1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* This is the per-socket lock. The spinlock provides a synchronization 80 * between user contexts and software interrupt processing, whereas the 81 * mini-semaphore synchronizes multiple users amongst themselves. 82 */ 83 typedef struct { 84 spinlock_t slock; 85 int owned; 86 wait_queue_head_t wq; 87 /* 88 * We express the mutex-alike socket_lock semantics 89 * to the lock validator by explicitly managing 90 * the slock as a lock variant (in addition to 91 * the slock itself): 92 */ 93 #ifdef CONFIG_DEBUG_LOCK_ALLOC 94 struct lockdep_map dep_map; 95 #endif 96 } socket_lock_t; 97 98 struct sock; 99 struct proto; 100 struct net; 101 102 typedef __u32 __bitwise __portpair; 103 typedef __u64 __bitwise __addrpair; 104 105 /** 106 * struct sock_common - minimal network layer representation of sockets 107 * @skc_daddr: Foreign IPv4 addr 108 * @skc_rcv_saddr: Bound local IPv4 addr 109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 110 * @skc_hash: hash value used with various protocol lookup tables 111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 112 * @skc_dport: placeholder for inet_dport/tw_dport 113 * @skc_num: placeholder for inet_num/tw_num 114 * @skc_portpair: __u32 union of @skc_dport & @skc_num 115 * @skc_family: network address family 116 * @skc_state: Connection state 117 * @skc_reuse: %SO_REUSEADDR setting 118 * @skc_reuseport: %SO_REUSEPORT setting 119 * @skc_ipv6only: socket is IPV6 only 120 * @skc_net_refcnt: socket is using net ref counting 121 * @skc_bound_dev_if: bound device index if != 0 122 * @skc_bind_node: bind hash linkage for various protocol lookup tables 123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 124 * @skc_prot: protocol handlers inside a network family 125 * @skc_net: reference to the network namespace of this socket 126 * @skc_v6_daddr: IPV6 destination address 127 * @skc_v6_rcv_saddr: IPV6 source address 128 * @skc_cookie: socket's cookie value 129 * @skc_node: main hash linkage for various protocol lookup tables 130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 131 * @skc_tx_queue_mapping: tx queue number for this connection 132 * @skc_rx_queue_mapping: rx queue number for this connection 133 * @skc_flags: place holder for sk_flags 134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 136 * @skc_listener: connection request listener socket (aka rsk_listener) 137 * [union with @skc_flags] 138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 139 * [union with @skc_flags] 140 * @skc_incoming_cpu: record/match cpu processing incoming packets 141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 142 * [union with @skc_incoming_cpu] 143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 144 * [union with @skc_incoming_cpu] 145 * @skc_refcnt: reference count 146 * 147 * This is the minimal network layer representation of sockets, the header 148 * for struct sock and struct inet_timewait_sock. 149 */ 150 struct sock_common { 151 union { 152 __addrpair skc_addrpair; 153 struct { 154 __be32 skc_daddr; 155 __be32 skc_rcv_saddr; 156 }; 157 }; 158 union { 159 unsigned int skc_hash; 160 __u16 skc_u16hashes[2]; 161 }; 162 /* skc_dport && skc_num must be grouped as well */ 163 union { 164 __portpair skc_portpair; 165 struct { 166 __be16 skc_dport; 167 __u16 skc_num; 168 }; 169 }; 170 171 unsigned short skc_family; 172 volatile unsigned char skc_state; 173 unsigned char skc_reuse:4; 174 unsigned char skc_reuseport:1; 175 unsigned char skc_ipv6only:1; 176 unsigned char skc_net_refcnt:1; 177 int skc_bound_dev_if; 178 union { 179 struct hlist_node skc_bind_node; 180 struct hlist_node skc_portaddr_node; 181 }; 182 struct proto *skc_prot; 183 possible_net_t skc_net; 184 185 #if IS_ENABLED(CONFIG_IPV6) 186 struct in6_addr skc_v6_daddr; 187 struct in6_addr skc_v6_rcv_saddr; 188 #endif 189 190 atomic64_t skc_cookie; 191 192 /* following fields are padding to force 193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 194 * assuming IPV6 is enabled. We use this padding differently 195 * for different kind of 'sockets' 196 */ 197 union { 198 unsigned long skc_flags; 199 struct sock *skc_listener; /* request_sock */ 200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 201 }; 202 /* 203 * fields between dontcopy_begin/dontcopy_end 204 * are not copied in sock_copy() 205 */ 206 /* private: */ 207 int skc_dontcopy_begin[0]; 208 /* public: */ 209 union { 210 struct hlist_node skc_node; 211 struct hlist_nulls_node skc_nulls_node; 212 }; 213 unsigned short skc_tx_queue_mapping; 214 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 215 unsigned short skc_rx_queue_mapping; 216 #endif 217 union { 218 int skc_incoming_cpu; 219 u32 skc_rcv_wnd; 220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 221 }; 222 223 refcount_t skc_refcnt; 224 /* private: */ 225 int skc_dontcopy_end[0]; 226 union { 227 u32 skc_rxhash; 228 u32 skc_window_clamp; 229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 230 }; 231 /* public: */ 232 }; 233 234 struct bpf_local_storage; 235 struct sk_filter; 236 237 /** 238 * struct sock - network layer representation of sockets 239 * @__sk_common: shared layout with inet_timewait_sock 240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 242 * @sk_lock: synchronizer 243 * @sk_kern_sock: True if sock is using kernel lock classes 244 * @sk_rcvbuf: size of receive buffer in bytes 245 * @sk_wq: sock wait queue and async head 246 * @sk_rx_dst: receive input route used by early demux 247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 249 * @sk_dst_cache: destination cache 250 * @sk_dst_pending_confirm: need to confirm neighbour 251 * @sk_policy: flow policy 252 * @sk_receive_queue: incoming packets 253 * @sk_wmem_alloc: transmit queue bytes committed 254 * @sk_tsq_flags: TCP Small Queues flags 255 * @sk_write_queue: Packet sending queue 256 * @sk_omem_alloc: "o" is "option" or "other" 257 * @sk_wmem_queued: persistent queue size 258 * @sk_forward_alloc: space allocated forward 259 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 260 * @sk_napi_id: id of the last napi context to receive data for sk 261 * @sk_ll_usec: usecs to busypoll when there is no data 262 * @sk_allocation: allocation mode 263 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 264 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 265 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 266 * @sk_sndbuf: size of send buffer in bytes 267 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 268 * @sk_no_check_rx: allow zero checksum in RX packets 269 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 270 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 271 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 272 * @sk_gso_max_size: Maximum GSO segment size to build 273 * @sk_gso_max_segs: Maximum number of GSO segments 274 * @sk_pacing_shift: scaling factor for TCP Small Queues 275 * @sk_lingertime: %SO_LINGER l_linger setting 276 * @sk_backlog: always used with the per-socket spinlock held 277 * @sk_callback_lock: used with the callbacks in the end of this struct 278 * @sk_error_queue: rarely used 279 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 280 * IPV6_ADDRFORM for instance) 281 * @sk_err: last error 282 * @sk_err_soft: errors that don't cause failure but are the cause of a 283 * persistent failure not just 'timed out' 284 * @sk_drops: raw/udp drops counter 285 * @sk_ack_backlog: current listen backlog 286 * @sk_max_ack_backlog: listen backlog set in listen() 287 * @sk_uid: user id of owner 288 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 289 * @sk_busy_poll_budget: napi processing budget when busypolling 290 * @sk_priority: %SO_PRIORITY setting 291 * @sk_type: socket type (%SOCK_STREAM, etc) 292 * @sk_protocol: which protocol this socket belongs in this network family 293 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 294 * @sk_peer_pid: &struct pid for this socket's peer 295 * @sk_peer_cred: %SO_PEERCRED setting 296 * @sk_rcvlowat: %SO_RCVLOWAT setting 297 * @sk_rcvtimeo: %SO_RCVTIMEO setting 298 * @sk_sndtimeo: %SO_SNDTIMEO setting 299 * @sk_txhash: computed flow hash for use on transmit 300 * @sk_txrehash: enable TX hash rethink 301 * @sk_filter: socket filtering instructions 302 * @sk_timer: sock cleanup timer 303 * @sk_stamp: time stamp of last packet received 304 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 305 * @sk_tsflags: SO_TIMESTAMPING flags 306 * @sk_bpf_cb_flags: used in bpf_setsockopt() 307 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 308 * Sockets that can be used under memory reclaim should 309 * set this to false. 310 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 311 * for timestamping 312 * @sk_tskey: counter to disambiguate concurrent tstamp requests 313 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 314 * @sk_socket: Identd and reporting IO signals 315 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 316 * @sk_frag: cached page frag 317 * @sk_peek_off: current peek_offset value 318 * @sk_send_head: front of stuff to transmit 319 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 320 * @sk_security: used by security modules 321 * @sk_mark: generic packet mark 322 * @sk_cgrp_data: cgroup data for this cgroup 323 * @sk_memcg: this socket's memory cgroup association 324 * @sk_write_pending: a write to stream socket waits to start 325 * @sk_disconnects: number of disconnect operations performed on this sock 326 * @sk_state_change: callback to indicate change in the state of the sock 327 * @sk_data_ready: callback to indicate there is data to be processed 328 * @sk_write_space: callback to indicate there is bf sending space available 329 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 330 * @sk_backlog_rcv: callback to process the backlog 331 * @sk_validate_xmit_skb: ptr to an optional validate function 332 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 333 * @sk_reuseport_cb: reuseport group container 334 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 335 * @sk_rcu: used during RCU grace period 336 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 337 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 338 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 339 * @sk_txtime_unused: unused txtime flags 340 * @sk_scm_recv_flags: all flags used by scm_recv() 341 * @sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS 342 * @sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY 343 * @sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD 344 * @sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS 345 * @sk_scm_unused: unused flags for scm_recv() 346 * @ns_tracker: tracker for netns reference 347 * @sk_user_frags: xarray of pages the user is holding a reference on. 348 * @sk_owner: reference to the real owner of the socket that calls 349 * sock_lock_init_class_and_name(). 350 */ 351 struct sock { 352 /* 353 * Now struct inet_timewait_sock also uses sock_common, so please just 354 * don't add nothing before this first member (__sk_common) --acme 355 */ 356 struct sock_common __sk_common; 357 #define sk_node __sk_common.skc_node 358 #define sk_nulls_node __sk_common.skc_nulls_node 359 #define sk_refcnt __sk_common.skc_refcnt 360 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 361 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 362 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 363 #endif 364 365 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 366 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 367 #define sk_hash __sk_common.skc_hash 368 #define sk_portpair __sk_common.skc_portpair 369 #define sk_num __sk_common.skc_num 370 #define sk_dport __sk_common.skc_dport 371 #define sk_addrpair __sk_common.skc_addrpair 372 #define sk_daddr __sk_common.skc_daddr 373 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 374 #define sk_family __sk_common.skc_family 375 #define sk_state __sk_common.skc_state 376 #define sk_reuse __sk_common.skc_reuse 377 #define sk_reuseport __sk_common.skc_reuseport 378 #define sk_ipv6only __sk_common.skc_ipv6only 379 #define sk_net_refcnt __sk_common.skc_net_refcnt 380 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 381 #define sk_bind_node __sk_common.skc_bind_node 382 #define sk_prot __sk_common.skc_prot 383 #define sk_net __sk_common.skc_net 384 #define sk_v6_daddr __sk_common.skc_v6_daddr 385 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 386 #define sk_cookie __sk_common.skc_cookie 387 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 388 #define sk_flags __sk_common.skc_flags 389 #define sk_rxhash __sk_common.skc_rxhash 390 391 __cacheline_group_begin(sock_write_rx); 392 393 atomic_t sk_drops; 394 __s32 sk_peek_off; 395 struct sk_buff_head sk_error_queue; 396 struct sk_buff_head sk_receive_queue; 397 /* 398 * The backlog queue is special, it is always used with 399 * the per-socket spinlock held and requires low latency 400 * access. Therefore we special case it's implementation. 401 * Note : rmem_alloc is in this structure to fill a hole 402 * on 64bit arches, not because its logically part of 403 * backlog. 404 */ 405 struct { 406 atomic_t rmem_alloc; 407 int len; 408 struct sk_buff *head; 409 struct sk_buff *tail; 410 } sk_backlog; 411 #define sk_rmem_alloc sk_backlog.rmem_alloc 412 413 __cacheline_group_end(sock_write_rx); 414 415 __cacheline_group_begin(sock_read_rx); 416 /* early demux fields */ 417 struct dst_entry __rcu *sk_rx_dst; 418 int sk_rx_dst_ifindex; 419 u32 sk_rx_dst_cookie; 420 421 #ifdef CONFIG_NET_RX_BUSY_POLL 422 unsigned int sk_ll_usec; 423 unsigned int sk_napi_id; 424 u16 sk_busy_poll_budget; 425 u8 sk_prefer_busy_poll; 426 #endif 427 u8 sk_userlocks; 428 int sk_rcvbuf; 429 430 struct sk_filter __rcu *sk_filter; 431 union { 432 struct socket_wq __rcu *sk_wq; 433 /* private: */ 434 struct socket_wq *sk_wq_raw; 435 /* public: */ 436 }; 437 438 void (*sk_data_ready)(struct sock *sk); 439 long sk_rcvtimeo; 440 int sk_rcvlowat; 441 __cacheline_group_end(sock_read_rx); 442 443 __cacheline_group_begin(sock_read_rxtx); 444 int sk_err; 445 struct socket *sk_socket; 446 struct mem_cgroup *sk_memcg; 447 #ifdef CONFIG_XFRM 448 struct xfrm_policy __rcu *sk_policy[2]; 449 #endif 450 __cacheline_group_end(sock_read_rxtx); 451 452 __cacheline_group_begin(sock_write_rxtx); 453 socket_lock_t sk_lock; 454 u32 sk_reserved_mem; 455 int sk_forward_alloc; 456 u32 sk_tsflags; 457 __cacheline_group_end(sock_write_rxtx); 458 459 __cacheline_group_begin(sock_write_tx); 460 int sk_write_pending; 461 atomic_t sk_omem_alloc; 462 int sk_sndbuf; 463 464 int sk_wmem_queued; 465 refcount_t sk_wmem_alloc; 466 unsigned long sk_tsq_flags; 467 union { 468 struct sk_buff *sk_send_head; 469 struct rb_root tcp_rtx_queue; 470 }; 471 struct sk_buff_head sk_write_queue; 472 u32 sk_dst_pending_confirm; 473 u32 sk_pacing_status; /* see enum sk_pacing */ 474 struct page_frag sk_frag; 475 struct timer_list sk_timer; 476 477 unsigned long sk_pacing_rate; /* bytes per second */ 478 atomic_t sk_zckey; 479 atomic_t sk_tskey; 480 __cacheline_group_end(sock_write_tx); 481 482 __cacheline_group_begin(sock_read_tx); 483 unsigned long sk_max_pacing_rate; 484 long sk_sndtimeo; 485 u32 sk_priority; 486 u32 sk_mark; 487 struct dst_entry __rcu *sk_dst_cache; 488 netdev_features_t sk_route_caps; 489 #ifdef CONFIG_SOCK_VALIDATE_XMIT 490 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 491 struct net_device *dev, 492 struct sk_buff *skb); 493 #endif 494 u16 sk_gso_type; 495 u16 sk_gso_max_segs; 496 unsigned int sk_gso_max_size; 497 gfp_t sk_allocation; 498 u32 sk_txhash; 499 u8 sk_pacing_shift; 500 bool sk_use_task_frag; 501 __cacheline_group_end(sock_read_tx); 502 503 /* 504 * Because of non atomicity rules, all 505 * changes are protected by socket lock. 506 */ 507 u8 sk_gso_disabled : 1, 508 sk_kern_sock : 1, 509 sk_no_check_tx : 1, 510 sk_no_check_rx : 1; 511 u8 sk_shutdown; 512 u16 sk_type; 513 u16 sk_protocol; 514 unsigned long sk_lingertime; 515 struct proto *sk_prot_creator; 516 rwlock_t sk_callback_lock; 517 int sk_err_soft; 518 u32 sk_ack_backlog; 519 u32 sk_max_ack_backlog; 520 kuid_t sk_uid; 521 spinlock_t sk_peer_lock; 522 int sk_bind_phc; 523 struct pid *sk_peer_pid; 524 const struct cred *sk_peer_cred; 525 526 ktime_t sk_stamp; 527 #if BITS_PER_LONG==32 528 seqlock_t sk_stamp_seq; 529 #endif 530 int sk_disconnects; 531 532 union { 533 u8 sk_txrehash; 534 u8 sk_scm_recv_flags; 535 struct { 536 u8 sk_scm_credentials : 1, 537 sk_scm_security : 1, 538 sk_scm_pidfd : 1, 539 sk_scm_rights : 1, 540 sk_scm_unused : 4; 541 }; 542 }; 543 u8 sk_clockid; 544 u8 sk_txtime_deadline_mode : 1, 545 sk_txtime_report_errors : 1, 546 sk_txtime_unused : 6; 547 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG)) 548 u8 sk_bpf_cb_flags; 549 550 void *sk_user_data; 551 #ifdef CONFIG_SECURITY 552 void *sk_security; 553 #endif 554 struct sock_cgroup_data sk_cgrp_data; 555 void (*sk_state_change)(struct sock *sk); 556 void (*sk_write_space)(struct sock *sk); 557 void (*sk_error_report)(struct sock *sk); 558 int (*sk_backlog_rcv)(struct sock *sk, 559 struct sk_buff *skb); 560 void (*sk_destruct)(struct sock *sk); 561 struct sock_reuseport __rcu *sk_reuseport_cb; 562 #ifdef CONFIG_BPF_SYSCALL 563 struct bpf_local_storage __rcu *sk_bpf_storage; 564 #endif 565 struct rcu_head sk_rcu; 566 netns_tracker ns_tracker; 567 struct xarray sk_user_frags; 568 569 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 570 struct module *sk_owner; 571 #endif 572 }; 573 574 struct sock_bh_locked { 575 struct sock *sock; 576 local_lock_t bh_lock; 577 }; 578 579 enum sk_pacing { 580 SK_PACING_NONE = 0, 581 SK_PACING_NEEDED = 1, 582 SK_PACING_FQ = 2, 583 }; 584 585 /* flag bits in sk_user_data 586 * 587 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 588 * not be suitable for copying when cloning the socket. For instance, 589 * it can point to a reference counted object. sk_user_data bottom 590 * bit is set if pointer must not be copied. 591 * 592 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 593 * managed/owned by a BPF reuseport array. This bit should be set 594 * when sk_user_data's sk is added to the bpf's reuseport_array. 595 * 596 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 597 * sk_user_data points to psock type. This bit should be set 598 * when sk_user_data is assigned to a psock object. 599 */ 600 #define SK_USER_DATA_NOCOPY 1UL 601 #define SK_USER_DATA_BPF 2UL 602 #define SK_USER_DATA_PSOCK 4UL 603 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 604 SK_USER_DATA_PSOCK) 605 606 /** 607 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 608 * @sk: socket 609 */ 610 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 611 { 612 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 613 } 614 615 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 616 617 /** 618 * __locked_read_sk_user_data_with_flags - return the pointer 619 * only if argument flags all has been set in sk_user_data. Otherwise 620 * return NULL 621 * 622 * @sk: socket 623 * @flags: flag bits 624 * 625 * The caller must be holding sk->sk_callback_lock. 626 */ 627 static inline void * 628 __locked_read_sk_user_data_with_flags(const struct sock *sk, 629 uintptr_t flags) 630 { 631 uintptr_t sk_user_data = 632 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 633 lockdep_is_held(&sk->sk_callback_lock)); 634 635 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 636 637 if ((sk_user_data & flags) == flags) 638 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 639 return NULL; 640 } 641 642 /** 643 * __rcu_dereference_sk_user_data_with_flags - return the pointer 644 * only if argument flags all has been set in sk_user_data. Otherwise 645 * return NULL 646 * 647 * @sk: socket 648 * @flags: flag bits 649 */ 650 static inline void * 651 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 652 uintptr_t flags) 653 { 654 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 655 656 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 657 658 if ((sk_user_data & flags) == flags) 659 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 660 return NULL; 661 } 662 663 #define rcu_dereference_sk_user_data(sk) \ 664 __rcu_dereference_sk_user_data_with_flags(sk, 0) 665 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 666 ({ \ 667 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 668 __tmp2 = (uintptr_t)(flags); \ 669 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 670 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 671 rcu_assign_pointer(__sk_user_data((sk)), \ 672 __tmp1 | __tmp2); \ 673 }) 674 #define rcu_assign_sk_user_data(sk, ptr) \ 675 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 676 677 static inline 678 struct net *sock_net(const struct sock *sk) 679 { 680 return read_pnet(&sk->sk_net); 681 } 682 683 static inline 684 void sock_net_set(struct sock *sk, struct net *net) 685 { 686 write_pnet(&sk->sk_net, net); 687 } 688 689 /* 690 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 691 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 692 * on a socket means that the socket will reuse everybody else's port 693 * without looking at the other's sk_reuse value. 694 */ 695 696 #define SK_NO_REUSE 0 697 #define SK_CAN_REUSE 1 698 #define SK_FORCE_REUSE 2 699 700 int sk_set_peek_off(struct sock *sk, int val); 701 702 static inline int sk_peek_offset(const struct sock *sk, int flags) 703 { 704 if (unlikely(flags & MSG_PEEK)) { 705 return READ_ONCE(sk->sk_peek_off); 706 } 707 708 return 0; 709 } 710 711 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 712 { 713 s32 off = READ_ONCE(sk->sk_peek_off); 714 715 if (unlikely(off >= 0)) { 716 off = max_t(s32, off - val, 0); 717 WRITE_ONCE(sk->sk_peek_off, off); 718 } 719 } 720 721 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 722 { 723 sk_peek_offset_bwd(sk, -val); 724 } 725 726 /* 727 * Hashed lists helper routines 728 */ 729 static inline struct sock *sk_entry(const struct hlist_node *node) 730 { 731 return hlist_entry(node, struct sock, sk_node); 732 } 733 734 static inline struct sock *__sk_head(const struct hlist_head *head) 735 { 736 return hlist_entry(head->first, struct sock, sk_node); 737 } 738 739 static inline struct sock *sk_head(const struct hlist_head *head) 740 { 741 return hlist_empty(head) ? NULL : __sk_head(head); 742 } 743 744 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 745 { 746 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 747 } 748 749 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 750 { 751 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 752 } 753 754 static inline struct sock *sk_next(const struct sock *sk) 755 { 756 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 757 } 758 759 static inline struct sock *sk_nulls_next(const struct sock *sk) 760 { 761 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 762 hlist_nulls_entry(sk->sk_nulls_node.next, 763 struct sock, sk_nulls_node) : 764 NULL; 765 } 766 767 static inline bool sk_unhashed(const struct sock *sk) 768 { 769 return hlist_unhashed(&sk->sk_node); 770 } 771 772 static inline bool sk_hashed(const struct sock *sk) 773 { 774 return !sk_unhashed(sk); 775 } 776 777 static inline void sk_node_init(struct hlist_node *node) 778 { 779 node->pprev = NULL; 780 } 781 782 static inline void __sk_del_node(struct sock *sk) 783 { 784 __hlist_del(&sk->sk_node); 785 } 786 787 /* NB: equivalent to hlist_del_init_rcu */ 788 static inline bool __sk_del_node_init(struct sock *sk) 789 { 790 if (sk_hashed(sk)) { 791 __sk_del_node(sk); 792 sk_node_init(&sk->sk_node); 793 return true; 794 } 795 return false; 796 } 797 798 /* Grab socket reference count. This operation is valid only 799 when sk is ALREADY grabbed f.e. it is found in hash table 800 or a list and the lookup is made under lock preventing hash table 801 modifications. 802 */ 803 804 static __always_inline void sock_hold(struct sock *sk) 805 { 806 refcount_inc(&sk->sk_refcnt); 807 } 808 809 /* Ungrab socket in the context, which assumes that socket refcnt 810 cannot hit zero, f.e. it is true in context of any socketcall. 811 */ 812 static __always_inline void __sock_put(struct sock *sk) 813 { 814 refcount_dec(&sk->sk_refcnt); 815 } 816 817 static inline bool sk_del_node_init(struct sock *sk) 818 { 819 bool rc = __sk_del_node_init(sk); 820 821 if (rc) { 822 /* paranoid for a while -acme */ 823 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 824 __sock_put(sk); 825 } 826 return rc; 827 } 828 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 829 830 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 831 { 832 if (sk_hashed(sk)) { 833 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 834 return true; 835 } 836 return false; 837 } 838 839 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 840 { 841 bool rc = __sk_nulls_del_node_init_rcu(sk); 842 843 if (rc) { 844 /* paranoid for a while -acme */ 845 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 846 __sock_put(sk); 847 } 848 return rc; 849 } 850 851 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 852 { 853 hlist_add_head(&sk->sk_node, list); 854 } 855 856 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 857 { 858 sock_hold(sk); 859 __sk_add_node(sk, list); 860 } 861 862 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 863 { 864 sock_hold(sk); 865 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 866 sk->sk_family == AF_INET6) 867 hlist_add_tail_rcu(&sk->sk_node, list); 868 else 869 hlist_add_head_rcu(&sk->sk_node, list); 870 } 871 872 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 873 { 874 sock_hold(sk); 875 hlist_add_tail_rcu(&sk->sk_node, list); 876 } 877 878 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 879 { 880 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 881 } 882 883 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 884 { 885 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 886 } 887 888 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 889 { 890 sock_hold(sk); 891 __sk_nulls_add_node_rcu(sk, list); 892 } 893 894 static inline void __sk_del_bind_node(struct sock *sk) 895 { 896 __hlist_del(&sk->sk_bind_node); 897 } 898 899 static inline void sk_add_bind_node(struct sock *sk, 900 struct hlist_head *list) 901 { 902 hlist_add_head(&sk->sk_bind_node, list); 903 } 904 905 #define sk_for_each(__sk, list) \ 906 hlist_for_each_entry(__sk, list, sk_node) 907 #define sk_for_each_rcu(__sk, list) \ 908 hlist_for_each_entry_rcu(__sk, list, sk_node) 909 #define sk_nulls_for_each(__sk, node, list) \ 910 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 911 #define sk_nulls_for_each_rcu(__sk, node, list) \ 912 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 913 #define sk_for_each_from(__sk) \ 914 hlist_for_each_entry_from(__sk, sk_node) 915 #define sk_nulls_for_each_from(__sk, node) \ 916 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 917 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 918 #define sk_for_each_safe(__sk, tmp, list) \ 919 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 920 #define sk_for_each_bound(__sk, list) \ 921 hlist_for_each_entry(__sk, list, sk_bind_node) 922 #define sk_for_each_bound_safe(__sk, tmp, list) \ 923 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node) 924 925 /** 926 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 927 * @tpos: the type * to use as a loop cursor. 928 * @pos: the &struct hlist_node to use as a loop cursor. 929 * @head: the head for your list. 930 * @offset: offset of hlist_node within the struct. 931 * 932 */ 933 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 934 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 935 pos != NULL && \ 936 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 937 pos = rcu_dereference(hlist_next_rcu(pos))) 938 939 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 940 { 941 /* Careful only use this in a context where these parameters 942 * can not change and must all be valid, such as recvmsg from 943 * userspace. 944 */ 945 return sk->sk_socket->file->f_cred->user_ns; 946 } 947 948 /* Sock flags */ 949 enum sock_flags { 950 SOCK_DEAD, 951 SOCK_DONE, 952 SOCK_URGINLINE, 953 SOCK_KEEPOPEN, 954 SOCK_LINGER, 955 SOCK_DESTROY, 956 SOCK_BROADCAST, 957 SOCK_TIMESTAMP, 958 SOCK_ZAPPED, 959 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 960 SOCK_DBG, /* %SO_DEBUG setting */ 961 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 962 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 963 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 964 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 965 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 966 SOCK_FASYNC, /* fasync() active */ 967 SOCK_RXQ_OVFL, 968 SOCK_ZEROCOPY, /* buffers from userspace */ 969 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 970 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 971 * Will use last 4 bytes of packet sent from 972 * user-space instead. 973 */ 974 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 975 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 976 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 977 SOCK_TXTIME, 978 SOCK_XDP, /* XDP is attached */ 979 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 980 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 981 SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */ 982 SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */ 983 }; 984 985 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 986 /* 987 * The highest bit of sk_tsflags is reserved for kernel-internal 988 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that 989 * SOF_TIMESTAMPING* values do not reach this reserved area 990 */ 991 #define SOCKCM_FLAG_TS_OPT_ID BIT(31) 992 993 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 994 { 995 nsk->sk_flags = osk->sk_flags; 996 } 997 998 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 999 { 1000 __set_bit(flag, &sk->sk_flags); 1001 } 1002 1003 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 1004 { 1005 __clear_bit(flag, &sk->sk_flags); 1006 } 1007 1008 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 1009 int valbool) 1010 { 1011 if (valbool) 1012 sock_set_flag(sk, bit); 1013 else 1014 sock_reset_flag(sk, bit); 1015 } 1016 1017 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 1018 { 1019 return test_bit(flag, &sk->sk_flags); 1020 } 1021 1022 #ifdef CONFIG_NET 1023 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 1024 static inline int sk_memalloc_socks(void) 1025 { 1026 return static_branch_unlikely(&memalloc_socks_key); 1027 } 1028 1029 void __receive_sock(struct file *file); 1030 #else 1031 1032 static inline int sk_memalloc_socks(void) 1033 { 1034 return 0; 1035 } 1036 1037 static inline void __receive_sock(struct file *file) 1038 { } 1039 #endif 1040 1041 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1042 { 1043 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1044 } 1045 1046 static inline void sk_acceptq_removed(struct sock *sk) 1047 { 1048 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1049 } 1050 1051 static inline void sk_acceptq_added(struct sock *sk) 1052 { 1053 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1054 } 1055 1056 /* Note: If you think the test should be: 1057 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1058 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1059 */ 1060 static inline bool sk_acceptq_is_full(const struct sock *sk) 1061 { 1062 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1063 } 1064 1065 /* 1066 * Compute minimal free write space needed to queue new packets. 1067 */ 1068 static inline int sk_stream_min_wspace(const struct sock *sk) 1069 { 1070 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1071 } 1072 1073 static inline int sk_stream_wspace(const struct sock *sk) 1074 { 1075 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1076 } 1077 1078 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1079 { 1080 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1081 } 1082 1083 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1084 { 1085 /* Paired with lockless reads of sk->sk_forward_alloc */ 1086 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1087 } 1088 1089 void sk_stream_write_space(struct sock *sk); 1090 1091 /* OOB backlog add */ 1092 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1093 { 1094 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1095 skb_dst_force(skb); 1096 1097 if (!sk->sk_backlog.tail) 1098 WRITE_ONCE(sk->sk_backlog.head, skb); 1099 else 1100 sk->sk_backlog.tail->next = skb; 1101 1102 WRITE_ONCE(sk->sk_backlog.tail, skb); 1103 skb->next = NULL; 1104 } 1105 1106 /* 1107 * Take into account size of receive queue and backlog queue 1108 * Do not take into account this skb truesize, 1109 * to allow even a single big packet to come. 1110 */ 1111 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1112 { 1113 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1114 1115 return qsize > limit; 1116 } 1117 1118 /* The per-socket spinlock must be held here. */ 1119 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1120 unsigned int limit) 1121 { 1122 if (sk_rcvqueues_full(sk, limit)) 1123 return -ENOBUFS; 1124 1125 /* 1126 * If the skb was allocated from pfmemalloc reserves, only 1127 * allow SOCK_MEMALLOC sockets to use it as this socket is 1128 * helping free memory 1129 */ 1130 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1131 return -ENOMEM; 1132 1133 __sk_add_backlog(sk, skb); 1134 sk->sk_backlog.len += skb->truesize; 1135 return 0; 1136 } 1137 1138 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1139 1140 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1141 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1142 1143 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1144 { 1145 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1146 return __sk_backlog_rcv(sk, skb); 1147 1148 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1149 tcp_v6_do_rcv, 1150 tcp_v4_do_rcv, 1151 sk, skb); 1152 } 1153 1154 static inline void sk_incoming_cpu_update(struct sock *sk) 1155 { 1156 int cpu = raw_smp_processor_id(); 1157 1158 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1159 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1160 } 1161 1162 1163 static inline void sock_rps_save_rxhash(struct sock *sk, 1164 const struct sk_buff *skb) 1165 { 1166 #ifdef CONFIG_RPS 1167 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1168 * here, and another one in sock_rps_record_flow(). 1169 */ 1170 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1171 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1172 #endif 1173 } 1174 1175 static inline void sock_rps_reset_rxhash(struct sock *sk) 1176 { 1177 #ifdef CONFIG_RPS 1178 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1179 WRITE_ONCE(sk->sk_rxhash, 0); 1180 #endif 1181 } 1182 1183 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1184 ({ int __rc, __dis = __sk->sk_disconnects; \ 1185 release_sock(__sk); \ 1186 __rc = __condition; \ 1187 if (!__rc) { \ 1188 *(__timeo) = wait_woken(__wait, \ 1189 TASK_INTERRUPTIBLE, \ 1190 *(__timeo)); \ 1191 } \ 1192 sched_annotate_sleep(); \ 1193 lock_sock(__sk); \ 1194 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1195 __rc; \ 1196 }) 1197 1198 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1199 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1200 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1201 int sk_stream_error(struct sock *sk, int flags, int err); 1202 void sk_stream_kill_queues(struct sock *sk); 1203 void sk_set_memalloc(struct sock *sk); 1204 void sk_clear_memalloc(struct sock *sk); 1205 1206 void __sk_flush_backlog(struct sock *sk); 1207 1208 static inline bool sk_flush_backlog(struct sock *sk) 1209 { 1210 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1211 __sk_flush_backlog(sk); 1212 return true; 1213 } 1214 return false; 1215 } 1216 1217 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1218 1219 struct request_sock_ops; 1220 struct timewait_sock_ops; 1221 struct inet_hashinfo; 1222 struct raw_hashinfo; 1223 struct smc_hashinfo; 1224 struct module; 1225 struct sk_psock; 1226 1227 /* 1228 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1229 * un-modified. Special care is taken when initializing object to zero. 1230 */ 1231 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1232 { 1233 if (offsetof(struct sock, sk_node.next) != 0) 1234 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1235 memset(&sk->sk_node.pprev, 0, 1236 size - offsetof(struct sock, sk_node.pprev)); 1237 } 1238 1239 struct proto_accept_arg { 1240 int flags; 1241 int err; 1242 int is_empty; 1243 bool kern; 1244 }; 1245 1246 /* Networking protocol blocks we attach to sockets. 1247 * socket layer -> transport layer interface 1248 */ 1249 struct proto { 1250 void (*close)(struct sock *sk, 1251 long timeout); 1252 int (*pre_connect)(struct sock *sk, 1253 struct sockaddr *uaddr, 1254 int addr_len); 1255 int (*connect)(struct sock *sk, 1256 struct sockaddr *uaddr, 1257 int addr_len); 1258 int (*disconnect)(struct sock *sk, int flags); 1259 1260 struct sock * (*accept)(struct sock *sk, 1261 struct proto_accept_arg *arg); 1262 1263 int (*ioctl)(struct sock *sk, int cmd, 1264 int *karg); 1265 int (*init)(struct sock *sk); 1266 void (*destroy)(struct sock *sk); 1267 void (*shutdown)(struct sock *sk, int how); 1268 int (*setsockopt)(struct sock *sk, int level, 1269 int optname, sockptr_t optval, 1270 unsigned int optlen); 1271 int (*getsockopt)(struct sock *sk, int level, 1272 int optname, char __user *optval, 1273 int __user *option); 1274 void (*keepalive)(struct sock *sk, int valbool); 1275 #ifdef CONFIG_COMPAT 1276 int (*compat_ioctl)(struct sock *sk, 1277 unsigned int cmd, unsigned long arg); 1278 #endif 1279 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1280 size_t len); 1281 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1282 size_t len, int flags, int *addr_len); 1283 void (*splice_eof)(struct socket *sock); 1284 int (*bind)(struct sock *sk, 1285 struct sockaddr *addr, int addr_len); 1286 int (*bind_add)(struct sock *sk, 1287 struct sockaddr *addr, int addr_len); 1288 1289 int (*backlog_rcv) (struct sock *sk, 1290 struct sk_buff *skb); 1291 bool (*bpf_bypass_getsockopt)(int level, 1292 int optname); 1293 1294 void (*release_cb)(struct sock *sk); 1295 1296 /* Keeping track of sk's, looking them up, and port selection methods. */ 1297 int (*hash)(struct sock *sk); 1298 void (*unhash)(struct sock *sk); 1299 void (*rehash)(struct sock *sk); 1300 int (*get_port)(struct sock *sk, unsigned short snum); 1301 void (*put_port)(struct sock *sk); 1302 #ifdef CONFIG_BPF_SYSCALL 1303 int (*psock_update_sk_prot)(struct sock *sk, 1304 struct sk_psock *psock, 1305 bool restore); 1306 #endif 1307 1308 /* Keeping track of sockets in use */ 1309 #ifdef CONFIG_PROC_FS 1310 unsigned int inuse_idx; 1311 #endif 1312 1313 bool (*stream_memory_free)(const struct sock *sk, int wake); 1314 bool (*sock_is_readable)(struct sock *sk); 1315 /* Memory pressure */ 1316 void (*enter_memory_pressure)(struct sock *sk); 1317 void (*leave_memory_pressure)(struct sock *sk); 1318 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1319 int __percpu *per_cpu_fw_alloc; 1320 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1321 1322 /* 1323 * Pressure flag: try to collapse. 1324 * Technical note: it is used by multiple contexts non atomically. 1325 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1326 * All the __sk_mem_schedule() is of this nature: accounting 1327 * is strict, actions are advisory and have some latency. 1328 */ 1329 unsigned long *memory_pressure; 1330 long *sysctl_mem; 1331 1332 int *sysctl_wmem; 1333 int *sysctl_rmem; 1334 u32 sysctl_wmem_offset; 1335 u32 sysctl_rmem_offset; 1336 1337 int max_header; 1338 bool no_autobind; 1339 1340 struct kmem_cache *slab; 1341 unsigned int obj_size; 1342 unsigned int ipv6_pinfo_offset; 1343 slab_flags_t slab_flags; 1344 unsigned int useroffset; /* Usercopy region offset */ 1345 unsigned int usersize; /* Usercopy region size */ 1346 1347 unsigned int __percpu *orphan_count; 1348 1349 struct request_sock_ops *rsk_prot; 1350 struct timewait_sock_ops *twsk_prot; 1351 1352 union { 1353 struct inet_hashinfo *hashinfo; 1354 struct udp_table *udp_table; 1355 struct raw_hashinfo *raw_hash; 1356 struct smc_hashinfo *smc_hash; 1357 } h; 1358 1359 struct module *owner; 1360 1361 char name[32]; 1362 1363 struct list_head node; 1364 int (*diag_destroy)(struct sock *sk, int err); 1365 } __randomize_layout; 1366 1367 int proto_register(struct proto *prot, int alloc_slab); 1368 void proto_unregister(struct proto *prot); 1369 int sock_load_diag_module(int family, int protocol); 1370 1371 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1372 1373 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1374 { 1375 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1376 return false; 1377 1378 return sk->sk_prot->stream_memory_free ? 1379 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1380 tcp_stream_memory_free, sk, wake) : true; 1381 } 1382 1383 static inline bool sk_stream_memory_free(const struct sock *sk) 1384 { 1385 return __sk_stream_memory_free(sk, 0); 1386 } 1387 1388 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1389 { 1390 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1391 __sk_stream_memory_free(sk, wake); 1392 } 1393 1394 static inline bool sk_stream_is_writeable(const struct sock *sk) 1395 { 1396 return __sk_stream_is_writeable(sk, 0); 1397 } 1398 1399 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1400 struct cgroup *ancestor) 1401 { 1402 #ifdef CONFIG_SOCK_CGROUP_DATA 1403 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1404 ancestor); 1405 #else 1406 return -ENOTSUPP; 1407 #endif 1408 } 1409 1410 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1411 1412 static inline void sk_sockets_allocated_dec(struct sock *sk) 1413 { 1414 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1415 SK_ALLOC_PERCPU_COUNTER_BATCH); 1416 } 1417 1418 static inline void sk_sockets_allocated_inc(struct sock *sk) 1419 { 1420 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1421 SK_ALLOC_PERCPU_COUNTER_BATCH); 1422 } 1423 1424 static inline u64 1425 sk_sockets_allocated_read_positive(struct sock *sk) 1426 { 1427 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1428 } 1429 1430 static inline int 1431 proto_sockets_allocated_sum_positive(struct proto *prot) 1432 { 1433 return percpu_counter_sum_positive(prot->sockets_allocated); 1434 } 1435 1436 #ifdef CONFIG_PROC_FS 1437 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1438 struct prot_inuse { 1439 int all; 1440 int val[PROTO_INUSE_NR]; 1441 }; 1442 1443 static inline void sock_prot_inuse_add(const struct net *net, 1444 const struct proto *prot, int val) 1445 { 1446 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1447 } 1448 1449 static inline void sock_inuse_add(const struct net *net, int val) 1450 { 1451 this_cpu_add(net->core.prot_inuse->all, val); 1452 } 1453 1454 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1455 int sock_inuse_get(struct net *net); 1456 #else 1457 static inline void sock_prot_inuse_add(const struct net *net, 1458 const struct proto *prot, int val) 1459 { 1460 } 1461 1462 static inline void sock_inuse_add(const struct net *net, int val) 1463 { 1464 } 1465 #endif 1466 1467 1468 /* With per-bucket locks this operation is not-atomic, so that 1469 * this version is not worse. 1470 */ 1471 static inline int __sk_prot_rehash(struct sock *sk) 1472 { 1473 sk->sk_prot->unhash(sk); 1474 return sk->sk_prot->hash(sk); 1475 } 1476 1477 /* About 10 seconds */ 1478 #define SOCK_DESTROY_TIME (10*HZ) 1479 1480 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1481 #define PROT_SOCK 1024 1482 1483 #define SHUTDOWN_MASK 3 1484 #define RCV_SHUTDOWN 1 1485 #define SEND_SHUTDOWN 2 1486 1487 #define SOCK_BINDADDR_LOCK 4 1488 #define SOCK_BINDPORT_LOCK 8 1489 1490 struct socket_alloc { 1491 struct socket socket; 1492 struct inode vfs_inode; 1493 }; 1494 1495 static inline struct socket *SOCKET_I(struct inode *inode) 1496 { 1497 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1498 } 1499 1500 static inline struct inode *SOCK_INODE(struct socket *socket) 1501 { 1502 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1503 } 1504 1505 /* 1506 * Functions for memory accounting 1507 */ 1508 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1509 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1510 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1511 void __sk_mem_reclaim(struct sock *sk, int amount); 1512 1513 #define SK_MEM_SEND 0 1514 #define SK_MEM_RECV 1 1515 1516 /* sysctl_mem values are in pages */ 1517 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1518 { 1519 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1520 } 1521 1522 static inline int sk_mem_pages(int amt) 1523 { 1524 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1525 } 1526 1527 static inline bool sk_has_account(struct sock *sk) 1528 { 1529 /* return true if protocol supports memory accounting */ 1530 return !!sk->sk_prot->memory_allocated; 1531 } 1532 1533 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1534 { 1535 int delta; 1536 1537 if (!sk_has_account(sk)) 1538 return true; 1539 delta = size - sk->sk_forward_alloc; 1540 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1541 } 1542 1543 static inline bool 1544 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc) 1545 { 1546 int delta; 1547 1548 if (!sk_has_account(sk)) 1549 return true; 1550 delta = size - sk->sk_forward_alloc; 1551 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1552 pfmemalloc; 1553 } 1554 1555 static inline bool 1556 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1557 { 1558 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb)); 1559 } 1560 1561 static inline int sk_unused_reserved_mem(const struct sock *sk) 1562 { 1563 int unused_mem; 1564 1565 if (likely(!sk->sk_reserved_mem)) 1566 return 0; 1567 1568 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1569 atomic_read(&sk->sk_rmem_alloc); 1570 1571 return unused_mem > 0 ? unused_mem : 0; 1572 } 1573 1574 static inline void sk_mem_reclaim(struct sock *sk) 1575 { 1576 int reclaimable; 1577 1578 if (!sk_has_account(sk)) 1579 return; 1580 1581 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1582 1583 if (reclaimable >= (int)PAGE_SIZE) 1584 __sk_mem_reclaim(sk, reclaimable); 1585 } 1586 1587 static inline void sk_mem_reclaim_final(struct sock *sk) 1588 { 1589 sk->sk_reserved_mem = 0; 1590 sk_mem_reclaim(sk); 1591 } 1592 1593 static inline void sk_mem_charge(struct sock *sk, int size) 1594 { 1595 if (!sk_has_account(sk)) 1596 return; 1597 sk_forward_alloc_add(sk, -size); 1598 } 1599 1600 static inline void sk_mem_uncharge(struct sock *sk, int size) 1601 { 1602 if (!sk_has_account(sk)) 1603 return; 1604 sk_forward_alloc_add(sk, size); 1605 sk_mem_reclaim(sk); 1606 } 1607 1608 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 1609 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1610 { 1611 __module_get(owner); 1612 sk->sk_owner = owner; 1613 } 1614 1615 static inline void sk_owner_clear(struct sock *sk) 1616 { 1617 sk->sk_owner = NULL; 1618 } 1619 1620 static inline void sk_owner_put(struct sock *sk) 1621 { 1622 module_put(sk->sk_owner); 1623 } 1624 #else 1625 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1626 { 1627 } 1628 1629 static inline void sk_owner_clear(struct sock *sk) 1630 { 1631 } 1632 1633 static inline void sk_owner_put(struct sock *sk) 1634 { 1635 } 1636 #endif 1637 /* 1638 * Macro so as to not evaluate some arguments when 1639 * lockdep is not enabled. 1640 * 1641 * Mark both the sk_lock and the sk_lock.slock as a 1642 * per-address-family lock class. 1643 */ 1644 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1645 do { \ 1646 sk_owner_set(sk, THIS_MODULE); \ 1647 sk->sk_lock.owned = 0; \ 1648 init_waitqueue_head(&sk->sk_lock.wq); \ 1649 spin_lock_init(&(sk)->sk_lock.slock); \ 1650 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1651 sizeof((sk)->sk_lock)); \ 1652 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1653 (skey), (sname)); \ 1654 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1655 } while (0) 1656 1657 static inline bool lockdep_sock_is_held(const struct sock *sk) 1658 { 1659 return lockdep_is_held(&sk->sk_lock) || 1660 lockdep_is_held(&sk->sk_lock.slock); 1661 } 1662 1663 void lock_sock_nested(struct sock *sk, int subclass); 1664 1665 static inline void lock_sock(struct sock *sk) 1666 { 1667 lock_sock_nested(sk, 0); 1668 } 1669 1670 void __lock_sock(struct sock *sk); 1671 void __release_sock(struct sock *sk); 1672 void release_sock(struct sock *sk); 1673 1674 /* BH context may only use the following locking interface. */ 1675 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1676 #define bh_lock_sock_nested(__sk) \ 1677 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1678 SINGLE_DEPTH_NESTING) 1679 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1680 1681 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1682 1683 /** 1684 * lock_sock_fast - fast version of lock_sock 1685 * @sk: socket 1686 * 1687 * This version should be used for very small section, where process won't block 1688 * return false if fast path is taken: 1689 * 1690 * sk_lock.slock locked, owned = 0, BH disabled 1691 * 1692 * return true if slow path is taken: 1693 * 1694 * sk_lock.slock unlocked, owned = 1, BH enabled 1695 */ 1696 static inline bool lock_sock_fast(struct sock *sk) 1697 { 1698 /* The sk_lock has mutex_lock() semantics here. */ 1699 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1700 1701 return __lock_sock_fast(sk); 1702 } 1703 1704 /* fast socket lock variant for caller already holding a [different] socket lock */ 1705 static inline bool lock_sock_fast_nested(struct sock *sk) 1706 { 1707 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1708 1709 return __lock_sock_fast(sk); 1710 } 1711 1712 /** 1713 * unlock_sock_fast - complement of lock_sock_fast 1714 * @sk: socket 1715 * @slow: slow mode 1716 * 1717 * fast unlock socket for user context. 1718 * If slow mode is on, we call regular release_sock() 1719 */ 1720 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1721 __releases(&sk->sk_lock.slock) 1722 { 1723 if (slow) { 1724 release_sock(sk); 1725 __release(&sk->sk_lock.slock); 1726 } else { 1727 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1728 spin_unlock_bh(&sk->sk_lock.slock); 1729 } 1730 } 1731 1732 void sockopt_lock_sock(struct sock *sk); 1733 void sockopt_release_sock(struct sock *sk); 1734 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1735 bool sockopt_capable(int cap); 1736 1737 /* Used by processes to "lock" a socket state, so that 1738 * interrupts and bottom half handlers won't change it 1739 * from under us. It essentially blocks any incoming 1740 * packets, so that we won't get any new data or any 1741 * packets that change the state of the socket. 1742 * 1743 * While locked, BH processing will add new packets to 1744 * the backlog queue. This queue is processed by the 1745 * owner of the socket lock right before it is released. 1746 * 1747 * Since ~2.3.5 it is also exclusive sleep lock serializing 1748 * accesses from user process context. 1749 */ 1750 1751 static inline void sock_owned_by_me(const struct sock *sk) 1752 { 1753 #ifdef CONFIG_LOCKDEP 1754 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1755 #endif 1756 } 1757 1758 static inline void sock_not_owned_by_me(const struct sock *sk) 1759 { 1760 #ifdef CONFIG_LOCKDEP 1761 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1762 #endif 1763 } 1764 1765 static inline bool sock_owned_by_user(const struct sock *sk) 1766 { 1767 sock_owned_by_me(sk); 1768 return sk->sk_lock.owned; 1769 } 1770 1771 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1772 { 1773 return sk->sk_lock.owned; 1774 } 1775 1776 static inline void sock_release_ownership(struct sock *sk) 1777 { 1778 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1779 sk->sk_lock.owned = 0; 1780 1781 /* The sk_lock has mutex_unlock() semantics: */ 1782 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1783 } 1784 1785 /* no reclassification while locks are held */ 1786 static inline bool sock_allow_reclassification(const struct sock *csk) 1787 { 1788 struct sock *sk = (struct sock *)csk; 1789 1790 return !sock_owned_by_user_nocheck(sk) && 1791 !spin_is_locked(&sk->sk_lock.slock); 1792 } 1793 1794 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1795 struct proto *prot, int kern); 1796 void sk_free(struct sock *sk); 1797 void sk_net_refcnt_upgrade(struct sock *sk); 1798 void sk_destruct(struct sock *sk); 1799 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1800 1801 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1802 gfp_t priority); 1803 void __sock_wfree(struct sk_buff *skb); 1804 void sock_wfree(struct sk_buff *skb); 1805 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1806 gfp_t priority); 1807 void skb_orphan_partial(struct sk_buff *skb); 1808 void sock_rfree(struct sk_buff *skb); 1809 void sock_efree(struct sk_buff *skb); 1810 #ifdef CONFIG_INET 1811 void sock_edemux(struct sk_buff *skb); 1812 void sock_pfree(struct sk_buff *skb); 1813 1814 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk) 1815 { 1816 skb_orphan(skb); 1817 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 1818 skb->sk = sk; 1819 skb->destructor = sock_edemux; 1820 } 1821 } 1822 #else 1823 #define sock_edemux sock_efree 1824 #endif 1825 1826 int sk_setsockopt(struct sock *sk, int level, int optname, 1827 sockptr_t optval, unsigned int optlen); 1828 int sock_setsockopt(struct socket *sock, int level, int op, 1829 sockptr_t optval, unsigned int optlen); 1830 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1831 int optname, sockptr_t optval, int optlen); 1832 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1833 int optname, sockptr_t optval, sockptr_t optlen); 1834 1835 int sk_getsockopt(struct sock *sk, int level, int optname, 1836 sockptr_t optval, sockptr_t optlen); 1837 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1838 bool timeval, bool time32); 1839 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1840 unsigned long data_len, int noblock, 1841 int *errcode, int max_page_order); 1842 1843 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1844 unsigned long size, 1845 int noblock, int *errcode) 1846 { 1847 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1848 } 1849 1850 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1851 void *sock_kmemdup(struct sock *sk, const void *src, 1852 int size, gfp_t priority); 1853 void sock_kfree_s(struct sock *sk, void *mem, int size); 1854 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1855 void sk_send_sigurg(struct sock *sk); 1856 1857 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1858 { 1859 if (sk->sk_socket) 1860 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1861 WRITE_ONCE(sk->sk_prot, proto); 1862 } 1863 1864 struct sockcm_cookie { 1865 u64 transmit_time; 1866 u32 mark; 1867 u32 tsflags; 1868 u32 ts_opt_id; 1869 u32 priority; 1870 u32 dmabuf_id; 1871 }; 1872 1873 static inline void sockcm_init(struct sockcm_cookie *sockc, 1874 const struct sock *sk) 1875 { 1876 *sockc = (struct sockcm_cookie) { 1877 .mark = READ_ONCE(sk->sk_mark), 1878 .tsflags = READ_ONCE(sk->sk_tsflags), 1879 .priority = READ_ONCE(sk->sk_priority), 1880 }; 1881 } 1882 1883 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1884 struct sockcm_cookie *sockc); 1885 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1886 struct sockcm_cookie *sockc); 1887 1888 /* 1889 * Functions to fill in entries in struct proto_ops when a protocol 1890 * does not implement a particular function. 1891 */ 1892 int sock_no_bind(struct socket *, struct sockaddr *, int); 1893 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1894 int sock_no_socketpair(struct socket *, struct socket *); 1895 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *); 1896 int sock_no_getname(struct socket *, struct sockaddr *, int); 1897 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1898 int sock_no_listen(struct socket *, int); 1899 int sock_no_shutdown(struct socket *, int); 1900 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1901 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1902 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1903 int sock_no_mmap(struct file *file, struct socket *sock, 1904 struct vm_area_struct *vma); 1905 1906 /* 1907 * Functions to fill in entries in struct proto_ops when a protocol 1908 * uses the inet style. 1909 */ 1910 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1911 char __user *optval, int __user *optlen); 1912 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1913 int flags); 1914 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1915 sockptr_t optval, unsigned int optlen); 1916 1917 void sk_common_release(struct sock *sk); 1918 1919 /* 1920 * Default socket callbacks and setup code 1921 */ 1922 1923 /* Initialise core socket variables using an explicit uid. */ 1924 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1925 1926 /* Initialise core socket variables. 1927 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1928 */ 1929 void sock_init_data(struct socket *sock, struct sock *sk); 1930 1931 /* 1932 * Socket reference counting postulates. 1933 * 1934 * * Each user of socket SHOULD hold a reference count. 1935 * * Each access point to socket (an hash table bucket, reference from a list, 1936 * running timer, skb in flight MUST hold a reference count. 1937 * * When reference count hits 0, it means it will never increase back. 1938 * * When reference count hits 0, it means that no references from 1939 * outside exist to this socket and current process on current CPU 1940 * is last user and may/should destroy this socket. 1941 * * sk_free is called from any context: process, BH, IRQ. When 1942 * it is called, socket has no references from outside -> sk_free 1943 * may release descendant resources allocated by the socket, but 1944 * to the time when it is called, socket is NOT referenced by any 1945 * hash tables, lists etc. 1946 * * Packets, delivered from outside (from network or from another process) 1947 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1948 * when they sit in queue. Otherwise, packets will leak to hole, when 1949 * socket is looked up by one cpu and unhasing is made by another CPU. 1950 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1951 * (leak to backlog). Packet socket does all the processing inside 1952 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1953 * use separate SMP lock, so that they are prone too. 1954 */ 1955 1956 /* Ungrab socket and destroy it, if it was the last reference. */ 1957 static inline void sock_put(struct sock *sk) 1958 { 1959 if (refcount_dec_and_test(&sk->sk_refcnt)) 1960 sk_free(sk); 1961 } 1962 /* Generic version of sock_put(), dealing with all sockets 1963 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1964 */ 1965 void sock_gen_put(struct sock *sk); 1966 1967 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1968 unsigned int trim_cap, bool refcounted); 1969 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1970 const int nested) 1971 { 1972 return __sk_receive_skb(sk, skb, nested, 1, true); 1973 } 1974 1975 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1976 { 1977 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1978 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1979 return; 1980 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1981 * other WRITE_ONCE() because socket lock might be not held. 1982 */ 1983 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 1984 } 1985 1986 #define NO_QUEUE_MAPPING USHRT_MAX 1987 1988 static inline void sk_tx_queue_clear(struct sock *sk) 1989 { 1990 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1991 * other WRITE_ONCE() because socket lock might be not held. 1992 */ 1993 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 1994 } 1995 1996 static inline int sk_tx_queue_get(const struct sock *sk) 1997 { 1998 if (sk) { 1999 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 2000 * and sk_tx_queue_set(). 2001 */ 2002 int val = READ_ONCE(sk->sk_tx_queue_mapping); 2003 2004 if (val != NO_QUEUE_MAPPING) 2005 return val; 2006 } 2007 return -1; 2008 } 2009 2010 static inline void __sk_rx_queue_set(struct sock *sk, 2011 const struct sk_buff *skb, 2012 bool force_set) 2013 { 2014 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2015 if (skb_rx_queue_recorded(skb)) { 2016 u16 rx_queue = skb_get_rx_queue(skb); 2017 2018 if (force_set || 2019 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2020 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2021 } 2022 #endif 2023 } 2024 2025 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2026 { 2027 __sk_rx_queue_set(sk, skb, true); 2028 } 2029 2030 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2031 { 2032 __sk_rx_queue_set(sk, skb, false); 2033 } 2034 2035 static inline void sk_rx_queue_clear(struct sock *sk) 2036 { 2037 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2038 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2039 #endif 2040 } 2041 2042 static inline int sk_rx_queue_get(const struct sock *sk) 2043 { 2044 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2045 if (sk) { 2046 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2047 2048 if (res != NO_QUEUE_MAPPING) 2049 return res; 2050 } 2051 #endif 2052 2053 return -1; 2054 } 2055 2056 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2057 { 2058 sk->sk_socket = sock; 2059 } 2060 2061 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2062 { 2063 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2064 return &rcu_dereference_raw(sk->sk_wq)->wait; 2065 } 2066 /* Detach socket from process context. 2067 * Announce socket dead, detach it from wait queue and inode. 2068 * Note that parent inode held reference count on this struct sock, 2069 * we do not release it in this function, because protocol 2070 * probably wants some additional cleanups or even continuing 2071 * to work with this socket (TCP). 2072 */ 2073 static inline void sock_orphan(struct sock *sk) 2074 { 2075 write_lock_bh(&sk->sk_callback_lock); 2076 sock_set_flag(sk, SOCK_DEAD); 2077 sk_set_socket(sk, NULL); 2078 sk->sk_wq = NULL; 2079 write_unlock_bh(&sk->sk_callback_lock); 2080 } 2081 2082 static inline void sock_graft(struct sock *sk, struct socket *parent) 2083 { 2084 WARN_ON(parent->sk); 2085 write_lock_bh(&sk->sk_callback_lock); 2086 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2087 parent->sk = sk; 2088 sk_set_socket(sk, parent); 2089 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2090 security_sock_graft(sk, parent); 2091 write_unlock_bh(&sk->sk_callback_lock); 2092 } 2093 2094 kuid_t sock_i_uid(struct sock *sk); 2095 unsigned long __sock_i_ino(struct sock *sk); 2096 unsigned long sock_i_ino(struct sock *sk); 2097 2098 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2099 { 2100 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2101 } 2102 2103 static inline u32 net_tx_rndhash(void) 2104 { 2105 u32 v = get_random_u32(); 2106 2107 return v ?: 1; 2108 } 2109 2110 static inline void sk_set_txhash(struct sock *sk) 2111 { 2112 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2113 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2114 } 2115 2116 static inline bool sk_rethink_txhash(struct sock *sk) 2117 { 2118 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2119 sk_set_txhash(sk); 2120 return true; 2121 } 2122 return false; 2123 } 2124 2125 static inline struct dst_entry * 2126 __sk_dst_get(const struct sock *sk) 2127 { 2128 return rcu_dereference_check(sk->sk_dst_cache, 2129 lockdep_sock_is_held(sk)); 2130 } 2131 2132 static inline struct dst_entry * 2133 sk_dst_get(const struct sock *sk) 2134 { 2135 struct dst_entry *dst; 2136 2137 rcu_read_lock(); 2138 dst = rcu_dereference(sk->sk_dst_cache); 2139 if (dst && !rcuref_get(&dst->__rcuref)) 2140 dst = NULL; 2141 rcu_read_unlock(); 2142 return dst; 2143 } 2144 2145 static inline void __dst_negative_advice(struct sock *sk) 2146 { 2147 struct dst_entry *dst = __sk_dst_get(sk); 2148 2149 if (dst && dst->ops->negative_advice) 2150 dst->ops->negative_advice(sk, dst); 2151 } 2152 2153 static inline void dst_negative_advice(struct sock *sk) 2154 { 2155 sk_rethink_txhash(sk); 2156 __dst_negative_advice(sk); 2157 } 2158 2159 static inline void 2160 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2161 { 2162 struct dst_entry *old_dst; 2163 2164 sk_tx_queue_clear(sk); 2165 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2166 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2167 lockdep_sock_is_held(sk)); 2168 rcu_assign_pointer(sk->sk_dst_cache, dst); 2169 dst_release(old_dst); 2170 } 2171 2172 static inline void 2173 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2174 { 2175 struct dst_entry *old_dst; 2176 2177 sk_tx_queue_clear(sk); 2178 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2179 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst))); 2180 dst_release(old_dst); 2181 } 2182 2183 static inline void 2184 __sk_dst_reset(struct sock *sk) 2185 { 2186 __sk_dst_set(sk, NULL); 2187 } 2188 2189 static inline void 2190 sk_dst_reset(struct sock *sk) 2191 { 2192 sk_dst_set(sk, NULL); 2193 } 2194 2195 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2196 2197 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2198 2199 static inline void sk_dst_confirm(struct sock *sk) 2200 { 2201 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2202 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2203 } 2204 2205 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2206 { 2207 if (skb_get_dst_pending_confirm(skb)) { 2208 struct sock *sk = skb->sk; 2209 2210 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2211 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2212 neigh_confirm(n); 2213 } 2214 } 2215 2216 bool sk_mc_loop(const struct sock *sk); 2217 2218 static inline bool sk_can_gso(const struct sock *sk) 2219 { 2220 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2221 } 2222 2223 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2224 2225 static inline void sk_gso_disable(struct sock *sk) 2226 { 2227 sk->sk_gso_disabled = 1; 2228 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2229 } 2230 2231 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2232 struct iov_iter *from, char *to, 2233 int copy, int offset) 2234 { 2235 if (skb->ip_summed == CHECKSUM_NONE) { 2236 __wsum csum = 0; 2237 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2238 return -EFAULT; 2239 skb->csum = csum_block_add(skb->csum, csum, offset); 2240 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2241 if (!copy_from_iter_full_nocache(to, copy, from)) 2242 return -EFAULT; 2243 } else if (!copy_from_iter_full(to, copy, from)) 2244 return -EFAULT; 2245 2246 return 0; 2247 } 2248 2249 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2250 struct iov_iter *from, int copy) 2251 { 2252 int err, offset = skb->len; 2253 2254 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2255 copy, offset); 2256 if (err) 2257 __skb_trim(skb, offset); 2258 2259 return err; 2260 } 2261 2262 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2263 struct sk_buff *skb, 2264 struct page *page, 2265 int off, int copy) 2266 { 2267 int err; 2268 2269 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2270 copy, skb->len); 2271 if (err) 2272 return err; 2273 2274 skb_len_add(skb, copy); 2275 sk_wmem_queued_add(sk, copy); 2276 sk_mem_charge(sk, copy); 2277 return 0; 2278 } 2279 2280 /** 2281 * sk_wmem_alloc_get - returns write allocations 2282 * @sk: socket 2283 * 2284 * Return: sk_wmem_alloc minus initial offset of one 2285 */ 2286 static inline int sk_wmem_alloc_get(const struct sock *sk) 2287 { 2288 return refcount_read(&sk->sk_wmem_alloc) - 1; 2289 } 2290 2291 /** 2292 * sk_rmem_alloc_get - returns read allocations 2293 * @sk: socket 2294 * 2295 * Return: sk_rmem_alloc 2296 */ 2297 static inline int sk_rmem_alloc_get(const struct sock *sk) 2298 { 2299 return atomic_read(&sk->sk_rmem_alloc); 2300 } 2301 2302 /** 2303 * sk_has_allocations - check if allocations are outstanding 2304 * @sk: socket 2305 * 2306 * Return: true if socket has write or read allocations 2307 */ 2308 static inline bool sk_has_allocations(const struct sock *sk) 2309 { 2310 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2311 } 2312 2313 /** 2314 * skwq_has_sleeper - check if there are any waiting processes 2315 * @wq: struct socket_wq 2316 * 2317 * Return: true if socket_wq has waiting processes 2318 * 2319 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2320 * barrier call. They were added due to the race found within the tcp code. 2321 * 2322 * Consider following tcp code paths:: 2323 * 2324 * CPU1 CPU2 2325 * sys_select receive packet 2326 * ... ... 2327 * __add_wait_queue update tp->rcv_nxt 2328 * ... ... 2329 * tp->rcv_nxt check sock_def_readable 2330 * ... { 2331 * schedule rcu_read_lock(); 2332 * wq = rcu_dereference(sk->sk_wq); 2333 * if (wq && waitqueue_active(&wq->wait)) 2334 * wake_up_interruptible(&wq->wait) 2335 * ... 2336 * } 2337 * 2338 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2339 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2340 * could then endup calling schedule and sleep forever if there are no more 2341 * data on the socket. 2342 * 2343 */ 2344 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2345 { 2346 return wq && wq_has_sleeper(&wq->wait); 2347 } 2348 2349 /** 2350 * sock_poll_wait - wrapper for the poll_wait call. 2351 * @filp: file 2352 * @sock: socket to wait on 2353 * @p: poll_table 2354 * 2355 * See the comments in the wq_has_sleeper function. 2356 */ 2357 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2358 poll_table *p) 2359 { 2360 /* Provides a barrier we need to be sure we are in sync 2361 * with the socket flags modification. 2362 * 2363 * This memory barrier is paired in the wq_has_sleeper. 2364 */ 2365 poll_wait(filp, &sock->wq.wait, p); 2366 } 2367 2368 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2369 { 2370 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2371 u32 txhash = READ_ONCE(sk->sk_txhash); 2372 2373 if (txhash) { 2374 skb->l4_hash = 1; 2375 skb->hash = txhash; 2376 } 2377 } 2378 2379 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2380 2381 /* 2382 * Queue a received datagram if it will fit. Stream and sequenced 2383 * protocols can't normally use this as they need to fit buffers in 2384 * and play with them. 2385 * 2386 * Inlined as it's very short and called for pretty much every 2387 * packet ever received. 2388 */ 2389 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2390 { 2391 skb_orphan(skb); 2392 skb->sk = sk; 2393 skb->destructor = sock_rfree; 2394 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2395 sk_mem_charge(sk, skb->truesize); 2396 } 2397 2398 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2399 { 2400 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2401 skb_orphan(skb); 2402 skb->destructor = sock_efree; 2403 skb->sk = sk; 2404 return true; 2405 } 2406 return false; 2407 } 2408 2409 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2410 { 2411 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2412 if (skb) { 2413 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2414 skb_set_owner_r(skb, sk); 2415 return skb; 2416 } 2417 __kfree_skb(skb); 2418 } 2419 return NULL; 2420 } 2421 2422 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2423 { 2424 if (skb->destructor != sock_wfree) { 2425 skb_orphan(skb); 2426 return; 2427 } 2428 skb->slow_gro = 1; 2429 } 2430 2431 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2432 unsigned long expires); 2433 2434 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2435 2436 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2437 2438 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2439 struct sk_buff *skb, unsigned int flags, 2440 void (*destructor)(struct sock *sk, 2441 struct sk_buff *skb)); 2442 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2443 2444 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2445 enum skb_drop_reason *reason); 2446 2447 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2448 { 2449 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2450 } 2451 2452 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2453 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2454 2455 /* 2456 * Recover an error report and clear atomically 2457 */ 2458 2459 static inline int sock_error(struct sock *sk) 2460 { 2461 int err; 2462 2463 /* Avoid an atomic operation for the common case. 2464 * This is racy since another cpu/thread can change sk_err under us. 2465 */ 2466 if (likely(data_race(!sk->sk_err))) 2467 return 0; 2468 2469 err = xchg(&sk->sk_err, 0); 2470 return -err; 2471 } 2472 2473 void sk_error_report(struct sock *sk); 2474 2475 static inline unsigned long sock_wspace(struct sock *sk) 2476 { 2477 int amt = 0; 2478 2479 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2480 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2481 if (amt < 0) 2482 amt = 0; 2483 } 2484 return amt; 2485 } 2486 2487 /* Note: 2488 * We use sk->sk_wq_raw, from contexts knowing this 2489 * pointer is not NULL and cannot disappear/change. 2490 */ 2491 static inline void sk_set_bit(int nr, struct sock *sk) 2492 { 2493 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2494 !sock_flag(sk, SOCK_FASYNC)) 2495 return; 2496 2497 set_bit(nr, &sk->sk_wq_raw->flags); 2498 } 2499 2500 static inline void sk_clear_bit(int nr, struct sock *sk) 2501 { 2502 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2503 !sock_flag(sk, SOCK_FASYNC)) 2504 return; 2505 2506 clear_bit(nr, &sk->sk_wq_raw->flags); 2507 } 2508 2509 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2510 { 2511 if (sock_flag(sk, SOCK_FASYNC)) { 2512 rcu_read_lock(); 2513 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2514 rcu_read_unlock(); 2515 } 2516 } 2517 2518 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band) 2519 { 2520 if (unlikely(sock_flag(sk, SOCK_FASYNC))) 2521 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2522 } 2523 2524 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2525 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2526 * Note: for send buffers, TCP works better if we can build two skbs at 2527 * minimum. 2528 */ 2529 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2530 2531 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2532 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2533 2534 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2535 { 2536 u32 val; 2537 2538 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2539 return; 2540 2541 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2542 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2543 2544 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2545 } 2546 2547 /** 2548 * sk_page_frag - return an appropriate page_frag 2549 * @sk: socket 2550 * 2551 * Use the per task page_frag instead of the per socket one for 2552 * optimization when we know that we're in process context and own 2553 * everything that's associated with %current. 2554 * 2555 * Both direct reclaim and page faults can nest inside other 2556 * socket operations and end up recursing into sk_page_frag() 2557 * while it's already in use: explicitly avoid task page_frag 2558 * when users disable sk_use_task_frag. 2559 * 2560 * Return: a per task page_frag if context allows that, 2561 * otherwise a per socket one. 2562 */ 2563 static inline struct page_frag *sk_page_frag(struct sock *sk) 2564 { 2565 if (sk->sk_use_task_frag) 2566 return ¤t->task_frag; 2567 2568 return &sk->sk_frag; 2569 } 2570 2571 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2572 2573 /* 2574 * Default write policy as shown to user space via poll/select/SIGIO 2575 */ 2576 static inline bool sock_writeable(const struct sock *sk) 2577 { 2578 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2579 } 2580 2581 static inline gfp_t gfp_any(void) 2582 { 2583 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2584 } 2585 2586 static inline gfp_t gfp_memcg_charge(void) 2587 { 2588 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2589 } 2590 2591 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2592 { 2593 return noblock ? 0 : sk->sk_rcvtimeo; 2594 } 2595 2596 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2597 { 2598 return noblock ? 0 : sk->sk_sndtimeo; 2599 } 2600 2601 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2602 { 2603 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2604 2605 return v ?: 1; 2606 } 2607 2608 /* Alas, with timeout socket operations are not restartable. 2609 * Compare this to poll(). 2610 */ 2611 static inline int sock_intr_errno(long timeo) 2612 { 2613 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2614 } 2615 2616 struct sock_skb_cb { 2617 u32 dropcount; 2618 }; 2619 2620 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2621 * using skb->cb[] would keep using it directly and utilize its 2622 * alignment guarantee. 2623 */ 2624 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \ 2625 sizeof(struct sock_skb_cb)) 2626 2627 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2628 SOCK_SKB_CB_OFFSET)) 2629 2630 #define sock_skb_cb_check_size(size) \ 2631 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2632 2633 static inline void 2634 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2635 { 2636 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2637 atomic_read(&sk->sk_drops) : 0; 2638 } 2639 2640 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2641 { 2642 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2643 2644 atomic_add(segs, &sk->sk_drops); 2645 } 2646 2647 static inline ktime_t sock_read_timestamp(struct sock *sk) 2648 { 2649 #if BITS_PER_LONG==32 2650 unsigned int seq; 2651 ktime_t kt; 2652 2653 do { 2654 seq = read_seqbegin(&sk->sk_stamp_seq); 2655 kt = sk->sk_stamp; 2656 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2657 2658 return kt; 2659 #else 2660 return READ_ONCE(sk->sk_stamp); 2661 #endif 2662 } 2663 2664 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2665 { 2666 #if BITS_PER_LONG==32 2667 write_seqlock(&sk->sk_stamp_seq); 2668 sk->sk_stamp = kt; 2669 write_sequnlock(&sk->sk_stamp_seq); 2670 #else 2671 WRITE_ONCE(sk->sk_stamp, kt); 2672 #endif 2673 } 2674 2675 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2676 struct sk_buff *skb); 2677 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2678 struct sk_buff *skb); 2679 2680 static inline void 2681 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2682 { 2683 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2684 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2685 ktime_t kt = skb->tstamp; 2686 /* 2687 * generate control messages if 2688 * - receive time stamping in software requested 2689 * - software time stamp available and wanted 2690 * - hardware time stamps available and wanted 2691 */ 2692 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2693 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2694 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2695 (hwtstamps->hwtstamp && 2696 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2697 __sock_recv_timestamp(msg, sk, skb); 2698 else 2699 sock_write_timestamp(sk, kt); 2700 2701 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2702 __sock_recv_wifi_status(msg, sk, skb); 2703 } 2704 2705 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2706 struct sk_buff *skb); 2707 2708 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2709 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2710 struct sk_buff *skb) 2711 { 2712 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2713 (1UL << SOCK_RCVTSTAMP) | \ 2714 (1UL << SOCK_RCVMARK) | \ 2715 (1UL << SOCK_RCVPRIORITY) | \ 2716 (1UL << SOCK_TIMESTAMPING_ANY)) 2717 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2718 SOF_TIMESTAMPING_RAW_HARDWARE) 2719 2720 if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS) 2721 __sock_recv_cmsgs(msg, sk, skb); 2722 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2723 sock_write_timestamp(sk, skb->tstamp); 2724 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2725 sock_write_timestamp(sk, 0); 2726 } 2727 2728 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags); 2729 2730 /** 2731 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2732 * @sk: socket sending this packet 2733 * @sockc: pointer to socket cmsg cookie to get timestamping info 2734 * @tx_flags: completed with instructions for time stamping 2735 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2736 * 2737 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2738 */ 2739 static inline void _sock_tx_timestamp(struct sock *sk, 2740 const struct sockcm_cookie *sockc, 2741 __u8 *tx_flags, __u32 *tskey) 2742 { 2743 __u32 tsflags = sockc->tsflags; 2744 2745 if (unlikely(tsflags)) { 2746 __sock_tx_timestamp(tsflags, tx_flags); 2747 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2748 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) { 2749 if (tsflags & SOCKCM_FLAG_TS_OPT_ID) 2750 *tskey = sockc->ts_opt_id; 2751 else 2752 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2753 } 2754 } 2755 } 2756 2757 static inline void sock_tx_timestamp(struct sock *sk, 2758 const struct sockcm_cookie *sockc, 2759 __u8 *tx_flags) 2760 { 2761 _sock_tx_timestamp(sk, sockc, tx_flags, NULL); 2762 } 2763 2764 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, 2765 const struct sockcm_cookie *sockc) 2766 { 2767 _sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags, 2768 &skb_shinfo(skb)->tskey); 2769 } 2770 2771 static inline bool sk_is_inet(const struct sock *sk) 2772 { 2773 int family = READ_ONCE(sk->sk_family); 2774 2775 return family == AF_INET || family == AF_INET6; 2776 } 2777 2778 static inline bool sk_is_tcp(const struct sock *sk) 2779 { 2780 return sk_is_inet(sk) && 2781 sk->sk_type == SOCK_STREAM && 2782 sk->sk_protocol == IPPROTO_TCP; 2783 } 2784 2785 static inline bool sk_is_udp(const struct sock *sk) 2786 { 2787 return sk_is_inet(sk) && 2788 sk->sk_type == SOCK_DGRAM && 2789 sk->sk_protocol == IPPROTO_UDP; 2790 } 2791 2792 static inline bool sk_is_unix(const struct sock *sk) 2793 { 2794 return sk->sk_family == AF_UNIX; 2795 } 2796 2797 static inline bool sk_is_stream_unix(const struct sock *sk) 2798 { 2799 return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM; 2800 } 2801 2802 static inline bool sk_is_vsock(const struct sock *sk) 2803 { 2804 return sk->sk_family == AF_VSOCK; 2805 } 2806 2807 static inline bool sk_may_scm_recv(const struct sock *sk) 2808 { 2809 return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) || 2810 sk->sk_family == AF_NETLINK || 2811 (IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH); 2812 } 2813 2814 /** 2815 * sk_eat_skb - Release a skb if it is no longer needed 2816 * @sk: socket to eat this skb from 2817 * @skb: socket buffer to eat 2818 * 2819 * This routine must be called with interrupts disabled or with the socket 2820 * locked so that the sk_buff queue operation is ok. 2821 */ 2822 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2823 { 2824 __skb_unlink(skb, &sk->sk_receive_queue); 2825 __kfree_skb(skb); 2826 } 2827 2828 static inline bool 2829 skb_sk_is_prefetched(struct sk_buff *skb) 2830 { 2831 #ifdef CONFIG_INET 2832 return skb->destructor == sock_pfree; 2833 #else 2834 return false; 2835 #endif /* CONFIG_INET */ 2836 } 2837 2838 /* This helper checks if a socket is a full socket, 2839 * ie _not_ a timewait or request socket. 2840 */ 2841 static inline bool sk_fullsock(const struct sock *sk) 2842 { 2843 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2844 } 2845 2846 static inline bool 2847 sk_is_refcounted(struct sock *sk) 2848 { 2849 /* Only full sockets have sk->sk_flags. */ 2850 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2851 } 2852 2853 static inline bool 2854 sk_requests_wifi_status(struct sock *sk) 2855 { 2856 return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS); 2857 } 2858 2859 /* Checks if this SKB belongs to an HW offloaded socket 2860 * and whether any SW fallbacks are required based on dev. 2861 * Check decrypted mark in case skb_orphan() cleared socket. 2862 */ 2863 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2864 struct net_device *dev) 2865 { 2866 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2867 struct sock *sk = skb->sk; 2868 2869 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2870 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2871 } else if (unlikely(skb_is_decrypted(skb))) { 2872 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2873 kfree_skb(skb); 2874 skb = NULL; 2875 } 2876 #endif 2877 2878 return skb; 2879 } 2880 2881 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2882 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2883 */ 2884 static inline bool sk_listener(const struct sock *sk) 2885 { 2886 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2887 } 2888 2889 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT 2890 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE) 2891 * TCP RST and ACK can be attached to TIME_WAIT. 2892 */ 2893 static inline bool sk_listener_or_tw(const struct sock *sk) 2894 { 2895 return (1 << READ_ONCE(sk->sk_state)) & 2896 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT); 2897 } 2898 2899 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2900 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2901 int type); 2902 2903 bool sk_ns_capable(const struct sock *sk, 2904 struct user_namespace *user_ns, int cap); 2905 bool sk_capable(const struct sock *sk, int cap); 2906 bool sk_net_capable(const struct sock *sk, int cap); 2907 2908 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2909 2910 /* Take into consideration the size of the struct sk_buff overhead in the 2911 * determination of these values, since that is non-constant across 2912 * platforms. This makes socket queueing behavior and performance 2913 * not depend upon such differences. 2914 */ 2915 #define _SK_MEM_PACKETS 256 2916 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2917 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2918 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2919 2920 extern __u32 sysctl_wmem_max; 2921 extern __u32 sysctl_rmem_max; 2922 2923 extern __u32 sysctl_wmem_default; 2924 extern __u32 sysctl_rmem_default; 2925 2926 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2927 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2928 2929 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2930 { 2931 /* Does this proto have per netns sysctl_wmem ? */ 2932 if (proto->sysctl_wmem_offset) 2933 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2934 2935 return READ_ONCE(*proto->sysctl_wmem); 2936 } 2937 2938 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2939 { 2940 /* Does this proto have per netns sysctl_rmem ? */ 2941 if (proto->sysctl_rmem_offset) 2942 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2943 2944 return READ_ONCE(*proto->sysctl_rmem); 2945 } 2946 2947 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2948 * Some wifi drivers need to tweak it to get more chunks. 2949 * They can use this helper from their ndo_start_xmit() 2950 */ 2951 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2952 { 2953 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2954 return; 2955 WRITE_ONCE(sk->sk_pacing_shift, val); 2956 } 2957 2958 /* if a socket is bound to a device, check that the given device 2959 * index is either the same or that the socket is bound to an L3 2960 * master device and the given device index is also enslaved to 2961 * that L3 master 2962 */ 2963 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2964 { 2965 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2966 int mdif; 2967 2968 if (!bound_dev_if || bound_dev_if == dif) 2969 return true; 2970 2971 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2972 if (mdif && mdif == bound_dev_if) 2973 return true; 2974 2975 return false; 2976 } 2977 2978 void sock_def_readable(struct sock *sk); 2979 2980 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 2981 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 2982 int sock_set_timestamping(struct sock *sk, int optname, 2983 struct so_timestamping timestamping); 2984 2985 void sock_enable_timestamps(struct sock *sk); 2986 #if defined(CONFIG_CGROUP_BPF) 2987 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op); 2988 #else 2989 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 2990 { 2991 } 2992 #endif 2993 void sock_no_linger(struct sock *sk); 2994 void sock_set_keepalive(struct sock *sk); 2995 void sock_set_priority(struct sock *sk, u32 priority); 2996 void sock_set_rcvbuf(struct sock *sk, int val); 2997 void sock_set_mark(struct sock *sk, u32 val); 2998 void sock_set_reuseaddr(struct sock *sk); 2999 void sock_set_reuseport(struct sock *sk); 3000 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3001 3002 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 3003 3004 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3005 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3006 sockptr_t optval, int optlen, bool old_timeval); 3007 3008 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3009 void __user *arg, void *karg, size_t size); 3010 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3011 static inline bool sk_is_readable(struct sock *sk) 3012 { 3013 const struct proto *prot = READ_ONCE(sk->sk_prot); 3014 3015 if (prot->sock_is_readable) 3016 return prot->sock_is_readable(sk); 3017 3018 return false; 3019 } 3020 #endif /* _SOCK_H */ 3021