1 /* SPDX-License-Identifier: GPL-2.0-or-later */ 2 /* 3 * INET An implementation of the TCP/IP protocol suite for the LINUX 4 * operating system. INET is implemented using the BSD Socket 5 * interface as the means of communication with the user level. 6 * 7 * Definitions for the AF_INET socket handler. 8 * 9 * Version: @(#)sock.h 1.0.4 05/13/93 10 * 11 * Authors: Ross Biro 12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 13 * Corey Minyard <wf-rch!minyard@relay.EU.net> 14 * Florian La Roche <flla@stud.uni-sb.de> 15 * 16 * Fixes: 17 * Alan Cox : Volatiles in skbuff pointers. See 18 * skbuff comments. May be overdone, 19 * better to prove they can be removed 20 * than the reverse. 21 * Alan Cox : Added a zapped field for tcp to note 22 * a socket is reset and must stay shut up 23 * Alan Cox : New fields for options 24 * Pauline Middelink : identd support 25 * Alan Cox : Eliminate low level recv/recvfrom 26 * David S. Miller : New socket lookup architecture. 27 * Steve Whitehouse: Default routines for sock_ops 28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made 29 * protinfo be just a void pointer, as the 30 * protocol specific parts were moved to 31 * respective headers and ipv4/v6, etc now 32 * use private slabcaches for its socks 33 * Pedro Hortas : New flags field for socket options 34 */ 35 #ifndef _SOCK_H 36 #define _SOCK_H 37 38 #include <linux/hardirq.h> 39 #include <linux/kernel.h> 40 #include <linux/list.h> 41 #include <linux/list_nulls.h> 42 #include <linux/timer.h> 43 #include <linux/cache.h> 44 #include <linux/bitops.h> 45 #include <linux/lockdep.h> 46 #include <linux/netdevice.h> 47 #include <linux/skbuff.h> /* struct sk_buff */ 48 #include <linux/mm.h> 49 #include <linux/security.h> 50 #include <linux/slab.h> 51 #include <linux/uaccess.h> 52 #include <linux/page_counter.h> 53 #include <linux/memcontrol.h> 54 #include <linux/static_key.h> 55 #include <linux/sched.h> 56 #include <linux/wait.h> 57 #include <linux/cgroup-defs.h> 58 #include <linux/rbtree.h> 59 #include <linux/rculist_nulls.h> 60 #include <linux/poll.h> 61 #include <linux/sockptr.h> 62 #include <linux/indirect_call_wrapper.h> 63 #include <linux/atomic.h> 64 #include <linux/refcount.h> 65 #include <linux/llist.h> 66 #include <net/dst.h> 67 #include <net/checksum.h> 68 #include <net/tcp_states.h> 69 #include <linux/net_tstamp.h> 70 #include <net/l3mdev.h> 71 #include <uapi/linux/socket.h> 72 73 /* 74 * This structure really needs to be cleaned up. 75 * Most of it is for TCP, and not used by any of 76 * the other protocols. 77 */ 78 79 /* This is the per-socket lock. The spinlock provides a synchronization 80 * between user contexts and software interrupt processing, whereas the 81 * mini-semaphore synchronizes multiple users amongst themselves. 82 */ 83 typedef struct { 84 spinlock_t slock; 85 int owned; 86 wait_queue_head_t wq; 87 /* 88 * We express the mutex-alike socket_lock semantics 89 * to the lock validator by explicitly managing 90 * the slock as a lock variant (in addition to 91 * the slock itself): 92 */ 93 #ifdef CONFIG_DEBUG_LOCK_ALLOC 94 struct lockdep_map dep_map; 95 #endif 96 } socket_lock_t; 97 98 struct sock; 99 struct proto; 100 struct net; 101 102 typedef __u32 __bitwise __portpair; 103 typedef __u64 __bitwise __addrpair; 104 105 /** 106 * struct sock_common - minimal network layer representation of sockets 107 * @skc_daddr: Foreign IPv4 addr 108 * @skc_rcv_saddr: Bound local IPv4 addr 109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr 110 * @skc_hash: hash value used with various protocol lookup tables 111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables 112 * @skc_dport: placeholder for inet_dport/tw_dport 113 * @skc_num: placeholder for inet_num/tw_num 114 * @skc_portpair: __u32 union of @skc_dport & @skc_num 115 * @skc_family: network address family 116 * @skc_state: Connection state 117 * @skc_reuse: %SO_REUSEADDR setting 118 * @skc_reuseport: %SO_REUSEPORT setting 119 * @skc_ipv6only: socket is IPV6 only 120 * @skc_net_refcnt: socket is using net ref counting 121 * @skc_bound_dev_if: bound device index if != 0 122 * @skc_bind_node: bind hash linkage for various protocol lookup tables 123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 124 * @skc_prot: protocol handlers inside a network family 125 * @skc_net: reference to the network namespace of this socket 126 * @skc_v6_daddr: IPV6 destination address 127 * @skc_v6_rcv_saddr: IPV6 source address 128 * @skc_cookie: socket's cookie value 129 * @skc_node: main hash linkage for various protocol lookup tables 130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 131 * @skc_tx_queue_mapping: tx queue number for this connection 132 * @skc_rx_queue_mapping: rx queue number for this connection 133 * @skc_flags: place holder for sk_flags 134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 136 * @skc_listener: connection request listener socket (aka rsk_listener) 137 * [union with @skc_flags] 138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 139 * [union with @skc_flags] 140 * @skc_incoming_cpu: record/match cpu processing incoming packets 141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 142 * [union with @skc_incoming_cpu] 143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 144 * [union with @skc_incoming_cpu] 145 * @skc_refcnt: reference count 146 * 147 * This is the minimal network layer representation of sockets, the header 148 * for struct sock and struct inet_timewait_sock. 149 */ 150 struct sock_common { 151 union { 152 __addrpair skc_addrpair; 153 struct { 154 __be32 skc_daddr; 155 __be32 skc_rcv_saddr; 156 }; 157 }; 158 union { 159 unsigned int skc_hash; 160 __u16 skc_u16hashes[2]; 161 }; 162 /* skc_dport && skc_num must be grouped as well */ 163 union { 164 __portpair skc_portpair; 165 struct { 166 __be16 skc_dport; 167 __u16 skc_num; 168 }; 169 }; 170 171 unsigned short skc_family; 172 volatile unsigned char skc_state; 173 unsigned char skc_reuse:4; 174 unsigned char skc_reuseport:1; 175 unsigned char skc_ipv6only:1; 176 unsigned char skc_net_refcnt:1; 177 int skc_bound_dev_if; 178 union { 179 struct hlist_node skc_bind_node; 180 struct hlist_node skc_portaddr_node; 181 }; 182 struct proto *skc_prot; 183 possible_net_t skc_net; 184 185 #if IS_ENABLED(CONFIG_IPV6) 186 struct in6_addr skc_v6_daddr; 187 struct in6_addr skc_v6_rcv_saddr; 188 #endif 189 190 atomic64_t skc_cookie; 191 192 /* following fields are padding to force 193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 194 * assuming IPV6 is enabled. We use this padding differently 195 * for different kind of 'sockets' 196 */ 197 union { 198 unsigned long skc_flags; 199 struct sock *skc_listener; /* request_sock */ 200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 201 }; 202 /* 203 * fields between dontcopy_begin/dontcopy_end 204 * are not copied in sock_copy() 205 */ 206 /* private: */ 207 int skc_dontcopy_begin[0]; 208 /* public: */ 209 union { 210 struct hlist_node skc_node; 211 struct hlist_nulls_node skc_nulls_node; 212 }; 213 unsigned short skc_tx_queue_mapping; 214 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 215 unsigned short skc_rx_queue_mapping; 216 #endif 217 union { 218 int skc_incoming_cpu; 219 u32 skc_rcv_wnd; 220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 221 }; 222 223 refcount_t skc_refcnt; 224 /* private: */ 225 int skc_dontcopy_end[0]; 226 union { 227 u32 skc_rxhash; 228 u32 skc_window_clamp; 229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 230 }; 231 /* public: */ 232 }; 233 234 struct bpf_local_storage; 235 struct sk_filter; 236 237 /** 238 * struct sock - network layer representation of sockets 239 * @__sk_common: shared layout with inet_timewait_sock 240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 242 * @sk_lock: synchronizer 243 * @sk_kern_sock: True if sock is using kernel lock classes 244 * @sk_rcvbuf: size of receive buffer in bytes 245 * @sk_wq: sock wait queue and async head 246 * @sk_rx_dst: receive input route used by early demux 247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 249 * @sk_dst_cache: destination cache 250 * @sk_dst_pending_confirm: need to confirm neighbour 251 * @sk_policy: flow policy 252 * @sk_receive_queue: incoming packets 253 * @sk_wmem_alloc: transmit queue bytes committed 254 * @sk_tsq_flags: TCP Small Queues flags 255 * @sk_write_queue: Packet sending queue 256 * @sk_omem_alloc: "o" is "option" or "other" 257 * @sk_wmem_queued: persistent queue size 258 * @sk_forward_alloc: space allocated forward 259 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 260 * @sk_napi_id: id of the last napi context to receive data for sk 261 * @sk_ll_usec: usecs to busypoll when there is no data 262 * @sk_allocation: allocation mode 263 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 264 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 265 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 266 * @sk_sndbuf: size of send buffer in bytes 267 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 268 * @sk_no_check_rx: allow zero checksum in RX packets 269 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 270 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 271 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 272 * @sk_gso_max_size: Maximum GSO segment size to build 273 * @sk_gso_max_segs: Maximum number of GSO segments 274 * @sk_pacing_shift: scaling factor for TCP Small Queues 275 * @sk_lingertime: %SO_LINGER l_linger setting 276 * @sk_backlog: always used with the per-socket spinlock held 277 * @sk_callback_lock: used with the callbacks in the end of this struct 278 * @sk_error_queue: rarely used 279 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 280 * IPV6_ADDRFORM for instance) 281 * @sk_err: last error 282 * @sk_err_soft: errors that don't cause failure but are the cause of a 283 * persistent failure not just 'timed out' 284 * @sk_drops: raw/udp drops counter 285 * @sk_ack_backlog: current listen backlog 286 * @sk_max_ack_backlog: listen backlog set in listen() 287 * @sk_uid: user id of owner 288 * @sk_ino: inode number (zero if orphaned) 289 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 290 * @sk_busy_poll_budget: napi processing budget when busypolling 291 * @sk_priority: %SO_PRIORITY setting 292 * @sk_type: socket type (%SOCK_STREAM, etc) 293 * @sk_protocol: which protocol this socket belongs in this network family 294 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 295 * @sk_peer_pid: &struct pid for this socket's peer 296 * @sk_peer_cred: %SO_PEERCRED setting 297 * @sk_rcvlowat: %SO_RCVLOWAT setting 298 * @sk_rcvtimeo: %SO_RCVTIMEO setting 299 * @sk_sndtimeo: %SO_SNDTIMEO setting 300 * @sk_txhash: computed flow hash for use on transmit 301 * @sk_txrehash: enable TX hash rethink 302 * @sk_filter: socket filtering instructions 303 * @sk_timer: sock cleanup timer 304 * @sk_stamp: time stamp of last packet received 305 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 306 * @sk_tsflags: SO_TIMESTAMPING flags 307 * @sk_bpf_cb_flags: used in bpf_setsockopt() 308 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 309 * Sockets that can be used under memory reclaim should 310 * set this to false. 311 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 312 * for timestamping 313 * @sk_tskey: counter to disambiguate concurrent tstamp requests 314 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 315 * @sk_socket: Identd and reporting IO signals 316 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 317 * @sk_frag: cached page frag 318 * @sk_peek_off: current peek_offset value 319 * @sk_send_head: front of stuff to transmit 320 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 321 * @sk_security: used by security modules 322 * @sk_mark: generic packet mark 323 * @sk_cgrp_data: cgroup data for this cgroup 324 * @sk_memcg: this socket's memory cgroup association 325 * @sk_write_pending: a write to stream socket waits to start 326 * @sk_disconnects: number of disconnect operations performed on this sock 327 * @sk_state_change: callback to indicate change in the state of the sock 328 * @sk_data_ready: callback to indicate there is data to be processed 329 * @sk_write_space: callback to indicate there is bf sending space available 330 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 331 * @sk_backlog_rcv: callback to process the backlog 332 * @sk_validate_xmit_skb: ptr to an optional validate function 333 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 334 * @sk_reuseport_cb: reuseport group container 335 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 336 * @sk_rcu: used during RCU grace period 337 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 338 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 339 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 340 * @sk_txtime_unused: unused txtime flags 341 * @sk_scm_recv_flags: all flags used by scm_recv() 342 * @sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS 343 * @sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY 344 * @sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD 345 * @sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS 346 * @sk_scm_unused: unused flags for scm_recv() 347 * @ns_tracker: tracker for netns reference 348 * @sk_user_frags: xarray of pages the user is holding a reference on. 349 * @sk_owner: reference to the real owner of the socket that calls 350 * sock_lock_init_class_and_name(). 351 */ 352 struct sock { 353 /* 354 * Now struct inet_timewait_sock also uses sock_common, so please just 355 * don't add nothing before this first member (__sk_common) --acme 356 */ 357 struct sock_common __sk_common; 358 #define sk_node __sk_common.skc_node 359 #define sk_nulls_node __sk_common.skc_nulls_node 360 #define sk_refcnt __sk_common.skc_refcnt 361 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 362 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 363 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 364 #endif 365 366 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 367 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 368 #define sk_hash __sk_common.skc_hash 369 #define sk_portpair __sk_common.skc_portpair 370 #define sk_num __sk_common.skc_num 371 #define sk_dport __sk_common.skc_dport 372 #define sk_addrpair __sk_common.skc_addrpair 373 #define sk_daddr __sk_common.skc_daddr 374 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 375 #define sk_family __sk_common.skc_family 376 #define sk_state __sk_common.skc_state 377 #define sk_reuse __sk_common.skc_reuse 378 #define sk_reuseport __sk_common.skc_reuseport 379 #define sk_ipv6only __sk_common.skc_ipv6only 380 #define sk_net_refcnt __sk_common.skc_net_refcnt 381 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 382 #define sk_bind_node __sk_common.skc_bind_node 383 #define sk_prot __sk_common.skc_prot 384 #define sk_net __sk_common.skc_net 385 #define sk_v6_daddr __sk_common.skc_v6_daddr 386 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 387 #define sk_cookie __sk_common.skc_cookie 388 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 389 #define sk_flags __sk_common.skc_flags 390 #define sk_rxhash __sk_common.skc_rxhash 391 392 __cacheline_group_begin(sock_write_rx); 393 394 atomic_t sk_drops; 395 __s32 sk_peek_off; 396 struct sk_buff_head sk_error_queue; 397 struct sk_buff_head sk_receive_queue; 398 /* 399 * The backlog queue is special, it is always used with 400 * the per-socket spinlock held and requires low latency 401 * access. Therefore we special case it's implementation. 402 * Note : rmem_alloc is in this structure to fill a hole 403 * on 64bit arches, not because its logically part of 404 * backlog. 405 */ 406 struct { 407 atomic_t rmem_alloc; 408 int len; 409 struct sk_buff *head; 410 struct sk_buff *tail; 411 } sk_backlog; 412 #define sk_rmem_alloc sk_backlog.rmem_alloc 413 414 __cacheline_group_end(sock_write_rx); 415 416 __cacheline_group_begin(sock_read_rx); 417 /* early demux fields */ 418 struct dst_entry __rcu *sk_rx_dst; 419 int sk_rx_dst_ifindex; 420 u32 sk_rx_dst_cookie; 421 422 #ifdef CONFIG_NET_RX_BUSY_POLL 423 unsigned int sk_ll_usec; 424 unsigned int sk_napi_id; 425 u16 sk_busy_poll_budget; 426 u8 sk_prefer_busy_poll; 427 #endif 428 u8 sk_userlocks; 429 int sk_rcvbuf; 430 431 struct sk_filter __rcu *sk_filter; 432 union { 433 struct socket_wq __rcu *sk_wq; 434 /* private: */ 435 struct socket_wq *sk_wq_raw; 436 /* public: */ 437 }; 438 439 void (*sk_data_ready)(struct sock *sk); 440 long sk_rcvtimeo; 441 int sk_rcvlowat; 442 __cacheline_group_end(sock_read_rx); 443 444 __cacheline_group_begin(sock_read_rxtx); 445 int sk_err; 446 struct socket *sk_socket; 447 struct mem_cgroup *sk_memcg; 448 #ifdef CONFIG_XFRM 449 struct xfrm_policy __rcu *sk_policy[2]; 450 #endif 451 __cacheline_group_end(sock_read_rxtx); 452 453 __cacheline_group_begin(sock_write_rxtx); 454 socket_lock_t sk_lock; 455 u32 sk_reserved_mem; 456 int sk_forward_alloc; 457 u32 sk_tsflags; 458 __cacheline_group_end(sock_write_rxtx); 459 460 __cacheline_group_begin(sock_write_tx); 461 int sk_write_pending; 462 atomic_t sk_omem_alloc; 463 int sk_sndbuf; 464 465 int sk_wmem_queued; 466 refcount_t sk_wmem_alloc; 467 unsigned long sk_tsq_flags; 468 union { 469 struct sk_buff *sk_send_head; 470 struct rb_root tcp_rtx_queue; 471 }; 472 struct sk_buff_head sk_write_queue; 473 u32 sk_dst_pending_confirm; 474 u32 sk_pacing_status; /* see enum sk_pacing */ 475 struct page_frag sk_frag; 476 struct timer_list sk_timer; 477 478 unsigned long sk_pacing_rate; /* bytes per second */ 479 atomic_t sk_zckey; 480 atomic_t sk_tskey; 481 __cacheline_group_end(sock_write_tx); 482 483 __cacheline_group_begin(sock_read_tx); 484 unsigned long sk_max_pacing_rate; 485 long sk_sndtimeo; 486 u32 sk_priority; 487 u32 sk_mark; 488 struct dst_entry __rcu *sk_dst_cache; 489 netdev_features_t sk_route_caps; 490 #ifdef CONFIG_SOCK_VALIDATE_XMIT 491 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 492 struct net_device *dev, 493 struct sk_buff *skb); 494 #endif 495 u16 sk_gso_type; 496 u16 sk_gso_max_segs; 497 unsigned int sk_gso_max_size; 498 gfp_t sk_allocation; 499 u32 sk_txhash; 500 u8 sk_pacing_shift; 501 bool sk_use_task_frag; 502 __cacheline_group_end(sock_read_tx); 503 504 /* 505 * Because of non atomicity rules, all 506 * changes are protected by socket lock. 507 */ 508 u8 sk_gso_disabled : 1, 509 sk_kern_sock : 1, 510 sk_no_check_tx : 1, 511 sk_no_check_rx : 1; 512 u8 sk_shutdown; 513 u16 sk_type; 514 u16 sk_protocol; 515 unsigned long sk_lingertime; 516 struct proto *sk_prot_creator; 517 rwlock_t sk_callback_lock; 518 int sk_err_soft; 519 u32 sk_ack_backlog; 520 u32 sk_max_ack_backlog; 521 kuid_t sk_uid; 522 unsigned long sk_ino; 523 spinlock_t sk_peer_lock; 524 int sk_bind_phc; 525 struct pid *sk_peer_pid; 526 const struct cred *sk_peer_cred; 527 528 ktime_t sk_stamp; 529 #if BITS_PER_LONG==32 530 seqlock_t sk_stamp_seq; 531 #endif 532 int sk_disconnects; 533 534 union { 535 u8 sk_txrehash; 536 u8 sk_scm_recv_flags; 537 struct { 538 u8 sk_scm_credentials : 1, 539 sk_scm_security : 1, 540 sk_scm_pidfd : 1, 541 sk_scm_rights : 1, 542 sk_scm_unused : 4; 543 }; 544 }; 545 u8 sk_clockid; 546 u8 sk_txtime_deadline_mode : 1, 547 sk_txtime_report_errors : 1, 548 sk_txtime_unused : 6; 549 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG)) 550 u8 sk_bpf_cb_flags; 551 552 void *sk_user_data; 553 #ifdef CONFIG_SECURITY 554 void *sk_security; 555 #endif 556 struct sock_cgroup_data sk_cgrp_data; 557 void (*sk_state_change)(struct sock *sk); 558 void (*sk_write_space)(struct sock *sk); 559 void (*sk_error_report)(struct sock *sk); 560 int (*sk_backlog_rcv)(struct sock *sk, 561 struct sk_buff *skb); 562 void (*sk_destruct)(struct sock *sk); 563 struct sock_reuseport __rcu *sk_reuseport_cb; 564 #ifdef CONFIG_BPF_SYSCALL 565 struct bpf_local_storage __rcu *sk_bpf_storage; 566 #endif 567 struct rcu_head sk_rcu; 568 netns_tracker ns_tracker; 569 struct xarray sk_user_frags; 570 571 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 572 struct module *sk_owner; 573 #endif 574 }; 575 576 struct sock_bh_locked { 577 struct sock *sock; 578 local_lock_t bh_lock; 579 }; 580 581 enum sk_pacing { 582 SK_PACING_NONE = 0, 583 SK_PACING_NEEDED = 1, 584 SK_PACING_FQ = 2, 585 }; 586 587 /* flag bits in sk_user_data 588 * 589 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 590 * not be suitable for copying when cloning the socket. For instance, 591 * it can point to a reference counted object. sk_user_data bottom 592 * bit is set if pointer must not be copied. 593 * 594 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 595 * managed/owned by a BPF reuseport array. This bit should be set 596 * when sk_user_data's sk is added to the bpf's reuseport_array. 597 * 598 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 599 * sk_user_data points to psock type. This bit should be set 600 * when sk_user_data is assigned to a psock object. 601 */ 602 #define SK_USER_DATA_NOCOPY 1UL 603 #define SK_USER_DATA_BPF 2UL 604 #define SK_USER_DATA_PSOCK 4UL 605 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 606 SK_USER_DATA_PSOCK) 607 608 /** 609 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 610 * @sk: socket 611 */ 612 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 613 { 614 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 615 } 616 617 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 618 619 /** 620 * __locked_read_sk_user_data_with_flags - return the pointer 621 * only if argument flags all has been set in sk_user_data. Otherwise 622 * return NULL 623 * 624 * @sk: socket 625 * @flags: flag bits 626 * 627 * The caller must be holding sk->sk_callback_lock. 628 */ 629 static inline void * 630 __locked_read_sk_user_data_with_flags(const struct sock *sk, 631 uintptr_t flags) 632 { 633 uintptr_t sk_user_data = 634 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 635 lockdep_is_held(&sk->sk_callback_lock)); 636 637 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 638 639 if ((sk_user_data & flags) == flags) 640 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 641 return NULL; 642 } 643 644 /** 645 * __rcu_dereference_sk_user_data_with_flags - return the pointer 646 * only if argument flags all has been set in sk_user_data. Otherwise 647 * return NULL 648 * 649 * @sk: socket 650 * @flags: flag bits 651 */ 652 static inline void * 653 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 654 uintptr_t flags) 655 { 656 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 657 658 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 659 660 if ((sk_user_data & flags) == flags) 661 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 662 return NULL; 663 } 664 665 #define rcu_dereference_sk_user_data(sk) \ 666 __rcu_dereference_sk_user_data_with_flags(sk, 0) 667 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 668 ({ \ 669 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 670 __tmp2 = (uintptr_t)(flags); \ 671 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 672 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 673 rcu_assign_pointer(__sk_user_data((sk)), \ 674 __tmp1 | __tmp2); \ 675 }) 676 #define rcu_assign_sk_user_data(sk, ptr) \ 677 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 678 679 static inline 680 struct net *sock_net(const struct sock *sk) 681 { 682 return read_pnet(&sk->sk_net); 683 } 684 685 static inline 686 void sock_net_set(struct sock *sk, struct net *net) 687 { 688 write_pnet(&sk->sk_net, net); 689 } 690 691 /* 692 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 693 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 694 * on a socket means that the socket will reuse everybody else's port 695 * without looking at the other's sk_reuse value. 696 */ 697 698 #define SK_NO_REUSE 0 699 #define SK_CAN_REUSE 1 700 #define SK_FORCE_REUSE 2 701 702 int sk_set_peek_off(struct sock *sk, int val); 703 704 static inline int sk_peek_offset(const struct sock *sk, int flags) 705 { 706 if (unlikely(flags & MSG_PEEK)) { 707 return READ_ONCE(sk->sk_peek_off); 708 } 709 710 return 0; 711 } 712 713 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 714 { 715 s32 off = READ_ONCE(sk->sk_peek_off); 716 717 if (unlikely(off >= 0)) { 718 off = max_t(s32, off - val, 0); 719 WRITE_ONCE(sk->sk_peek_off, off); 720 } 721 } 722 723 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 724 { 725 sk_peek_offset_bwd(sk, -val); 726 } 727 728 /* 729 * Hashed lists helper routines 730 */ 731 static inline struct sock *sk_entry(const struct hlist_node *node) 732 { 733 return hlist_entry(node, struct sock, sk_node); 734 } 735 736 static inline struct sock *__sk_head(const struct hlist_head *head) 737 { 738 return hlist_entry(head->first, struct sock, sk_node); 739 } 740 741 static inline struct sock *sk_head(const struct hlist_head *head) 742 { 743 return hlist_empty(head) ? NULL : __sk_head(head); 744 } 745 746 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 747 { 748 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 749 } 750 751 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 752 { 753 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 754 } 755 756 static inline struct sock *sk_next(const struct sock *sk) 757 { 758 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 759 } 760 761 static inline struct sock *sk_nulls_next(const struct sock *sk) 762 { 763 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 764 hlist_nulls_entry(sk->sk_nulls_node.next, 765 struct sock, sk_nulls_node) : 766 NULL; 767 } 768 769 static inline bool sk_unhashed(const struct sock *sk) 770 { 771 return hlist_unhashed(&sk->sk_node); 772 } 773 774 static inline bool sk_hashed(const struct sock *sk) 775 { 776 return !sk_unhashed(sk); 777 } 778 779 static inline void sk_node_init(struct hlist_node *node) 780 { 781 node->pprev = NULL; 782 } 783 784 static inline void __sk_del_node(struct sock *sk) 785 { 786 __hlist_del(&sk->sk_node); 787 } 788 789 /* NB: equivalent to hlist_del_init_rcu */ 790 static inline bool __sk_del_node_init(struct sock *sk) 791 { 792 if (sk_hashed(sk)) { 793 __sk_del_node(sk); 794 sk_node_init(&sk->sk_node); 795 return true; 796 } 797 return false; 798 } 799 800 /* Grab socket reference count. This operation is valid only 801 when sk is ALREADY grabbed f.e. it is found in hash table 802 or a list and the lookup is made under lock preventing hash table 803 modifications. 804 */ 805 806 static __always_inline void sock_hold(struct sock *sk) 807 { 808 refcount_inc(&sk->sk_refcnt); 809 } 810 811 /* Ungrab socket in the context, which assumes that socket refcnt 812 cannot hit zero, f.e. it is true in context of any socketcall. 813 */ 814 static __always_inline void __sock_put(struct sock *sk) 815 { 816 refcount_dec(&sk->sk_refcnt); 817 } 818 819 static inline bool sk_del_node_init(struct sock *sk) 820 { 821 bool rc = __sk_del_node_init(sk); 822 823 if (rc) { 824 /* paranoid for a while -acme */ 825 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 826 __sock_put(sk); 827 } 828 return rc; 829 } 830 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 831 832 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 833 { 834 if (sk_hashed(sk)) { 835 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 836 return true; 837 } 838 return false; 839 } 840 841 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 842 { 843 bool rc = __sk_nulls_del_node_init_rcu(sk); 844 845 if (rc) { 846 /* paranoid for a while -acme */ 847 WARN_ON(refcount_read(&sk->sk_refcnt) == 1); 848 __sock_put(sk); 849 } 850 return rc; 851 } 852 853 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 854 { 855 hlist_add_head(&sk->sk_node, list); 856 } 857 858 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 859 { 860 sock_hold(sk); 861 __sk_add_node(sk, list); 862 } 863 864 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 865 { 866 sock_hold(sk); 867 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 868 sk->sk_family == AF_INET6) 869 hlist_add_tail_rcu(&sk->sk_node, list); 870 else 871 hlist_add_head_rcu(&sk->sk_node, list); 872 } 873 874 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 875 { 876 sock_hold(sk); 877 hlist_add_tail_rcu(&sk->sk_node, list); 878 } 879 880 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 881 { 882 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 883 } 884 885 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 886 { 887 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 888 } 889 890 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 891 { 892 sock_hold(sk); 893 __sk_nulls_add_node_rcu(sk, list); 894 } 895 896 static inline void __sk_del_bind_node(struct sock *sk) 897 { 898 __hlist_del(&sk->sk_bind_node); 899 } 900 901 static inline void sk_add_bind_node(struct sock *sk, 902 struct hlist_head *list) 903 { 904 hlist_add_head(&sk->sk_bind_node, list); 905 } 906 907 #define sk_for_each(__sk, list) \ 908 hlist_for_each_entry(__sk, list, sk_node) 909 #define sk_for_each_rcu(__sk, list) \ 910 hlist_for_each_entry_rcu(__sk, list, sk_node) 911 #define sk_nulls_for_each(__sk, node, list) \ 912 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 913 #define sk_nulls_for_each_rcu(__sk, node, list) \ 914 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 915 #define sk_for_each_from(__sk) \ 916 hlist_for_each_entry_from(__sk, sk_node) 917 #define sk_nulls_for_each_from(__sk, node) \ 918 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 919 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 920 #define sk_for_each_safe(__sk, tmp, list) \ 921 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 922 #define sk_for_each_bound(__sk, list) \ 923 hlist_for_each_entry(__sk, list, sk_bind_node) 924 #define sk_for_each_bound_safe(__sk, tmp, list) \ 925 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node) 926 927 /** 928 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 929 * @tpos: the type * to use as a loop cursor. 930 * @pos: the &struct hlist_node to use as a loop cursor. 931 * @head: the head for your list. 932 * @offset: offset of hlist_node within the struct. 933 * 934 */ 935 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 936 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 937 pos != NULL && \ 938 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 939 pos = rcu_dereference(hlist_next_rcu(pos))) 940 941 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 942 { 943 /* Careful only use this in a context where these parameters 944 * can not change and must all be valid, such as recvmsg from 945 * userspace. 946 */ 947 return sk->sk_socket->file->f_cred->user_ns; 948 } 949 950 /* Sock flags */ 951 enum sock_flags { 952 SOCK_DEAD, 953 SOCK_DONE, 954 SOCK_URGINLINE, 955 SOCK_KEEPOPEN, 956 SOCK_LINGER, 957 SOCK_DESTROY, 958 SOCK_BROADCAST, 959 SOCK_TIMESTAMP, 960 SOCK_ZAPPED, 961 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 962 SOCK_DBG, /* %SO_DEBUG setting */ 963 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 964 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 965 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 966 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 967 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 968 SOCK_FASYNC, /* fasync() active */ 969 SOCK_RXQ_OVFL, 970 SOCK_ZEROCOPY, /* buffers from userspace */ 971 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 972 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 973 * Will use last 4 bytes of packet sent from 974 * user-space instead. 975 */ 976 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 977 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 978 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 979 SOCK_TXTIME, 980 SOCK_XDP, /* XDP is attached */ 981 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 982 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 983 SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */ 984 SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */ 985 }; 986 987 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 988 /* 989 * The highest bit of sk_tsflags is reserved for kernel-internal 990 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that 991 * SOF_TIMESTAMPING* values do not reach this reserved area 992 */ 993 #define SOCKCM_FLAG_TS_OPT_ID BIT(31) 994 995 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 996 { 997 nsk->sk_flags = osk->sk_flags; 998 } 999 1000 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 1001 { 1002 __set_bit(flag, &sk->sk_flags); 1003 } 1004 1005 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 1006 { 1007 __clear_bit(flag, &sk->sk_flags); 1008 } 1009 1010 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 1011 int valbool) 1012 { 1013 if (valbool) 1014 sock_set_flag(sk, bit); 1015 else 1016 sock_reset_flag(sk, bit); 1017 } 1018 1019 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 1020 { 1021 return test_bit(flag, &sk->sk_flags); 1022 } 1023 1024 #ifdef CONFIG_NET 1025 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 1026 static inline int sk_memalloc_socks(void) 1027 { 1028 return static_branch_unlikely(&memalloc_socks_key); 1029 } 1030 1031 void __receive_sock(struct file *file); 1032 #else 1033 1034 static inline int sk_memalloc_socks(void) 1035 { 1036 return 0; 1037 } 1038 1039 static inline void __receive_sock(struct file *file) 1040 { } 1041 #endif 1042 1043 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1044 { 1045 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1046 } 1047 1048 static inline void sk_acceptq_removed(struct sock *sk) 1049 { 1050 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1051 } 1052 1053 static inline void sk_acceptq_added(struct sock *sk) 1054 { 1055 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1056 } 1057 1058 /* Note: If you think the test should be: 1059 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1060 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1061 */ 1062 static inline bool sk_acceptq_is_full(const struct sock *sk) 1063 { 1064 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1065 } 1066 1067 /* 1068 * Compute minimal free write space needed to queue new packets. 1069 */ 1070 static inline int sk_stream_min_wspace(const struct sock *sk) 1071 { 1072 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1073 } 1074 1075 static inline int sk_stream_wspace(const struct sock *sk) 1076 { 1077 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1078 } 1079 1080 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1081 { 1082 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1083 } 1084 1085 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1086 { 1087 /* Paired with lockless reads of sk->sk_forward_alloc */ 1088 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1089 } 1090 1091 void sk_stream_write_space(struct sock *sk); 1092 1093 /* OOB backlog add */ 1094 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1095 { 1096 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1097 skb_dst_force(skb); 1098 1099 if (!sk->sk_backlog.tail) 1100 WRITE_ONCE(sk->sk_backlog.head, skb); 1101 else 1102 sk->sk_backlog.tail->next = skb; 1103 1104 WRITE_ONCE(sk->sk_backlog.tail, skb); 1105 skb->next = NULL; 1106 } 1107 1108 /* 1109 * Take into account size of receive queue and backlog queue 1110 * Do not take into account this skb truesize, 1111 * to allow even a single big packet to come. 1112 */ 1113 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1114 { 1115 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1116 1117 return qsize > limit; 1118 } 1119 1120 /* The per-socket spinlock must be held here. */ 1121 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1122 unsigned int limit) 1123 { 1124 if (sk_rcvqueues_full(sk, limit)) 1125 return -ENOBUFS; 1126 1127 /* 1128 * If the skb was allocated from pfmemalloc reserves, only 1129 * allow SOCK_MEMALLOC sockets to use it as this socket is 1130 * helping free memory 1131 */ 1132 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1133 return -ENOMEM; 1134 1135 __sk_add_backlog(sk, skb); 1136 sk->sk_backlog.len += skb->truesize; 1137 return 0; 1138 } 1139 1140 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1141 1142 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1143 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1144 1145 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1146 { 1147 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1148 return __sk_backlog_rcv(sk, skb); 1149 1150 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1151 tcp_v6_do_rcv, 1152 tcp_v4_do_rcv, 1153 sk, skb); 1154 } 1155 1156 static inline void sk_incoming_cpu_update(struct sock *sk) 1157 { 1158 int cpu = raw_smp_processor_id(); 1159 1160 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1161 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1162 } 1163 1164 1165 static inline void sock_rps_save_rxhash(struct sock *sk, 1166 const struct sk_buff *skb) 1167 { 1168 #ifdef CONFIG_RPS 1169 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1170 * here, and another one in sock_rps_record_flow(). 1171 */ 1172 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1173 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1174 #endif 1175 } 1176 1177 static inline void sock_rps_reset_rxhash(struct sock *sk) 1178 { 1179 #ifdef CONFIG_RPS 1180 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1181 WRITE_ONCE(sk->sk_rxhash, 0); 1182 #endif 1183 } 1184 1185 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1186 ({ int __rc, __dis = __sk->sk_disconnects; \ 1187 release_sock(__sk); \ 1188 __rc = __condition; \ 1189 if (!__rc) { \ 1190 *(__timeo) = wait_woken(__wait, \ 1191 TASK_INTERRUPTIBLE, \ 1192 *(__timeo)); \ 1193 } \ 1194 sched_annotate_sleep(); \ 1195 lock_sock(__sk); \ 1196 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1197 __rc; \ 1198 }) 1199 1200 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1201 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1202 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1203 int sk_stream_error(struct sock *sk, int flags, int err); 1204 void sk_stream_kill_queues(struct sock *sk); 1205 void sk_set_memalloc(struct sock *sk); 1206 void sk_clear_memalloc(struct sock *sk); 1207 1208 void __sk_flush_backlog(struct sock *sk); 1209 1210 static inline bool sk_flush_backlog(struct sock *sk) 1211 { 1212 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1213 __sk_flush_backlog(sk); 1214 return true; 1215 } 1216 return false; 1217 } 1218 1219 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1220 1221 struct request_sock_ops; 1222 struct timewait_sock_ops; 1223 struct inet_hashinfo; 1224 struct raw_hashinfo; 1225 struct smc_hashinfo; 1226 struct module; 1227 struct sk_psock; 1228 1229 /* 1230 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1231 * un-modified. Special care is taken when initializing object to zero. 1232 */ 1233 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1234 { 1235 if (offsetof(struct sock, sk_node.next) != 0) 1236 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1237 memset(&sk->sk_node.pprev, 0, 1238 size - offsetof(struct sock, sk_node.pprev)); 1239 } 1240 1241 struct proto_accept_arg { 1242 int flags; 1243 int err; 1244 int is_empty; 1245 bool kern; 1246 }; 1247 1248 /* Networking protocol blocks we attach to sockets. 1249 * socket layer -> transport layer interface 1250 */ 1251 struct proto { 1252 void (*close)(struct sock *sk, 1253 long timeout); 1254 int (*pre_connect)(struct sock *sk, 1255 struct sockaddr *uaddr, 1256 int addr_len); 1257 int (*connect)(struct sock *sk, 1258 struct sockaddr *uaddr, 1259 int addr_len); 1260 int (*disconnect)(struct sock *sk, int flags); 1261 1262 struct sock * (*accept)(struct sock *sk, 1263 struct proto_accept_arg *arg); 1264 1265 int (*ioctl)(struct sock *sk, int cmd, 1266 int *karg); 1267 int (*init)(struct sock *sk); 1268 void (*destroy)(struct sock *sk); 1269 void (*shutdown)(struct sock *sk, int how); 1270 int (*setsockopt)(struct sock *sk, int level, 1271 int optname, sockptr_t optval, 1272 unsigned int optlen); 1273 int (*getsockopt)(struct sock *sk, int level, 1274 int optname, char __user *optval, 1275 int __user *option); 1276 void (*keepalive)(struct sock *sk, int valbool); 1277 #ifdef CONFIG_COMPAT 1278 int (*compat_ioctl)(struct sock *sk, 1279 unsigned int cmd, unsigned long arg); 1280 #endif 1281 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1282 size_t len); 1283 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1284 size_t len, int flags, int *addr_len); 1285 void (*splice_eof)(struct socket *sock); 1286 int (*bind)(struct sock *sk, 1287 struct sockaddr *addr, int addr_len); 1288 int (*bind_add)(struct sock *sk, 1289 struct sockaddr *addr, int addr_len); 1290 1291 int (*backlog_rcv) (struct sock *sk, 1292 struct sk_buff *skb); 1293 bool (*bpf_bypass_getsockopt)(int level, 1294 int optname); 1295 1296 void (*release_cb)(struct sock *sk); 1297 1298 /* Keeping track of sk's, looking them up, and port selection methods. */ 1299 int (*hash)(struct sock *sk); 1300 void (*unhash)(struct sock *sk); 1301 void (*rehash)(struct sock *sk); 1302 int (*get_port)(struct sock *sk, unsigned short snum); 1303 void (*put_port)(struct sock *sk); 1304 #ifdef CONFIG_BPF_SYSCALL 1305 int (*psock_update_sk_prot)(struct sock *sk, 1306 struct sk_psock *psock, 1307 bool restore); 1308 #endif 1309 1310 /* Keeping track of sockets in use */ 1311 #ifdef CONFIG_PROC_FS 1312 unsigned int inuse_idx; 1313 #endif 1314 1315 bool (*stream_memory_free)(const struct sock *sk, int wake); 1316 bool (*sock_is_readable)(struct sock *sk); 1317 /* Memory pressure */ 1318 void (*enter_memory_pressure)(struct sock *sk); 1319 void (*leave_memory_pressure)(struct sock *sk); 1320 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1321 int __percpu *per_cpu_fw_alloc; 1322 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1323 1324 /* 1325 * Pressure flag: try to collapse. 1326 * Technical note: it is used by multiple contexts non atomically. 1327 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1328 * All the __sk_mem_schedule() is of this nature: accounting 1329 * is strict, actions are advisory and have some latency. 1330 */ 1331 unsigned long *memory_pressure; 1332 long *sysctl_mem; 1333 1334 int *sysctl_wmem; 1335 int *sysctl_rmem; 1336 u32 sysctl_wmem_offset; 1337 u32 sysctl_rmem_offset; 1338 1339 int max_header; 1340 bool no_autobind; 1341 1342 struct kmem_cache *slab; 1343 unsigned int obj_size; 1344 unsigned int ipv6_pinfo_offset; 1345 slab_flags_t slab_flags; 1346 unsigned int useroffset; /* Usercopy region offset */ 1347 unsigned int usersize; /* Usercopy region size */ 1348 1349 unsigned int __percpu *orphan_count; 1350 1351 struct request_sock_ops *rsk_prot; 1352 struct timewait_sock_ops *twsk_prot; 1353 1354 union { 1355 struct inet_hashinfo *hashinfo; 1356 struct udp_table *udp_table; 1357 struct raw_hashinfo *raw_hash; 1358 struct smc_hashinfo *smc_hash; 1359 } h; 1360 1361 struct module *owner; 1362 1363 char name[32]; 1364 1365 struct list_head node; 1366 int (*diag_destroy)(struct sock *sk, int err); 1367 } __randomize_layout; 1368 1369 int proto_register(struct proto *prot, int alloc_slab); 1370 void proto_unregister(struct proto *prot); 1371 int sock_load_diag_module(int family, int protocol); 1372 1373 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1374 1375 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1376 { 1377 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1378 return false; 1379 1380 return sk->sk_prot->stream_memory_free ? 1381 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1382 tcp_stream_memory_free, sk, wake) : true; 1383 } 1384 1385 static inline bool sk_stream_memory_free(const struct sock *sk) 1386 { 1387 return __sk_stream_memory_free(sk, 0); 1388 } 1389 1390 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1391 { 1392 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1393 __sk_stream_memory_free(sk, wake); 1394 } 1395 1396 static inline bool sk_stream_is_writeable(const struct sock *sk) 1397 { 1398 return __sk_stream_is_writeable(sk, 0); 1399 } 1400 1401 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1402 struct cgroup *ancestor) 1403 { 1404 #ifdef CONFIG_SOCK_CGROUP_DATA 1405 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1406 ancestor); 1407 #else 1408 return -ENOTSUPP; 1409 #endif 1410 } 1411 1412 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1413 1414 static inline void sk_sockets_allocated_dec(struct sock *sk) 1415 { 1416 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1417 SK_ALLOC_PERCPU_COUNTER_BATCH); 1418 } 1419 1420 static inline void sk_sockets_allocated_inc(struct sock *sk) 1421 { 1422 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1423 SK_ALLOC_PERCPU_COUNTER_BATCH); 1424 } 1425 1426 static inline u64 1427 sk_sockets_allocated_read_positive(struct sock *sk) 1428 { 1429 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1430 } 1431 1432 static inline int 1433 proto_sockets_allocated_sum_positive(struct proto *prot) 1434 { 1435 return percpu_counter_sum_positive(prot->sockets_allocated); 1436 } 1437 1438 #ifdef CONFIG_PROC_FS 1439 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1440 struct prot_inuse { 1441 int all; 1442 int val[PROTO_INUSE_NR]; 1443 }; 1444 1445 static inline void sock_prot_inuse_add(const struct net *net, 1446 const struct proto *prot, int val) 1447 { 1448 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1449 } 1450 1451 static inline void sock_inuse_add(const struct net *net, int val) 1452 { 1453 this_cpu_add(net->core.prot_inuse->all, val); 1454 } 1455 1456 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1457 int sock_inuse_get(struct net *net); 1458 #else 1459 static inline void sock_prot_inuse_add(const struct net *net, 1460 const struct proto *prot, int val) 1461 { 1462 } 1463 1464 static inline void sock_inuse_add(const struct net *net, int val) 1465 { 1466 } 1467 #endif 1468 1469 1470 /* With per-bucket locks this operation is not-atomic, so that 1471 * this version is not worse. 1472 */ 1473 static inline int __sk_prot_rehash(struct sock *sk) 1474 { 1475 sk->sk_prot->unhash(sk); 1476 return sk->sk_prot->hash(sk); 1477 } 1478 1479 /* About 10 seconds */ 1480 #define SOCK_DESTROY_TIME (10*HZ) 1481 1482 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1483 #define PROT_SOCK 1024 1484 1485 #define SHUTDOWN_MASK 3 1486 #define RCV_SHUTDOWN 1 1487 #define SEND_SHUTDOWN 2 1488 1489 #define SOCK_BINDADDR_LOCK 4 1490 #define SOCK_BINDPORT_LOCK 8 1491 1492 struct socket_alloc { 1493 struct socket socket; 1494 struct inode vfs_inode; 1495 }; 1496 1497 static inline struct socket *SOCKET_I(struct inode *inode) 1498 { 1499 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1500 } 1501 1502 static inline struct inode *SOCK_INODE(struct socket *socket) 1503 { 1504 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1505 } 1506 1507 /* 1508 * Functions for memory accounting 1509 */ 1510 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1511 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1512 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1513 void __sk_mem_reclaim(struct sock *sk, int amount); 1514 1515 #define SK_MEM_SEND 0 1516 #define SK_MEM_RECV 1 1517 1518 /* sysctl_mem values are in pages */ 1519 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1520 { 1521 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1522 } 1523 1524 static inline int sk_mem_pages(int amt) 1525 { 1526 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1527 } 1528 1529 static inline bool sk_has_account(struct sock *sk) 1530 { 1531 /* return true if protocol supports memory accounting */ 1532 return !!sk->sk_prot->memory_allocated; 1533 } 1534 1535 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1536 { 1537 int delta; 1538 1539 if (!sk_has_account(sk)) 1540 return true; 1541 delta = size - sk->sk_forward_alloc; 1542 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1543 } 1544 1545 static inline bool 1546 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc) 1547 { 1548 int delta; 1549 1550 if (!sk_has_account(sk)) 1551 return true; 1552 delta = size - sk->sk_forward_alloc; 1553 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1554 pfmemalloc; 1555 } 1556 1557 static inline bool 1558 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size) 1559 { 1560 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb)); 1561 } 1562 1563 static inline int sk_unused_reserved_mem(const struct sock *sk) 1564 { 1565 int unused_mem; 1566 1567 if (likely(!sk->sk_reserved_mem)) 1568 return 0; 1569 1570 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1571 atomic_read(&sk->sk_rmem_alloc); 1572 1573 return unused_mem > 0 ? unused_mem : 0; 1574 } 1575 1576 static inline void sk_mem_reclaim(struct sock *sk) 1577 { 1578 int reclaimable; 1579 1580 if (!sk_has_account(sk)) 1581 return; 1582 1583 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1584 1585 if (reclaimable >= (int)PAGE_SIZE) 1586 __sk_mem_reclaim(sk, reclaimable); 1587 } 1588 1589 static inline void sk_mem_reclaim_final(struct sock *sk) 1590 { 1591 sk->sk_reserved_mem = 0; 1592 sk_mem_reclaim(sk); 1593 } 1594 1595 static inline void sk_mem_charge(struct sock *sk, int size) 1596 { 1597 if (!sk_has_account(sk)) 1598 return; 1599 sk_forward_alloc_add(sk, -size); 1600 } 1601 1602 static inline void sk_mem_uncharge(struct sock *sk, int size) 1603 { 1604 if (!sk_has_account(sk)) 1605 return; 1606 sk_forward_alloc_add(sk, size); 1607 sk_mem_reclaim(sk); 1608 } 1609 1610 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 1611 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1612 { 1613 __module_get(owner); 1614 sk->sk_owner = owner; 1615 } 1616 1617 static inline void sk_owner_clear(struct sock *sk) 1618 { 1619 sk->sk_owner = NULL; 1620 } 1621 1622 static inline void sk_owner_put(struct sock *sk) 1623 { 1624 module_put(sk->sk_owner); 1625 } 1626 #else 1627 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1628 { 1629 } 1630 1631 static inline void sk_owner_clear(struct sock *sk) 1632 { 1633 } 1634 1635 static inline void sk_owner_put(struct sock *sk) 1636 { 1637 } 1638 #endif 1639 /* 1640 * Macro so as to not evaluate some arguments when 1641 * lockdep is not enabled. 1642 * 1643 * Mark both the sk_lock and the sk_lock.slock as a 1644 * per-address-family lock class. 1645 */ 1646 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1647 do { \ 1648 sk_owner_set(sk, THIS_MODULE); \ 1649 sk->sk_lock.owned = 0; \ 1650 init_waitqueue_head(&sk->sk_lock.wq); \ 1651 spin_lock_init(&(sk)->sk_lock.slock); \ 1652 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1653 sizeof((sk)->sk_lock)); \ 1654 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1655 (skey), (sname)); \ 1656 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1657 } while (0) 1658 1659 static inline bool lockdep_sock_is_held(const struct sock *sk) 1660 { 1661 return lockdep_is_held(&sk->sk_lock) || 1662 lockdep_is_held(&sk->sk_lock.slock); 1663 } 1664 1665 void lock_sock_nested(struct sock *sk, int subclass); 1666 1667 static inline void lock_sock(struct sock *sk) 1668 { 1669 lock_sock_nested(sk, 0); 1670 } 1671 1672 void __lock_sock(struct sock *sk); 1673 void __release_sock(struct sock *sk); 1674 void release_sock(struct sock *sk); 1675 1676 /* BH context may only use the following locking interface. */ 1677 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1678 #define bh_lock_sock_nested(__sk) \ 1679 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1680 SINGLE_DEPTH_NESTING) 1681 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1682 1683 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1684 1685 /** 1686 * lock_sock_fast - fast version of lock_sock 1687 * @sk: socket 1688 * 1689 * This version should be used for very small section, where process won't block 1690 * return false if fast path is taken: 1691 * 1692 * sk_lock.slock locked, owned = 0, BH disabled 1693 * 1694 * return true if slow path is taken: 1695 * 1696 * sk_lock.slock unlocked, owned = 1, BH enabled 1697 */ 1698 static inline bool lock_sock_fast(struct sock *sk) 1699 { 1700 /* The sk_lock has mutex_lock() semantics here. */ 1701 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1702 1703 return __lock_sock_fast(sk); 1704 } 1705 1706 /* fast socket lock variant for caller already holding a [different] socket lock */ 1707 static inline bool lock_sock_fast_nested(struct sock *sk) 1708 { 1709 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1710 1711 return __lock_sock_fast(sk); 1712 } 1713 1714 /** 1715 * unlock_sock_fast - complement of lock_sock_fast 1716 * @sk: socket 1717 * @slow: slow mode 1718 * 1719 * fast unlock socket for user context. 1720 * If slow mode is on, we call regular release_sock() 1721 */ 1722 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1723 __releases(&sk->sk_lock.slock) 1724 { 1725 if (slow) { 1726 release_sock(sk); 1727 __release(&sk->sk_lock.slock); 1728 } else { 1729 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1730 spin_unlock_bh(&sk->sk_lock.slock); 1731 } 1732 } 1733 1734 void sockopt_lock_sock(struct sock *sk); 1735 void sockopt_release_sock(struct sock *sk); 1736 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1737 bool sockopt_capable(int cap); 1738 1739 /* Used by processes to "lock" a socket state, so that 1740 * interrupts and bottom half handlers won't change it 1741 * from under us. It essentially blocks any incoming 1742 * packets, so that we won't get any new data or any 1743 * packets that change the state of the socket. 1744 * 1745 * While locked, BH processing will add new packets to 1746 * the backlog queue. This queue is processed by the 1747 * owner of the socket lock right before it is released. 1748 * 1749 * Since ~2.3.5 it is also exclusive sleep lock serializing 1750 * accesses from user process context. 1751 */ 1752 1753 static inline void sock_owned_by_me(const struct sock *sk) 1754 { 1755 #ifdef CONFIG_LOCKDEP 1756 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1757 #endif 1758 } 1759 1760 static inline void sock_not_owned_by_me(const struct sock *sk) 1761 { 1762 #ifdef CONFIG_LOCKDEP 1763 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1764 #endif 1765 } 1766 1767 static inline bool sock_owned_by_user(const struct sock *sk) 1768 { 1769 sock_owned_by_me(sk); 1770 return sk->sk_lock.owned; 1771 } 1772 1773 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1774 { 1775 return sk->sk_lock.owned; 1776 } 1777 1778 static inline void sock_release_ownership(struct sock *sk) 1779 { 1780 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1781 sk->sk_lock.owned = 0; 1782 1783 /* The sk_lock has mutex_unlock() semantics: */ 1784 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1785 } 1786 1787 /* no reclassification while locks are held */ 1788 static inline bool sock_allow_reclassification(const struct sock *csk) 1789 { 1790 struct sock *sk = (struct sock *)csk; 1791 1792 return !sock_owned_by_user_nocheck(sk) && 1793 !spin_is_locked(&sk->sk_lock.slock); 1794 } 1795 1796 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1797 struct proto *prot, int kern); 1798 void sk_free(struct sock *sk); 1799 void sk_net_refcnt_upgrade(struct sock *sk); 1800 void sk_destruct(struct sock *sk); 1801 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority); 1802 1803 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1804 gfp_t priority); 1805 void __sock_wfree(struct sk_buff *skb); 1806 void sock_wfree(struct sk_buff *skb); 1807 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1808 gfp_t priority); 1809 void skb_orphan_partial(struct sk_buff *skb); 1810 void sock_rfree(struct sk_buff *skb); 1811 void sock_efree(struct sk_buff *skb); 1812 #ifdef CONFIG_INET 1813 void sock_edemux(struct sk_buff *skb); 1814 void sock_pfree(struct sk_buff *skb); 1815 1816 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk) 1817 { 1818 skb_orphan(skb); 1819 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 1820 skb->sk = sk; 1821 skb->destructor = sock_edemux; 1822 } 1823 } 1824 #else 1825 #define sock_edemux sock_efree 1826 #endif 1827 1828 int sk_setsockopt(struct sock *sk, int level, int optname, 1829 sockptr_t optval, unsigned int optlen); 1830 int sock_setsockopt(struct socket *sock, int level, int op, 1831 sockptr_t optval, unsigned int optlen); 1832 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1833 int optname, sockptr_t optval, int optlen); 1834 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1835 int optname, sockptr_t optval, sockptr_t optlen); 1836 1837 int sk_getsockopt(struct sock *sk, int level, int optname, 1838 sockptr_t optval, sockptr_t optlen); 1839 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1840 bool timeval, bool time32); 1841 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1842 unsigned long data_len, int noblock, 1843 int *errcode, int max_page_order); 1844 1845 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1846 unsigned long size, 1847 int noblock, int *errcode) 1848 { 1849 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1850 } 1851 1852 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1853 void *sock_kmemdup(struct sock *sk, const void *src, 1854 int size, gfp_t priority); 1855 void sock_kfree_s(struct sock *sk, void *mem, int size); 1856 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1857 void sk_send_sigurg(struct sock *sk); 1858 1859 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1860 { 1861 if (sk->sk_socket) 1862 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1863 WRITE_ONCE(sk->sk_prot, proto); 1864 } 1865 1866 struct sockcm_cookie { 1867 u64 transmit_time; 1868 u32 mark; 1869 u32 tsflags; 1870 u32 ts_opt_id; 1871 u32 priority; 1872 u32 dmabuf_id; 1873 }; 1874 1875 static inline void sockcm_init(struct sockcm_cookie *sockc, 1876 const struct sock *sk) 1877 { 1878 *sockc = (struct sockcm_cookie) { 1879 .mark = READ_ONCE(sk->sk_mark), 1880 .tsflags = READ_ONCE(sk->sk_tsflags), 1881 .priority = READ_ONCE(sk->sk_priority), 1882 }; 1883 } 1884 1885 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1886 struct sockcm_cookie *sockc); 1887 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1888 struct sockcm_cookie *sockc); 1889 1890 /* 1891 * Functions to fill in entries in struct proto_ops when a protocol 1892 * does not implement a particular function. 1893 */ 1894 int sock_no_bind(struct socket *, struct sockaddr *, int); 1895 int sock_no_connect(struct socket *, struct sockaddr *, int, int); 1896 int sock_no_socketpair(struct socket *, struct socket *); 1897 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *); 1898 int sock_no_getname(struct socket *, struct sockaddr *, int); 1899 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1900 int sock_no_listen(struct socket *, int); 1901 int sock_no_shutdown(struct socket *, int); 1902 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1903 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1904 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1905 int sock_no_mmap(struct file *file, struct socket *sock, 1906 struct vm_area_struct *vma); 1907 1908 /* 1909 * Functions to fill in entries in struct proto_ops when a protocol 1910 * uses the inet style. 1911 */ 1912 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1913 char __user *optval, int __user *optlen); 1914 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1915 int flags); 1916 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1917 sockptr_t optval, unsigned int optlen); 1918 1919 void sk_common_release(struct sock *sk); 1920 1921 /* 1922 * Default socket callbacks and setup code 1923 */ 1924 1925 /* Initialise core socket variables using an explicit uid. */ 1926 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1927 1928 /* Initialise core socket variables. 1929 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1930 */ 1931 void sock_init_data(struct socket *sock, struct sock *sk); 1932 1933 /* 1934 * Socket reference counting postulates. 1935 * 1936 * * Each user of socket SHOULD hold a reference count. 1937 * * Each access point to socket (an hash table bucket, reference from a list, 1938 * running timer, skb in flight MUST hold a reference count. 1939 * * When reference count hits 0, it means it will never increase back. 1940 * * When reference count hits 0, it means that no references from 1941 * outside exist to this socket and current process on current CPU 1942 * is last user and may/should destroy this socket. 1943 * * sk_free is called from any context: process, BH, IRQ. When 1944 * it is called, socket has no references from outside -> sk_free 1945 * may release descendant resources allocated by the socket, but 1946 * to the time when it is called, socket is NOT referenced by any 1947 * hash tables, lists etc. 1948 * * Packets, delivered from outside (from network or from another process) 1949 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1950 * when they sit in queue. Otherwise, packets will leak to hole, when 1951 * socket is looked up by one cpu and unhasing is made by another CPU. 1952 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1953 * (leak to backlog). Packet socket does all the processing inside 1954 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1955 * use separate SMP lock, so that they are prone too. 1956 */ 1957 1958 /* Ungrab socket and destroy it, if it was the last reference. */ 1959 static inline void sock_put(struct sock *sk) 1960 { 1961 if (refcount_dec_and_test(&sk->sk_refcnt)) 1962 sk_free(sk); 1963 } 1964 /* Generic version of sock_put(), dealing with all sockets 1965 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1966 */ 1967 void sock_gen_put(struct sock *sk); 1968 1969 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1970 unsigned int trim_cap, bool refcounted); 1971 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 1972 const int nested) 1973 { 1974 return __sk_receive_skb(sk, skb, nested, 1, true); 1975 } 1976 1977 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 1978 { 1979 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 1980 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 1981 return; 1982 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1983 * other WRITE_ONCE() because socket lock might be not held. 1984 */ 1985 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 1986 } 1987 1988 #define NO_QUEUE_MAPPING USHRT_MAX 1989 1990 static inline void sk_tx_queue_clear(struct sock *sk) 1991 { 1992 /* Paired with READ_ONCE() in sk_tx_queue_get() and 1993 * other WRITE_ONCE() because socket lock might be not held. 1994 */ 1995 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 1996 } 1997 1998 static inline int sk_tx_queue_get(const struct sock *sk) 1999 { 2000 if (sk) { 2001 /* Paired with WRITE_ONCE() in sk_tx_queue_clear() 2002 * and sk_tx_queue_set(). 2003 */ 2004 int val = READ_ONCE(sk->sk_tx_queue_mapping); 2005 2006 if (val != NO_QUEUE_MAPPING) 2007 return val; 2008 } 2009 return -1; 2010 } 2011 2012 static inline void __sk_rx_queue_set(struct sock *sk, 2013 const struct sk_buff *skb, 2014 bool force_set) 2015 { 2016 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2017 if (skb_rx_queue_recorded(skb)) { 2018 u16 rx_queue = skb_get_rx_queue(skb); 2019 2020 if (force_set || 2021 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2022 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2023 } 2024 #endif 2025 } 2026 2027 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2028 { 2029 __sk_rx_queue_set(sk, skb, true); 2030 } 2031 2032 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2033 { 2034 __sk_rx_queue_set(sk, skb, false); 2035 } 2036 2037 static inline void sk_rx_queue_clear(struct sock *sk) 2038 { 2039 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2040 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2041 #endif 2042 } 2043 2044 static inline int sk_rx_queue_get(const struct sock *sk) 2045 { 2046 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2047 if (sk) { 2048 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2049 2050 if (res != NO_QUEUE_MAPPING) 2051 return res; 2052 } 2053 #endif 2054 2055 return -1; 2056 } 2057 2058 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2059 { 2060 sk->sk_socket = sock; 2061 if (sock) { 2062 WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid); 2063 WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino); 2064 } else { 2065 /* Note: sk_uid is unchanged. */ 2066 WRITE_ONCE(sk->sk_ino, 0); 2067 } 2068 } 2069 2070 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2071 { 2072 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2073 return &rcu_dereference_raw(sk->sk_wq)->wait; 2074 } 2075 /* Detach socket from process context. 2076 * Announce socket dead, detach it from wait queue and inode. 2077 * Note that parent inode held reference count on this struct sock, 2078 * we do not release it in this function, because protocol 2079 * probably wants some additional cleanups or even continuing 2080 * to work with this socket (TCP). 2081 */ 2082 static inline void sock_orphan(struct sock *sk) 2083 { 2084 write_lock_bh(&sk->sk_callback_lock); 2085 sock_set_flag(sk, SOCK_DEAD); 2086 sk_set_socket(sk, NULL); 2087 sk->sk_wq = NULL; 2088 write_unlock_bh(&sk->sk_callback_lock); 2089 } 2090 2091 static inline void sock_graft(struct sock *sk, struct socket *parent) 2092 { 2093 WARN_ON(parent->sk); 2094 write_lock_bh(&sk->sk_callback_lock); 2095 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2096 parent->sk = sk; 2097 sk_set_socket(sk, parent); 2098 security_sock_graft(sk, parent); 2099 write_unlock_bh(&sk->sk_callback_lock); 2100 } 2101 2102 static inline unsigned long sock_i_ino(const struct sock *sk) 2103 { 2104 /* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */ 2105 return READ_ONCE(sk->sk_ino); 2106 } 2107 2108 static inline kuid_t sk_uid(const struct sock *sk) 2109 { 2110 /* Paired with WRITE_ONCE() in sockfs_setattr() */ 2111 return READ_ONCE(sk->sk_uid); 2112 } 2113 2114 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2115 { 2116 return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0); 2117 } 2118 2119 static inline u32 net_tx_rndhash(void) 2120 { 2121 u32 v = get_random_u32(); 2122 2123 return v ?: 1; 2124 } 2125 2126 static inline void sk_set_txhash(struct sock *sk) 2127 { 2128 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2129 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2130 } 2131 2132 static inline bool sk_rethink_txhash(struct sock *sk) 2133 { 2134 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2135 sk_set_txhash(sk); 2136 return true; 2137 } 2138 return false; 2139 } 2140 2141 static inline struct dst_entry * 2142 __sk_dst_get(const struct sock *sk) 2143 { 2144 return rcu_dereference_check(sk->sk_dst_cache, 2145 lockdep_sock_is_held(sk)); 2146 } 2147 2148 static inline struct dst_entry * 2149 sk_dst_get(const struct sock *sk) 2150 { 2151 struct dst_entry *dst; 2152 2153 rcu_read_lock(); 2154 dst = rcu_dereference(sk->sk_dst_cache); 2155 if (dst && !rcuref_get(&dst->__rcuref)) 2156 dst = NULL; 2157 rcu_read_unlock(); 2158 return dst; 2159 } 2160 2161 static inline void __dst_negative_advice(struct sock *sk) 2162 { 2163 struct dst_entry *dst = __sk_dst_get(sk); 2164 2165 if (dst && dst->ops->negative_advice) 2166 dst->ops->negative_advice(sk, dst); 2167 } 2168 2169 static inline void dst_negative_advice(struct sock *sk) 2170 { 2171 sk_rethink_txhash(sk); 2172 __dst_negative_advice(sk); 2173 } 2174 2175 static inline void 2176 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2177 { 2178 struct dst_entry *old_dst; 2179 2180 sk_tx_queue_clear(sk); 2181 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2182 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2183 lockdep_sock_is_held(sk)); 2184 rcu_assign_pointer(sk->sk_dst_cache, dst); 2185 dst_release(old_dst); 2186 } 2187 2188 static inline void 2189 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2190 { 2191 struct dst_entry *old_dst; 2192 2193 sk_tx_queue_clear(sk); 2194 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2195 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst))); 2196 dst_release(old_dst); 2197 } 2198 2199 static inline void 2200 __sk_dst_reset(struct sock *sk) 2201 { 2202 __sk_dst_set(sk, NULL); 2203 } 2204 2205 static inline void 2206 sk_dst_reset(struct sock *sk) 2207 { 2208 sk_dst_set(sk, NULL); 2209 } 2210 2211 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2212 2213 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2214 2215 static inline void sk_dst_confirm(struct sock *sk) 2216 { 2217 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2218 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2219 } 2220 2221 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2222 { 2223 if (skb_get_dst_pending_confirm(skb)) { 2224 struct sock *sk = skb->sk; 2225 2226 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2227 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2228 neigh_confirm(n); 2229 } 2230 } 2231 2232 bool sk_mc_loop(const struct sock *sk); 2233 2234 static inline bool sk_can_gso(const struct sock *sk) 2235 { 2236 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2237 } 2238 2239 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2240 2241 static inline void sk_gso_disable(struct sock *sk) 2242 { 2243 sk->sk_gso_disabled = 1; 2244 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2245 } 2246 2247 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2248 struct iov_iter *from, char *to, 2249 int copy, int offset) 2250 { 2251 if (skb->ip_summed == CHECKSUM_NONE) { 2252 __wsum csum = 0; 2253 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2254 return -EFAULT; 2255 skb->csum = csum_block_add(skb->csum, csum, offset); 2256 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2257 if (!copy_from_iter_full_nocache(to, copy, from)) 2258 return -EFAULT; 2259 } else if (!copy_from_iter_full(to, copy, from)) 2260 return -EFAULT; 2261 2262 return 0; 2263 } 2264 2265 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2266 struct iov_iter *from, int copy) 2267 { 2268 int err, offset = skb->len; 2269 2270 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2271 copy, offset); 2272 if (err) 2273 __skb_trim(skb, offset); 2274 2275 return err; 2276 } 2277 2278 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2279 struct sk_buff *skb, 2280 struct page *page, 2281 int off, int copy) 2282 { 2283 int err; 2284 2285 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2286 copy, skb->len); 2287 if (err) 2288 return err; 2289 2290 skb_len_add(skb, copy); 2291 sk_wmem_queued_add(sk, copy); 2292 sk_mem_charge(sk, copy); 2293 return 0; 2294 } 2295 2296 /** 2297 * sk_wmem_alloc_get - returns write allocations 2298 * @sk: socket 2299 * 2300 * Return: sk_wmem_alloc minus initial offset of one 2301 */ 2302 static inline int sk_wmem_alloc_get(const struct sock *sk) 2303 { 2304 return refcount_read(&sk->sk_wmem_alloc) - 1; 2305 } 2306 2307 /** 2308 * sk_rmem_alloc_get - returns read allocations 2309 * @sk: socket 2310 * 2311 * Return: sk_rmem_alloc 2312 */ 2313 static inline int sk_rmem_alloc_get(const struct sock *sk) 2314 { 2315 return atomic_read(&sk->sk_rmem_alloc); 2316 } 2317 2318 /** 2319 * sk_has_allocations - check if allocations are outstanding 2320 * @sk: socket 2321 * 2322 * Return: true if socket has write or read allocations 2323 */ 2324 static inline bool sk_has_allocations(const struct sock *sk) 2325 { 2326 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2327 } 2328 2329 /** 2330 * skwq_has_sleeper - check if there are any waiting processes 2331 * @wq: struct socket_wq 2332 * 2333 * Return: true if socket_wq has waiting processes 2334 * 2335 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2336 * barrier call. They were added due to the race found within the tcp code. 2337 * 2338 * Consider following tcp code paths:: 2339 * 2340 * CPU1 CPU2 2341 * sys_select receive packet 2342 * ... ... 2343 * __add_wait_queue update tp->rcv_nxt 2344 * ... ... 2345 * tp->rcv_nxt check sock_def_readable 2346 * ... { 2347 * schedule rcu_read_lock(); 2348 * wq = rcu_dereference(sk->sk_wq); 2349 * if (wq && waitqueue_active(&wq->wait)) 2350 * wake_up_interruptible(&wq->wait) 2351 * ... 2352 * } 2353 * 2354 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2355 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2356 * could then endup calling schedule and sleep forever if there are no more 2357 * data on the socket. 2358 * 2359 */ 2360 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2361 { 2362 return wq && wq_has_sleeper(&wq->wait); 2363 } 2364 2365 /** 2366 * sock_poll_wait - wrapper for the poll_wait call. 2367 * @filp: file 2368 * @sock: socket to wait on 2369 * @p: poll_table 2370 * 2371 * See the comments in the wq_has_sleeper function. 2372 */ 2373 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2374 poll_table *p) 2375 { 2376 /* Provides a barrier we need to be sure we are in sync 2377 * with the socket flags modification. 2378 * 2379 * This memory barrier is paired in the wq_has_sleeper. 2380 */ 2381 poll_wait(filp, &sock->wq.wait, p); 2382 } 2383 2384 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2385 { 2386 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2387 u32 txhash = READ_ONCE(sk->sk_txhash); 2388 2389 if (txhash) { 2390 skb->l4_hash = 1; 2391 skb->hash = txhash; 2392 } 2393 } 2394 2395 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2396 2397 /* 2398 * Queue a received datagram if it will fit. Stream and sequenced 2399 * protocols can't normally use this as they need to fit buffers in 2400 * and play with them. 2401 * 2402 * Inlined as it's very short and called for pretty much every 2403 * packet ever received. 2404 */ 2405 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2406 { 2407 skb_orphan(skb); 2408 skb->sk = sk; 2409 skb->destructor = sock_rfree; 2410 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2411 sk_mem_charge(sk, skb->truesize); 2412 } 2413 2414 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2415 { 2416 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2417 skb_orphan(skb); 2418 skb->destructor = sock_efree; 2419 skb->sk = sk; 2420 return true; 2421 } 2422 return false; 2423 } 2424 2425 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2426 { 2427 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2428 if (skb) { 2429 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2430 skb_set_owner_r(skb, sk); 2431 return skb; 2432 } 2433 __kfree_skb(skb); 2434 } 2435 return NULL; 2436 } 2437 2438 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2439 { 2440 if (skb->destructor != sock_wfree) { 2441 skb_orphan(skb); 2442 return; 2443 } 2444 skb->slow_gro = 1; 2445 } 2446 2447 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2448 unsigned long expires); 2449 2450 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2451 2452 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2453 2454 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2455 struct sk_buff *skb, unsigned int flags, 2456 void (*destructor)(struct sock *sk, 2457 struct sk_buff *skb)); 2458 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2459 2460 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2461 enum skb_drop_reason *reason); 2462 2463 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2464 { 2465 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2466 } 2467 2468 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2469 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2470 2471 /* 2472 * Recover an error report and clear atomically 2473 */ 2474 2475 static inline int sock_error(struct sock *sk) 2476 { 2477 int err; 2478 2479 /* Avoid an atomic operation for the common case. 2480 * This is racy since another cpu/thread can change sk_err under us. 2481 */ 2482 if (likely(data_race(!sk->sk_err))) 2483 return 0; 2484 2485 err = xchg(&sk->sk_err, 0); 2486 return -err; 2487 } 2488 2489 void sk_error_report(struct sock *sk); 2490 2491 static inline unsigned long sock_wspace(struct sock *sk) 2492 { 2493 int amt = 0; 2494 2495 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2496 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2497 if (amt < 0) 2498 amt = 0; 2499 } 2500 return amt; 2501 } 2502 2503 /* Note: 2504 * We use sk->sk_wq_raw, from contexts knowing this 2505 * pointer is not NULL and cannot disappear/change. 2506 */ 2507 static inline void sk_set_bit(int nr, struct sock *sk) 2508 { 2509 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2510 !sock_flag(sk, SOCK_FASYNC)) 2511 return; 2512 2513 set_bit(nr, &sk->sk_wq_raw->flags); 2514 } 2515 2516 static inline void sk_clear_bit(int nr, struct sock *sk) 2517 { 2518 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2519 !sock_flag(sk, SOCK_FASYNC)) 2520 return; 2521 2522 clear_bit(nr, &sk->sk_wq_raw->flags); 2523 } 2524 2525 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2526 { 2527 if (sock_flag(sk, SOCK_FASYNC)) { 2528 rcu_read_lock(); 2529 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2530 rcu_read_unlock(); 2531 } 2532 } 2533 2534 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band) 2535 { 2536 if (unlikely(sock_flag(sk, SOCK_FASYNC))) 2537 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2538 } 2539 2540 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2541 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2542 * Note: for send buffers, TCP works better if we can build two skbs at 2543 * minimum. 2544 */ 2545 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2546 2547 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2548 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2549 2550 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2551 { 2552 u32 val; 2553 2554 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2555 return; 2556 2557 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2558 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2559 2560 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2561 } 2562 2563 /** 2564 * sk_page_frag - return an appropriate page_frag 2565 * @sk: socket 2566 * 2567 * Use the per task page_frag instead of the per socket one for 2568 * optimization when we know that we're in process context and own 2569 * everything that's associated with %current. 2570 * 2571 * Both direct reclaim and page faults can nest inside other 2572 * socket operations and end up recursing into sk_page_frag() 2573 * while it's already in use: explicitly avoid task page_frag 2574 * when users disable sk_use_task_frag. 2575 * 2576 * Return: a per task page_frag if context allows that, 2577 * otherwise a per socket one. 2578 */ 2579 static inline struct page_frag *sk_page_frag(struct sock *sk) 2580 { 2581 if (sk->sk_use_task_frag) 2582 return ¤t->task_frag; 2583 2584 return &sk->sk_frag; 2585 } 2586 2587 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2588 2589 /* 2590 * Default write policy as shown to user space via poll/select/SIGIO 2591 */ 2592 static inline bool sock_writeable(const struct sock *sk) 2593 { 2594 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1); 2595 } 2596 2597 static inline gfp_t gfp_any(void) 2598 { 2599 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2600 } 2601 2602 static inline gfp_t gfp_memcg_charge(void) 2603 { 2604 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2605 } 2606 2607 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2608 { 2609 return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo); 2610 } 2611 2612 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2613 { 2614 return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo); 2615 } 2616 2617 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2618 { 2619 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2620 2621 return v ?: 1; 2622 } 2623 2624 /* Alas, with timeout socket operations are not restartable. 2625 * Compare this to poll(). 2626 */ 2627 static inline int sock_intr_errno(long timeo) 2628 { 2629 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2630 } 2631 2632 struct sock_skb_cb { 2633 u32 dropcount; 2634 }; 2635 2636 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2637 * using skb->cb[] would keep using it directly and utilize its 2638 * alignment guarantee. 2639 */ 2640 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \ 2641 sizeof(struct sock_skb_cb)) 2642 2643 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2644 SOCK_SKB_CB_OFFSET)) 2645 2646 #define sock_skb_cb_check_size(size) \ 2647 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2648 2649 static inline void 2650 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2651 { 2652 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2653 atomic_read(&sk->sk_drops) : 0; 2654 } 2655 2656 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb) 2657 { 2658 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2659 2660 atomic_add(segs, &sk->sk_drops); 2661 } 2662 2663 static inline ktime_t sock_read_timestamp(struct sock *sk) 2664 { 2665 #if BITS_PER_LONG==32 2666 unsigned int seq; 2667 ktime_t kt; 2668 2669 do { 2670 seq = read_seqbegin(&sk->sk_stamp_seq); 2671 kt = sk->sk_stamp; 2672 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2673 2674 return kt; 2675 #else 2676 return READ_ONCE(sk->sk_stamp); 2677 #endif 2678 } 2679 2680 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2681 { 2682 #if BITS_PER_LONG==32 2683 write_seqlock(&sk->sk_stamp_seq); 2684 sk->sk_stamp = kt; 2685 write_sequnlock(&sk->sk_stamp_seq); 2686 #else 2687 WRITE_ONCE(sk->sk_stamp, kt); 2688 #endif 2689 } 2690 2691 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2692 struct sk_buff *skb); 2693 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2694 struct sk_buff *skb); 2695 2696 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk); 2697 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk, 2698 struct timespec64 *ts); 2699 2700 static inline void 2701 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2702 { 2703 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2704 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2705 ktime_t kt = skb->tstamp; 2706 /* 2707 * generate control messages if 2708 * - receive time stamping in software requested 2709 * - software time stamp available and wanted 2710 * - hardware time stamps available and wanted 2711 */ 2712 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2713 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2714 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2715 (hwtstamps->hwtstamp && 2716 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2717 __sock_recv_timestamp(msg, sk, skb); 2718 else 2719 sock_write_timestamp(sk, kt); 2720 2721 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2722 __sock_recv_wifi_status(msg, sk, skb); 2723 } 2724 2725 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2726 struct sk_buff *skb); 2727 2728 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2729 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2730 struct sk_buff *skb) 2731 { 2732 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2733 (1UL << SOCK_RCVTSTAMP) | \ 2734 (1UL << SOCK_RCVMARK) | \ 2735 (1UL << SOCK_RCVPRIORITY) | \ 2736 (1UL << SOCK_TIMESTAMPING_ANY)) 2737 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2738 SOF_TIMESTAMPING_RAW_HARDWARE) 2739 2740 if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS) 2741 __sock_recv_cmsgs(msg, sk, skb); 2742 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2743 sock_write_timestamp(sk, skb->tstamp); 2744 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2745 sock_write_timestamp(sk, 0); 2746 } 2747 2748 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags); 2749 2750 /** 2751 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2752 * @sk: socket sending this packet 2753 * @sockc: pointer to socket cmsg cookie to get timestamping info 2754 * @tx_flags: completed with instructions for time stamping 2755 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2756 * 2757 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2758 */ 2759 static inline void _sock_tx_timestamp(struct sock *sk, 2760 const struct sockcm_cookie *sockc, 2761 __u8 *tx_flags, __u32 *tskey) 2762 { 2763 __u32 tsflags = sockc->tsflags; 2764 2765 if (unlikely(tsflags)) { 2766 __sock_tx_timestamp(tsflags, tx_flags); 2767 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2768 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) { 2769 if (tsflags & SOCKCM_FLAG_TS_OPT_ID) 2770 *tskey = sockc->ts_opt_id; 2771 else 2772 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2773 } 2774 } 2775 } 2776 2777 static inline void sock_tx_timestamp(struct sock *sk, 2778 const struct sockcm_cookie *sockc, 2779 __u8 *tx_flags) 2780 { 2781 _sock_tx_timestamp(sk, sockc, tx_flags, NULL); 2782 } 2783 2784 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, 2785 const struct sockcm_cookie *sockc) 2786 { 2787 _sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags, 2788 &skb_shinfo(skb)->tskey); 2789 } 2790 2791 static inline bool sk_is_inet(const struct sock *sk) 2792 { 2793 int family = READ_ONCE(sk->sk_family); 2794 2795 return family == AF_INET || family == AF_INET6; 2796 } 2797 2798 static inline bool sk_is_tcp(const struct sock *sk) 2799 { 2800 return sk_is_inet(sk) && 2801 sk->sk_type == SOCK_STREAM && 2802 sk->sk_protocol == IPPROTO_TCP; 2803 } 2804 2805 static inline bool sk_is_udp(const struct sock *sk) 2806 { 2807 return sk_is_inet(sk) && 2808 sk->sk_type == SOCK_DGRAM && 2809 sk->sk_protocol == IPPROTO_UDP; 2810 } 2811 2812 static inline bool sk_is_unix(const struct sock *sk) 2813 { 2814 return sk->sk_family == AF_UNIX; 2815 } 2816 2817 static inline bool sk_is_stream_unix(const struct sock *sk) 2818 { 2819 return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM; 2820 } 2821 2822 static inline bool sk_is_vsock(const struct sock *sk) 2823 { 2824 return sk->sk_family == AF_VSOCK; 2825 } 2826 2827 static inline bool sk_may_scm_recv(const struct sock *sk) 2828 { 2829 return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) || 2830 sk->sk_family == AF_NETLINK || 2831 (IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH); 2832 } 2833 2834 /** 2835 * sk_eat_skb - Release a skb if it is no longer needed 2836 * @sk: socket to eat this skb from 2837 * @skb: socket buffer to eat 2838 * 2839 * This routine must be called with interrupts disabled or with the socket 2840 * locked so that the sk_buff queue operation is ok. 2841 */ 2842 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2843 { 2844 __skb_unlink(skb, &sk->sk_receive_queue); 2845 __kfree_skb(skb); 2846 } 2847 2848 static inline bool 2849 skb_sk_is_prefetched(struct sk_buff *skb) 2850 { 2851 #ifdef CONFIG_INET 2852 return skb->destructor == sock_pfree; 2853 #else 2854 return false; 2855 #endif /* CONFIG_INET */ 2856 } 2857 2858 /* This helper checks if a socket is a full socket, 2859 * ie _not_ a timewait or request socket. 2860 */ 2861 static inline bool sk_fullsock(const struct sock *sk) 2862 { 2863 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2864 } 2865 2866 static inline bool 2867 sk_is_refcounted(struct sock *sk) 2868 { 2869 /* Only full sockets have sk->sk_flags. */ 2870 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2871 } 2872 2873 static inline bool 2874 sk_requests_wifi_status(struct sock *sk) 2875 { 2876 return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS); 2877 } 2878 2879 /* Checks if this SKB belongs to an HW offloaded socket 2880 * and whether any SW fallbacks are required based on dev. 2881 * Check decrypted mark in case skb_orphan() cleared socket. 2882 */ 2883 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb, 2884 struct net_device *dev) 2885 { 2886 #ifdef CONFIG_SOCK_VALIDATE_XMIT 2887 struct sock *sk = skb->sk; 2888 2889 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) { 2890 skb = sk->sk_validate_xmit_skb(sk, dev, skb); 2891 } else if (unlikely(skb_is_decrypted(skb))) { 2892 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n"); 2893 kfree_skb(skb); 2894 skb = NULL; 2895 } 2896 #endif 2897 2898 return skb; 2899 } 2900 2901 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2902 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2903 */ 2904 static inline bool sk_listener(const struct sock *sk) 2905 { 2906 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2907 } 2908 2909 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT 2910 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE) 2911 * TCP RST and ACK can be attached to TIME_WAIT. 2912 */ 2913 static inline bool sk_listener_or_tw(const struct sock *sk) 2914 { 2915 return (1 << READ_ONCE(sk->sk_state)) & 2916 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT); 2917 } 2918 2919 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 2920 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 2921 int type); 2922 2923 bool sk_ns_capable(const struct sock *sk, 2924 struct user_namespace *user_ns, int cap); 2925 bool sk_capable(const struct sock *sk, int cap); 2926 bool sk_net_capable(const struct sock *sk, int cap); 2927 2928 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 2929 2930 /* Take into consideration the size of the struct sk_buff overhead in the 2931 * determination of these values, since that is non-constant across 2932 * platforms. This makes socket queueing behavior and performance 2933 * not depend upon such differences. 2934 */ 2935 #define _SK_MEM_PACKETS 256 2936 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 2937 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2938 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 2939 2940 extern __u32 sysctl_wmem_max; 2941 extern __u32 sysctl_rmem_max; 2942 2943 extern __u32 sysctl_wmem_default; 2944 extern __u32 sysctl_rmem_default; 2945 2946 #define SKB_FRAG_PAGE_ORDER get_order(32768) 2947 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 2948 2949 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 2950 { 2951 /* Does this proto have per netns sysctl_wmem ? */ 2952 if (proto->sysctl_wmem_offset) 2953 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 2954 2955 return READ_ONCE(*proto->sysctl_wmem); 2956 } 2957 2958 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 2959 { 2960 /* Does this proto have per netns sysctl_rmem ? */ 2961 if (proto->sysctl_rmem_offset) 2962 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 2963 2964 return READ_ONCE(*proto->sysctl_rmem); 2965 } 2966 2967 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 2968 * Some wifi drivers need to tweak it to get more chunks. 2969 * They can use this helper from their ndo_start_xmit() 2970 */ 2971 static inline void sk_pacing_shift_update(struct sock *sk, int val) 2972 { 2973 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 2974 return; 2975 WRITE_ONCE(sk->sk_pacing_shift, val); 2976 } 2977 2978 /* if a socket is bound to a device, check that the given device 2979 * index is either the same or that the socket is bound to an L3 2980 * master device and the given device index is also enslaved to 2981 * that L3 master 2982 */ 2983 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 2984 { 2985 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 2986 int mdif; 2987 2988 if (!bound_dev_if || bound_dev_if == dif) 2989 return true; 2990 2991 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 2992 if (mdif && mdif == bound_dev_if) 2993 return true; 2994 2995 return false; 2996 } 2997 2998 void sock_def_readable(struct sock *sk); 2999 3000 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 3001 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 3002 int sock_set_timestamping(struct sock *sk, int optname, 3003 struct so_timestamping timestamping); 3004 3005 #if defined(CONFIG_CGROUP_BPF) 3006 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op); 3007 #else 3008 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 3009 { 3010 } 3011 #endif 3012 void sock_no_linger(struct sock *sk); 3013 void sock_set_keepalive(struct sock *sk); 3014 void sock_set_priority(struct sock *sk, u32 priority); 3015 void sock_set_rcvbuf(struct sock *sk, int val); 3016 void sock_set_mark(struct sock *sk, u32 val); 3017 void sock_set_reuseaddr(struct sock *sk); 3018 void sock_set_reuseport(struct sock *sk); 3019 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3020 3021 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len); 3022 3023 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3024 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3025 sockptr_t optval, int optlen, bool old_timeval); 3026 3027 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3028 void __user *arg, void *karg, size_t size); 3029 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3030 static inline bool sk_is_readable(struct sock *sk) 3031 { 3032 const struct proto *prot = READ_ONCE(sk->sk_prot); 3033 3034 if (prot->sock_is_readable) 3035 return prot->sock_is_readable(sk); 3036 3037 return false; 3038 } 3039 #endif /* _SOCK_H */ 3040