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