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