priority, bool lock); 1847 1848 static inline struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1849 { 1850 return sk_clone(sk, priority, true); 1851 } 1852 1853 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1854 gfp_t priority); 1855 void sock_wfree(struct sk_buff *skb); 1856 void __sock_wfree(struct sk_buff *skb); 1857 void tcp_wfree(struct sk_buff *skb); 1858 1859 static inline bool is_skb_wmem(const struct sk_buff *skb) 1860 { 1861 return skb->destructor == sock_wfree || 1862 (IS_ENABLED(CONFIG_INET) && skb->destructor == __sock_wfree) || 1863 (IS_ENABLED(CONFIG_INET) && skb->destructor == tcp_wfree); 1864 } 1865 1866 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1867 gfp_t priority); 1868 void skb_orphan_partial(struct sk_buff *skb); 1869 void sock_rfree(struct sk_buff *skb); 1870 void sock_rmem_free(struct sk_buff *skb); 1871 void sock_efree(struct sk_buff *skb); 1872 #ifdef CONFIG_INET 1873 void sock_edemux(struct sk_buff *skb); 1874 void sock_pfree(struct sk_buff *skb); 1875 1876 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk) 1877 { 1878 skb_orphan(skb); 1879 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 1880 skb->sk = sk; 1881 skb->destructor = sock_edemux; 1882 } 1883 } 1884 #else 1885 #define sock_edemux sock_efree 1886 #endif 1887 1888 int sk_setsockopt(struct sock *sk, int level, int optname, 1889 sockptr_t optval, unsigned int optlen); 1890 int sock_setsockopt(struct socket *sock, int level, int op, 1891 sockptr_t optval, unsigned int optlen); 1892 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1893 int optname, sockptr_t optval, int optlen); 1894 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1895 int optname, sockptr_t optval, sockptr_t optlen); 1896 1897 int sk_getsockopt(struct sock *sk, int level, int optname, 1898 sockptr_t optval, sockptr_t optlen); 1899 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1900 bool timeval, bool time32); 1901 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1902 unsigned long data_len, int noblock, 1903 int *errcode, int max_page_order); 1904 1905 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1906 unsigned long size, 1907 int noblock, int *errcode) 1908 { 1909 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1910 } 1911 1912 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1913 void *sock_kmemdup(struct sock *sk, const void *src, 1914 int size, gfp_t priority); 1915 void sock_kfree_s(struct sock *sk, void *mem, int size); 1916 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1917 void sk_send_sigurg(struct sock *sk); 1918 1919 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1920 { 1921 if (sk->sk_socket) 1922 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1923 WRITE_ONCE(sk->sk_prot, proto); 1924 } 1925 1926 struct sockcm_cookie { 1927 u64 transmit_time; 1928 u32 mark; 1929 u32 tsflags; 1930 u32 ts_opt_id; 1931 u32 priority; 1932 u32 dmabuf_id; 1933 }; 1934 1935 static inline void sockcm_init(struct sockcm_cookie *sockc, 1936 const struct sock *sk) 1937 { 1938 *sockc = (struct sockcm_cookie) { 1939 .mark = READ_ONCE(sk->sk_mark), 1940 .tsflags = READ_ONCE(sk->sk_tsflags), 1941 .priority = READ_ONCE(sk->sk_priority), 1942 }; 1943 } 1944 1945 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1946 struct sockcm_cookie *sockc); 1947 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1948 struct sockcm_cookie *sockc); 1949 1950 /* 1951 * Functions to fill in entries in struct proto_ops when a protocol 1952 * does not implement a particular function. 1953 */ 1954 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len); 1955 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr, int len, int flags); 1956 int sock_no_socketpair(struct socket *, struct socket *); 1957 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *); 1958 int sock_no_getname(struct socket *, struct sockaddr *, int); 1959 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1960 int sock_no_listen(struct socket *, int); 1961 int sock_no_shutdown(struct socket *, int); 1962 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1963 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1964 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1965 int sock_no_mmap(struct file *file, struct socket *sock, 1966 struct vm_area_struct *vma); 1967 1968 /* 1969 * Functions to fill in entries in struct proto_ops when a protocol 1970 * uses the inet style. 1971 */ 1972 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1973 char __user *optval, int __user *optlen); 1974 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1975 int flags); 1976 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1977 sockptr_t optval, unsigned int optlen); 1978 1979 void sk_common_release(struct sock *sk); 1980 1981 /* 1982 * Default socket callbacks and setup code 1983 */ 1984 1985 /* Initialise core socket variables using an explicit uid. */ 1986 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1987 1988 /* Initialise core socket variables. 1989 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1990 */ 1991 void sock_init_data(struct socket *sock, struct sock *sk); 1992 1993 /* 1994 * Socket reference counting postulates. 1995 * 1996 * * Each user of socket SHOULD hold a reference count. 1997 * * Each access point to socket (an hash table bucket, reference from a list, 1998 * running timer, skb in flight MUST hold a reference count. 1999 * * When reference count hits 0, it means it will never increase back. 2000 * * When reference count hits 0, it means that no references from 2001 * outside exist to this socket and current process on current CPU 2002 * is last user and may/should destroy this socket. 2003 * * sk_free is called from any context: process, BH, IRQ. When 2004 * it is called, socket has no references from outside -> sk_free 2005 * may release descendant resources allocated by the socket, but 2006 * to the time when it is called, socket is NOT referenced by any 2007 * hash tables, lists etc. 2008 * * Packets, delivered from outside (from network or from another process) 2009 * and enqueued on receive/error queues SHOULD NOT grab reference count, 2010 * when they sit in queue. Otherwise, packets will leak to hole, when 2011 * socket is looked up by one cpu and unhasing is made by another CPU. 2012 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 2013 * (leak to backlog). Packet socket does all the processing inside 2014 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 2015 * use separate SMP lock, so that they are prone too. 2016 */ 2017 2018 /* Ungrab socket and destroy it, if it was the last reference. */ 2019 static inline void sock_put(struct sock *sk) 2020 { 2021 if (refcount_dec_and_test(&sk->sk_refcnt)) 2022 sk_free(sk); 2023 } 2024 /* Generic version of sock_put(), dealing with all sockets 2025 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 2026 */ 2027 void sock_gen_put(struct sock *sk); 2028 2029 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 2030 unsigned int trim_cap, bool refcounted); 2031 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 2032 const int nested) 2033 { 2034 return __sk_receive_skb(sk, skb, nested, 1, true); 2035 } 2036 2037 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 2038 { 2039 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2040 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2041 return; 2042 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2043 * other WRITE_ONCE() because socket lock might be not held. 2044 */ 2045 if (READ_ONCE(sk->sk_tx_queue_mapping) != tx_queue) { 2046 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 2047 WRITE_ONCE(sk->sk_tx_queue_mapping_jiffies, jiffies); 2048 return; 2049 } 2050 2051 /* Refresh sk_tx_queue_mapping_jiffies if too old. */ 2052 if (time_is_before_jiffies(READ_ONCE(sk->sk_tx_queue_mapping_jiffies) + HZ)) 2053 WRITE_ONCE(sk->sk_tx_queue_mapping_jiffies, jiffies); 2054 } 2055 2056 #define NO_QUEUE_MAPPING USHRT_MAX 2057 2058 static inline void sk_tx_queue_clear(struct sock *sk) 2059 { 2060 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2061 * other WRITE_ONCE() because socket lock might be not held. 2062 */ 2063 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 2064 } 2065 2066 int sk_tx_queue_get(const struct sock *sk); 2067 2068 static inline void __sk_rx_queue_set(struct sock *sk, 2069 const struct sk_buff *skb, 2070 bool force_set) 2071 { 2072 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2073 if (skb_rx_queue_recorded(skb)) { 2074 u16 rx_queue = skb_get_rx_queue(skb); 2075 2076 if (force_set || 2077 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2078 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2079 } 2080 #endif 2081 } 2082 2083 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2084 { 2085 __sk_rx_queue_set(sk, skb, true); 2086 } 2087 2088 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2089 { 2090 __sk_rx_queue_set(sk, skb, false); 2091 } 2092 2093 static inline void sk_rx_queue_clear(struct sock *sk) 2094 { 2095 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2096 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2097 #endif 2098 } 2099 2100 static inline int sk_rx_queue_get(const struct sock *sk) 2101 { 2102 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2103 if (sk) { 2104 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2105 2106 if (res != NO_QUEUE_MAPPING) 2107 return res; 2108 } 2109 #endif 2110 2111 return -1; 2112 } 2113 2114 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2115 { 2116 WRITE_ONCE(sk->sk_socket, sock); 2117 if (sock) { 2118 WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid); 2119 WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino); 2120 } else { 2121 /* Note: sk_uid is unchanged. */ 2122 WRITE_ONCE(sk->sk_ino, 0); 2123 } 2124 } 2125 2126 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2127 { 2128 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2129 return &rcu_dereference_raw(sk->sk_wq)->wait; 2130 } 2131 /* Detach socket from process context. 2132 * Announce socket dead, detach it from wait queue and inode. 2133 * Note that parent inode held reference count on this struct sock, 2134 * we do not release it in this function, because protocol 2135 * probably wants some additional cleanups or even continuing 2136 * to work with this socket (TCP). 2137 */ 2138 static inline void sock_orphan(struct sock *sk) 2139 { 2140 write_lock_bh(&sk->sk_callback_lock); 2141 sock_set_flag(sk, SOCK_DEAD); 2142 sk_set_socket(sk, NULL); 2143 sk->sk_wq = NULL; 2144 write_unlock_bh(&sk->sk_callback_lock); 2145 } 2146 2147 static inline void sock_graft(struct sock *sk, struct socket *parent) 2148 { 2149 WARN_ON(parent->sk); 2150 write_lock_bh(&sk->sk_callback_lock); 2151 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2152 parent->sk = sk; 2153 sk_set_socket(sk, parent); 2154 security_sock_graft(sk, parent); 2155 write_unlock_bh(&sk->sk_callback_lock); 2156 } 2157 2158 static inline u64 sock_i_ino(const struct sock *sk) 2159 { 2160 /* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */ 2161 return READ_ONCE(sk->sk_ino); 2162 } 2163 2164 static inline kuid_t sk_uid(const struct sock *sk) 2165 { 2166 /* Paired with WRITE_ONCE() in sockfs_setattr() */ 2167 return READ_ONCE(sk->sk_uid); 2168 } 2169 2170 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2171 { 2172 return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0); 2173 } 2174 2175 static inline u32 net_tx_rndhash(void) 2176 { 2177 u32 v = get_random_u32(); 2178 2179 return v ?: 1; 2180 } 2181 2182 static inline void sk_set_txhash(struct sock *sk) 2183 { 2184 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2185 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2186 } 2187 2188 static inline bool sk_rethink_txhash(struct sock *sk) 2189 { 2190 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2191 sk_set_txhash(sk); 2192 return true; 2193 } 2194 return false; 2195 } 2196 2197 static inline struct dst_entry * 2198 __sk_dst_get(const struct sock *sk) 2199 { 2200 return rcu_dereference_check(sk->sk_dst_cache, 2201 lockdep_sock_is_held(sk)); 2202 } 2203 2204 static inline struct dst_entry * 2205 sk_dst_get(const struct sock *sk) 2206 { 2207 struct dst_entry *dst; 2208 2209 rcu_read_lock(); 2210 dst = rcu_dereference(sk->sk_dst_cache); 2211 if (dst && !rcuref_get(&dst->__rcuref)) 2212 dst = NULL; 2213 rcu_read_unlock(); 2214 return dst; 2215 } 2216 2217 static inline void __dst_negative_advice(struct sock *sk) 2218 { 2219 struct dst_entry *dst = __sk_dst_get(sk); 2220 2221 if (dst && dst->ops->negative_advice) 2222 dst->ops->negative_advice(sk, dst); 2223 } 2224 2225 static inline void dst_negative_advice(struct sock *sk) 2226 { 2227 sk_rethink_txhash(sk); 2228 __dst_negative_advice(sk); 2229 } 2230 2231 static inline void 2232 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2233 { 2234 struct dst_entry *old_dst; 2235 2236 sk_tx_queue_clear(sk); 2237 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2238 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2239 lockdep_sock_is_held(sk)); 2240 rcu_assign_pointer(sk->sk_dst_cache, dst); 2241 dst_release(old_dst); 2242 } 2243 2244 static inline void 2245 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2246 { 2247 struct dst_entry *old_dst; 2248 2249 sk_tx_queue_clear(sk); 2250 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2251 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst))); 2252 dst_release(old_dst); 2253 } 2254 2255 static inline void 2256 __sk_dst_reset(struct sock *sk) 2257 { 2258 __sk_dst_set(sk, NULL); 2259 } 2260 2261 static inline void 2262 sk_dst_reset(struct sock *sk) 2263 { 2264 sk_dst_set(sk, NULL); 2265 } 2266 2267 /* Re-roll the socket txhash. On a rehash, IPv6 also drops the cached route 2268 * so the next transmit re-selects an ECMP path; IPv4 keeps its route, since 2269 * IPv4 ECMP path selection does not use sk_txhash. 2270 */ 2271 static inline bool __sk_rethink_txhash_reset_dst(struct sock *sk) 2272 { 2273 if (sk_rethink_txhash(sk)) { 2274 if (sk->sk_family == AF_INET6) 2275 __sk_dst_reset(sk); 2276 return true; 2277 } 2278 return false; 2279 } 2280 2281 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2282 2283 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2284 2285 static inline void sk_dst_confirm(struct sock *sk) 2286 { 2287 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2288 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2289 } 2290 2291 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2292 { 2293 if (skb_get_dst_pending_confirm(skb)) { 2294 struct sock *sk = skb->sk; 2295 2296 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2297 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2298 neigh_confirm(n); 2299 } 2300 } 2301 2302 bool sk_mc_loop(const struct sock *sk); 2303 2304 static inline bool sk_can_gso(const struct sock *sk) 2305 { 2306 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2307 } 2308 2309 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2310 2311 static inline void sk_gso_disable(struct sock *sk) 2312 { 2313 sk->sk_gso_disabled = 1; 2314 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2315 } 2316 2317 bool sk_has_decrypt_user(const struct sock *sk); 2318 2319 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2320 struct iov_iter *from, char *to, 2321 int copy, int offset) 2322 { 2323 if (skb->ip_summed == CHECKSUM_NONE) { 2324 __wsum csum = 0; 2325 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2326 return -EFAULT; 2327 skb->csum = csum_block_add(skb->csum, csum, offset); 2328 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2329 if (!copy_from_iter_full_nocache(to, copy, from)) 2330 return -EFAULT; 2331 } else if (!copy_from_iter_full(to, copy, from)) 2332 return -EFAULT; 2333 2334 return 0; 2335 } 2336 2337 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2338 struct iov_iter *from, int copy) 2339 { 2340 int err, offset = skb->len; 2341 2342 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2343 copy, offset); 2344 if (err) 2345 __skb_trim(skb, offset); 2346 2347 return err; 2348 } 2349 2350 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2351 struct sk_buff *skb, 2352 struct page *page, 2353 int off, int copy) 2354 { 2355 int err; 2356 2357 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2358 copy, skb->len); 2359 if (err) 2360 return err; 2361 2362 skb_len_add(skb, copy); 2363 sk_wmem_queued_add(sk, copy); 2364 sk_mem_charge(sk, copy); 2365 return 0; 2366 } 2367 2368 #define SK_WMEM_ALLOC_BIAS 1 2369 /** 2370 * sk_wmem_alloc_get - returns write allocations 2371 * @sk: socket 2372 * 2373 * Return: sk_wmem_alloc minus initial offset of one 2374 */ 2375 static inline int sk_wmem_alloc_get(const struct sock *sk) 2376 { 2377 return refcount_read(&sk->sk_wmem_alloc) - SK_WMEM_ALLOC_BIAS; 2378 } 2379 2380 /** 2381 * sk_rmem_alloc_get - returns read allocations 2382 * @sk: socket 2383 * 2384 * Return: sk_rmem_alloc 2385 */ 2386 static inline int sk_rmem_alloc_get(const struct sock *sk) 2387 { 2388 return atomic_read(&sk->sk_rmem_alloc); 2389 } 2390 2391 /** 2392 * sk_has_allocations - check if allocations are outstanding 2393 * @sk: socket 2394 * 2395 * Return: true if socket has write or read allocations 2396 */ 2397 static inline bool sk_has_allocations(const struct sock *sk) 2398 { 2399 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2400 } 2401 2402 /** 2403 * skwq_has_sleeper - check if there are any waiting processes 2404 * @wq: struct socket_wq 2405 * 2406 * Return: true if socket_wq has waiting processes 2407 * 2408 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2409 * barrier call. They were added due to the race found within the tcp code. 2410 * 2411 * Consider following tcp code paths:: 2412 * 2413 * CPU1 CPU2 2414 * sys_select receive packet 2415 * ... ... 2416 * __add_wait_queue update tp->rcv_nxt 2417 * ... ... 2418 * tp->rcv_nxt check sock_def_readable 2419 * ... { 2420 * schedule rcu_read_lock(); 2421 * wq = rcu_dereference(sk->sk_wq); 2422 * if (wq && waitqueue_active(&wq->wait)) 2423 * wake_up_interruptible(&wq->wait) 2424 * ... 2425 * } 2426 * 2427 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2428 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2429 * could then endup calling schedule and sleep forever if there are no more 2430 * data on the socket. 2431 * 2432 */ 2433 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2434 { 2435 return wq && wq_has_sleeper(&wq->wait); 2436 } 2437 2438 /** 2439 * sock_poll_wait - wrapper for the poll_wait call. 2440 * @filp: file 2441 * @sock: socket to wait on 2442 * @p: poll_table 2443 * 2444 * See the comments in the wq_has_sleeper function. 2445 */ 2446 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2447 poll_table *p) 2448 { 2449 /* Provides a barrier we need to be sure we are in sync 2450 * with the socket flags modification. 2451 * 2452 * This memory barrier is paired in the wq_has_sleeper. 2453 */ 2454 poll_wait(filp, &sock->wq.wait, p); 2455 } 2456 2457 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2458 { 2459 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2460 u32 txhash = READ_ONCE(sk->sk_txhash); 2461 2462 if (txhash) { 2463 skb->l4_hash = 1; 2464 skb->hash = txhash; 2465 } 2466 } 2467 2468 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2469 2470 /* 2471 * Queue a received datagram if it will fit. Stream and sequenced 2472 * protocols can't normally use this as they need to fit buffers in 2473 * and play with them. 2474 * 2475 * Inlined as it's very short and called for pretty much every 2476 * packet ever received. 2477 */ 2478 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2479 { 2480 skb_orphan(skb); 2481 skb->sk = sk; 2482 skb->destructor = sock_rfree; 2483 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2484 sk_mem_charge(sk, skb->truesize); 2485 } 2486 2487 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2488 { 2489 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2490 skb_orphan(skb); 2491 skb->destructor = sock_efree; 2492 skb->sk = sk; 2493 return true; 2494 } 2495 return false; 2496 } 2497 2498 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2499 { 2500 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2501 if (skb) { 2502 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2503 skb_set_owner_r(skb, sk); 2504 return skb; 2505 } 2506 __kfree_skb(skb); 2507 } 2508 return NULL; 2509 } 2510 2511 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2512 { 2513 if (skb->destructor != sock_wfree) { 2514 skb_orphan(skb); 2515 return; 2516 } 2517 skb->slow_gro = 1; 2518 } 2519 2520 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2521 unsigned long expires); 2522 2523 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2524 2525 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2526 2527 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2528 struct sk_buff *skb, unsigned int flags, 2529 void (*destructor)(struct sock *sk, 2530 struct sk_buff *skb)); 2531 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2532 2533 enum skb_drop_reason 2534 sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb); 2535 2536 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2537 { 2538 enum skb_drop_reason drop_reason = sock_queue_rcv_skb_reason(sk, skb); 2539 2540 switch (drop_reason) { 2541 case SKB_DROP_REASON_SOCKET_RCVBUFF: 2542 return -ENOMEM; 2543 case SKB_DROP_REASON_PROTO_MEM: 2544 return -ENOBUFS; 2545 case 0: 2546 return 0; 2547 default: 2548 return -EPERM; 2549 } 2550 } 2551 2552 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2553 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2554 2555 /* 2556 * Recover an error report and clear atomically 2557 */ 2558 2559 static inline int sock_error(struct sock *sk) 2560 { 2561 int err; 2562 2563 /* Avoid an atomic operation for the common case. 2564 * This is racy since another cpu/thread can change sk_err under us. 2565 */ 2566 if (likely(data_race(!sk->sk_err))) 2567 return 0; 2568 2569 err = xchg(&sk->sk_err, 0); 2570 return -err; 2571 } 2572 2573 void sk_error_report(struct sock *sk); 2574 2575 static inline unsigned long sock_wspace(struct sock *sk) 2576 { 2577 int amt = 0; 2578 2579 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2580 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2581 if (amt < 0) 2582 amt = 0; 2583 } 2584 return amt; 2585 } 2586 2587 /* Note: 2588 * We use sk->sk_wq_raw, from contexts knowing this 2589 * pointer is not NULL and cannot disappear/change. 2590 */ 2591 static inline void sk_set_bit(int nr, struct sock *sk) 2592 { 2593 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2594 !sock_flag(sk, SOCK_FASYNC)) 2595 return; 2596 2597 set_bit(nr, &sk->sk_wq_raw->flags); 2598 } 2599 2600 static inline void sk_clear_bit(int nr, struct sock *sk) 2601 { 2602 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2603 !sock_flag(sk, SOCK_FASYNC)) 2604 return; 2605 2606 clear_bit(nr, &sk->sk_wq_raw->flags); 2607 } 2608 2609 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2610 { 2611 if (sock_flag(sk, SOCK_FASYNC)) { 2612 rcu_read_lock(); 2613 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2614 rcu_read_unlock(); 2615 } 2616 } 2617 2618 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band) 2619 { 2620 if (unlikely(sock_flag(sk, SOCK_FASYNC))) 2621 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2622 } 2623 2624 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2625 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2626 * Note: for send buffers, TCP works better if we can build two skbs at 2627 * minimum. 2628 */ 2629 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2630 2631 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2632 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2633 2634 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2635 { 2636 u32 val; 2637 2638 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2639 return; 2640 2641 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2642 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2643 2644 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2645 } 2646 2647 /** 2648 * sk_page_frag - return an appropriate page_frag 2649 * @sk: socket 2650 * 2651 * Use the per task page_frag instead of the per socket one for 2652 * optimization when we know that we're in process context and own 2653 * everything that's associated with %current. 2654 * 2655 * Both direct reclaim and page faults can nest inside other 2656 * socket operations and end up recursing into sk_page_frag() 2657 * while it's already in use: explicitly avoid task page_frag 2658 * when users disable sk_use_task_frag. 2659 * 2660 * Return: a per task page_frag if context allows that, 2661 * otherwise a per socket one. 2662 */ 2663 static inline struct page_frag *sk_page_frag(struct sock *sk) 2664 { 2665 if (sk->sk_use_task_frag) 2666 return ¤t->task_frag; 2667 2668 return &sk->sk_frag; 2669 } 2670 2671 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2672 2673 static inline bool __sock_writeable(const struct sock *sk, int wmem_alloc) 2674 { 2675 return wmem_alloc < (READ_ONCE(sk->sk_sndbuf) >> 1); 2676 } 2677 /* 2678 * Default write policy as shown to user space via poll/select/SIGIO 2679 */ 2680 static inline bool sock_writeable(const struct sock *sk) 2681 { 2682 return __sock_writeable(sk, refcount_read(&sk->sk_wmem_alloc)); 2683 } 2684 2685 static inline gfp_t gfp_any(void) 2686 { 2687 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2688 } 2689 2690 static inline gfp_t gfp_memcg_charge(void) 2691 { 2692 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2693 } 2694 2695 #ifdef CONFIG_MEMCG 2696 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk) 2697 { 2698 return sk->sk_memcg; 2699 } 2700 2701 static inline bool mem_cgroup_sk_enabled(const struct sock *sk) 2702 { 2703 return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk); 2704 } 2705 2706 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk) 2707 { 2708 struct mem_cgroup *memcg = mem_cgroup_from_sk(sk); 2709 2710 #ifdef CONFIG_MEMCG_V1 2711 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 2712 return !!memcg->tcpmem_pressure; 2713 #endif /* CONFIG_MEMCG_V1 */ 2714 2715 do { 2716 if (time_before64(get_jiffies_64(), 2717 mem_cgroup_get_socket_pressure(memcg))) { 2718 memcg_memory_event(mem_cgroup_from_sk(sk), 2719 MEMCG_SOCK_THROTTLED); 2720 return true; 2721 } 2722 } while ((memcg = parent_mem_cgroup(memcg))); 2723 2724 return false; 2725 } 2726 #else 2727 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk) 2728 { 2729 return NULL; 2730 } 2731 2732 static inline bool mem_cgroup_sk_enabled(const struct sock *sk) 2733 { 2734 return false; 2735 } 2736 2737 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk) 2738 { 2739 return false; 2740 } 2741 #endif 2742 2743 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2744 { 2745 return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo); 2746 } 2747 2748 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2749 { 2750 return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo); 2751 } 2752 2753 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2754 { 2755 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2756 2757 return v ?: 1; 2758 } 2759 2760 /* Alas, with timeout socket operations are not restartable. 2761 * Compare this to poll(). 2762 */ 2763 static inline int sock_intr_errno(long timeo) 2764 { 2765 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2766 } 2767 2768 struct sock_skb_cb { 2769 u32 dropcount; 2770 }; 2771 2772 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2773 * using skb->cb[] would keep using it directly and utilize its 2774 * alignment guarantee. 2775 */ 2776 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \ 2777 sizeof(struct sock_skb_cb)) 2778 2779 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2780 SOCK_SKB_CB_OFFSET)) 2781 2782 #define sock_skb_cb_check_size(size) \ 2783 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2784 2785 static inline void sk_drops_add(struct sock *sk, int segs) 2786 { 2787 struct numa_drop_counters *ndc = sk->sk_drop_counters; 2788 2789 if (ndc) 2790 numa_drop_add(ndc, segs); 2791 else 2792 atomic_add(segs, &sk->sk_drops); 2793 } 2794 2795 static inline void sk_drops_inc(struct sock *sk) 2796 { 2797 sk_drops_add(sk, 1); 2798 } 2799 2800 static inline int sk_drops_read(const struct sock *sk) 2801 { 2802 const struct numa_drop_counters *ndc = sk->sk_drop_counters; 2803 2804 if (ndc) { 2805 DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops)); 2806 return numa_drop_read(ndc); 2807 } 2808 return atomic_read(&sk->sk_drops); 2809 } 2810 2811 static inline void sk_drops_reset(struct sock *sk) 2812 { 2813 struct numa_drop_counters *ndc = sk->sk_drop_counters; 2814 2815 if (ndc) 2816 numa_drop_reset(ndc); 2817 atomic_set(&sk->sk_drops, 0); 2818 } 2819 2820 static inline void 2821 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2822 { 2823 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2824 sk_drops_read(sk) : 0; 2825 } 2826 2827 static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb) 2828 { 2829 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2830 2831 sk_drops_add(sk, segs); 2832 } 2833 2834 static inline ktime_t sock_read_timestamp(struct sock *sk) 2835 { 2836 #if BITS_PER_LONG==32 2837 unsigned int seq; 2838 ktime_t kt; 2839 2840 do { 2841 seq = read_seqbegin(&sk->sk_stamp_seq); 2842 kt = sk->sk_stamp; 2843 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2844 2845 return kt; 2846 #else 2847 return READ_ONCE(sk->sk_stamp); 2848 #endif 2849 } 2850 2851 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2852 { 2853 #if BITS_PER_LONG==32 2854 write_seqlock(&sk->sk_stamp_seq); 2855 sk->sk_stamp = kt; 2856 write_sequnlock(&sk->sk_stamp_seq); 2857 #else 2858 WRITE_ONCE(sk->sk_stamp, kt); 2859 #endif 2860 } 2861 2862 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2863 struct sk_buff *skb); 2864 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2865 struct sk_buff *skb); 2866 2867 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk); 2868 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk, 2869 struct timespec64 *ts); 2870 2871 static inline void 2872 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2873 { 2874 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2875 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2876 ktime_t kt = skb->tstamp; 2877 /* 2878 * generate control messages if 2879 * - receive time stamping in software requested 2880 * - software time stamp available and wanted 2881 * - hardware time stamps available and wanted 2882 */ 2883 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2884 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2885 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2886 (hwtstamps->hwtstamp && 2887 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2888 __sock_recv_timestamp(msg, sk, skb); 2889 else 2890 sock_write_timestamp(sk, kt); 2891 2892 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2893 __sock_recv_wifi_status(msg, sk, skb); 2894 } 2895 2896 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2897 struct sk_buff *skb); 2898 2899 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2900 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2901 struct sk_buff *skb) 2902 { 2903 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2904 (1UL << SOCK_RCVTSTAMP) | \ 2905 (1UL << SOCK_RCVMARK) | \ 2906 (1UL << SOCK_RCVPRIORITY) | \ 2907 (1UL << SOCK_TIMESTAMPING_ANY)) 2908 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2909 SOF_TIMESTAMPING_RAW_HARDWARE) 2910 2911 if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS) 2912 __sock_recv_cmsgs(msg, sk, skb); 2913 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2914 sock_write_timestamp(sk, skb->tstamp); 2915 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2916 sock_write_timestamp(sk, 0); 2917 } 2918 2919 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags); 2920 2921 /** 2922 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2923 * @sk: socket sending this packet 2924 * @sockc: pointer to socket cmsg cookie to get timestamping info 2925 * @tx_flags: completed with instructions for time stamping 2926 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2927 * 2928 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2929 */ 2930 static inline void _sock_tx_timestamp(struct sock *sk, 2931 const struct sockcm_cookie *sockc, 2932 __u8 *tx_flags, __u32 *tskey) 2933 { 2934 __u32 tsflags = sockc->tsflags; 2935 2936 if (unlikely(tsflags)) { 2937 __sock_tx_timestamp(tsflags, tx_flags); 2938 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2939 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) { 2940 if (tsflags & SOCKCM_FLAG_TS_OPT_ID) 2941 *tskey = sockc->ts_opt_id; 2942 else 2943 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2944 } 2945 } 2946 } 2947 2948 static inline void sock_tx_timestamp(struct sock *sk, 2949 const struct sockcm_cookie *sockc, 2950 __u8 *tx_flags) 2951 { 2952 _sock_tx_timestamp(sk, sockc, tx_flags, NULL); 2953 } 2954 2955 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, 2956 const struct sockcm_cookie *sockc) 2957 { 2958 _sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags, 2959 &skb_shinfo(skb)->tskey); 2960 } 2961 2962 static inline bool sk_is_inet(const struct sock *sk) 2963 { 2964 int family = READ_ONCE(sk->sk_family); 2965 2966 return family == AF_INET || family == AF_INET6; 2967 } 2968 2969 static inline bool sk_is_tcp(const struct sock *sk) 2970 { 2971 return sk_is_inet(sk) && 2972 sk->sk_type == SOCK_STREAM && 2973 sk->sk_protocol == IPPROTO_TCP; 2974 } 2975 2976 static inline bool sk_is_udp(const struct sock *sk) 2977 { 2978 return sk_is_inet(sk) && 2979 sk->sk_type == SOCK_DGRAM && 2980 sk->sk_protocol == IPPROTO_UDP; 2981 } 2982 2983 static inline bool sk_is_unix(const struct sock *sk) 2984 { 2985 return sk->sk_family == AF_UNIX; 2986 } 2987 2988 static inline bool sk_is_stream_unix(const struct sock *sk) 2989 { 2990 return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM; 2991 } 2992 2993 static inline bool sk_is_vsock(const struct sock *sk) 2994 { 2995 return sk->sk_family == AF_VSOCK; 2996 } 2997 2998 static inline bool sk_may_scm_recv(const struct sock *sk) 2999 { 3000 return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) || 3001 sk->sk_family == AF_NETLINK || 3002 (IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH); 3003 } 3004 3005 /** 3006 * sk_eat_skb - Release a skb if it is no longer needed 3007 * @sk: socket to eat this skb from 3008 * @skb: socket buffer to eat 3009 * 3010 * This routine must be called with interrupts disabled or with the socket 3011 * locked so that the sk_buff queue operation is ok. 3012 */ 3013 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 3014 { 3015 __skb_unlink(skb, &sk->sk_receive_queue); 3016 __kfree_skb(skb); 3017 } 3018 3019 static inline bool 3020 skb_sk_is_prefetched(struct sk_buff *skb) 3021 { 3022 #ifdef CONFIG_INET 3023 return skb->destructor == sock_pfree; 3024 #else 3025 return false; 3026 #endif /* CONFIG_INET */ 3027 } 3028 3029 /* This helper checks if a socket is a full socket, 3030 * ie _not_ a timewait or request socket. 3031 */ 3032 static inline bool sk_fullsock(const struct sock *sk) 3033 { 3034 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 3035 } 3036 3037 static inline bool 3038 sk_is_refcounted(struct sock *sk) 3039 { 3040 /* Only full sockets have sk->sk_flags. */ 3041 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 3042 } 3043 3044 static inline bool 3045 sk_requests_wifi_status(struct sock *sk) 3046 { 3047 return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS); 3048 } 3049 3050 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 3051 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 3052 */ 3053 static inline bool sk_listener(const struct sock *sk) 3054 { 3055 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 3056 } 3057 3058 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT 3059 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE) 3060 * TCP RST and ACK can be attached to TIME_WAIT. 3061 */ 3062 static inline bool sk_listener_or_tw(const struct sock *sk) 3063 { 3064 return (1 << READ_ONCE(sk->sk_state)) & 3065 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT); 3066 } 3067 3068 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 3069 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 3070 int type); 3071 3072 bool sk_ns_capable(const struct sock *sk, 3073 struct user_namespace *user_ns, int cap); 3074 bool sk_capable(const struct sock *sk, int cap); 3075 bool sk_net_capable(const struct sock *sk, int cap); 3076 3077 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 3078 3079 /* Take into consideration the size of the struct sk_buff overhead in the 3080 * determination of these values, since that is non-constant across 3081 * platforms. This makes socket queueing behavior and performance 3082 * not depend upon such differences. 3083 */ 3084 #define _SK_MEM_PACKETS 256 3085 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 3086 #define SK_WMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 3087 #define SK_RMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 3088 3089 extern __u32 sysctl_wmem_max; 3090 extern __u32 sysctl_rmem_max; 3091 3092 extern __u32 sysctl_wmem_default; 3093 extern __u32 sysctl_rmem_default; 3094 3095 #define SKB_FRAG_PAGE_ORDER get_order(32768) 3096 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3097 3098 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 3099 { 3100 /* Does this proto have per netns sysctl_wmem ? */ 3101 if (proto->sysctl_wmem_offset) 3102 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 3103 3104 return READ_ONCE(*proto->sysctl_wmem); 3105 } 3106 3107 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 3108 { 3109 /* Does this proto have per netns sysctl_rmem ? */ 3110 if (proto->sysctl_rmem_offset) 3111 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 3112 3113 return READ_ONCE(*proto->sysctl_rmem); 3114 } 3115 3116 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 3117 * Some wifi drivers need to tweak it to get more chunks. 3118 * They can use this helper from their ndo_start_xmit() 3119 */ 3120 static inline void sk_pacing_shift_update(struct sock *sk, int val) 3121 { 3122 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 3123 return; 3124 WRITE_ONCE(sk->sk_pacing_shift, val); 3125 } 3126 3127 /* if a socket is bound to a device, check that the given device 3128 * index is either the same or that the socket is bound to an L3 3129 * master device and the given device index is also enslaved to 3130 * that L3 master 3131 */ 3132 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 3133 { 3134 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 3135 int mdif; 3136 3137 if (!bound_dev_if || bound_dev_if == dif) 3138 return true; 3139 3140 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 3141 if (mdif && mdif == bound_dev_if) 3142 return true; 3143 3144 return false; 3145 } 3146 3147 void sock_def_readable(struct sock *sk); 3148 3149 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 3150 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 3151 int sock_set_timestamping(struct sock *sk, int optname, 3152 struct so_timestamping timestamping); 3153 3154 #if defined(CONFIG_CGROUP_BPF) 3155 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op); 3156 #else 3157 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 3158 { 3159 } 3160 #endif 3161 void sock_no_linger(struct sock *sk); 3162 void sock_set_keepalive(struct sock *sk); 3163 void sock_set_priority(struct sock *sk, u32 priority); 3164 void sock_set_rcvbuf(struct sock *sk, int val); 3165 void sock_set_mark(struct sock *sk, u32 val); 3166 void sock_set_reuseaddr(struct sock *sk); 3167 void sock_set_reuseport(struct sock *sk); 3168 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3169 3170 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len); 3171 3172 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3173 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3174 sockptr_t optval, int optlen, bool old_timeval); 3175 3176 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3177 void __user *arg, void *karg, size_t size); 3178 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3179 static inline bool sk_is_readable(struct sock *sk) 3180 { 3181 const struct proto *prot = READ_ONCE(sk->sk_prot); 3182 3183 if (prot->sock_is_readable) 3184 return prot->sock_is_readable(sk); 3185 3186 return false; 3187 } 3188 #endif /* _SOCK_H */ 3189