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 /* Define this to get the SOCK_DBG debugging facility. */ 80 #define SOCK_DEBUGGING 81 #ifdef SOCK_DEBUGGING 82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \ 83 printk(KERN_DEBUG msg); } while (0) 84 #else 85 /* Validate arguments and do nothing */ 86 static inline __printf(2, 3) 87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...) 88 { 89 } 90 #endif 91 92 /* This is the per-socket lock. The spinlock provides a synchronization 93 * between user contexts and software interrupt processing, whereas the 94 * mini-semaphore synchronizes multiple users amongst themselves. 95 */ 96 typedef struct { 97 spinlock_t slock; 98 int owned; 99 wait_queue_head_t wq; 100 /* 101 * We express the mutex-alike socket_lock semantics 102 * to the lock validator by explicitly managing 103 * the slock as a lock variant (in addition to 104 * the slock itself): 105 */ 106 #ifdef CONFIG_DEBUG_LOCK_ALLOC 107 struct lockdep_map dep_map; 108 #endif 109 } socket_lock_t; 110 111 struct sock; 112 struct proto; 113 struct net; 114 115 typedef __u32 __bitwise __portpair; 116 typedef __u64 __bitwise __addrpair; 117 118 /** 119 * struct sock_common - minimal network layer representation of sockets 120 * @skc_daddr: Foreign IPv4 addr 121 * @skc_rcv_saddr: Bound local IPv4 addr 122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 123 * @skc_hash: hash value used with various protocol lookup tables 124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 125 * @skc_dport: placeholder for inet_dport/tw_dport 126 * @skc_num: placeholder for inet_num/tw_num 127 * @skc_portpair: __u32 union of @skc_dport & @skc_num 128 * @skc_family: network address family 129 * @skc_state: Connection state 130 * @skc_reuse: %SO_REUSEADDR setting 131 * @skc_reuseport: %SO_REUSEPORT setting 132 * @skc_ipv6only: socket is IPV6 only 133 * @skc_net_refcnt: socket is using net ref counting 134 * @skc_bound_dev_if: bound device index if != 0 135 * @skc_bind_node: bind hash linkage for various protocol lookup tables 136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 137 * @skc_prot: protocol handlers inside a network family 138 * @skc_net: reference to the network namespace of this socket 139 * @skc_v6_daddr: IPV6 destination address 140 * @skc_v6_rcv_saddr: IPV6 source address 141 * @skc_cookie: socket's cookie value 142 * @skc_node: main hash linkage for various protocol lookup tables 143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 144 * @skc_tx_queue_mapping: tx queue number for this connection 145 * @skc_rx_queue_mapping: rx queue number for this connection 146 * @skc_flags: place holder for sk_flags 147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 149 * @skc_listener: connection request listener socket (aka rsk_listener) 150 * [union with @skc_flags] 151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 152 * [union with @skc_flags] 153 * @skc_incoming_cpu: record/match cpu processing incoming packets 154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 155 * [union with @skc_incoming_cpu] 156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 157 * [union with @skc_incoming_cpu] 158 * @skc_refcnt: reference count 159 * 160 * This is the minimal network layer representation of sockets, the header 161 * for struct sock and struct inet_timewait_sock. 162 */ 163 struct sock_common { 164 union { 165 __addrpair skc_addrpair; 166 struct { 167 __be32 skc_daddr; 168 __be32 skc_rcv_saddr; 169 }; 170 }; 171 union { 172 unsigned int skc_hash; 173 __u16 skc_u16hashes[2]; 174 }; 175 /* skc_dport && skc_num must be grouped as well */ 176 union { 177 __portpair skc_portpair; 178 struct { 179 __be16 skc_dport; 180 __u16 skc_num; 181 }; 182 }; 183 184 unsigned short skc_family; 185 volatile unsigned char skc_state; 186 unsigned char skc_reuse:4; 187 unsigned char skc_reuseport:1; 188 unsigned char skc_ipv6only:1; 189 unsigned char skc_net_refcnt:1; 190 int skc_bound_dev_if; 191 union { 192 struct hlist_node skc_bind_node; 193 struct hlist_node skc_portaddr_node; 194 }; 195 struct proto *skc_prot; 196 possible_net_t skc_net; 197 198 #if IS_ENABLED(CONFIG_IPV6) 199 struct in6_addr skc_v6_daddr; 200 struct in6_addr skc_v6_rcv_saddr; 201 #endif 202 203 atomic64_t skc_cookie; 204 205 /* following fields are padding to force 206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 207 * assuming IPV6 is enabled. We use this padding differently 208 * for different kind of 'sockets' 209 */ 210 union { 211 unsigned long skc_flags; 212 struct sock *skc_listener; /* request_sock */ 213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 214 }; 215 /* 216 * fields between dontcopy_begin/dontcopy_end 217 * are not copied in sock_copy() 218 */ 219 /* private: */ 220 int skc_dontcopy_begin[0]; 221 /* public: */ 222 union { 223 struct hlist_node skc_node; 224 struct hlist_nulls_node skc_nulls_node; 225 }; 226 unsigned short skc_tx_queue_mapping; 227 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 228 unsigned short skc_rx_queue_mapping; 229 #endif 230 union { 231 int skc_incoming_cpu; 232 u32 skc_rcv_wnd; 233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 234 }; 235 236 refcount_t skc_refcnt; 237 /* private: */ 238 int skc_dontcopy_end[0]; 239 union { 240 u32 skc_rxhash; 241 u32 skc_window_clamp; 242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 243 }; 244 /* public: */ 245 }; 246 247 struct bpf_local_storage; 248 struct sk_filter; 249 250 /** 251 * struct sock - network layer representation of sockets 252 * @__sk_common: shared layout with inet_timewait_sock 253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 255 * @sk_lock: synchronizer 256 * @sk_kern_sock: True if sock is using kernel lock classes 257 * @sk_rcvbuf: size of receive buffer in bytes 258 * @sk_wq: sock wait queue and async head 259 * @sk_rx_dst: receive input route used by early demux 260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 262 * @sk_dst_cache: destination cache 263 * @sk_dst_pending_confirm: need to confirm neighbour 264 * @sk_policy: flow policy 265 * @sk_receive_queue: incoming packets 266 * @sk_wmem_alloc: transmit queue bytes committed 267 * @sk_tsq_flags: TCP Small Queues flags 268 * @sk_write_queue: Packet sending queue 269 * @sk_omem_alloc: "o" is "option" or "other" 270 * @sk_wmem_queued: persistent queue size 271 * @sk_forward_alloc: space allocated forward 272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 273 * @sk_napi_id: id of the last napi context to receive data for sk 274 * @sk_ll_usec: usecs to busypoll when there is no data 275 * @sk_allocation: allocation mode 276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 279 * @sk_sndbuf: size of send buffer in bytes 280 * @__sk_flags_offset: empty field used to determine location of bitfield 281 * @sk_padding: unused element for alignment 282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 283 * @sk_no_check_rx: allow zero checksum in RX packets 284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 287 * @sk_gso_max_size: Maximum GSO segment size to build 288 * @sk_gso_max_segs: Maximum number of GSO segments 289 * @sk_pacing_shift: scaling factor for TCP Small Queues 290 * @sk_lingertime: %SO_LINGER l_linger setting 291 * @sk_backlog: always used with the per-socket spinlock held 292 * @sk_callback_lock: used with the callbacks in the end of this struct 293 * @sk_error_queue: rarely used 294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 295 * IPV6_ADDRFORM for instance) 296 * @sk_err: last error 297 * @sk_err_soft: errors that don't cause failure but are the cause of a 298 * persistent failure not just 'timed out' 299 * @sk_drops: raw/udp drops counter 300 * @sk_ack_backlog: current listen backlog 301 * @sk_max_ack_backlog: listen backlog set in listen() 302 * @sk_uid: user id of owner 303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 304 * @sk_busy_poll_budget: napi processing budget when busypolling 305 * @sk_priority: %SO_PRIORITY setting 306 * @sk_type: socket type (%SOCK_STREAM, etc) 307 * @sk_protocol: which protocol this socket belongs in this network family 308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 309 * @sk_peer_pid: &struct pid for this socket's peer 310 * @sk_peer_cred: %SO_PEERCRED setting 311 * @sk_rcvlowat: %SO_RCVLOWAT setting 312 * @sk_rcvtimeo: %SO_RCVTIMEO setting 313 * @sk_sndtimeo: %SO_SNDTIMEO setting 314 * @sk_txhash: computed flow hash for use on transmit 315 * @sk_txrehash: enable TX hash rethink 316 * @sk_filter: socket filtering instructions 317 * @sk_timer: sock cleanup timer 318 * @sk_stamp: time stamp of last packet received 319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 320 * @sk_tsflags: SO_TIMESTAMPING flags 321 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 322 * Sockets that can be used under memory reclaim should 323 * set this to false. 324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 325 * for timestamping 326 * @sk_tskey: counter to disambiguate concurrent tstamp requests 327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 328 * @sk_socket: Identd and reporting IO signals 329 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 330 * @sk_frag: cached page frag 331 * @sk_peek_off: current peek_offset value 332 * @sk_send_head: front of stuff to transmit 333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 334 * @sk_security: used by security modules 335 * @sk_mark: generic packet mark 336 * @sk_cgrp_data: cgroup data for this cgroup 337 * @sk_memcg: this socket's memory cgroup association 338 * @sk_write_pending: a write to stream socket waits to start 339 * @sk_disconnects: number of disconnect operations performed on this sock 340 * @sk_state_change: callback to indicate change in the state of the sock 341 * @sk_data_ready: callback to indicate there is data to be processed 342 * @sk_write_space: callback to indicate there is bf sending space available 343 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 344 * @sk_backlog_rcv: callback to process the backlog 345 * @sk_validate_xmit_skb: ptr to an optional validate function 346 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 347 * @sk_reuseport_cb: reuseport group container 348 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 349 * @sk_rcu: used during RCU grace period 350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 353 * @sk_txtime_unused: unused txtime flags 354 * @ns_tracker: tracker for netns reference 355 * @sk_bind2_node: bind node in the bhash2 table 356 */ 357 struct sock { 358 /* 359 * Now struct inet_timewait_sock also uses sock_common, so please just 360 * don't add nothing before this first member (__sk_common) --acme 361 */ 362 struct sock_common __sk_common; 363 #define sk_node __sk_common.skc_node 364 #define sk_nulls_node __sk_common.skc_nulls_node 365 #define sk_refcnt __sk_common.skc_refcnt 366 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 367 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 368 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 369 #endif 370 371 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 372 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 373 #define sk_hash __sk_common.skc_hash 374 #define sk_portpair __sk_common.skc_portpair 375 #define sk_num __sk_common.skc_num 376 #define sk_dport __sk_common.skc_dport 377 #define sk_addrpair __sk_common.skc_addrpair 378 #define sk_daddr __sk_common.skc_daddr 379 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 380 #define sk_family __sk_common.skc_family 381 #define sk_state __sk_common.skc_state 382 #define sk_reuse __sk_common.skc_reuse 383 #define sk_reuseport __sk_common.skc_reuseport 384 #define sk_ipv6only __sk_common.skc_ipv6only 385 #define sk_net_refcnt __sk_common.skc_net_refcnt 386 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 387 #define sk_bind_node __sk_common.skc_bind_node 388 #define sk_prot __sk_common.skc_prot 389 #define sk_net __sk_common.skc_net 390 #define sk_v6_daddr __sk_common.skc_v6_daddr 391 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 392 #define sk_cookie __sk_common.skc_cookie 393 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 394 #define sk_flags __sk_common.skc_flags 395 #define sk_rxhash __sk_common.skc_rxhash 396 397 /* early demux fields */ 398 struct dst_entry __rcu *sk_rx_dst; 399 int sk_rx_dst_ifindex; 400 u32 sk_rx_dst_cookie; 401 402 socket_lock_t sk_lock; 403 atomic_t sk_drops; 404 int sk_rcvlowat; 405 struct sk_buff_head sk_error_queue; 406 struct sk_buff_head sk_receive_queue; 407 /* 408 * The backlog queue is special, it is always used with 409 * the per-socket spinlock held and requires low latency 410 * access. Therefore we special case it's implementation. 411 * Note : rmem_alloc is in this structure to fill a hole 412 * on 64bit arches, not because its logically part of 413 * backlog. 414 */ 415 struct { 416 atomic_t rmem_alloc; 417 int len; 418 struct sk_buff *head; 419 struct sk_buff *tail; 420 } sk_backlog; 421 422 #define sk_rmem_alloc sk_backlog.rmem_alloc 423 424 int sk_forward_alloc; 425 u32 sk_reserved_mem; 426 #ifdef CONFIG_NET_RX_BUSY_POLL 427 unsigned int sk_ll_usec; 428 /* ===== mostly read cache line ===== */ 429 unsigned int sk_napi_id; 430 #endif 431 int sk_rcvbuf; 432 int sk_disconnects; 433 434 struct sk_filter __rcu *sk_filter; 435 union { 436 struct socket_wq __rcu *sk_wq; 437 /* private: */ 438 struct socket_wq *sk_wq_raw; 439 /* public: */ 440 }; 441 #ifdef CONFIG_XFRM 442 struct xfrm_policy __rcu *sk_policy[2]; 443 #endif 444 445 struct dst_entry __rcu *sk_dst_cache; 446 atomic_t sk_omem_alloc; 447 int sk_sndbuf; 448 449 /* ===== cache line for TX ===== */ 450 int sk_wmem_queued; 451 refcount_t sk_wmem_alloc; 452 unsigned long sk_tsq_flags; 453 union { 454 struct sk_buff *sk_send_head; 455 struct rb_root tcp_rtx_queue; 456 }; 457 struct sk_buff_head sk_write_queue; 458 __s32 sk_peek_off; 459 int sk_write_pending; 460 __u32 sk_dst_pending_confirm; 461 u32 sk_pacing_status; /* see enum sk_pacing */ 462 long sk_sndtimeo; 463 struct timer_list sk_timer; 464 __u32 sk_priority; 465 __u32 sk_mark; 466 unsigned long sk_pacing_rate; /* bytes per second */ 467 unsigned long sk_max_pacing_rate; 468 struct page_frag sk_frag; 469 netdev_features_t sk_route_caps; 470 int sk_gso_type; 471 unsigned int sk_gso_max_size; 472 gfp_t sk_allocation; 473 __u32 sk_txhash; 474 475 /* 476 * Because of non atomicity rules, all 477 * changes are protected by socket lock. 478 */ 479 u8 sk_gso_disabled : 1, 480 sk_kern_sock : 1, 481 sk_no_check_tx : 1, 482 sk_no_check_rx : 1, 483 sk_userlocks : 4; 484 u8 sk_pacing_shift; 485 u16 sk_type; 486 u16 sk_protocol; 487 u16 sk_gso_max_segs; 488 unsigned long sk_lingertime; 489 struct proto *sk_prot_creator; 490 rwlock_t sk_callback_lock; 491 int sk_err, 492 sk_err_soft; 493 u32 sk_ack_backlog; 494 u32 sk_max_ack_backlog; 495 kuid_t sk_uid; 496 u8 sk_txrehash; 497 #ifdef CONFIG_NET_RX_BUSY_POLL 498 u8 sk_prefer_busy_poll; 499 u16 sk_busy_poll_budget; 500 #endif 501 spinlock_t sk_peer_lock; 502 int sk_bind_phc; 503 struct pid *sk_peer_pid; 504 const struct cred *sk_peer_cred; 505 506 long sk_rcvtimeo; 507 ktime_t sk_stamp; 508 #if BITS_PER_LONG==32 509 seqlock_t sk_stamp_seq; 510 #endif 511 atomic_t sk_tskey; 512 atomic_t sk_zckey; 513 u32 sk_tsflags; 514 u8 sk_shutdown; 515 516 u8 sk_clockid; 517 u8 sk_txtime_deadline_mode : 1, 518 sk_txtime_report_errors : 1, 519 sk_txtime_unused : 6; 520 bool sk_use_task_frag; 521 522 struct socket *sk_socket; 523 void *sk_user_data; 524 #ifdef CONFIG_SECURITY 525 void *sk_security; 526 #endif 527 struct sock_cgroup_data sk_cgrp_data; 528 struct mem_cgroup *sk_memcg; 529 void (*sk_state_change)(struct sock *sk); 530 void (*sk_data_ready)(struct sock *sk); 531 void (*sk_write_space)(struct sock *sk); 532 void (*sk_error_report)(struct sock *sk); 533 int (*sk_backlog_rcv)(struct sock *sk, 534 struct sk_buff *skb); 535 #ifdef CONFIG_SOCK_VALIDATE_XMIT 536 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 537 struct net_device *dev, 538 struct sk_buff *skb); 539 #endif 540 void (*sk_destruct)(struct sock *sk); 541 struct sock_reuseport __rcu *sk_reuseport_cb; 542 #ifdef CONFIG_BPF_SYSCALL 543 struct bpf_local_storage __rcu *sk_bpf_storage; 544 #endif 545 struct rcu_head sk_rcu; 546 netns_tracker ns_tracker; 547 struct hlist_node sk_bind2_node; 548 }; 549 550 enum sk_pacing { 551 SK_PACING_NONE = 0, 552 SK_PACING_NEEDED = 1, 553 SK_PACING_FQ = 2, 554 }; 555 556 /* flag bits in sk_user_data 557 * 558 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 559 * not be suitable for copying when cloning the socket. For instance, 560 * it can point to a reference counted object. sk_user_data bottom 561 * bit is set if pointer must not be copied. 562 * 563 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 564 * managed/owned by a BPF reuseport array. This bit should be set 565 * when sk_user_data's sk is added to the bpf's reuseport_array. 566 * 567 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 568 * sk_user_data points to psock type. This bit should be set 569 * when sk_user_data is assigned to a psock object. 570 */ 571 #define SK_USER_DATA_NOCOPY 1UL 572 #define SK_USER_DATA_BPF 2UL 573 #define SK_USER_DATA_PSOCK 4UL 574 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 575 SK_USER_DATA_PSOCK) 576 577 /** 578 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 579 * @sk: socket 580 */ 581 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 582 { 583 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 584 } 585 586 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 587 588 /** 589 * __locked_read_sk_user_data_with_flags - return the pointer 590 * only if argument flags all has been set in sk_user_data. Otherwise 591 * return NULL 592 * 593 * @sk: socket 594 * @flags: flag bits 595 * 596 * The caller must be holding sk->sk_callback_lock. 597 */ 598 static inline void * 599 __locked_read_sk_user_data_with_flags(const struct sock *sk, 600 uintptr_t flags) 601 { 602 uintptr_t sk_user_data = 603 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 604 lockdep_is_held(&sk->sk_callback_lock)); 605 606 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 607 608 if ((sk_user_data & flags) == flags) 609 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 610 return NULL; 611 } 612 613 /** 614 * __rcu_dereference_sk_user_data_with_flags - return the pointer 615 * only if argument flags all has been set in sk_user_data. Otherwise 616 * return NULL 617 * 618 * @sk: socket 619 * @flags: flag bits 620 */ 621 static inline void * 622 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 623 uintptr_t flags) 624 { 625 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 626 627 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 628 629 if ((sk_user_data & flags) == flags) 630 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 631 return NULL; 632 } 633 634 #define rcu_dereference_sk_user_data(sk) \ 635 __rcu_dereference_sk_user_data_with_flags(sk, 0) 636 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 637 ({ \ 638 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 639 __tmp2 = (uintptr_t)(flags); \ 640 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 641 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 642 rcu_assign_pointer(__sk_user_data((sk)), \ 643 __tmp1 | __tmp2); \ 644 }) 645 #define rcu_assign_sk_user_data(sk, ptr) \ 646 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 647 648 static inline 649 struct net *sock_net(const struct sock *sk) 650 { 651 return read_pnet(&sk->sk_net); 652 } 653 654 static inline 655 void sock_net_set(struct sock *sk, struct net *net) 656 { 657 write_pnet(&sk->sk_net, net); 658 } 659 660 /* 661 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 662 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 663 * on a socket means that the socket will reuse everybody else's port 664 * without looking at the other's sk_reuse value. 665 */ 666 667 #define SK_NO_REUSE 0 668 #define SK_CAN_REUSE 1 669 #define SK_FORCE_REUSE 2 670 671 int sk_set_peek_off(struct sock *sk, int val); 672 673 static inline int sk_peek_offset(const struct sock *sk, int flags) 674 { 675 if (unlikely(flags & MSG_PEEK)) { 676 return READ_ONCE(sk->sk_peek_off); 677 } 678 679 return 0; 680 } 681 682 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 683 { 684 s32 off = READ_ONCE(sk->sk_peek_off); 685 686 if (unlikely(off >= 0)) { 687 off = max_t(s32, off - val, 0); 688 WRITE_ONCE(sk->sk_peek_off, off); 689 } 690 } 691 692 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 693 { 694 sk_peek_offset_bwd(sk, -val); 695 } 696 697 /* 698 * Hashed lists helper routines 699 */ 700 static inline struct sock *sk_entry(const struct hlist_node *node) 701 { 702 return hlist_entry(node, struct sock, sk_node); 703 } 704 705 static inline struct sock *__sk_head(const struct hlist_head *head) 706 { 707 return hlist_entry(head->first, struct sock, sk_node); 708 } 709 710 static inline struct sock *sk_head(const struct hlist_head *head) 711 { 712 return hlist_empty(head) ? NULL : __sk_head(head); 713 } 714 715 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 716 { 717 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 718 } 719 720 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 721 { 722 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 723 } 724 725 static inline struct sock *sk_next(const struct sock *sk) 726 { 727 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 728 } 729 730 static inline struct sock *sk_nulls_next(const struct sock *sk) 731 { 732 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 733 hlist_nulls_entry(sk->sk_nulls_node.next, 734 struct sock, sk_nulls_node) : 735 NULL; 736 } 737 738 static inline bool sk_unhashed(const struct sock *sk) 739 { 740 return hlist_unhashed(&sk->sk_node); 741 } 742 743 static inline bool sk_hashed(const struct sock *sk) 744 { 745 return !sk_unhashed(sk); 746 } 747 748 static inline void sk_node_init(struct hlist_node *node) 749 { 750 node->pprev = NULL; 751 } 752 753 static inline void __sk_del_node(struct sock *sk) 754 { 755 __hlist_del(&sk->sk_node); 756 } 757 758 /* NB: equivalent to hlist_del_init_rcu */ 759 static inline bool __sk_del_node_init(struct sock *sk) 760 { 761 if (sk_hashed(sk)) { 762 __sk_del_node(sk); 763 sk_node_init(&sk->sk_node); 764 return true; 765 } 766 return false; 767 } 768 769 /* Grab socket reference count. This operation is valid only 770 when sk is ALREADY grabbed f.e. it is found in hash table 771 or a list and the lookup is made under lock preventing hash table 772 modifications. 773 */ 774 775 static __always_inline void sock_hold(struct sock *sk) 776 { 777 refcount_inc(&sk->sk_refcnt); 778 } 779 780 /* Ungrab socket in the context, which assumes that socket refcnt 781 cannot hit zero, f.e. it is true in context of any socketcall. 782 */ 783 static __always_inline void __sock_put(struct sock *sk) 784 { 785 refcount_dec(&sk->sk_refcnt); 786 } 787 788 static inline bool sk_del_node_init(struct sock *sk) 789 { 790 bool rc = __sk_del_node_init(sk); 791 792 if (rc) { 793 /* paranoid for a while -acme */ 794 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 795 __sock_put(sk); 796 } 797 return rc; 798 } 799 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 800 801 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 802 { 803 if (sk_hashed(sk)) { 804 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 805 return true; 806 } 807 return false; 808 } 809 810 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 811 { 812 bool rc = __sk_nulls_del_node_init_rcu(sk); 813 814 if (rc) { 815 /* paranoid for a while -acme */ 816 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 817 __sock_put(sk); 818 } 819 return rc; 820 } 821 822 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 823 { 824 hlist_add_head(&sk->sk_node, list); 825 } 826 827 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 828 { 829 sock_hold(sk); 830 __sk_add_node(sk, list); 831 } 832 833 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 834 { 835 sock_hold(sk); 836 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 837 sk->sk_family == AF_INET6) 838 hlist_add_tail_rcu(&sk->sk_node, list); 839 else 840 hlist_add_head_rcu(&sk->sk_node, list); 841 } 842 843 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 844 { 845 sock_hold(sk); 846 hlist_add_tail_rcu(&sk->sk_node, list); 847 } 848 849 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 850 { 851 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 852 } 853 854 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 855 { 856 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 857 } 858 859 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 860 { 861 sock_hold(sk); 862 __sk_nulls_add_node_rcu(sk, list); 863 } 864 865 static inline void __sk_del_bind_node(struct sock *sk) 866 { 867 __hlist_del(&sk->sk_bind_node); 868 } 869 870 static inline void sk_add_bind_node(struct sock *sk, 871 struct hlist_head *list) 872 { 873 hlist_add_head(&sk->sk_bind_node, list); 874 } 875 876 static inline void __sk_del_bind2_node(struct sock *sk) 877 { 878 __hlist_del(&sk->sk_bind2_node); 879 } 880 881 static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list) 882 { 883 hlist_add_head(&sk->sk_bind2_node, list); 884 } 885 886 #define sk_for_each(__sk, list) \ 887 hlist_for_each_entry(__sk, list, sk_node) 888 #define sk_for_each_rcu(__sk, list) \ 889 hlist_for_each_entry_rcu(__sk, list, sk_node) 890 #define sk_nulls_for_each(__sk, node, list) \ 891 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 892 #define sk_nulls_for_each_rcu(__sk, node, list) \ 893 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 894 #define sk_for_each_from(__sk) \ 895 hlist_for_each_entry_from(__sk, sk_node) 896 #define sk_nulls_for_each_from(__sk, node) \ 897 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 898 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 899 #define sk_for_each_safe(__sk, tmp, list) \ 900 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 901 #define sk_for_each_bound(__sk, list) \ 902 hlist_for_each_entry(__sk, list, sk_bind_node) 903 #define sk_for_each_bound_bhash2(__sk, list) \ 904 hlist_for_each_entry(__sk, list, sk_bind2_node) 905 #define sk_for_each_bound_safe(__sk, tmp, list) \ 906 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node) 907 908 /** 909 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 910 * @tpos: the type * to use as a loop cursor. 911 * @pos: the &struct hlist_node to use as a loop cursor. 912 * @head: the head for your list. 913 * @offset: offset of hlist_node within the struct. 914 * 915 */ 916 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 917 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 918 pos != NULL && \ 919 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 920 pos = rcu_dereference(hlist_next_rcu(pos))) 921 922 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 923 { 924 /* Careful only use this in a context where these parameters 925 * can not change and must all be valid, such as recvmsg from 926 * userspace. 927 */ 928 return sk->sk_socket->file->f_cred->user_ns; 929 } 930 931 /* Sock flags */ 932 enum sock_flags { 933 SOCK_DEAD, 934 SOCK_DONE, 935 SOCK_URGINLINE, 936 SOCK_KEEPOPEN, 937 SOCK_LINGER, 938 SOCK_DESTROY, 939 SOCK_BROADCAST, 940 SOCK_TIMESTAMP, 941 SOCK_ZAPPED, 942 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 943 SOCK_DBG, /* %SO_DEBUG setting */ 944 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 945 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 946 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 947 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 948 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 949 SOCK_FASYNC, /* fasync() active */ 950 SOCK_RXQ_OVFL, 951 SOCK_ZEROCOPY, /* buffers from userspace */ 952 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 953 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 954 * Will use last 4 bytes of packet sent from 955 * user-space instead. 956 */ 957 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 958 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 959 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 960 SOCK_TXTIME, 961 SOCK_XDP, /* XDP is attached */ 962 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 963 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 964 }; 965 966 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 967 968 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 969 { 970 nsk->sk_flags = osk->sk_flags; 971 } 972 973 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 974 { 975 __set_bit(flag, &sk->sk_flags); 976 } 977 978 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 979 { 980 __clear_bit(flag, &sk->sk_flags); 981 } 982 983 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 984 int valbool) 985 { 986 if (valbool) 987 sock_set_flag(sk, bit); 988 else 989 sock_reset_flag(sk, bit); 990 } 991 992 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 993 { 994 return test_bit(flag, &sk->sk_flags); 995 } 996 997 #ifdef CONFIG_NET 998 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 999 static inline int sk_memalloc_socks(void) 1000 { 1001 return static_branch_unlikely(&memalloc_socks_key); 1002 } 1003 1004 void __receive_sock(struct file *file); 1005 #else 1006 1007 static inline int sk_memalloc_socks(void) 1008 { 1009 return 0; 1010 } 1011 1012 static inline void __receive_sock(struct file *file) 1013 { } 1014 #endif 1015 1016 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1017 { 1018 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1019 } 1020 1021 static inline void sk_acceptq_removed(struct sock *sk) 1022 { 1023 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1024 } 1025 1026 static inline void sk_acceptq_added(struct sock *sk) 1027 { 1028 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1029 } 1030 1031 /* Note: If you think the test should be: 1032 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1033 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1034 */ 1035 static inline bool sk_acceptq_is_full(const struct sock *sk) 1036 { 1037 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1038 } 1039 1040 /* 1041 * Compute minimal free write space needed to queue new packets. 1042 */ 1043 static inline int sk_stream_min_wspace(const struct sock *sk) 1044 { 1045 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1046 } 1047 1048 static inline int sk_stream_wspace(const struct sock *sk) 1049 { 1050 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1051 } 1052 1053 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1054 { 1055 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1056 } 1057 1058 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1059 { 1060 /* Paired with lockless reads of sk->sk_forward_alloc */ 1061 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1062 } 1063 1064 void sk_stream_write_space(struct sock *sk); 1065 1066 /* OOB backlog add */ 1067 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1068 { 1069 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1070 skb_dst_force(skb); 1071 1072 if (!sk->sk_backlog.tail) 1073 WRITE_ONCE(sk->sk_backlog.head, skb); 1074 else 1075 sk->sk_backlog.tail->next = skb; 1076 1077 WRITE_ONCE(sk->sk_backlog.tail, skb); 1078 skb->next = NULL; 1079 } 1080 1081 /* 1082 * Take into account size of receive queue and backlog queue 1083 * Do not take into account this skb truesize, 1084 * to allow even a single big packet to come. 1085 */ 1086 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1087 { 1088 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1089 1090 return qsize > limit; 1091 } 1092 1093 /* The per-socket spinlock must be held here. */ 1094 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1095 unsigned int limit) 1096 { 1097 if (sk_rcvqueues_full(sk, limit)) 1098 return -ENOBUFS; 1099 1100 /* 1101 * If the skb was allocated from pfmemalloc reserves, only 1102 * allow SOCK_MEMALLOC sockets to use it as this socket is 1103 * helping free memory 1104 */ 1105 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1106 return -ENOMEM; 1107 1108 __sk_add_backlog(sk, skb); 1109 sk->sk_backlog.len += skb->truesize; 1110 return 0; 1111 } 1112 1113 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1114 1115 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1116 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1117 1118 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1119 { 1120 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1121 return __sk_backlog_rcv(sk, skb); 1122 1123 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1124 tcp_v6_do_rcv, 1125 tcp_v4_do_rcv, 1126 sk, skb); 1127 } 1128 1129 static inline void sk_incoming_cpu_update(struct sock *sk) 1130 { 1131 int cpu = raw_smp_processor_id(); 1132 1133 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1134 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1135 } 1136 1137 static inline void sock_rps_record_flow_hash(__u32 hash) 1138 { 1139 #ifdef CONFIG_RPS 1140 struct rps_sock_flow_table *sock_flow_table; 1141 1142 rcu_read_lock(); 1143 sock_flow_table = rcu_dereference(rps_sock_flow_table); 1144 rps_record_sock_flow(sock_flow_table, hash); 1145 rcu_read_unlock(); 1146 #endif 1147 } 1148 1149 static inline void sock_rps_record_flow(const struct sock *sk) 1150 { 1151 #ifdef CONFIG_RPS 1152 if (static_branch_unlikely(&rfs_needed)) { 1153 /* Reading sk->sk_rxhash might incur an expensive cache line 1154 * miss. 1155 * 1156 * TCP_ESTABLISHED does cover almost all states where RFS 1157 * might be useful, and is cheaper [1] than testing : 1158 * IPv4: inet_sk(sk)->inet_daddr 1159 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr) 1160 * OR an additional socket flag 1161 * [1] : sk_state and sk_prot are in the same cache line. 1162 */ 1163 if (sk->sk_state == TCP_ESTABLISHED) { 1164 /* This READ_ONCE() is paired with the WRITE_ONCE() 1165 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash(). 1166 */ 1167 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash)); 1168 } 1169 } 1170 #endif 1171 } 1172 1173 static inline void sock_rps_save_rxhash(struct sock *sk, 1174 const struct sk_buff *skb) 1175 { 1176 #ifdef CONFIG_RPS 1177 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1178 * here, and another one in sock_rps_record_flow(). 1179 */ 1180 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1181 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1182 #endif 1183 } 1184 1185 static inline void sock_rps_reset_rxhash(struct sock *sk) 1186 { 1187 #ifdef CONFIG_RPS 1188 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1189 WRITE_ONCE(sk->sk_rxhash, 0); 1190 #endif 1191 } 1192 1193 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1194 ({ int __rc, __dis = __sk->sk_disconnects; \ 1195 release_sock(__sk); \ 1196 __rc = __condition; \ 1197 if (!__rc) { \ 1198 *(__timeo) = wait_woken(__wait, \ 1199 TASK_INTERRUPTIBLE, \ 1200 *(__timeo)); \ 1201 } \ 1202 sched_annotate_sleep(); \ 1203 lock_sock(__sk); \ 1204 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1205 __rc; \ 1206 }) 1207 1208 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1209 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1210 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1211 int sk_stream_error(struct sock *sk, int flags, int err); 1212 void sk_stream_kill_queues(struct sock *sk); 1213 void sk_set_memalloc(struct sock *sk); 1214 void sk_clear_memalloc(struct sock *sk); 1215 1216 void __sk_flush_backlog(struct sock *sk); 1217 1218 static inline bool sk_flush_backlog(struct sock *sk) 1219 { 1220 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1221 __sk_flush_backlog(sk); 1222 return true; 1223 } 1224 return false; 1225 } 1226 1227 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1228 1229 struct request_sock_ops; 1230 struct timewait_sock_ops; 1231 struct inet_hashinfo; 1232 struct raw_hashinfo; 1233 struct smc_hashinfo; 1234 struct module; 1235 struct sk_psock; 1236 1237 /* 1238 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1239 * un-modified. Special care is taken when initializing object to zero. 1240 */ 1241 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1242 { 1243 if (offsetof(struct sock, sk_node.next) != 0) 1244 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1245 memset(&sk->sk_node.pprev, 0, 1246 size - offsetof(struct sock, sk_node.pprev)); 1247 } 1248 1249 /* Networking protocol blocks we attach to sockets. 1250 * socket layer -> transport layer interface 1251 */ 1252 struct proto { 1253 void (*close)(struct sock *sk, 1254 long timeout); 1255 int (*pre_connect)(struct sock *sk, 1256 struct sockaddr *uaddr, 1257 int addr_len); 1258 int (*connect)(struct sock *sk, 1259 struct sockaddr *uaddr, 1260 int addr_len); 1261 int (*disconnect)(struct sock *sk, int flags); 1262 1263 struct sock * (*accept)(struct sock *sk, int flags, int *err, 1264 bool kern); 1265 1266 int (*ioctl)(struct sock *sk, int cmd, 1267 int *karg); 1268 int (*init)(struct sock *sk); 1269 void (*destroy)(struct sock *sk); 1270 void (*shutdown)(struct sock *sk, int how); 1271 int (*setsockopt)(struct sock *sk, int level, 1272 int optname, sockptr_t optval, 1273 unsigned int optlen); 1274 int (*getsockopt)(struct sock *sk, int level, 1275 int optname, char __user *optval, 1276 int __user *option); 1277 void (*keepalive)(struct sock *sk, int valbool); 1278 #ifdef CONFIG_COMPAT 1279 int (*compat_ioctl)(struct sock *sk, 1280 unsigned int cmd, unsigned long arg); 1281 #endif 1282 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1283 size_t len); 1284 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1285 size_t len, int flags, int *addr_len); 1286 void (*splice_eof)(struct socket *sock); 1287 int (*bind)(struct sock *sk, 1288 struct sockaddr *addr, int addr_len); 1289 int (*bind_add)(struct sock *sk, 1290 struct sockaddr *addr, int addr_len); 1291 1292 int (*backlog_rcv) (struct sock *sk, 1293 struct sk_buff *skb); 1294 bool (*bpf_bypass_getsockopt)(int level, 1295 int optname); 1296 1297 void (*release_cb)(struct sock *sk); 1298 1299 /* Keeping track of sk's, looking them up, and port selection methods. */ 1300 int (*hash)(struct sock *sk); 1301 void (*unhash)(struct sock *sk); 1302 void (*rehash)(struct sock *sk); 1303 int (*get_port)(struct sock *sk, unsigned short snum); 1304 void (*put_port)(struct sock *sk); 1305 #ifdef CONFIG_BPF_SYSCALL 1306 int (*psock_update_sk_prot)(struct sock *sk, 1307 struct sk_psock *psock, 1308 bool restore); 1309 #endif 1310 1311 /* Keeping track of sockets in use */ 1312 #ifdef CONFIG_PROC_FS 1313 unsigned int inuse_idx; 1314 #endif 1315 1316 #if IS_ENABLED(CONFIG_MPTCP) 1317 int (*forward_alloc_get)(const struct sock *sk); 1318 #endif 1319 1320 bool (*stream_memory_free)(const struct sock *sk, int wake); 1321 bool (*sock_is_readable)(struct sock *sk); 1322 /* Memory pressure */ 1323 void (*enter_memory_pressure)(struct sock *sk); 1324 void (*leave_memory_pressure)(struct sock *sk); 1325 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1326 int __percpu *per_cpu_fw_alloc; 1327 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1328 1329 /* 1330 * Pressure flag: try to collapse. 1331 * Technical note: it is used by multiple contexts non atomically. 1332 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1333 * All the __sk_mem_schedule() is of this nature: accounting 1334 * is strict, actions are advisory and have some latency. 1335 */ 1336 unsigned long *memory_pressure; 1337 long *sysctl_mem; 1338 1339 int *sysctl_wmem; 1340 int *sysctl_rmem; 1341 u32 sysctl_wmem_offset; 1342 u32 sysctl_rmem_offset; 1343 1344 int max_header; 1345 bool no_autobind; 1346 1347 struct kmem_cache *slab; 1348 unsigned int obj_size; 1349 unsigned int ipv6_pinfo_offset; 1350 slab_flags_t slab_flags; 1351 unsigned int useroffset; /* Usercopy region offset */ 1352 unsigned int usersize; /* Usercopy region size */ 1353 1354 unsigned int __percpu *orphan_count; 1355 1356 struct request_sock_ops *rsk_prot; 1357 struct timewait_sock_ops *twsk_prot; 1358 1359 union { 1360 struct inet_hashinfo *hashinfo; 1361 struct udp_table *udp_table; 1362 struct raw_hashinfo *raw_hash; 1363 struct smc_hashinfo *smc_hash; 1364 } h; 1365 1366 struct module *owner; 1367 1368 char name[32]; 1369 1370 struct list_head node; 1371 int (*diag_destroy)(struct sock *sk, int err); 1372 } __randomize_layout; 1373 1374 int proto_register(struct proto *prot, int alloc_slab); 1375 void proto_unregister(struct proto *prot); 1376 int sock_load_diag_module(int family, int protocol); 1377 1378 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1379 1380 static inline int sk_forward_alloc_get(const struct sock *sk) 1381 { 1382 #if IS_ENABLED(CONFIG_MPTCP) 1383 if (sk->sk_prot->forward_alloc_get) 1384 return sk->sk_prot->forward_alloc_get(sk); 1385 #endif 1386 return READ_ONCE(sk->sk_forward_alloc); 1387 } 1388 1389 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1390 { 1391 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1392 return false; 1393 1394 return sk->sk_prot->stream_memory_free ? 1395 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1396 tcp_stream_memory_free, sk, wake) : true; 1397 } 1398 1399 static inline bool sk_stream_memory_free(const struct sock *sk) 1400 { 1401 return __sk_stream_memory_free(sk, 0); 1402 } 1403 1404 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1405 { 1406 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1407 __sk_stream_memory_free(sk, wake); 1408 } 1409 1410 static inline bool sk_stream_is_writeable(const struct sock *sk) 1411 { 1412 return __sk_stream_is_writeable(sk, 0); 1413 } 1414 1415 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1416 struct cgroup *ancestor) 1417 { 1418 #ifdef CONFIG_SOCK_CGROUP_DATA 1419 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1420 ancestor); 1421 #else 1422 return -ENOTSUPP; 1423 #endif 1424 } 1425 1426 static inline bool sk_has_memory_pressure(const struct sock *sk) 1427 { 1428 return sk->sk_prot->memory_pressure != NULL; 1429 } 1430 1431 static inline bool sk_under_global_memory_pressure(const struct sock *sk) 1432 { 1433 return sk->sk_prot->memory_pressure && 1434 !!READ_ONCE(*sk->sk_prot->memory_pressure); 1435 } 1436 1437 static inline bool sk_under_memory_pressure(const struct sock *sk) 1438 { 1439 if (!sk->sk_prot->memory_pressure) 1440 return false; 1441 1442 if (mem_cgroup_sockets_enabled && sk->sk_memcg && 1443 mem_cgroup_under_socket_pressure(sk->sk_memcg)) 1444 return true; 1445 1446 return !!READ_ONCE(*sk->sk_prot->memory_pressure); 1447 } 1448 1449 static inline long 1450 proto_memory_allocated(const struct proto *prot) 1451 { 1452 return max(0L, atomic_long_read(prot->memory_allocated)); 1453 } 1454 1455 static inline long 1456 sk_memory_allocated(const struct sock *sk) 1457 { 1458 return proto_memory_allocated(sk->sk_prot); 1459 } 1460 1461 /* 1 MB per cpu, in page units */ 1462 #define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT)) 1463 extern int sysctl_mem_pcpu_rsv; 1464 1465 static inline void proto_memory_pcpu_drain(struct proto *proto) 1466 { 1467 int val = this_cpu_xchg(*proto->per_cpu_fw_alloc, 0); 1468 1469 if (val) 1470 atomic_long_add(val, proto->memory_allocated); 1471 } 1472 1473 static inline void 1474 sk_memory_allocated_add(const struct sock *sk, int val) 1475 { 1476 struct proto *proto = sk->sk_prot; 1477 1478 val = this_cpu_add_return(*proto->per_cpu_fw_alloc, val); 1479 1480 if (unlikely(val >= READ_ONCE(sysctl_mem_pcpu_rsv))) 1481 proto_memory_pcpu_drain(proto); 1482 } 1483 1484 static inline void 1485 sk_memory_allocated_sub(const struct sock *sk, int val) 1486 { 1487 struct proto *proto = sk->sk_prot; 1488 1489 val = this_cpu_sub_return(*proto->per_cpu_fw_alloc, val); 1490 1491 if (unlikely(val <= -READ_ONCE(sysctl_mem_pcpu_rsv))) 1492 proto_memory_pcpu_drain(proto); 1493 } 1494 1495 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1496 1497 static inline void sk_sockets_allocated_dec(struct sock *sk) 1498 { 1499 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1500 SK_ALLOC_PERCPU_COUNTER_BATCH); 1501 } 1502 1503 static inline void sk_sockets_allocated_inc(struct sock *sk) 1504 { 1505 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1506 SK_ALLOC_PERCPU_COUNTER_BATCH); 1507 } 1508 1509 static inline u64 1510 sk_sockets_allocated_read_positive(struct sock *sk) 1511 { 1512 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1513 } 1514 1515 static inline int 1516 proto_sockets_allocated_sum_positive(struct proto *prot) 1517 { 1518 return percpu_counter_sum_positive(prot->sockets_allocated); 1519 } 1520 1521 static inline bool 1522 proto_memory_pressure(struct proto *prot) 1523 { 1524 if (!prot->memory_pressure) 1525 return false; 1526 return !!READ_ONCE(*prot->memory_pressure); 1527 } 1528 1529 1530 #ifdef CONFIG_PROC_FS 1531 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1532 struct prot_inuse { 1533 int all; 1534 int val[PROTO_INUSE_NR]; 1535 }; 1536 1537 static inline void sock_prot_inuse_add(const struct net *net, 1538 const struct proto *prot, int val) 1539 { 1540 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1541 } 1542 1543 static inline void sock_inuse_add(const struct net *net, int val) 1544 { 1545 this_cpu_add(net->core.prot_inuse->all, val); 1546 } 1547 1548 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1549 int sock_inuse_get(struct net *net); 1550 #else 1551 static inline void sock_prot_inuse_add(const struct net *net, 1552 const struct proto *prot, int val) 1553 { 1554 } 1555 1556 static inline void sock_inuse_add(const struct net *net, int val) 1557 { 1558 } 1559 #endif 1560 1561 1562 /* With per-bucket locks this operation is not-atomic, so that 1563 * this version is not worse. 1564 */ 1565 static inline int __sk_prot_rehash(struct sock *sk) 1566 { 1567 sk->sk_prot->unhash(sk); 1568 return sk->sk_prot->hash(sk); 1569 } 1570 1571 /* About 10 seconds */ 1572 #define SOCK_DESTROY_TIME (10*HZ) 1573 1574 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1575 #define PROT_SOCK 1024 1576 1577 #define SHUTDOWN_MASK 3 1578 #define RCV_SHUTDOWN 1 1579 #define SEND_SHUTDOWN 2 1580 1581 #define SOCK_BINDADDR_LOCK 4 1582 #define SOCK_BINDPORT_LOCK 8 1583 1584 struct socket_alloc { 1585 struct socket socket; 1586 struct inode vfs_inode; 1587 }; 1588 1589 static inline struct socket *SOCKET_I(struct inode *inode) 1590 { 1591 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1592 } 1593 1594 static inline struct inode *SOCK_INODE(struct socket *socket) 1595 { 1596 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1597 } 1598 1599 /* 1600 * Functions for memory accounting 1601 */ 1602 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1603 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1604 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1605 void __sk_mem_reclaim(struct sock *sk, int amount); 1606 1607 #define SK_MEM_SEND 0 1608 #define SK_MEM_RECV 1 1609 1610 /* sysctl_mem values are in pages */ 1611 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1612 { 1613 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1614 } 1615 1616 static inline int sk_mem_pages(int amt) 1617 { 1618 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1619 } 1620 1621 static inline bool sk_has_account(struct sock *sk) 1622 { 1623 /* return true if protocol supports memory accounting */ 1624 return !!sk->sk_prot->memory_allocated; 1625 } 1626 1627 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1628 { 1629 int delta; 1630 1631 if (!sk_has_account(sk)) 1632 return true; 1633 delta = size - sk->sk_forward_alloc; 1634 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1635 } 1636 1637 static inline bool 1638 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc) 1639 { 1640 int delta; 1641 1642 if (!sk_has_account(sk)) 1643 return true; 1644 delta = size - sk->sk_forward_alloc; 1645 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1646 pfmemalloc; 1647 } 1648 1649 static inline bool 1650 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size) 1651 { 1652 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb)); 1653 } 1654 1655 static inline int sk_unused_reserved_mem(const struct sock *sk) 1656 { 1657 int unused_mem; 1658 1659 if (likely(!sk->sk_reserved_mem)) 1660 return 0; 1661 1662 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1663 atomic_read(&sk->sk_rmem_alloc); 1664 1665 return unused_mem > 0 ? unused_mem : 0; 1666 } 1667 1668 static inline void sk_mem_reclaim(struct sock *sk) 1669 { 1670 int reclaimable; 1671 1672 if (!sk_has_account(sk)) 1673 return; 1674 1675 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1676 1677 if (reclaimable >= (int)PAGE_SIZE) 1678 __sk_mem_reclaim(sk, reclaimable); 1679 } 1680 1681 static inline void sk_mem_reclaim_final(struct sock *sk) 1682 { 1683 sk->sk_reserved_mem = 0; 1684 sk_mem_reclaim(sk); 1685 } 1686 1687 static inline void sk_mem_charge(struct sock *sk, int size) 1688 { 1689 if (!sk_has_account(sk)) 1690 return; 1691 sk_forward_alloc_add(sk, -size); 1692 } 1693 1694 static inline void sk_mem_uncharge(struct sock *sk, int size) 1695 { 1696 if (!sk_has_account(sk)) 1697 return; 1698 sk_forward_alloc_add(sk, size); 1699 sk_mem_reclaim(sk); 1700 } 1701 1702 /* 1703 * Macro so as to not evaluate some arguments when 1704 * lockdep is not enabled. 1705 * 1706 * Mark both the sk_lock and the sk_lock.slock as a 1707 * per-address-family lock class. 1708 */ 1709 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1710 do { \ 1711 sk->sk_lock.owned = 0; \ 1712 init_waitqueue_head(&sk->sk_lock.wq); \ 1713 spin_lock_init(&(sk)->sk_lock.slock); \ 1714 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1715 sizeof((sk)->sk_lock)); \ 1716 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1717 (skey), (sname)); \ 1718 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1719 } while (0) 1720 1721 static inline bool lockdep_sock_is_held(const struct sock *sk) 1722 { 1723 return lockdep_is_held(&sk->sk_lock) || 1724 lockdep_is_held(&sk->sk_lock.slock); 1725 } 1726 1727 void lock_sock_nested(struct sock *sk, int subclass); 1728 1729 static inline void lock_sock(struct sock *sk) 1730 { 1731 lock_sock_nested(sk, 0); 1732 } 1733 1734 void __lock_sock(struct sock *sk); 1735 void __release_sock(struct sock *sk); 1736 void release_sock(struct sock *sk); 1737 1738 /* BH context may only use the following locking interface. */ 1739 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1740 #define bh_lock_sock_nested(__sk) \ 1741 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1742 SINGLE_DEPTH_NESTING) 1743 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1744 1745 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1746 1747 /** 1748 * lock_sock_fast - fast version of lock_sock 1749 * @sk: socket 1750 * 1751 * This version should be used for very small section, where process wont block 1752 * return false if fast path is taken: 1753 * 1754 * sk_lock.slock locked, owned = 0, BH disabled 1755 * 1756 * return true if slow path is taken: 1757 * 1758 * sk_lock.slock unlocked, owned = 1, BH enabled 1759 */ 1760 static inline bool lock_sock_fast(struct sock *sk) 1761 { 1762 /* The sk_lock has mutex_lock() semantics here. */ 1763 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1764 1765 return __lock_sock_fast(sk); 1766 } 1767 1768 /* fast socket lock variant for caller already holding a [different] socket lock */ 1769 static inline bool lock_sock_fast_nested(struct sock *sk) 1770 { 1771 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1772 1773 return __lock_sock_fast(sk); 1774 } 1775 1776 /** 1777 * unlock_sock_fast - complement of lock_sock_fast 1778 * @sk: socket 1779 * @slow: slow mode 1780 * 1781 * fast unlock socket for user context. 1782 * If slow mode is on, we call regular release_sock() 1783 */ 1784 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1785 __releases(&sk->sk_lock.slock) 1786 { 1787 if (slow) { 1788 release_sock(sk); 1789 __release(&sk->sk_lock.slock); 1790 } else { 1791 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1792 spin_unlock_bh(&sk->sk_lock.slock); 1793 } 1794 } 1795 1796 void sockopt_lock_sock(struct sock *sk); 1797 void sockopt_release_sock(struct sock *sk); 1798 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1799 bool sockopt_capable(int cap); 1800 1801 /* Used by processes to "lock" a socket state, so that 1802 * interrupts and bottom half handlers won't change it 1803 * from under us. It essentially blocks any incoming 1804 * packets, so that we won't get any new data or any 1805 * packets that change the state of the socket. 1806 * 1807 * While locked, BH processing will add new packets to 1808 * the backlog queue. This queue is processed by the 1809 * owner of the socket lock right before it is released. 1810 * 1811 * Since ~2.3.5 it is also exclusive sleep lock serializing 1812 * accesses from user process context. 1813 */ 1814 1815 static inline void sock_owned_by_me(const struct sock *sk) 1816 { 1817 #ifdef CONFIG_LOCKDEP 1818 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1819 #endif 1820 } 1821 1822 static inline void sock_not_owned_by_me(const struct sock *sk) 1823 { 1824 #ifdef CONFIG_LOCKDEP 1825 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1826 #endif 1827 } 1828 1829 static inline bool sock_owned_by_user(const struct sock *sk) 1830 { 1831 sock_owned_by_me(sk); 1832 return sk->sk_lock.owned; 1833 } 1834 1835 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1836 { 1837 return sk->sk_lock.owned; 1838 } 1839 1840 static inline void sock_release_ownership(struct sock *sk) 1841 { 1842 if (sock_owned_by_user_nocheck(sk)) { 1843 sk->sk_lock.owned = 0; 1844 1845 /* The sk_lock has mutex_unlock() semantics: */ 1846 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1847 } 1848 } 1849 1850 /* no reclassification while locks are held */ 1851 static inline bool sock_allow_reclassification(const struct sock *csk) 1852 { 1853 struct sock *sk = (struct sock *)csk; 1854 1855 return !sock_owned_by_user_nocheck(sk) && 1856 !spin_is_locked(&sk->sk_lock.slock); 1857 } 1858 1859 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1860 struct proto *prot, int kern); 1861 void sk_free(struct sock *sk); 1862 void sk_destruct(struct sock *sk); 1863 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1864 void sk_free_unlock_clone(struct sock *sk); 1865 1866 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1867 gfp_t priority); 1868 void __sock_wfree(struct sk_buff *skb); 1869 void sock_wfree(struct sk_buff *skb); 1870 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1871 gfp_t priority); 1872 void skb_orphan_partial(struct sk_buff *skb); 1873 void sock_rfree(struct sk_buff *skb); 1874 void sock_efree(struct sk_buff *skb); 1875 #ifdef CONFIG_INET 1876 void sock_edemux(struct sk_buff *skb); 1877 void sock_pfree(struct sk_buff *skb); 1878 #else 1879 #define sock_edemux sock_efree 1880 #endif 1881 1882 int sk_setsockopt(struct sock *sk, int level, int optname, 1883 sockptr_t optval, unsigned int optlen); 1884 int sock_setsockopt(struct socket *sock, int level, int op, 1885 sockptr_t optval, unsigned int optlen); 1886 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1887 int optname, sockptr_t optval, int optlen); 1888 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1889 int optname, sockptr_t optval, sockptr_t optlen); 1890 1891 int sk_getsockopt(struct sock *sk, int level, int optname, 1892 sockptr_t optval, sockptr_t optlen); 1893 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1894 bool timeval, bool time32); 1895 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1896 unsigned long data_len, int noblock, 1897 int *errcode, int max_page_order); 1898 1899 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1900 unsigned long size, 1901 int noblock, int *errcode) 1902 { 1903 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1904 } 1905 1906 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1907 void sock_kfree_s(struct sock *sk, void *mem, int size); 1908 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1909 void sk_send_sigurg(struct sock *sk); 1910 1911 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1912 { 1913 if (sk->sk_socket) 1914 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1915 WRITE_ONCE(sk->sk_prot, proto); 1916 } 1917 1918 struct sockcm_cookie { 1919 u64 transmit_time; 1920 u32 mark; 1921 u32 tsflags; 1922 }; 1923 1924 static inline void sockcm_init(struct sockcm_cookie *sockc, 1925 const struct sock *sk) 1926 { 1927 *sockc = (struct sockcm_cookie) { 1928 .tsflags = READ_ONCE(sk->sk_tsflags) 1929 }; 1930 } 1931 1932 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1933 struct sockcm_cookie *sockc); 1934 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1935 struct sockcm_cookie *sockc); 1936 1937 /* 1938 * Functions to fill in entries in struct proto_ops when a protocol 1939 * does not implement a particular function. 1940 */ 1941 int sock_no_bind(struct socket *, struct sockaddr *, int); 1942 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1943 int sock_no_socketpair(struct socket *, struct socket *); 1944 int sock_no_accept(struct socket *, struct socket *, int, bool); 1945 int sock_no_getname(struct socket *, struct sockaddr *, int); 1946 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1947 int sock_no_listen(struct socket *, int); 1948 int sock_no_shutdown(struct socket *, int); 1949 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1950 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1951 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1952 int sock_no_mmap(struct file *file, struct socket *sock, 1953 struct vm_area_struct *vma); 1954 1955 /* 1956 * Functions to fill in entries in struct proto_ops when a protocol 1957 * uses the inet style. 1958 */ 1959 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1960 char __user *optval, int __user *optlen); 1961 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1962 int flags); 1963 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1964 sockptr_t optval, unsigned int optlen); 1965 1966 void sk_common_release(struct sock *sk); 1967 1968 /* 1969 * Default socket callbacks and setup code 1970 */ 1971 1972 /* Initialise core socket variables using an explicit uid. */ 1973 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1974 1975 /* Initialise core socket variables. 1976 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1977 */ 1978 void sock_init_data(struct socket *sock, struct sock *sk); 1979 1980 /* 1981 * Socket reference counting postulates. 1982 * 1983 * * Each user of socket SHOULD hold a reference count. 1984 * * Each access point to socket (an hash table bucket, reference from a list, 1985 * running timer, skb in flight MUST hold a reference count. 1986 * * When reference count hits 0, it means it will never increase back. 1987 * * When reference count hits 0, it means that no references from 1988 * outside exist to this socket and current process on current CPU 1989 * is last user and may/should destroy this socket. 1990 * * sk_free is called from any context: process, BH, IRQ. When 1991 * it is called, socket has no references from outside -> sk_free 1992 * may release descendant resources allocated by the socket, but 1993 * to the time when it is called, socket is NOT referenced by any 1994 * hash tables, lists etc. 1995 * * Packets, delivered from outside (from network or from another process) 1996 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1997 * when they sit in queue. Otherwise, packets will leak to hole, when 1998 * socket is looked up by one cpu and unhasing is made by another CPU. 1999 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 2000 * (leak to backlog). Packet socket does all the processing inside 2001 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 2002 * use separate SMP lock, so that they are prone too. 2003 */ 2004 2005 /* Ungrab socket and destroy it, if it was the last reference. */ 2006 static inline void sock_put(struct sock *sk) 2007 { 2008 if (refcount_dec_and_test(&sk->sk_refcnt)) 2009 sk_free(sk); 2010 } 2011 /* Generic version of sock_put(), dealing with all sockets 2012 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 2013 */ 2014 void sock_gen_put(struct sock *sk); 2015 2016 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 2017 unsigned int trim_cap, bool refcounted); 2018 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 2019 const int nested) 2020 { 2021 return __sk_receive_skb(sk, skb, nested, 1, true); 2022 } 2023 2024 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 2025 { 2026 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2027 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2028 return; 2029 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2030 * other WRITE_ONCE() because socket lock might be not held. 2031 */ 2032 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 2033 } 2034 2035 #define NO_QUEUE_MAPPING USHRT_MAX 2036 2037 static inline void sk_tx_queue_clear(struct sock *sk) 2038 { 2039 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2040 * other WRITE_ONCE() because socket lock might be not held. 2041 */ 2042 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 2043 } 2044 2045 static inline int sk_tx_queue_get(const struct sock *sk) 2046 { 2047 if (sk) { 2048 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 2049 * and sk_tx_queue_set(). 2050 */ 2051 int val = READ_ONCE(sk->sk_tx_queue_mapping); 2052 2053 if (val != NO_QUEUE_MAPPING) 2054 return val; 2055 } 2056 return -1; 2057 } 2058 2059 static inline void __sk_rx_queue_set(struct sock *sk, 2060 const struct sk_buff *skb, 2061 bool force_set) 2062 { 2063 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2064 if (skb_rx_queue_recorded(skb)) { 2065 u16 rx_queue = skb_get_rx_queue(skb); 2066 2067 if (force_set || 2068 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2069 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2070 } 2071 #endif 2072 } 2073 2074 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2075 { 2076 __sk_rx_queue_set(sk, skb, true); 2077 } 2078 2079 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2080 { 2081 __sk_rx_queue_set(sk, skb, false); 2082 } 2083 2084 static inline void sk_rx_queue_clear(struct sock *sk) 2085 { 2086 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2087 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2088 #endif 2089 } 2090 2091 static inline int sk_rx_queue_get(const struct sock *sk) 2092 { 2093 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2094 if (sk) { 2095 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2096 2097 if (res != NO_QUEUE_MAPPING) 2098 return res; 2099 } 2100 #endif 2101 2102 return -1; 2103 } 2104 2105 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2106 { 2107 sk->sk_socket = sock; 2108 } 2109 2110 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2111 { 2112 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2113 return &rcu_dereference_raw(sk->sk_wq)->wait; 2114 } 2115 /* Detach socket from process context. 2116 * Announce socket dead, detach it from wait queue and inode. 2117 * Note that parent inode held reference count on this struct sock, 2118 * we do not release it in this function, because protocol 2119 * probably wants some additional cleanups or even continuing 2120 * to work with this socket (TCP). 2121 */ 2122 static inline void sock_orphan(struct sock *sk) 2123 { 2124 write_lock_bh(&sk->sk_callback_lock); 2125 sock_set_flag(sk, SOCK_DEAD); 2126 sk_set_socket(sk, NULL); 2127 sk->sk_wq = NULL; 2128 write_unlock_bh(&sk->sk_callback_lock); 2129 } 2130 2131 static inline void sock_graft(struct sock *sk, struct socket *parent) 2132 { 2133 WARN_ON(parent->sk); 2134 write_lock_bh(&sk->sk_callback_lock); 2135 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2136 parent->sk = sk; 2137 sk_set_socket(sk, parent); 2138 sk->sk_uid = SOCK_INODE(parent)->i_uid; 2139 security_sock_graft(sk, parent); 2140 write_unlock_bh(&sk->sk_callback_lock); 2141 } 2142 2143 kuid_t sock_i_uid(struct sock *sk); 2144 unsigned long __sock_i_ino(struct sock *sk); 2145 unsigned long sock_i_ino(struct sock *sk); 2146 2147 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2148 { 2149 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0); 2150 } 2151 2152 static inline u32 net_tx_rndhash(void) 2153 { 2154 u32 v = get_random_u32(); 2155 2156 return v ?: 1; 2157 } 2158 2159 static inline void sk_set_txhash(struct sock *sk) 2160 { 2161 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2162 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2163 } 2164 2165 static inline bool sk_rethink_txhash(struct sock *sk) 2166 { 2167 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2168 sk_set_txhash(sk); 2169 return true; 2170 } 2171 return false; 2172 } 2173 2174 static inline struct dst_entry * 2175 __sk_dst_get(const struct sock *sk) 2176 { 2177 return rcu_dereference_check(sk->sk_dst_cache, 2178 lockdep_sock_is_held(sk)); 2179 } 2180 2181 static inline struct dst_entry * 2182 sk_dst_get(const struct sock *sk) 2183 { 2184 struct dst_entry *dst; 2185 2186 rcu_read_lock(); 2187 dst = rcu_dereference(sk->sk_dst_cache); 2188 if (dst && !rcuref_get(&dst->__rcuref)) 2189 dst = NULL; 2190 rcu_read_unlock(); 2191 return dst; 2192 } 2193 2194 static inline void __dst_negative_advice(struct sock *sk) 2195 { 2196 struct dst_entry *dst = __sk_dst_get(sk); 2197 2198 if (dst && dst->ops->negative_advice) 2199 dst->ops->negative_advice(sk, dst); 2200 } 2201 2202 static inline void dst_negative_advice(struct sock *sk) 2203 { 2204 sk_rethink_txhash(sk); 2205 __dst_negative_advice(sk); 2206 } 2207 2208 static inline void 2209 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2210 { 2211 struct dst_entry *old_dst; 2212 2213 sk_tx_queue_clear(sk); 2214 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2215 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2216 lockdep_sock_is_held(sk)); 2217 rcu_assign_pointer(sk->sk_dst_cache, dst); 2218 dst_release(old_dst); 2219 } 2220 2221 static inline void 2222 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2223 { 2224 struct dst_entry *old_dst; 2225 2226 sk_tx_queue_clear(sk); 2227 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2228 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst))); 2229 dst_release(old_dst); 2230 } 2231 2232 static inline void 2233 __sk_dst_reset(struct sock *sk) 2234 { 2235 __sk_dst_set(sk, NULL); 2236 } 2237 2238 static inline void 2239 sk_dst_reset(struct sock *sk) 2240 { 2241 sk_dst_set(sk, NULL); 2242 } 2243 2244 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2245 2246 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2247 2248 static inline void sk_dst_confirm(struct sock *sk) 2249 { 2250 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2251 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2252 } 2253 2254 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2255 { 2256 if (skb_get_dst_pending_confirm(skb)) { 2257 struct sock *sk = skb->sk; 2258 2259 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2260 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2261 neigh_confirm(n); 2262 } 2263 } 2264 2265 bool sk_mc_loop(struct sock *sk); 2266 2267 static inline bool sk_can_gso(const struct sock *sk) 2268 { 2269 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2270 } 2271 2272 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2273 2274 static inline void sk_gso_disable(struct sock *sk) 2275 { 2276 sk->sk_gso_disabled = 1; 2277 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2278 } 2279 2280 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2281 struct iov_iter *from, char *to, 2282 int copy, int offset) 2283 { 2284 if (skb->ip_summed == CHECKSUM_NONE) { 2285 __wsum csum = 0; 2286 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2287 return -EFAULT; 2288 skb->csum = csum_block_add(skb->csum, csum, offset); 2289 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2290 if (!copy_from_iter_full_nocache(to, copy, from)) 2291 return -EFAULT; 2292 } else if (!copy_from_iter_full(to, copy, from)) 2293 return -EFAULT; 2294 2295 return 0; 2296 } 2297 2298 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2299 struct iov_iter *from, int copy) 2300 { 2301 int err, offset = skb->len; 2302 2303 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2304 copy, offset); 2305 if (err) 2306 __skb_trim(skb, offset); 2307 2308 return err; 2309 } 2310 2311 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2312 struct sk_buff *skb, 2313 struct page *page, 2314 int off, int copy) 2315 { 2316 int err; 2317 2318 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2319 copy, skb->len); 2320 if (err) 2321 return err; 2322 2323 skb_len_add(skb, copy); 2324 sk_wmem_queued_add(sk, copy); 2325 sk_mem_charge(sk, copy); 2326 return 0; 2327 } 2328 2329 /** 2330 * sk_wmem_alloc_get - returns write allocations 2331 * @sk: socket 2332 * 2333 * Return: sk_wmem_alloc minus initial offset of one 2334 */ 2335 static inline int sk_wmem_alloc_get(const struct sock *sk) 2336 { 2337 return refcount_read(&sk->sk_wmem_alloc) - 1; 2338 } 2339 2340 /** 2341 * sk_rmem_alloc_get - returns read allocations 2342 * @sk: socket 2343 * 2344 * Return: sk_rmem_alloc 2345 */ 2346 static inline int sk_rmem_alloc_get(const struct sock *sk) 2347 { 2348 return atomic_read(&sk->sk_rmem_alloc); 2349 } 2350 2351 /** 2352 * sk_has_allocations - check if allocations are outstanding 2353 * @sk: socket 2354 * 2355 * Return: true if socket has write or read allocations 2356 */ 2357 static inline bool sk_has_allocations(const struct sock *sk) 2358 { 2359 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2360 } 2361 2362 /** 2363 * skwq_has_sleeper - check if there are any waiting processes 2364 * @wq: struct socket_wq 2365 * 2366 * Return: true if socket_wq has waiting processes 2367 * 2368 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2369 * barrier call. They were added due to the race found within the tcp code. 2370 * 2371 * Consider following tcp code paths:: 2372 * 2373 * CPU1 CPU2 2374 * sys_select receive packet 2375 * ... ... 2376 * __add_wait_queue update tp->rcv_nxt 2377 * ... ... 2378 * tp->rcv_nxt check sock_def_readable 2379 * ... { 2380 * schedule rcu_read_lock(); 2381 * wq = rcu_dereference(sk->sk_wq); 2382 * if (wq && waitqueue_active(&wq->wait)) 2383 * wake_up_interruptible(&wq->wait) 2384 * ... 2385 * } 2386 * 2387 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2388 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2389 * could then endup calling schedule and sleep forever if there are no more 2390 * data on the socket. 2391 * 2392 */ 2393 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2394 { 2395 return wq && wq_has_sleeper(&wq->wait); 2396 } 2397 2398 /** 2399 * sock_poll_wait - place memory barrier behind the poll_wait call. 2400 * @filp: file 2401 * @sock: socket to wait on 2402 * @p: poll_table 2403 * 2404 * See the comments in the wq_has_sleeper function. 2405 */ 2406 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2407 poll_table *p) 2408 { 2409 if (!poll_does_not_wait(p)) { 2410 poll_wait(filp, &sock->wq.wait, p); 2411 /* We need to be sure we are in sync with the 2412 * socket flags modification. 2413 * 2414 * This memory barrier is paired in the wq_has_sleeper. 2415 */ 2416 smp_mb(); 2417 } 2418 } 2419 2420 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2421 { 2422 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2423 u32 txhash = READ_ONCE(sk->sk_txhash); 2424 2425 if (txhash) { 2426 skb->l4_hash = 1; 2427 skb->hash = txhash; 2428 } 2429 } 2430 2431 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2432 2433 /* 2434 * Queue a received datagram if it will fit. Stream and sequenced 2435 * protocols can't normally use this as they need to fit buffers in 2436 * and play with them. 2437 * 2438 * Inlined as it's very short and called for pretty much every 2439 * packet ever received. 2440 */ 2441 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2442 { 2443 skb_orphan(skb); 2444 skb->sk = sk; 2445 skb->destructor = sock_rfree; 2446 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2447 sk_mem_charge(sk, skb->truesize); 2448 } 2449 2450 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2451 { 2452 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2453 skb_orphan(skb); 2454 skb->destructor = sock_efree; 2455 skb->sk = sk; 2456 return true; 2457 } 2458 return false; 2459 } 2460 2461 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2462 { 2463 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2464 if (skb) { 2465 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2466 skb_set_owner_r(skb, sk); 2467 return skb; 2468 } 2469 __kfree_skb(skb); 2470 } 2471 return NULL; 2472 } 2473 2474 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2475 { 2476 if (skb->destructor != sock_wfree) { 2477 skb_orphan(skb); 2478 return; 2479 } 2480 skb->slow_gro = 1; 2481 } 2482 2483 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2484 unsigned long expires); 2485 2486 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2487 2488 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2489 2490 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2491 struct sk_buff *skb, unsigned int flags, 2492 void (*destructor)(struct sock *sk, 2493 struct sk_buff *skb)); 2494 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2495 2496 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2497 enum skb_drop_reason *reason); 2498 2499 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2500 { 2501 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2502 } 2503 2504 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2505 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2506 2507 /* 2508 * Recover an error report and clear atomically 2509 */ 2510 2511 static inline int sock_error(struct sock *sk) 2512 { 2513 int err; 2514 2515 /* Avoid an atomic operation for the common case. 2516 * This is racy since another cpu/thread can change sk_err under us. 2517 */ 2518 if (likely(data_race(!sk->sk_err))) 2519 return 0; 2520 2521 err = xchg(&sk->sk_err, 0); 2522 return -err; 2523 } 2524 2525 void sk_error_report(struct sock *sk); 2526 2527 static inline unsigned long sock_wspace(struct sock *sk) 2528 { 2529 int amt = 0; 2530 2531 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2532 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2533 if (amt < 0) 2534 amt = 0; 2535 } 2536 return amt; 2537 } 2538 2539 /* Note: 2540 * We use sk->sk_wq_raw, from contexts knowing this 2541 * pointer is not NULL and cannot disappear/change. 2542 */ 2543 static inline void sk_set_bit(int nr, struct sock *sk) 2544 { 2545 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2546 !sock_flag(sk, SOCK_FASYNC)) 2547 return; 2548 2549 set_bit(nr, &sk->sk_wq_raw->flags); 2550 } 2551 2552 static inline void sk_clear_bit(int nr, struct sock *sk) 2553 { 2554 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2555 !sock_flag(sk, SOCK_FASYNC)) 2556 return; 2557 2558 clear_bit(nr, &sk->sk_wq_raw->flags); 2559 } 2560 2561 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2562 { 2563 if (sock_flag(sk, SOCK_FASYNC)) { 2564 rcu_read_lock(); 2565 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2566 rcu_read_unlock(); 2567 } 2568 } 2569 2570 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2571 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2572 * Note: for send buffers, TCP works better if we can build two skbs at 2573 * minimum. 2574 */ 2575 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2576 2577 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2578 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2579 2580 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2581 { 2582 u32 val; 2583 2584 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2585 return; 2586 2587 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2588 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2589 2590 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2591 } 2592 2593 /** 2594 * sk_page_frag - return an appropriate page_frag 2595 * @sk: socket 2596 * 2597 * Use the per task page_frag instead of the per socket one for 2598 * optimization when we know that we're in process context and own 2599 * everything that's associated with %current. 2600 * 2601 * Both direct reclaim and page faults can nest inside other 2602 * socket operations and end up recursing into sk_page_frag() 2603 * while it's already in use: explicitly avoid task page_frag 2604 * when users disable sk_use_task_frag. 2605 * 2606 * Return: a per task page_frag if context allows that, 2607 * otherwise a per socket one. 2608 */ 2609 static inline struct page_frag *sk_page_frag(struct sock *sk) 2610 { 2611 if (sk->sk_use_task_frag) 2612 return ¤t->task_frag; 2613 2614 return &sk->sk_frag; 2615 } 2616 2617 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2618 2619 /* 2620 * Default write policy as shown to user space via poll/select/SIGIO 2621 */ 2622 static inline bool sock_writeable(const struct sock *sk) 2623 { 2624 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2625 } 2626 2627 static inline gfp_t gfp_any(void) 2628 { 2629 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2630 } 2631 2632 static inline gfp_t gfp_memcg_charge(void) 2633 { 2634 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2635 } 2636 2637 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2638 { 2639 return noblock ? 0 : sk->sk_rcvtimeo; 2640 } 2641 2642 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2643 { 2644 return noblock ? 0 : sk->sk_sndtimeo; 2645 } 2646 2647 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2648 { 2649 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2650 2651 return v ?: 1; 2652 } 2653 2654 /* Alas, with timeout socket operations are not restartable. 2655 * Compare this to poll(). 2656 */ 2657 static inline int sock_intr_errno(long timeo) 2658 { 2659 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2660 } 2661 2662 struct sock_skb_cb { 2663 u32 dropcount; 2664 }; 2665 2666 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2667 * using skb->cb[] would keep using it directly and utilize its 2668 * alignement guarantee. 2669 */ 2670 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \ 2671 sizeof(struct sock_skb_cb))) 2672 2673 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2674 SOCK_SKB_CB_OFFSET)) 2675 2676 #define sock_skb_cb_check_size(size) \ 2677 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2678 2679 static inline void 2680 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2681 { 2682 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2683 atomic_read(&sk->sk_drops) : 0; 2684 } 2685 2686 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2687 { 2688 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2689 2690 atomic_add(segs, &sk->sk_drops); 2691 } 2692 2693 static inline ktime_t sock_read_timestamp(struct sock *sk) 2694 { 2695 #if BITS_PER_LONG==32 2696 unsigned int seq; 2697 ktime_t kt; 2698 2699 do { 2700 seq = read_seqbegin(&sk->sk_stamp_seq); 2701 kt = sk->sk_stamp; 2702 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2703 2704 return kt; 2705 #else 2706 return READ_ONCE(sk->sk_stamp); 2707 #endif 2708 } 2709 2710 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2711 { 2712 #if BITS_PER_LONG==32 2713 write_seqlock(&sk->sk_stamp_seq); 2714 sk->sk_stamp = kt; 2715 write_sequnlock(&sk->sk_stamp_seq); 2716 #else 2717 WRITE_ONCE(sk->sk_stamp, kt); 2718 #endif 2719 } 2720 2721 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2722 struct sk_buff *skb); 2723 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2724 struct sk_buff *skb); 2725 2726 static inline void 2727 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2728 { 2729 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2730 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2731 ktime_t kt = skb->tstamp; 2732 /* 2733 * generate control messages if 2734 * - receive time stamping in software requested 2735 * - software time stamp available and wanted 2736 * - hardware time stamps available and wanted 2737 */ 2738 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2739 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2740 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2741 (hwtstamps->hwtstamp && 2742 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2743 __sock_recv_timestamp(msg, sk, skb); 2744 else 2745 sock_write_timestamp(sk, kt); 2746 2747 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2748 __sock_recv_wifi_status(msg, sk, skb); 2749 } 2750 2751 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2752 struct sk_buff *skb); 2753 2754 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2755 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2756 struct sk_buff *skb) 2757 { 2758 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2759 (1UL << SOCK_RCVTSTAMP) | \ 2760 (1UL << SOCK_RCVMARK)) 2761 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2762 SOF_TIMESTAMPING_RAW_HARDWARE) 2763 2764 if (sk->sk_flags & FLAGS_RECV_CMSGS || 2765 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY) 2766 __sock_recv_cmsgs(msg, sk, skb); 2767 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2768 sock_write_timestamp(sk, skb->tstamp); 2769 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2770 sock_write_timestamp(sk, 0); 2771 } 2772 2773 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags); 2774 2775 /** 2776 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2777 * @sk: socket sending this packet 2778 * @tsflags: timestamping flags to use 2779 * @tx_flags: completed with instructions for time stamping 2780 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2781 * 2782 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2783 */ 2784 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2785 __u8 *tx_flags, __u32 *tskey) 2786 { 2787 if (unlikely(tsflags)) { 2788 __sock_tx_timestamp(tsflags, tx_flags); 2789 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2790 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) 2791 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2792 } 2793 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS))) 2794 *tx_flags |= SKBTX_WIFI_STATUS; 2795 } 2796 2797 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags, 2798 __u8 *tx_flags) 2799 { 2800 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL); 2801 } 2802 2803 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags) 2804 { 2805 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags, 2806 &skb_shinfo(skb)->tskey); 2807 } 2808 2809 static inline bool sk_is_inet(const struct sock *sk) 2810 { 2811 int family = READ_ONCE(sk->sk_family); 2812 2813 return family == AF_INET || family == AF_INET6; 2814 } 2815 2816 static inline bool sk_is_tcp(const struct sock *sk) 2817 { 2818 return sk_is_inet(sk) && 2819 sk->sk_type == SOCK_STREAM && 2820 sk->sk_protocol == IPPROTO_TCP; 2821 } 2822 2823 static inline bool sk_is_udp(const struct sock *sk) 2824 { 2825 return sk_is_inet(sk) && 2826 sk->sk_type == SOCK_DGRAM && 2827 sk->sk_protocol == IPPROTO_UDP; 2828 } 2829 2830 static inline bool sk_is_stream_unix(const struct sock *sk) 2831 { 2832 return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM; 2833 } 2834 2835 static inline bool sk_is_vsock(const struct sock *sk) 2836 { 2837 return sk->sk_family == AF_VSOCK; 2838 } 2839 2840 /** 2841 * sk_eat_skb - Release a skb if it is no longer needed 2842 * @sk: socket to eat this skb from 2843 * @skb: socket buffer to eat 2844 * 2845 * This routine must be called with interrupts disabled or with the socket 2846 * locked so that the sk_buff queue operation is ok. 2847 */ 2848 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2849 { 2850 __skb_unlink(skb, &sk->sk_receive_queue); 2851 __kfree_skb(skb); 2852 } 2853 2854 static inline bool 2855 skb_sk_is_prefetched(struct sk_buff *skb) 2856 { 2857 #ifdef CONFIG_INET 2858 return skb->destructor == sock_pfree; 2859 #else 2860 return false; 2861 #endif /* CONFIG_INET */ 2862 } 2863 2864 /* This helper checks if a socket is a full socket, 2865 * ie _not_ a timewait or request socket. 2866 */ 2867 static inline bool sk_fullsock(const struct sock *sk) 2868 { 2869 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2870 } 2871 2872 static inline bool 2873 sk_is_refcounted(struct sock *sk) 2874 { 2875 /* Only full sockets have sk->sk_flags. */ 2876 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2877 } 2878 2879 /** 2880 * skb_steal_sock - steal a socket from an sk_buff 2881 * @skb: sk_buff to steal the socket from 2882 * @refcounted: is set to true if the socket is reference-counted 2883 * @prefetched: is set to true if the socket was assigned from bpf 2884 */ 2885 static inline struct sock * 2886 skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched) 2887 { 2888 if (skb->sk) { 2889 struct sock *sk = skb->sk; 2890 2891 *refcounted = true; 2892 *prefetched = skb_sk_is_prefetched(skb); 2893 if (*prefetched) 2894 *refcounted = sk_is_refcounted(sk); 2895 skb->destructor = NULL; 2896 skb->sk = NULL; 2897 return sk; 2898 } 2899 *prefetched = false; 2900 *refcounted = false; 2901 return NULL; 2902 } 2903 2904 /* Checks if this SKB belongs to an HW offloaded socket 2905 * and whether any SW fallbacks are required based on dev. 2906 * Check decrypted mark in case skb_orphan() cleared socket. 2907 */ 2908 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2909 struct net_device *dev) 2910 { 2911 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2912 struct sock *sk = skb->sk; 2913 2914 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2915 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2916 #ifdef CONFIG_TLS_DEVICE 2917 } else if (unlikely(skb->decrypted)) { 2918 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2919 kfree_skb(skb); 2920 skb = NULL; 2921 #endif 2922 } 2923 #endif 2924 2925 return skb; 2926 } 2927 2928 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2929 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2930 */ 2931 static inline bool sk_listener(const struct sock *sk) 2932 { 2933 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2934 } 2935 2936 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2937 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2938 int type); 2939 2940 bool sk_ns_capable(const struct sock *sk, 2941 struct user_namespace *user_ns, int cap); 2942 bool sk_capable(const struct sock *sk, int cap); 2943 bool sk_net_capable(const struct sock *sk, int cap); 2944 2945 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2946 2947 /* Take into consideration the size of the struct sk_buff overhead in the 2948 * determination of these values, since that is non-constant across 2949 * platforms. This makes socket queueing behavior and performance 2950 * not depend upon such differences. 2951 */ 2952 #define _SK_MEM_PACKETS 256 2953 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2954 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2955 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2956 2957 extern __u32 sysctl_wmem_max; 2958 extern __u32 sysctl_rmem_max; 2959 2960 extern int sysctl_tstamp_allow_data; 2961 extern int sysctl_optmem_max; 2962 2963 extern __u32 sysctl_wmem_default; 2964 extern __u32 sysctl_rmem_default; 2965 2966 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2967 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2968 2969 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2970 { 2971 /* Does this proto have per netns sysctl_wmem ? */ 2972 if (proto->sysctl_wmem_offset) 2973 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2974 2975 return READ_ONCE(*proto->sysctl_wmem); 2976 } 2977 2978 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2979 { 2980 /* Does this proto have per netns sysctl_rmem ? */ 2981 if (proto->sysctl_rmem_offset) 2982 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2983 2984 return READ_ONCE(*proto->sysctl_rmem); 2985 } 2986 2987 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2988 * Some wifi drivers need to tweak it to get more chunks. 2989 * They can use this helper from their ndo_start_xmit() 2990 */ 2991 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2992 { 2993 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2994 return; 2995 WRITE_ONCE(sk->sk_pacing_shift, val); 2996 } 2997 2998 /* if a socket is bound to a device, check that the given device 2999 * index is either the same or that the socket is bound to an L3 3000 * master device and the given device index is also enslaved to 3001 * that L3 master 3002 */ 3003 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 3004 { 3005 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 3006 int mdif; 3007 3008 if (!bound_dev_if || bound_dev_if == dif) 3009 return true; 3010 3011 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 3012 if (mdif && mdif == bound_dev_if) 3013 return true; 3014 3015 return false; 3016 } 3017 3018 void sock_def_readable(struct sock *sk); 3019 3020 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 3021 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 3022 int sock_set_timestamping(struct sock *sk, int optname, 3023 struct so_timestamping timestamping); 3024 3025 void sock_enable_timestamps(struct sock *sk); 3026 void sock_no_linger(struct sock *sk); 3027 void sock_set_keepalive(struct sock *sk); 3028 void sock_set_priority(struct sock *sk, u32 priority); 3029 void sock_set_rcvbuf(struct sock *sk, int val); 3030 void sock_set_mark(struct sock *sk, u32 val); 3031 void sock_set_reuseaddr(struct sock *sk); 3032 void sock_set_reuseport(struct sock *sk); 3033 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3034 3035 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 3036 3037 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3038 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3039 sockptr_t optval, int optlen, bool old_timeval); 3040 3041 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3042 void __user *arg, void *karg, size_t size); 3043 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3044 static inline bool sk_is_readable(struct sock *sk) 3045 { 3046 if (sk->sk_prot->sock_is_readable) 3047 return sk->sk_prot->sock_is_readable(sk); 3048 return false; 3049 } 3050 #endif /* _SOCK_H */ 3051