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_bypass_prot_mem: bypass the per-protocol memory accounting for skb 122 * @skc_bound_dev_if: bound device index if != 0 123 * @skc_bind_node: bind hash linkage for various protocol lookup tables 124 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol 125 * @skc_prot: protocol handlers inside a network family 126 * @skc_net: reference to the network namespace of this socket 127 * @skc_v6_daddr: IPV6 destination address 128 * @skc_v6_rcv_saddr: IPV6 source address 129 * @skc_cookie: socket's cookie value 130 * @skc_node: main hash linkage for various protocol lookup tables 131 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol 132 * @skc_tx_queue_mapping: tx queue number for this connection 133 * @skc_rx_queue_mapping: rx queue number for this connection 134 * @skc_flags: place holder for sk_flags 135 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, 136 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings 137 * @skc_listener: connection request listener socket (aka rsk_listener) 138 * [union with @skc_flags] 139 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row 140 * [union with @skc_flags] 141 * @skc_incoming_cpu: record/match cpu processing incoming packets 142 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled) 143 * [union with @skc_incoming_cpu] 144 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number 145 * [union with @skc_incoming_cpu] 146 * @skc_refcnt: reference count 147 * 148 * This is the minimal network layer representation of sockets, the header 149 * for struct sock and struct inet_timewait_sock. 150 */ 151 struct sock_common { 152 union { 153 __addrpair skc_addrpair; 154 struct { 155 __be32 skc_daddr; 156 __be32 skc_rcv_saddr; 157 }; 158 }; 159 union { 160 unsigned int skc_hash; 161 __u16 skc_u16hashes[2]; 162 }; 163 /* skc_dport && skc_num must be grouped as well */ 164 union { 165 __portpair skc_portpair; 166 struct { 167 __be16 skc_dport; 168 __u16 skc_num; 169 }; 170 }; 171 172 unsigned short skc_family; 173 volatile unsigned char skc_state; 174 unsigned char skc_reuse:4; 175 unsigned char skc_reuseport:1; 176 unsigned char skc_ipv6only:1; 177 unsigned char skc_net_refcnt:1; 178 unsigned char skc_bypass_prot_mem:1; 179 int skc_bound_dev_if; 180 union { 181 struct hlist_node skc_bind_node; 182 struct hlist_node skc_portaddr_node; 183 }; 184 struct proto *skc_prot; 185 possible_net_t skc_net; 186 187 #if IS_ENABLED(CONFIG_IPV6) 188 struct in6_addr skc_v6_daddr; 189 struct in6_addr skc_v6_rcv_saddr; 190 #endif 191 192 atomic64_t skc_cookie; 193 194 /* following fields are padding to force 195 * offset(struct sock, sk_refcnt) == 128 on 64bit arches 196 * assuming IPV6 is enabled. We use this padding differently 197 * for different kind of 'sockets' 198 */ 199 union { 200 unsigned long skc_flags; 201 struct sock *skc_listener; /* request_sock */ 202 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */ 203 }; 204 /* 205 * fields between dontcopy_begin/dontcopy_end 206 * are not copied in sock_copy() 207 */ 208 /* private: */ 209 int skc_dontcopy_begin[0]; 210 /* public: */ 211 union { 212 struct hlist_node skc_node; 213 struct hlist_nulls_node skc_nulls_node; 214 }; 215 unsigned short skc_tx_queue_mapping; 216 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 217 unsigned short skc_rx_queue_mapping; 218 #endif 219 union { 220 int skc_incoming_cpu; 221 u32 skc_rcv_wnd; 222 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */ 223 }; 224 225 refcount_t skc_refcnt; 226 /* private: */ 227 int skc_dontcopy_end[0]; 228 union { 229 u32 skc_rxhash; 230 u32 skc_window_clamp; 231 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */ 232 }; 233 /* public: */ 234 }; 235 236 struct bpf_local_storage; 237 struct sk_filter; 238 239 /** 240 * struct sock - network layer representation of sockets 241 * @__sk_common: shared layout with inet_timewait_sock 242 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN 243 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings 244 * @sk_lock: synchronizer 245 * @sk_kern_sock: True if sock is using kernel lock classes 246 * @sk_rcvbuf: size of receive buffer in bytes 247 * @sk_wq: sock wait queue and async head 248 * @sk_rx_dst: receive input route used by early demux 249 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst 250 * @sk_rx_dst_cookie: cookie for @sk_rx_dst 251 * @sk_dst_cache: destination cache 252 * @sk_dst_pending_confirm: need to confirm neighbour 253 * @sk_policy: flow policy 254 * @psp_assoc: PSP association, if socket is PSP-secured 255 * @sk_receive_queue: incoming packets 256 * @sk_wmem_alloc: transmit queue bytes committed 257 * @sk_tsq_flags: TCP Small Queues flags 258 * @sk_write_queue: Packet sending queue 259 * @sk_omem_alloc: "o" is "option" or "other" 260 * @sk_wmem_queued: persistent queue size 261 * @sk_forward_alloc: space allocated forward 262 * @sk_reserved_mem: space reserved and non-reclaimable for the socket 263 * @sk_napi_id: id of the last napi context to receive data for sk 264 * @sk_ll_usec: usecs to busypoll when there is no data 265 * @sk_allocation: allocation mode 266 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler) 267 * @sk_pacing_status: Pacing status (requested, handled by sch_fq) 268 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE) 269 * @sk_sndbuf: size of send buffer in bytes 270 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets 271 * @sk_no_check_rx: allow zero checksum in RX packets 272 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO) 273 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden. 274 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4) 275 * @sk_gso_max_size: Maximum GSO segment size to build 276 * @sk_gso_max_segs: Maximum number of GSO segments 277 * @sk_pacing_shift: scaling factor for TCP Small Queues 278 * @sk_lingertime: %SO_LINGER l_linger setting 279 * @sk_backlog: always used with the per-socket spinlock held 280 * @sk_callback_lock: used with the callbacks in the end of this struct 281 * @sk_error_queue: rarely used 282 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, 283 * IPV6_ADDRFORM for instance) 284 * @sk_err: last error 285 * @sk_err_soft: errors that don't cause failure but are the cause of a 286 * persistent failure not just 'timed out' 287 * @sk_drops: raw/udp drops counter 288 * @sk_drop_counters: optional pointer to numa_drop_counters 289 * @sk_ack_backlog: current listen backlog 290 * @sk_max_ack_backlog: listen backlog set in listen() 291 * @sk_uid: user id of owner 292 * @sk_ino: inode number (zero if orphaned) 293 * @sk_prefer_busy_poll: prefer busypolling over softirq processing 294 * @sk_busy_poll_budget: napi processing budget when busypolling 295 * @sk_priority: %SO_PRIORITY setting 296 * @sk_type: socket type (%SOCK_STREAM, etc) 297 * @sk_protocol: which protocol this socket belongs in this network family 298 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred 299 * @sk_peer_pid: &struct pid for this socket's peer 300 * @sk_peer_cred: %SO_PEERCRED setting 301 * @sk_rcvlowat: %SO_RCVLOWAT setting 302 * @sk_rcvtimeo: %SO_RCVTIMEO setting 303 * @sk_sndtimeo: %SO_SNDTIMEO setting 304 * @sk_txhash: computed flow hash for use on transmit 305 * @sk_txrehash: enable TX hash rethink 306 * @sk_filter: socket filtering instructions 307 * @sk_timer: sock cleanup timer 308 * @sk_stamp: time stamp of last packet received 309 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only 310 * @sk_tsflags: SO_TIMESTAMPING flags 311 * @sk_bpf_cb_flags: used in bpf_setsockopt() 312 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag. 313 * Sockets that can be used under memory reclaim should 314 * set this to false. 315 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock 316 * for timestamping 317 * @sk_tskey: counter to disambiguate concurrent tstamp requests 318 * @sk_tx_queue_mapping_jiffies: time in jiffies of last @sk_tx_queue_mapping refresh. 319 * @sk_zckey: counter to order MSG_ZEROCOPY notifications 320 * @sk_socket: Identd and reporting IO signals 321 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock. 322 * @sk_frag: cached page frag 323 * @sk_peek_off: current peek_offset value 324 * @sk_send_head: front of stuff to transmit 325 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head] 326 * @sk_security: used by security modules 327 * @sk_mark: generic packet mark 328 * @sk_cgrp_data: cgroup data for this cgroup 329 * @sk_memcg: this socket's memory cgroup association 330 * @sk_write_pending: a write to stream socket waits to start 331 * @sk_disconnects: number of disconnect operations performed on this sock 332 * @sk_state_change: callback to indicate change in the state of the sock 333 * @sk_data_ready: callback to indicate there is data to be processed 334 * @sk_write_space: callback to indicate there is bf sending space available 335 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE) 336 * @sk_backlog_rcv: callback to process the backlog 337 * @sk_validate_xmit_skb: ptr to an optional validate function 338 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0 339 * @sk_reuseport_cb: reuseport group container 340 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage 341 * @sk_rcu: used during RCU grace period 342 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME) 343 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME 344 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME 345 * @sk_txtime_unused: unused txtime flags 346 * @sk_scm_recv_flags: all flags used by scm_recv() 347 * @sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS 348 * @sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY 349 * @sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD 350 * @sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS 351 * @sk_scm_unused: unused flags for scm_recv() 352 * @ns_tracker: tracker for netns reference 353 * @sk_user_frags: xarray of pages the user is holding a reference on. 354 * @sk_owner: reference to the real owner of the socket that calls 355 * sock_lock_init_class_and_name(). 356 */ 357 struct sock { 358 /* 359 * Now struct inet_timewait_sock also uses sock_common, so please just 360 * don't add nothing before this first member (__sk_common) --acme 361 */ 362 struct sock_common __sk_common; 363 #define sk_node __sk_common.skc_node 364 #define sk_nulls_node __sk_common.skc_nulls_node 365 #define sk_refcnt __sk_common.skc_refcnt 366 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping 367 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 368 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping 369 #endif 370 371 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin 372 #define sk_dontcopy_end __sk_common.skc_dontcopy_end 373 #define sk_hash __sk_common.skc_hash 374 #define sk_portpair __sk_common.skc_portpair 375 #define sk_num __sk_common.skc_num 376 #define sk_dport __sk_common.skc_dport 377 #define sk_addrpair __sk_common.skc_addrpair 378 #define sk_daddr __sk_common.skc_daddr 379 #define sk_rcv_saddr __sk_common.skc_rcv_saddr 380 #define sk_family __sk_common.skc_family 381 #define sk_state __sk_common.skc_state 382 #define sk_reuse __sk_common.skc_reuse 383 #define sk_reuseport __sk_common.skc_reuseport 384 #define sk_ipv6only __sk_common.skc_ipv6only 385 #define sk_net_refcnt __sk_common.skc_net_refcnt 386 #define sk_bypass_prot_mem __sk_common.skc_bypass_prot_mem 387 #define sk_bound_dev_if __sk_common.skc_bound_dev_if 388 #define sk_bind_node __sk_common.skc_bind_node 389 #define sk_prot __sk_common.skc_prot 390 #define sk_net __sk_common.skc_net 391 #define sk_v6_daddr __sk_common.skc_v6_daddr 392 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr 393 #define sk_cookie __sk_common.skc_cookie 394 #define sk_incoming_cpu __sk_common.skc_incoming_cpu 395 #define sk_flags __sk_common.skc_flags 396 #define sk_rxhash __sk_common.skc_rxhash 397 398 __cacheline_group_begin(sock_write_rx); 399 400 atomic_t sk_drops; 401 __s32 sk_peek_off; 402 struct sk_buff_head sk_error_queue; 403 struct sk_buff_head sk_receive_queue; 404 /* 405 * The backlog queue is special, it is always used with 406 * the per-socket spinlock held and requires low latency 407 * access. Therefore we special case it's implementation. 408 * Note : rmem_alloc is in this structure to fill a hole 409 * on 64bit arches, not because its logically part of 410 * backlog. 411 */ 412 struct { 413 atomic_t rmem_alloc; 414 int len; 415 struct sk_buff *head; 416 struct sk_buff *tail; 417 } sk_backlog; 418 #define sk_rmem_alloc sk_backlog.rmem_alloc 419 420 __cacheline_group_end(sock_write_rx); 421 422 __cacheline_group_begin(sock_read_rx); 423 /* early demux fields */ 424 struct dst_entry __rcu *sk_rx_dst; 425 int sk_rx_dst_ifindex; 426 u32 sk_rx_dst_cookie; 427 428 #ifdef CONFIG_NET_RX_BUSY_POLL 429 unsigned int sk_ll_usec; 430 unsigned int sk_napi_id; 431 u16 sk_busy_poll_budget; 432 u8 sk_prefer_busy_poll; 433 #endif 434 u8 sk_userlocks; 435 int sk_rcvbuf; 436 437 struct sk_filter __rcu *sk_filter; 438 union { 439 struct socket_wq __rcu *sk_wq; 440 /* private: */ 441 struct socket_wq *sk_wq_raw; 442 /* public: */ 443 }; 444 445 void (*sk_data_ready)(struct sock *sk); 446 long sk_rcvtimeo; 447 int sk_rcvlowat; 448 __cacheline_group_end(sock_read_rx); 449 450 __cacheline_group_begin(sock_read_rxtx); 451 int sk_err; 452 struct socket *sk_socket; 453 #ifdef CONFIG_MEMCG 454 struct mem_cgroup *sk_memcg; 455 #endif 456 #ifdef CONFIG_XFRM 457 struct xfrm_policy __rcu *sk_policy[2]; 458 #endif 459 #if IS_ENABLED(CONFIG_INET_PSP) 460 struct psp_assoc __rcu *psp_assoc; 461 #endif 462 __cacheline_group_end(sock_read_rxtx); 463 464 __cacheline_group_begin(sock_write_rxtx); 465 socket_lock_t sk_lock; 466 u32 sk_reserved_mem; 467 int sk_forward_alloc; 468 u32 sk_tsflags; 469 __cacheline_group_end(sock_write_rxtx); 470 471 __cacheline_group_begin(sock_write_tx); 472 int sk_write_pending; 473 atomic_t sk_omem_alloc; 474 int sk_err_soft; 475 476 int sk_wmem_queued; 477 refcount_t sk_wmem_alloc; 478 unsigned long sk_tsq_flags; 479 union { 480 struct sk_buff *sk_send_head; 481 struct rb_root tcp_rtx_queue; 482 }; 483 struct sk_buff_head sk_write_queue; 484 u32 sk_dst_pending_confirm; 485 u32 sk_pacing_status; /* see enum sk_pacing */ 486 struct page_frag sk_frag; 487 struct timer_list sk_timer; 488 489 unsigned long sk_pacing_rate; /* bytes per second */ 490 atomic_t sk_zckey; 491 atomic_t sk_tskey; 492 unsigned long sk_tx_queue_mapping_jiffies; 493 __cacheline_group_end(sock_write_tx); 494 495 __cacheline_group_begin(sock_read_tx); 496 unsigned long sk_max_pacing_rate; 497 long sk_sndtimeo; 498 u32 sk_priority; 499 u32 sk_mark; 500 kuid_t sk_uid; 501 u16 sk_protocol; 502 u16 sk_type; 503 struct dst_entry __rcu *sk_dst_cache; 504 netdev_features_t sk_route_caps; 505 #ifdef CONFIG_SOCK_VALIDATE_XMIT 506 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk, 507 struct net_device *dev, 508 struct sk_buff *skb); 509 #endif 510 u16 sk_gso_type; 511 u16 sk_gso_max_segs; 512 unsigned int sk_gso_max_size; 513 gfp_t sk_allocation; 514 u32 sk_txhash; 515 int sk_sndbuf; 516 u8 sk_pacing_shift; 517 bool sk_use_task_frag; 518 __cacheline_group_end(sock_read_tx); 519 520 /* 521 * Because of non atomicity rules, all 522 * changes are protected by socket lock. 523 */ 524 u8 sk_gso_disabled : 1, 525 sk_kern_sock : 1, 526 sk_no_check_tx : 1, 527 sk_no_check_rx : 1; 528 u8 sk_shutdown; 529 unsigned long sk_lingertime; 530 struct proto *sk_prot_creator; 531 rwlock_t sk_callback_lock; 532 u32 sk_ack_backlog; 533 u32 sk_max_ack_backlog; 534 unsigned long sk_ino; 535 spinlock_t sk_peer_lock; 536 int sk_bind_phc; 537 struct pid *sk_peer_pid; 538 const struct cred *sk_peer_cred; 539 540 ktime_t sk_stamp; 541 #if BITS_PER_LONG==32 542 seqlock_t sk_stamp_seq; 543 #endif 544 int sk_disconnects; 545 546 union { 547 u8 sk_txrehash; 548 u8 sk_scm_recv_flags; 549 struct { 550 u8 sk_scm_credentials : 1, 551 sk_scm_security : 1, 552 sk_scm_pidfd : 1, 553 sk_scm_rights : 1, 554 sk_scm_unused : 4; 555 }; 556 }; 557 u8 sk_clockid; 558 u8 sk_txtime_deadline_mode : 1, 559 sk_txtime_report_errors : 1, 560 sk_txtime_unused : 6; 561 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG)) 562 u8 sk_bpf_cb_flags; 563 564 void *sk_user_data; 565 #ifdef CONFIG_SECURITY 566 void *sk_security; 567 #endif 568 struct sock_cgroup_data sk_cgrp_data; 569 void (*sk_state_change)(struct sock *sk); 570 void (*sk_write_space)(struct sock *sk); 571 void (*sk_error_report)(struct sock *sk); 572 int (*sk_backlog_rcv)(struct sock *sk, 573 struct sk_buff *skb); 574 void (*sk_destruct)(struct sock *sk); 575 struct sock_reuseport __rcu *sk_reuseport_cb; 576 #ifdef CONFIG_BPF_SYSCALL 577 struct bpf_local_storage __rcu *sk_bpf_storage; 578 #endif 579 struct numa_drop_counters *sk_drop_counters; 580 struct rcu_head sk_rcu; 581 netns_tracker ns_tracker; 582 struct xarray sk_user_frags; 583 584 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 585 struct module *sk_owner; 586 #endif 587 }; 588 589 struct sock_bh_locked { 590 struct sock *sock; 591 local_lock_t bh_lock; 592 }; 593 594 enum sk_pacing { 595 SK_PACING_NONE = 0, 596 SK_PACING_NEEDED = 1, 597 SK_PACING_FQ = 2, 598 }; 599 600 /* flag bits in sk_user_data 601 * 602 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might 603 * not be suitable for copying when cloning the socket. For instance, 604 * it can point to a reference counted object. sk_user_data bottom 605 * bit is set if pointer must not be copied. 606 * 607 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is 608 * managed/owned by a BPF reuseport array. This bit should be set 609 * when sk_user_data's sk is added to the bpf's reuseport_array. 610 * 611 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in 612 * sk_user_data points to psock type. This bit should be set 613 * when sk_user_data is assigned to a psock object. 614 */ 615 #define SK_USER_DATA_NOCOPY 1UL 616 #define SK_USER_DATA_BPF 2UL 617 #define SK_USER_DATA_PSOCK 4UL 618 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\ 619 SK_USER_DATA_PSOCK) 620 621 /** 622 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied 623 * @sk: socket 624 */ 625 static inline bool sk_user_data_is_nocopy(const struct sock *sk) 626 { 627 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY); 628 } 629 630 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data))) 631 632 /** 633 * __locked_read_sk_user_data_with_flags - return the pointer 634 * only if argument flags all has been set in sk_user_data. Otherwise 635 * return NULL 636 * 637 * @sk: socket 638 * @flags: flag bits 639 * 640 * The caller must be holding sk->sk_callback_lock. 641 */ 642 static inline void * 643 __locked_read_sk_user_data_with_flags(const struct sock *sk, 644 uintptr_t flags) 645 { 646 uintptr_t sk_user_data = 647 (uintptr_t)rcu_dereference_check(__sk_user_data(sk), 648 lockdep_is_held(&sk->sk_callback_lock)); 649 650 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 651 652 if ((sk_user_data & flags) == flags) 653 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 654 return NULL; 655 } 656 657 /** 658 * __rcu_dereference_sk_user_data_with_flags - return the pointer 659 * only if argument flags all has been set in sk_user_data. Otherwise 660 * return NULL 661 * 662 * @sk: socket 663 * @flags: flag bits 664 */ 665 static inline void * 666 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk, 667 uintptr_t flags) 668 { 669 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk)); 670 671 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK); 672 673 if ((sk_user_data & flags) == flags) 674 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK); 675 return NULL; 676 } 677 678 #define rcu_dereference_sk_user_data(sk) \ 679 __rcu_dereference_sk_user_data_with_flags(sk, 0) 680 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \ 681 ({ \ 682 uintptr_t __tmp1 = (uintptr_t)(ptr), \ 683 __tmp2 = (uintptr_t)(flags); \ 684 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \ 685 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \ 686 rcu_assign_pointer(__sk_user_data((sk)), \ 687 __tmp1 | __tmp2); \ 688 }) 689 #define rcu_assign_sk_user_data(sk, ptr) \ 690 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0) 691 692 static inline 693 struct net *sock_net(const struct sock *sk) 694 { 695 return read_pnet(&sk->sk_net); 696 } 697 698 static inline 699 void sock_net_set(struct sock *sk, struct net *net) 700 { 701 write_pnet(&sk->sk_net, net); 702 } 703 704 /* 705 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK 706 * or not whether his port will be reused by someone else. SK_FORCE_REUSE 707 * on a socket means that the socket will reuse everybody else's port 708 * without looking at the other's sk_reuse value. 709 */ 710 711 #define SK_NO_REUSE 0 712 #define SK_CAN_REUSE 1 713 #define SK_FORCE_REUSE 2 714 715 int sk_set_peek_off(struct sock *sk, int val); 716 717 static inline int sk_peek_offset(const struct sock *sk, int flags) 718 { 719 if (unlikely(flags & MSG_PEEK)) { 720 return READ_ONCE(sk->sk_peek_off); 721 } 722 723 return 0; 724 } 725 726 static inline void sk_peek_offset_bwd(struct sock *sk, int val) 727 { 728 s32 off = READ_ONCE(sk->sk_peek_off); 729 730 if (unlikely(off >= 0)) { 731 off = max_t(s32, off - val, 0); 732 WRITE_ONCE(sk->sk_peek_off, off); 733 } 734 } 735 736 static inline void sk_peek_offset_fwd(struct sock *sk, int val) 737 { 738 sk_peek_offset_bwd(sk, -val); 739 } 740 741 /* 742 * Hashed lists helper routines 743 */ 744 static inline struct sock *sk_entry(const struct hlist_node *node) 745 { 746 return hlist_entry(node, struct sock, sk_node); 747 } 748 749 static inline struct sock *__sk_head(const struct hlist_head *head) 750 { 751 return hlist_entry(head->first, struct sock, sk_node); 752 } 753 754 static inline struct sock *sk_head(const struct hlist_head *head) 755 { 756 return hlist_empty(head) ? NULL : __sk_head(head); 757 } 758 759 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head) 760 { 761 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node); 762 } 763 764 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head) 765 { 766 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head); 767 } 768 769 static inline struct sock *sk_next(const struct sock *sk) 770 { 771 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node); 772 } 773 774 static inline struct sock *sk_nulls_next(const struct sock *sk) 775 { 776 return (!is_a_nulls(sk->sk_nulls_node.next)) ? 777 hlist_nulls_entry(sk->sk_nulls_node.next, 778 struct sock, sk_nulls_node) : 779 NULL; 780 } 781 782 static inline bool sk_unhashed(const struct sock *sk) 783 { 784 return hlist_unhashed(&sk->sk_node); 785 } 786 787 static inline bool sk_hashed(const struct sock *sk) 788 { 789 return !sk_unhashed(sk); 790 } 791 792 static inline void sk_node_init(struct hlist_node *node) 793 { 794 node->pprev = NULL; 795 } 796 797 static inline void __sk_del_node(struct sock *sk) 798 { 799 __hlist_del(&sk->sk_node); 800 } 801 802 /* NB: equivalent to hlist_del_init_rcu */ 803 static inline bool __sk_del_node_init(struct sock *sk) 804 { 805 if (sk_hashed(sk)) { 806 __sk_del_node(sk); 807 sk_node_init(&sk->sk_node); 808 return true; 809 } 810 return false; 811 } 812 813 /* Grab socket reference count. This operation is valid only 814 when sk is ALREADY grabbed f.e. it is found in hash table 815 or a list and the lookup is made under lock preventing hash table 816 modifications. 817 */ 818 819 static __always_inline void sock_hold(struct sock *sk) 820 { 821 refcount_inc(&sk->sk_refcnt); 822 } 823 824 /* Ungrab socket in the context, which assumes that socket refcnt 825 cannot hit zero, f.e. it is true in context of any socketcall. 826 */ 827 static __always_inline void __sock_put(struct sock *sk) 828 { 829 refcount_dec(&sk->sk_refcnt); 830 } 831 832 static inline bool sk_del_node_init(struct sock *sk) 833 { 834 bool rc = __sk_del_node_init(sk); 835 836 if (rc) 837 __sock_put(sk); 838 839 return rc; 840 } 841 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk) 842 843 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk) 844 { 845 if (sk_hashed(sk)) { 846 hlist_nulls_del_init_rcu(&sk->sk_nulls_node); 847 return true; 848 } 849 return false; 850 } 851 852 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk) 853 { 854 bool rc = __sk_nulls_del_node_init_rcu(sk); 855 856 if (rc) 857 __sock_put(sk); 858 859 return rc; 860 } 861 862 static inline bool sk_nulls_replace_node_init_rcu(struct sock *old, 863 struct sock *new) 864 { 865 if (sk_hashed(old)) { 866 hlist_nulls_replace_init_rcu(&old->sk_nulls_node, 867 &new->sk_nulls_node); 868 __sock_put(old); 869 return true; 870 } 871 872 return false; 873 } 874 875 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list) 876 { 877 hlist_add_head(&sk->sk_node, list); 878 } 879 880 static inline void sk_add_node(struct sock *sk, struct hlist_head *list) 881 { 882 sock_hold(sk); 883 __sk_add_node(sk, list); 884 } 885 886 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list) 887 { 888 sock_hold(sk); 889 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport && 890 sk->sk_family == AF_INET6) 891 hlist_add_tail_rcu(&sk->sk_node, list); 892 else 893 hlist_add_head_rcu(&sk->sk_node, list); 894 } 895 896 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list) 897 { 898 sock_hold(sk); 899 hlist_add_tail_rcu(&sk->sk_node, list); 900 } 901 902 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 903 { 904 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list); 905 } 906 907 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list) 908 { 909 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list); 910 } 911 912 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list) 913 { 914 sock_hold(sk); 915 __sk_nulls_add_node_rcu(sk, list); 916 } 917 918 static inline void __sk_del_bind_node(struct sock *sk) 919 { 920 __hlist_del(&sk->sk_bind_node); 921 } 922 923 static inline void sk_add_bind_node(struct sock *sk, 924 struct hlist_head *list) 925 { 926 hlist_add_head(&sk->sk_bind_node, list); 927 } 928 929 #define sk_for_each(__sk, list) \ 930 hlist_for_each_entry(__sk, list, sk_node) 931 #define sk_for_each_rcu(__sk, list) \ 932 hlist_for_each_entry_rcu(__sk, list, sk_node) 933 #define sk_nulls_for_each(__sk, node, list) \ 934 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node) 935 #define sk_nulls_for_each_rcu(__sk, node, list) \ 936 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node) 937 #define sk_for_each_from(__sk) \ 938 hlist_for_each_entry_from(__sk, sk_node) 939 #define sk_nulls_for_each_from(__sk, node) \ 940 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \ 941 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node) 942 #define sk_for_each_safe(__sk, tmp, list) \ 943 hlist_for_each_entry_safe(__sk, tmp, list, sk_node) 944 #define sk_for_each_bound(__sk, list) \ 945 hlist_for_each_entry(__sk, list, sk_bind_node) 946 #define sk_for_each_bound_safe(__sk, tmp, list) \ 947 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node) 948 949 /** 950 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset 951 * @tpos: the type * to use as a loop cursor. 952 * @pos: the &struct hlist_node to use as a loop cursor. 953 * @head: the head for your list. 954 * @offset: offset of hlist_node within the struct. 955 * 956 */ 957 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \ 958 for (pos = rcu_dereference(hlist_first_rcu(head)); \ 959 pos != NULL && \ 960 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \ 961 pos = rcu_dereference(hlist_next_rcu(pos))) 962 963 static inline struct user_namespace *sk_user_ns(const struct sock *sk) 964 { 965 /* Careful only use this in a context where these parameters 966 * can not change and must all be valid, such as recvmsg from 967 * userspace. 968 */ 969 return sk->sk_socket->file->f_cred->user_ns; 970 } 971 972 /* Sock flags */ 973 enum sock_flags { 974 SOCK_DEAD, 975 SOCK_DONE, 976 SOCK_URGINLINE, 977 SOCK_KEEPOPEN, 978 SOCK_LINGER, 979 SOCK_DESTROY, 980 SOCK_BROADCAST, 981 SOCK_TIMESTAMP, 982 SOCK_ZAPPED, 983 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */ 984 SOCK_DBG, /* %SO_DEBUG setting */ 985 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */ 986 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */ 987 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */ 988 SOCK_MEMALLOC, /* VM depends on this socket for swapping */ 989 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */ 990 SOCK_FASYNC, /* fasync() active */ 991 SOCK_RXQ_OVFL, 992 SOCK_ZEROCOPY, /* buffers from userspace */ 993 SOCK_WIFI_STATUS, /* push wifi status to userspace */ 994 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS. 995 * Will use last 4 bytes of packet sent from 996 * user-space instead. 997 */ 998 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */ 999 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */ 1000 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */ 1001 SOCK_TXTIME, 1002 SOCK_XDP, /* XDP is attached */ 1003 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */ 1004 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */ 1005 SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */ 1006 SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */ 1007 }; 1008 1009 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)) 1010 /* 1011 * The highest bit of sk_tsflags is reserved for kernel-internal 1012 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that 1013 * SOF_TIMESTAMPING* values do not reach this reserved area 1014 */ 1015 #define SOCKCM_FLAG_TS_OPT_ID BIT(31) 1016 1017 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk) 1018 { 1019 nsk->sk_flags = osk->sk_flags; 1020 } 1021 1022 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag) 1023 { 1024 __set_bit(flag, &sk->sk_flags); 1025 } 1026 1027 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag) 1028 { 1029 __clear_bit(flag, &sk->sk_flags); 1030 } 1031 1032 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit, 1033 int valbool) 1034 { 1035 if (valbool) 1036 sock_set_flag(sk, bit); 1037 else 1038 sock_reset_flag(sk, bit); 1039 } 1040 1041 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag) 1042 { 1043 return test_bit(flag, &sk->sk_flags); 1044 } 1045 1046 #ifdef CONFIG_NET 1047 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key); 1048 static inline int sk_memalloc_socks(void) 1049 { 1050 return static_branch_unlikely(&memalloc_socks_key); 1051 } 1052 1053 void __receive_sock(struct file *file); 1054 #else 1055 1056 static inline int sk_memalloc_socks(void) 1057 { 1058 return 0; 1059 } 1060 1061 static inline void __receive_sock(struct file *file) 1062 { } 1063 #endif 1064 1065 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask) 1066 { 1067 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC); 1068 } 1069 1070 static inline void sk_acceptq_removed(struct sock *sk) 1071 { 1072 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1); 1073 } 1074 1075 static inline void sk_acceptq_added(struct sock *sk) 1076 { 1077 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1); 1078 } 1079 1080 /* Note: If you think the test should be: 1081 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog); 1082 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.") 1083 */ 1084 static inline bool sk_acceptq_is_full(const struct sock *sk) 1085 { 1086 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog); 1087 } 1088 1089 /* 1090 * Compute minimal free write space needed to queue new packets. 1091 */ 1092 static inline int sk_stream_min_wspace(const struct sock *sk) 1093 { 1094 return READ_ONCE(sk->sk_wmem_queued) >> 1; 1095 } 1096 1097 static inline int sk_stream_wspace(const struct sock *sk) 1098 { 1099 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued); 1100 } 1101 1102 static inline void sk_wmem_queued_add(struct sock *sk, int val) 1103 { 1104 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val); 1105 } 1106 1107 static inline void sk_forward_alloc_add(struct sock *sk, int val) 1108 { 1109 /* Paired with lockless reads of sk->sk_forward_alloc */ 1110 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val); 1111 } 1112 1113 void sk_stream_write_space(struct sock *sk); 1114 1115 /* OOB backlog add */ 1116 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb) 1117 { 1118 /* dont let skb dst not refcounted, we are going to leave rcu lock */ 1119 skb_dst_force(skb); 1120 1121 if (!sk->sk_backlog.tail) 1122 WRITE_ONCE(sk->sk_backlog.head, skb); 1123 else 1124 sk->sk_backlog.tail->next = skb; 1125 1126 WRITE_ONCE(sk->sk_backlog.tail, skb); 1127 skb->next = NULL; 1128 } 1129 1130 /* 1131 * Take into account size of receive queue and backlog queue 1132 * Do not take into account this skb truesize, 1133 * to allow even a single big packet to come. 1134 */ 1135 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit) 1136 { 1137 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc); 1138 1139 return qsize > limit; 1140 } 1141 1142 /* The per-socket spinlock must be held here. */ 1143 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb, 1144 unsigned int limit) 1145 { 1146 if (sk_rcvqueues_full(sk, limit)) 1147 return -ENOBUFS; 1148 1149 /* 1150 * If the skb was allocated from pfmemalloc reserves, only 1151 * allow SOCK_MEMALLOC sockets to use it as this socket is 1152 * helping free memory 1153 */ 1154 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) 1155 return -ENOMEM; 1156 1157 __sk_add_backlog(sk, skb); 1158 sk->sk_backlog.len += skb->truesize; 1159 return 0; 1160 } 1161 1162 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb); 1163 1164 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb)); 1165 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb)); 1166 1167 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb) 1168 { 1169 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) 1170 return __sk_backlog_rcv(sk, skb); 1171 1172 return INDIRECT_CALL_INET(sk->sk_backlog_rcv, 1173 tcp_v6_do_rcv, 1174 tcp_v4_do_rcv, 1175 sk, skb); 1176 } 1177 1178 static inline void sk_incoming_cpu_update(struct sock *sk) 1179 { 1180 int cpu = raw_smp_processor_id(); 1181 1182 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu)) 1183 WRITE_ONCE(sk->sk_incoming_cpu, cpu); 1184 } 1185 1186 1187 static inline void sock_rps_save_rxhash(struct sock *sk, 1188 const struct sk_buff *skb) 1189 { 1190 #ifdef CONFIG_RPS 1191 /* The following WRITE_ONCE() is paired with the READ_ONCE() 1192 * here, and another one in sock_rps_record_flow(). 1193 */ 1194 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash)) 1195 WRITE_ONCE(sk->sk_rxhash, skb->hash); 1196 #endif 1197 } 1198 1199 static inline void sock_rps_reset_rxhash(struct sock *sk) 1200 { 1201 #ifdef CONFIG_RPS 1202 /* Paired with READ_ONCE() in sock_rps_record_flow() */ 1203 WRITE_ONCE(sk->sk_rxhash, 0); 1204 #endif 1205 } 1206 1207 #define sk_wait_event(__sk, __timeo, __condition, __wait) \ 1208 ({ int __rc, __dis = __sk->sk_disconnects; \ 1209 release_sock(__sk); \ 1210 __rc = __condition; \ 1211 if (!__rc) { \ 1212 *(__timeo) = wait_woken(__wait, \ 1213 TASK_INTERRUPTIBLE, \ 1214 *(__timeo)); \ 1215 } \ 1216 sched_annotate_sleep(); \ 1217 lock_sock(__sk); \ 1218 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \ 1219 __rc; \ 1220 }) 1221 1222 int sk_stream_wait_connect(struct sock *sk, long *timeo_p); 1223 int sk_stream_wait_memory(struct sock *sk, long *timeo_p); 1224 void sk_stream_wait_close(struct sock *sk, long timeo_p); 1225 int sk_stream_error(struct sock *sk, int flags, int err); 1226 void sk_stream_kill_queues(struct sock *sk); 1227 void sk_set_memalloc(struct sock *sk); 1228 void sk_clear_memalloc(struct sock *sk); 1229 1230 void __sk_flush_backlog(struct sock *sk); 1231 1232 static inline bool sk_flush_backlog(struct sock *sk) 1233 { 1234 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) { 1235 __sk_flush_backlog(sk); 1236 return true; 1237 } 1238 return false; 1239 } 1240 1241 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb); 1242 1243 struct request_sock_ops; 1244 struct timewait_sock_ops; 1245 struct inet_hashinfo; 1246 struct raw_hashinfo; 1247 struct smc_hashinfo; 1248 struct module; 1249 struct sk_psock; 1250 1251 /* 1252 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes 1253 * un-modified. Special care is taken when initializing object to zero. 1254 */ 1255 static inline void sk_prot_clear_nulls(struct sock *sk, int size) 1256 { 1257 if (offsetof(struct sock, sk_node.next) != 0) 1258 memset(sk, 0, offsetof(struct sock, sk_node.next)); 1259 memset(&sk->sk_node.pprev, 0, 1260 size - offsetof(struct sock, sk_node.pprev)); 1261 } 1262 1263 struct proto_accept_arg { 1264 int flags; 1265 int err; 1266 int is_empty; 1267 bool kern; 1268 }; 1269 1270 /* Networking protocol blocks we attach to sockets. 1271 * socket layer -> transport layer interface 1272 */ 1273 struct proto { 1274 void (*close)(struct sock *sk, 1275 long timeout); 1276 int (*pre_connect)(struct sock *sk, 1277 struct sockaddr_unsized *uaddr, 1278 int addr_len); 1279 int (*connect)(struct sock *sk, 1280 struct sockaddr_unsized *uaddr, 1281 int addr_len); 1282 int (*disconnect)(struct sock *sk, int flags); 1283 1284 struct sock * (*accept)(struct sock *sk, 1285 struct proto_accept_arg *arg); 1286 1287 int (*ioctl)(struct sock *sk, int cmd, 1288 int *karg); 1289 int (*init)(struct sock *sk); 1290 void (*destroy)(struct sock *sk); 1291 void (*shutdown)(struct sock *sk, int how); 1292 int (*setsockopt)(struct sock *sk, int level, 1293 int optname, sockptr_t optval, 1294 unsigned int optlen); 1295 int (*getsockopt)(struct sock *sk, int level, 1296 int optname, char __user *optval, 1297 int __user *option); 1298 void (*keepalive)(struct sock *sk, int valbool); 1299 #ifdef CONFIG_COMPAT 1300 int (*compat_ioctl)(struct sock *sk, 1301 unsigned int cmd, unsigned long arg); 1302 #endif 1303 int (*sendmsg)(struct sock *sk, struct msghdr *msg, 1304 size_t len); 1305 int (*recvmsg)(struct sock *sk, struct msghdr *msg, 1306 size_t len, int flags, int *addr_len); 1307 void (*splice_eof)(struct socket *sock); 1308 int (*bind)(struct sock *sk, 1309 struct sockaddr_unsized *addr, int addr_len); 1310 int (*bind_add)(struct sock *sk, 1311 struct sockaddr_unsized *addr, int addr_len); 1312 1313 int (*backlog_rcv) (struct sock *sk, 1314 struct sk_buff *skb); 1315 bool (*bpf_bypass_getsockopt)(int level, 1316 int optname); 1317 1318 void (*release_cb)(struct sock *sk); 1319 1320 /* Keeping track of sk's, looking them up, and port selection methods. */ 1321 int (*hash)(struct sock *sk); 1322 void (*unhash)(struct sock *sk); 1323 void (*rehash)(struct sock *sk); 1324 int (*get_port)(struct sock *sk, unsigned short snum); 1325 void (*put_port)(struct sock *sk); 1326 #ifdef CONFIG_BPF_SYSCALL 1327 int (*psock_update_sk_prot)(struct sock *sk, 1328 struct sk_psock *psock, 1329 bool restore); 1330 #endif 1331 1332 /* Keeping track of sockets in use */ 1333 #ifdef CONFIG_PROC_FS 1334 unsigned int inuse_idx; 1335 #endif 1336 1337 bool (*stream_memory_free)(const struct sock *sk, int wake); 1338 bool (*sock_is_readable)(struct sock *sk); 1339 /* Memory pressure */ 1340 void (*enter_memory_pressure)(struct sock *sk); 1341 void (*leave_memory_pressure)(struct sock *sk); 1342 atomic_long_t *memory_allocated; /* Current allocated memory. */ 1343 int __percpu *per_cpu_fw_alloc; 1344 struct percpu_counter *sockets_allocated; /* Current number of sockets. */ 1345 1346 /* 1347 * Pressure flag: try to collapse. 1348 * Technical note: it is used by multiple contexts non atomically. 1349 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes. 1350 * All the __sk_mem_schedule() is of this nature: accounting 1351 * is strict, actions are advisory and have some latency. 1352 */ 1353 unsigned long *memory_pressure; 1354 long *sysctl_mem; 1355 1356 int *sysctl_wmem; 1357 int *sysctl_rmem; 1358 u32 sysctl_wmem_offset; 1359 u32 sysctl_rmem_offset; 1360 1361 int max_header; 1362 bool no_autobind; 1363 1364 struct kmem_cache *slab; 1365 unsigned int obj_size; 1366 unsigned int ipv6_pinfo_offset; 1367 slab_flags_t slab_flags; 1368 unsigned int useroffset; /* Usercopy region offset */ 1369 unsigned int usersize; /* Usercopy region size */ 1370 1371 struct request_sock_ops *rsk_prot; 1372 struct timewait_sock_ops *twsk_prot; 1373 1374 union { 1375 struct inet_hashinfo *hashinfo; 1376 struct udp_table *udp_table; 1377 struct raw_hashinfo *raw_hash; 1378 struct smc_hashinfo *smc_hash; 1379 } h; 1380 1381 struct module *owner; 1382 1383 char name[32]; 1384 1385 struct list_head node; 1386 int (*diag_destroy)(struct sock *sk, int err); 1387 } __randomize_layout; 1388 1389 int proto_register(struct proto *prot, int alloc_slab); 1390 void proto_unregister(struct proto *prot); 1391 int sock_load_diag_module(int family, int protocol); 1392 1393 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake)); 1394 1395 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake) 1396 { 1397 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf)) 1398 return false; 1399 1400 return sk->sk_prot->stream_memory_free ? 1401 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free, 1402 tcp_stream_memory_free, sk, wake) : true; 1403 } 1404 1405 static inline bool sk_stream_memory_free(const struct sock *sk) 1406 { 1407 return __sk_stream_memory_free(sk, 0); 1408 } 1409 1410 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake) 1411 { 1412 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) && 1413 __sk_stream_memory_free(sk, wake); 1414 } 1415 1416 static inline bool sk_stream_is_writeable(const struct sock *sk) 1417 { 1418 return __sk_stream_is_writeable(sk, 0); 1419 } 1420 1421 static inline int sk_under_cgroup_hierarchy(struct sock *sk, 1422 struct cgroup *ancestor) 1423 { 1424 #ifdef CONFIG_SOCK_CGROUP_DATA 1425 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data), 1426 ancestor); 1427 #else 1428 return -ENOTSUPP; 1429 #endif 1430 } 1431 1432 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16 1433 1434 static inline void sk_sockets_allocated_dec(struct sock *sk) 1435 { 1436 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1, 1437 SK_ALLOC_PERCPU_COUNTER_BATCH); 1438 } 1439 1440 static inline void sk_sockets_allocated_inc(struct sock *sk) 1441 { 1442 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1, 1443 SK_ALLOC_PERCPU_COUNTER_BATCH); 1444 } 1445 1446 static inline u64 1447 sk_sockets_allocated_read_positive(struct sock *sk) 1448 { 1449 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated); 1450 } 1451 1452 static inline int 1453 proto_sockets_allocated_sum_positive(struct proto *prot) 1454 { 1455 return percpu_counter_sum_positive(prot->sockets_allocated); 1456 } 1457 1458 #ifdef CONFIG_PROC_FS 1459 #define PROTO_INUSE_NR 64 /* should be enough for the first time */ 1460 struct prot_inuse { 1461 int all; 1462 int val[PROTO_INUSE_NR]; 1463 }; 1464 1465 static inline void sock_prot_inuse_add(const struct net *net, 1466 const struct proto *prot, int val) 1467 { 1468 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val); 1469 } 1470 1471 static inline void sock_inuse_add(const struct net *net, int val) 1472 { 1473 this_cpu_add(net->core.prot_inuse->all, val); 1474 } 1475 1476 int sock_prot_inuse_get(struct net *net, struct proto *proto); 1477 int sock_inuse_get(struct net *net); 1478 #else 1479 static inline void sock_prot_inuse_add(const struct net *net, 1480 const struct proto *prot, int val) 1481 { 1482 } 1483 1484 static inline void sock_inuse_add(const struct net *net, int val) 1485 { 1486 } 1487 #endif 1488 1489 1490 /* With per-bucket locks this operation is not-atomic, so that 1491 * this version is not worse. 1492 */ 1493 static inline int __sk_prot_rehash(struct sock *sk) 1494 { 1495 sk->sk_prot->unhash(sk); 1496 return sk->sk_prot->hash(sk); 1497 } 1498 1499 /* About 10 seconds */ 1500 #define SOCK_DESTROY_TIME (10*HZ) 1501 1502 /* Sockets 0-1023 can't be bound to unless you are superuser */ 1503 #define PROT_SOCK 1024 1504 1505 #define SHUTDOWN_MASK 3 1506 #define RCV_SHUTDOWN 1 1507 #define SEND_SHUTDOWN 2 1508 1509 #define SOCK_BINDADDR_LOCK 4 1510 #define SOCK_BINDPORT_LOCK 8 1511 /** 1512 * define SOCK_CONNECT_BIND - &sock->sk_userlocks flag for auto-bind at connect() time 1513 */ 1514 #define SOCK_CONNECT_BIND 16 1515 1516 struct socket_alloc { 1517 struct socket socket; 1518 struct inode vfs_inode; 1519 }; 1520 1521 static inline struct socket *SOCKET_I(struct inode *inode) 1522 { 1523 return &container_of(inode, struct socket_alloc, vfs_inode)->socket; 1524 } 1525 1526 static inline struct inode *SOCK_INODE(struct socket *socket) 1527 { 1528 return &container_of(socket, struct socket_alloc, socket)->vfs_inode; 1529 } 1530 1531 /* 1532 * Functions for memory accounting 1533 */ 1534 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind); 1535 int __sk_mem_schedule(struct sock *sk, int size, int kind); 1536 void __sk_mem_reduce_allocated(struct sock *sk, int amount); 1537 void __sk_mem_reclaim(struct sock *sk, int amount); 1538 1539 #define SK_MEM_SEND 0 1540 #define SK_MEM_RECV 1 1541 1542 /* sysctl_mem values are in pages */ 1543 static inline long sk_prot_mem_limits(const struct sock *sk, int index) 1544 { 1545 return READ_ONCE(sk->sk_prot->sysctl_mem[index]); 1546 } 1547 1548 static inline int sk_mem_pages(int amt) 1549 { 1550 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT; 1551 } 1552 1553 static inline bool sk_has_account(struct sock *sk) 1554 { 1555 /* return true if protocol supports memory accounting */ 1556 return !!sk->sk_prot->memory_allocated; 1557 } 1558 1559 static inline bool sk_wmem_schedule(struct sock *sk, int size) 1560 { 1561 int delta; 1562 1563 if (!sk_has_account(sk)) 1564 return true; 1565 delta = size - sk->sk_forward_alloc; 1566 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND); 1567 } 1568 1569 static inline bool 1570 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc) 1571 { 1572 int delta; 1573 1574 if (!sk_has_account(sk)) 1575 return true; 1576 delta = size - sk->sk_forward_alloc; 1577 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) || 1578 pfmemalloc; 1579 } 1580 1581 static inline bool 1582 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size) 1583 { 1584 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb)); 1585 } 1586 1587 static inline int sk_unused_reserved_mem(const struct sock *sk) 1588 { 1589 int unused_mem; 1590 1591 if (likely(!sk->sk_reserved_mem)) 1592 return 0; 1593 1594 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued - 1595 atomic_read(&sk->sk_rmem_alloc); 1596 1597 return unused_mem > 0 ? unused_mem : 0; 1598 } 1599 1600 static inline void sk_mem_reclaim(struct sock *sk) 1601 { 1602 int reclaimable; 1603 1604 if (!sk_has_account(sk)) 1605 return; 1606 1607 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk); 1608 1609 if (reclaimable >= (int)PAGE_SIZE) 1610 __sk_mem_reclaim(sk, reclaimable); 1611 } 1612 1613 static inline void sk_mem_reclaim_final(struct sock *sk) 1614 { 1615 sk->sk_reserved_mem = 0; 1616 sk_mem_reclaim(sk); 1617 } 1618 1619 static inline void sk_mem_charge(struct sock *sk, int size) 1620 { 1621 if (!sk_has_account(sk)) 1622 return; 1623 sk_forward_alloc_add(sk, -size); 1624 } 1625 1626 static inline void sk_mem_uncharge(struct sock *sk, int size) 1627 { 1628 if (!sk_has_account(sk)) 1629 return; 1630 sk_forward_alloc_add(sk, size); 1631 sk_mem_reclaim(sk); 1632 } 1633 1634 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES) 1635 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1636 { 1637 __module_get(owner); 1638 sk->sk_owner = owner; 1639 } 1640 1641 static inline void sk_owner_clear(struct sock *sk) 1642 { 1643 sk->sk_owner = NULL; 1644 } 1645 1646 static inline void sk_owner_put(struct sock *sk) 1647 { 1648 module_put(sk->sk_owner); 1649 } 1650 #else 1651 static inline void sk_owner_set(struct sock *sk, struct module *owner) 1652 { 1653 } 1654 1655 static inline void sk_owner_clear(struct sock *sk) 1656 { 1657 } 1658 1659 static inline void sk_owner_put(struct sock *sk) 1660 { 1661 } 1662 #endif 1663 /* 1664 * Macro so as to not evaluate some arguments when 1665 * lockdep is not enabled. 1666 * 1667 * Mark both the sk_lock and the sk_lock.slock as a 1668 * per-address-family lock class. 1669 */ 1670 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \ 1671 do { \ 1672 sk_owner_set(sk, THIS_MODULE); \ 1673 sk->sk_lock.owned = 0; \ 1674 init_waitqueue_head(&sk->sk_lock.wq); \ 1675 spin_lock_init(&(sk)->sk_lock.slock); \ 1676 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \ 1677 sizeof((sk)->sk_lock)); \ 1678 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \ 1679 (skey), (sname)); \ 1680 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \ 1681 } while (0) 1682 1683 static inline bool lockdep_sock_is_held(const struct sock *sk) 1684 { 1685 return lockdep_is_held(&sk->sk_lock) || 1686 lockdep_is_held(&sk->sk_lock.slock); 1687 } 1688 1689 void lock_sock_nested(struct sock *sk, int subclass); 1690 1691 static inline void lock_sock(struct sock *sk) 1692 { 1693 lock_sock_nested(sk, 0); 1694 } 1695 1696 void __lock_sock(struct sock *sk); 1697 void __release_sock(struct sock *sk); 1698 void release_sock(struct sock *sk); 1699 1700 /* BH context may only use the following locking interface. */ 1701 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock)) 1702 #define bh_lock_sock_nested(__sk) \ 1703 spin_lock_nested(&((__sk)->sk_lock.slock), \ 1704 SINGLE_DEPTH_NESTING) 1705 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock)) 1706 1707 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock); 1708 1709 /** 1710 * lock_sock_fast - fast version of lock_sock 1711 * @sk: socket 1712 * 1713 * This version should be used for very small section, where process won't block 1714 * return false if fast path is taken: 1715 * 1716 * sk_lock.slock locked, owned = 0, BH disabled 1717 * 1718 * return true if slow path is taken: 1719 * 1720 * sk_lock.slock unlocked, owned = 1, BH enabled 1721 */ 1722 static inline bool lock_sock_fast(struct sock *sk) 1723 { 1724 /* The sk_lock has mutex_lock() semantics here. */ 1725 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_); 1726 1727 return __lock_sock_fast(sk); 1728 } 1729 1730 /* fast socket lock variant for caller already holding a [different] socket lock */ 1731 static inline bool lock_sock_fast_nested(struct sock *sk) 1732 { 1733 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_); 1734 1735 return __lock_sock_fast(sk); 1736 } 1737 1738 /** 1739 * unlock_sock_fast - complement of lock_sock_fast 1740 * @sk: socket 1741 * @slow: slow mode 1742 * 1743 * fast unlock socket for user context. 1744 * If slow mode is on, we call regular release_sock() 1745 */ 1746 static inline void unlock_sock_fast(struct sock *sk, bool slow) 1747 __releases(&sk->sk_lock.slock) 1748 { 1749 if (slow) { 1750 release_sock(sk); 1751 __release(&sk->sk_lock.slock); 1752 } else { 1753 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1754 spin_unlock_bh(&sk->sk_lock.slock); 1755 } 1756 } 1757 1758 void sockopt_lock_sock(struct sock *sk); 1759 void sockopt_release_sock(struct sock *sk); 1760 bool sockopt_ns_capable(struct user_namespace *ns, int cap); 1761 bool sockopt_capable(int cap); 1762 1763 /* Used by processes to "lock" a socket state, so that 1764 * interrupts and bottom half handlers won't change it 1765 * from under us. It essentially blocks any incoming 1766 * packets, so that we won't get any new data or any 1767 * packets that change the state of the socket. 1768 * 1769 * While locked, BH processing will add new packets to 1770 * the backlog queue. This queue is processed by the 1771 * owner of the socket lock right before it is released. 1772 * 1773 * Since ~2.3.5 it is also exclusive sleep lock serializing 1774 * accesses from user process context. 1775 */ 1776 1777 static inline void sock_owned_by_me(const struct sock *sk) 1778 { 1779 #ifdef CONFIG_LOCKDEP 1780 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks); 1781 #endif 1782 } 1783 1784 static inline void sock_not_owned_by_me(const struct sock *sk) 1785 { 1786 #ifdef CONFIG_LOCKDEP 1787 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks); 1788 #endif 1789 } 1790 1791 static inline bool sock_owned_by_user(const struct sock *sk) 1792 { 1793 sock_owned_by_me(sk); 1794 return sk->sk_lock.owned; 1795 } 1796 1797 static inline bool sock_owned_by_user_nocheck(const struct sock *sk) 1798 { 1799 return sk->sk_lock.owned; 1800 } 1801 1802 static inline void sock_release_ownership(struct sock *sk) 1803 { 1804 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk)); 1805 sk->sk_lock.owned = 0; 1806 1807 /* The sk_lock has mutex_unlock() semantics: */ 1808 mutex_release(&sk->sk_lock.dep_map, _RET_IP_); 1809 } 1810 1811 /* no reclassification while locks are held */ 1812 static inline bool sock_allow_reclassification(const struct sock *csk) 1813 { 1814 struct sock *sk = (struct sock *)csk; 1815 1816 return !sock_owned_by_user_nocheck(sk) && 1817 !spin_is_locked(&sk->sk_lock.slock); 1818 } 1819 1820 struct sock *sk_alloc(struct net *net, int family, gfp_t priority, 1821 struct proto *prot, int kern); 1822 void sk_free(struct sock *sk); 1823 void sk_net_refcnt_upgrade(struct sock *sk); 1824 void sk_destruct(struct sock *sk); 1825 struct sock *sk_clone(const struct sock *sk, const gfp_t priority, bool lock); 1826 1827 static inline struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority) 1828 { 1829 return sk_clone(sk, priority, true); 1830 } 1831 1832 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, 1833 gfp_t priority); 1834 void __sock_wfree(struct sk_buff *skb); 1835 void sock_wfree(struct sk_buff *skb); 1836 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size, 1837 gfp_t priority); 1838 void skb_orphan_partial(struct sk_buff *skb); 1839 void sock_rfree(struct sk_buff *skb); 1840 void sock_efree(struct sk_buff *skb); 1841 #ifdef CONFIG_INET 1842 void sock_edemux(struct sk_buff *skb); 1843 void sock_pfree(struct sk_buff *skb); 1844 1845 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk) 1846 { 1847 skb_orphan(skb); 1848 if (refcount_inc_not_zero(&sk->sk_refcnt)) { 1849 skb->sk = sk; 1850 skb->destructor = sock_edemux; 1851 } 1852 } 1853 #else 1854 #define sock_edemux sock_efree 1855 #endif 1856 1857 int sk_setsockopt(struct sock *sk, int level, int optname, 1858 sockptr_t optval, unsigned int optlen); 1859 int sock_setsockopt(struct socket *sock, int level, int op, 1860 sockptr_t optval, unsigned int optlen); 1861 int do_sock_setsockopt(struct socket *sock, bool compat, int level, 1862 int optname, sockptr_t optval, int optlen); 1863 int do_sock_getsockopt(struct socket *sock, bool compat, int level, 1864 int optname, sockptr_t optval, sockptr_t optlen); 1865 1866 int sk_getsockopt(struct sock *sk, int level, int optname, 1867 sockptr_t optval, sockptr_t optlen); 1868 int sock_gettstamp(struct socket *sock, void __user *userstamp, 1869 bool timeval, bool time32); 1870 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len, 1871 unsigned long data_len, int noblock, 1872 int *errcode, int max_page_order); 1873 1874 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk, 1875 unsigned long size, 1876 int noblock, int *errcode) 1877 { 1878 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0); 1879 } 1880 1881 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority); 1882 void *sock_kmemdup(struct sock *sk, const void *src, 1883 int size, gfp_t priority); 1884 void sock_kfree_s(struct sock *sk, void *mem, int size); 1885 void sock_kzfree_s(struct sock *sk, void *mem, int size); 1886 void sk_send_sigurg(struct sock *sk); 1887 1888 static inline void sock_replace_proto(struct sock *sk, struct proto *proto) 1889 { 1890 if (sk->sk_socket) 1891 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags); 1892 WRITE_ONCE(sk->sk_prot, proto); 1893 } 1894 1895 struct sockcm_cookie { 1896 u64 transmit_time; 1897 u32 mark; 1898 u32 tsflags; 1899 u32 ts_opt_id; 1900 u32 priority; 1901 u32 dmabuf_id; 1902 }; 1903 1904 static inline void sockcm_init(struct sockcm_cookie *sockc, 1905 const struct sock *sk) 1906 { 1907 *sockc = (struct sockcm_cookie) { 1908 .mark = READ_ONCE(sk->sk_mark), 1909 .tsflags = READ_ONCE(sk->sk_tsflags), 1910 .priority = READ_ONCE(sk->sk_priority), 1911 }; 1912 } 1913 1914 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg, 1915 struct sockcm_cookie *sockc); 1916 int sock_cmsg_send(struct sock *sk, struct msghdr *msg, 1917 struct sockcm_cookie *sockc); 1918 1919 /* 1920 * Functions to fill in entries in struct proto_ops when a protocol 1921 * does not implement a particular function. 1922 */ 1923 int sock_no_bind(struct socket *sock, struct sockaddr_unsized *saddr, int len); 1924 int sock_no_connect(struct socket *sock, struct sockaddr_unsized *saddr, int len, int flags); 1925 int sock_no_socketpair(struct socket *, struct socket *); 1926 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *); 1927 int sock_no_getname(struct socket *, struct sockaddr *, int); 1928 int sock_no_ioctl(struct socket *, unsigned int, unsigned long); 1929 int sock_no_listen(struct socket *, int); 1930 int sock_no_shutdown(struct socket *, int); 1931 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t); 1932 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len); 1933 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int); 1934 int sock_no_mmap(struct file *file, struct socket *sock, 1935 struct vm_area_struct *vma); 1936 1937 /* 1938 * Functions to fill in entries in struct proto_ops when a protocol 1939 * uses the inet style. 1940 */ 1941 int sock_common_getsockopt(struct socket *sock, int level, int optname, 1942 char __user *optval, int __user *optlen); 1943 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 1944 int flags); 1945 int sock_common_setsockopt(struct socket *sock, int level, int optname, 1946 sockptr_t optval, unsigned int optlen); 1947 1948 void sk_common_release(struct sock *sk); 1949 1950 /* 1951 * Default socket callbacks and setup code 1952 */ 1953 1954 /* Initialise core socket variables using an explicit uid. */ 1955 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid); 1956 1957 /* Initialise core socket variables. 1958 * Assumes struct socket *sock is embedded in a struct socket_alloc. 1959 */ 1960 void sock_init_data(struct socket *sock, struct sock *sk); 1961 1962 /* 1963 * Socket reference counting postulates. 1964 * 1965 * * Each user of socket SHOULD hold a reference count. 1966 * * Each access point to socket (an hash table bucket, reference from a list, 1967 * running timer, skb in flight MUST hold a reference count. 1968 * * When reference count hits 0, it means it will never increase back. 1969 * * When reference count hits 0, it means that no references from 1970 * outside exist to this socket and current process on current CPU 1971 * is last user and may/should destroy this socket. 1972 * * sk_free is called from any context: process, BH, IRQ. When 1973 * it is called, socket has no references from outside -> sk_free 1974 * may release descendant resources allocated by the socket, but 1975 * to the time when it is called, socket is NOT referenced by any 1976 * hash tables, lists etc. 1977 * * Packets, delivered from outside (from network or from another process) 1978 * and enqueued on receive/error queues SHOULD NOT grab reference count, 1979 * when they sit in queue. Otherwise, packets will leak to hole, when 1980 * socket is looked up by one cpu and unhasing is made by another CPU. 1981 * It is true for udp/raw, netlink (leak to receive and error queues), tcp 1982 * (leak to backlog). Packet socket does all the processing inside 1983 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets 1984 * use separate SMP lock, so that they are prone too. 1985 */ 1986 1987 /* Ungrab socket and destroy it, if it was the last reference. */ 1988 static inline void sock_put(struct sock *sk) 1989 { 1990 if (refcount_dec_and_test(&sk->sk_refcnt)) 1991 sk_free(sk); 1992 } 1993 /* Generic version of sock_put(), dealing with all sockets 1994 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...) 1995 */ 1996 void sock_gen_put(struct sock *sk); 1997 1998 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested, 1999 unsigned int trim_cap, bool refcounted); 2000 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb, 2001 const int nested) 2002 { 2003 return __sk_receive_skb(sk, skb, nested, 1, true); 2004 } 2005 2006 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue) 2007 { 2008 /* sk_tx_queue_mapping accept only upto a 16-bit value */ 2009 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX)) 2010 return; 2011 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2012 * other WRITE_ONCE() because socket lock might be not held. 2013 */ 2014 if (READ_ONCE(sk->sk_tx_queue_mapping) != tx_queue) { 2015 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue); 2016 WRITE_ONCE(sk->sk_tx_queue_mapping_jiffies, jiffies); 2017 return; 2018 } 2019 2020 /* Refresh sk_tx_queue_mapping_jiffies if too old. */ 2021 if (time_is_before_jiffies(READ_ONCE(sk->sk_tx_queue_mapping_jiffies) + HZ)) 2022 WRITE_ONCE(sk->sk_tx_queue_mapping_jiffies, jiffies); 2023 } 2024 2025 #define NO_QUEUE_MAPPING USHRT_MAX 2026 2027 static inline void sk_tx_queue_clear(struct sock *sk) 2028 { 2029 /* Paired with READ_ONCE() in sk_tx_queue_get() and 2030 * other WRITE_ONCE() because socket lock might be not held. 2031 */ 2032 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING); 2033 } 2034 2035 int sk_tx_queue_get(const struct sock *sk); 2036 2037 static inline void __sk_rx_queue_set(struct sock *sk, 2038 const struct sk_buff *skb, 2039 bool force_set) 2040 { 2041 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2042 if (skb_rx_queue_recorded(skb)) { 2043 u16 rx_queue = skb_get_rx_queue(skb); 2044 2045 if (force_set || 2046 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue)) 2047 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue); 2048 } 2049 #endif 2050 } 2051 2052 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb) 2053 { 2054 __sk_rx_queue_set(sk, skb, true); 2055 } 2056 2057 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb) 2058 { 2059 __sk_rx_queue_set(sk, skb, false); 2060 } 2061 2062 static inline void sk_rx_queue_clear(struct sock *sk) 2063 { 2064 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2065 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING); 2066 #endif 2067 } 2068 2069 static inline int sk_rx_queue_get(const struct sock *sk) 2070 { 2071 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 2072 if (sk) { 2073 int res = READ_ONCE(sk->sk_rx_queue_mapping); 2074 2075 if (res != NO_QUEUE_MAPPING) 2076 return res; 2077 } 2078 #endif 2079 2080 return -1; 2081 } 2082 2083 static inline void sk_set_socket(struct sock *sk, struct socket *sock) 2084 { 2085 sk->sk_socket = sock; 2086 if (sock) { 2087 WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid); 2088 WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino); 2089 } else { 2090 /* Note: sk_uid is unchanged. */ 2091 WRITE_ONCE(sk->sk_ino, 0); 2092 } 2093 } 2094 2095 static inline wait_queue_head_t *sk_sleep(struct sock *sk) 2096 { 2097 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0); 2098 return &rcu_dereference_raw(sk->sk_wq)->wait; 2099 } 2100 /* Detach socket from process context. 2101 * Announce socket dead, detach it from wait queue and inode. 2102 * Note that parent inode held reference count on this struct sock, 2103 * we do not release it in this function, because protocol 2104 * probably wants some additional cleanups or even continuing 2105 * to work with this socket (TCP). 2106 */ 2107 static inline void sock_orphan(struct sock *sk) 2108 { 2109 write_lock_bh(&sk->sk_callback_lock); 2110 sock_set_flag(sk, SOCK_DEAD); 2111 sk_set_socket(sk, NULL); 2112 sk->sk_wq = NULL; 2113 write_unlock_bh(&sk->sk_callback_lock); 2114 } 2115 2116 static inline void sock_graft(struct sock *sk, struct socket *parent) 2117 { 2118 WARN_ON(parent->sk); 2119 write_lock_bh(&sk->sk_callback_lock); 2120 rcu_assign_pointer(sk->sk_wq, &parent->wq); 2121 parent->sk = sk; 2122 sk_set_socket(sk, parent); 2123 security_sock_graft(sk, parent); 2124 write_unlock_bh(&sk->sk_callback_lock); 2125 } 2126 2127 static inline unsigned long sock_i_ino(const struct sock *sk) 2128 { 2129 /* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */ 2130 return READ_ONCE(sk->sk_ino); 2131 } 2132 2133 static inline kuid_t sk_uid(const struct sock *sk) 2134 { 2135 /* Paired with WRITE_ONCE() in sockfs_setattr() */ 2136 return READ_ONCE(sk->sk_uid); 2137 } 2138 2139 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk) 2140 { 2141 return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0); 2142 } 2143 2144 static inline u32 net_tx_rndhash(void) 2145 { 2146 u32 v = get_random_u32(); 2147 2148 return v ?: 1; 2149 } 2150 2151 static inline void sk_set_txhash(struct sock *sk) 2152 { 2153 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */ 2154 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash()); 2155 } 2156 2157 static inline bool sk_rethink_txhash(struct sock *sk) 2158 { 2159 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) { 2160 sk_set_txhash(sk); 2161 return true; 2162 } 2163 return false; 2164 } 2165 2166 static inline struct dst_entry * 2167 __sk_dst_get(const struct sock *sk) 2168 { 2169 return rcu_dereference_check(sk->sk_dst_cache, 2170 lockdep_sock_is_held(sk)); 2171 } 2172 2173 static inline struct dst_entry * 2174 sk_dst_get(const struct sock *sk) 2175 { 2176 struct dst_entry *dst; 2177 2178 rcu_read_lock(); 2179 dst = rcu_dereference(sk->sk_dst_cache); 2180 if (dst && !rcuref_get(&dst->__rcuref)) 2181 dst = NULL; 2182 rcu_read_unlock(); 2183 return dst; 2184 } 2185 2186 static inline void __dst_negative_advice(struct sock *sk) 2187 { 2188 struct dst_entry *dst = __sk_dst_get(sk); 2189 2190 if (dst && dst->ops->negative_advice) 2191 dst->ops->negative_advice(sk, dst); 2192 } 2193 2194 static inline void dst_negative_advice(struct sock *sk) 2195 { 2196 sk_rethink_txhash(sk); 2197 __dst_negative_advice(sk); 2198 } 2199 2200 static inline void 2201 __sk_dst_set(struct sock *sk, struct dst_entry *dst) 2202 { 2203 struct dst_entry *old_dst; 2204 2205 sk_tx_queue_clear(sk); 2206 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2207 old_dst = rcu_dereference_protected(sk->sk_dst_cache, 2208 lockdep_sock_is_held(sk)); 2209 rcu_assign_pointer(sk->sk_dst_cache, dst); 2210 dst_release(old_dst); 2211 } 2212 2213 static inline void 2214 sk_dst_set(struct sock *sk, struct dst_entry *dst) 2215 { 2216 struct dst_entry *old_dst; 2217 2218 sk_tx_queue_clear(sk); 2219 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2220 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst))); 2221 dst_release(old_dst); 2222 } 2223 2224 static inline void 2225 __sk_dst_reset(struct sock *sk) 2226 { 2227 __sk_dst_set(sk, NULL); 2228 } 2229 2230 static inline void 2231 sk_dst_reset(struct sock *sk) 2232 { 2233 sk_dst_set(sk, NULL); 2234 } 2235 2236 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie); 2237 2238 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie); 2239 2240 static inline void sk_dst_confirm(struct sock *sk) 2241 { 2242 if (!READ_ONCE(sk->sk_dst_pending_confirm)) 2243 WRITE_ONCE(sk->sk_dst_pending_confirm, 1); 2244 } 2245 2246 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n) 2247 { 2248 if (skb_get_dst_pending_confirm(skb)) { 2249 struct sock *sk = skb->sk; 2250 2251 if (sk && READ_ONCE(sk->sk_dst_pending_confirm)) 2252 WRITE_ONCE(sk->sk_dst_pending_confirm, 0); 2253 neigh_confirm(n); 2254 } 2255 } 2256 2257 bool sk_mc_loop(const struct sock *sk); 2258 2259 static inline bool sk_can_gso(const struct sock *sk) 2260 { 2261 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type); 2262 } 2263 2264 void sk_setup_caps(struct sock *sk, struct dst_entry *dst); 2265 2266 static inline void sk_gso_disable(struct sock *sk) 2267 { 2268 sk->sk_gso_disabled = 1; 2269 sk->sk_route_caps &= ~NETIF_F_GSO_MASK; 2270 } 2271 2272 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb, 2273 struct iov_iter *from, char *to, 2274 int copy, int offset) 2275 { 2276 if (skb->ip_summed == CHECKSUM_NONE) { 2277 __wsum csum = 0; 2278 if (!csum_and_copy_from_iter_full(to, copy, &csum, from)) 2279 return -EFAULT; 2280 skb->csum = csum_block_add(skb->csum, csum, offset); 2281 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) { 2282 if (!copy_from_iter_full_nocache(to, copy, from)) 2283 return -EFAULT; 2284 } else if (!copy_from_iter_full(to, copy, from)) 2285 return -EFAULT; 2286 2287 return 0; 2288 } 2289 2290 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb, 2291 struct iov_iter *from, int copy) 2292 { 2293 int err, offset = skb->len; 2294 2295 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy), 2296 copy, offset); 2297 if (err) 2298 __skb_trim(skb, offset); 2299 2300 return err; 2301 } 2302 2303 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from, 2304 struct sk_buff *skb, 2305 struct page *page, 2306 int off, int copy) 2307 { 2308 int err; 2309 2310 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off, 2311 copy, skb->len); 2312 if (err) 2313 return err; 2314 2315 skb_len_add(skb, copy); 2316 sk_wmem_queued_add(sk, copy); 2317 sk_mem_charge(sk, copy); 2318 return 0; 2319 } 2320 2321 #define SK_WMEM_ALLOC_BIAS 1 2322 /** 2323 * sk_wmem_alloc_get - returns write allocations 2324 * @sk: socket 2325 * 2326 * Return: sk_wmem_alloc minus initial offset of one 2327 */ 2328 static inline int sk_wmem_alloc_get(const struct sock *sk) 2329 { 2330 return refcount_read(&sk->sk_wmem_alloc) - SK_WMEM_ALLOC_BIAS; 2331 } 2332 2333 /** 2334 * sk_rmem_alloc_get - returns read allocations 2335 * @sk: socket 2336 * 2337 * Return: sk_rmem_alloc 2338 */ 2339 static inline int sk_rmem_alloc_get(const struct sock *sk) 2340 { 2341 return atomic_read(&sk->sk_rmem_alloc); 2342 } 2343 2344 /** 2345 * sk_has_allocations - check if allocations are outstanding 2346 * @sk: socket 2347 * 2348 * Return: true if socket has write or read allocations 2349 */ 2350 static inline bool sk_has_allocations(const struct sock *sk) 2351 { 2352 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk); 2353 } 2354 2355 /** 2356 * skwq_has_sleeper - check if there are any waiting processes 2357 * @wq: struct socket_wq 2358 * 2359 * Return: true if socket_wq has waiting processes 2360 * 2361 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory 2362 * barrier call. They were added due to the race found within the tcp code. 2363 * 2364 * Consider following tcp code paths:: 2365 * 2366 * CPU1 CPU2 2367 * sys_select receive packet 2368 * ... ... 2369 * __add_wait_queue update tp->rcv_nxt 2370 * ... ... 2371 * tp->rcv_nxt check sock_def_readable 2372 * ... { 2373 * schedule rcu_read_lock(); 2374 * wq = rcu_dereference(sk->sk_wq); 2375 * if (wq && waitqueue_active(&wq->wait)) 2376 * wake_up_interruptible(&wq->wait) 2377 * ... 2378 * } 2379 * 2380 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay 2381 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1 2382 * could then endup calling schedule and sleep forever if there are no more 2383 * data on the socket. 2384 * 2385 */ 2386 static inline bool skwq_has_sleeper(struct socket_wq *wq) 2387 { 2388 return wq && wq_has_sleeper(&wq->wait); 2389 } 2390 2391 /** 2392 * sock_poll_wait - wrapper for the poll_wait call. 2393 * @filp: file 2394 * @sock: socket to wait on 2395 * @p: poll_table 2396 * 2397 * See the comments in the wq_has_sleeper function. 2398 */ 2399 static inline void sock_poll_wait(struct file *filp, struct socket *sock, 2400 poll_table *p) 2401 { 2402 /* Provides a barrier we need to be sure we are in sync 2403 * with the socket flags modification. 2404 * 2405 * This memory barrier is paired in the wq_has_sleeper. 2406 */ 2407 poll_wait(filp, &sock->wq.wait, p); 2408 } 2409 2410 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk) 2411 { 2412 /* This pairs with WRITE_ONCE() in sk_set_txhash() */ 2413 u32 txhash = READ_ONCE(sk->sk_txhash); 2414 2415 if (txhash) { 2416 skb->l4_hash = 1; 2417 skb->hash = txhash; 2418 } 2419 } 2420 2421 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk); 2422 2423 /* 2424 * Queue a received datagram if it will fit. Stream and sequenced 2425 * protocols can't normally use this as they need to fit buffers in 2426 * and play with them. 2427 * 2428 * Inlined as it's very short and called for pretty much every 2429 * packet ever received. 2430 */ 2431 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk) 2432 { 2433 skb_orphan(skb); 2434 skb->sk = sk; 2435 skb->destructor = sock_rfree; 2436 atomic_add(skb->truesize, &sk->sk_rmem_alloc); 2437 sk_mem_charge(sk, skb->truesize); 2438 } 2439 2440 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk) 2441 { 2442 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) { 2443 skb_orphan(skb); 2444 skb->destructor = sock_efree; 2445 skb->sk = sk; 2446 return true; 2447 } 2448 return false; 2449 } 2450 2451 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk) 2452 { 2453 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC)); 2454 if (skb) { 2455 if (sk_rmem_schedule(sk, skb, skb->truesize)) { 2456 skb_set_owner_r(skb, sk); 2457 return skb; 2458 } 2459 __kfree_skb(skb); 2460 } 2461 return NULL; 2462 } 2463 2464 static inline void skb_prepare_for_gro(struct sk_buff *skb) 2465 { 2466 if (skb->destructor != sock_wfree) { 2467 skb_orphan(skb); 2468 return; 2469 } 2470 skb->slow_gro = 1; 2471 } 2472 2473 void sk_reset_timer(struct sock *sk, struct timer_list *timer, 2474 unsigned long expires); 2475 2476 void sk_stop_timer(struct sock *sk, struct timer_list *timer); 2477 2478 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer); 2479 2480 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 2481 struct sk_buff *skb, unsigned int flags, 2482 void (*destructor)(struct sock *sk, 2483 struct sk_buff *skb)); 2484 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb); 2485 2486 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb, 2487 enum skb_drop_reason *reason); 2488 2489 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 2490 { 2491 return sock_queue_rcv_skb_reason(sk, skb, NULL); 2492 } 2493 2494 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb); 2495 struct sk_buff *sock_dequeue_err_skb(struct sock *sk); 2496 2497 /* 2498 * Recover an error report and clear atomically 2499 */ 2500 2501 static inline int sock_error(struct sock *sk) 2502 { 2503 int err; 2504 2505 /* Avoid an atomic operation for the common case. 2506 * This is racy since another cpu/thread can change sk_err under us. 2507 */ 2508 if (likely(data_race(!sk->sk_err))) 2509 return 0; 2510 2511 err = xchg(&sk->sk_err, 0); 2512 return -err; 2513 } 2514 2515 void sk_error_report(struct sock *sk); 2516 2517 static inline unsigned long sock_wspace(struct sock *sk) 2518 { 2519 int amt = 0; 2520 2521 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) { 2522 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc); 2523 if (amt < 0) 2524 amt = 0; 2525 } 2526 return amt; 2527 } 2528 2529 /* Note: 2530 * We use sk->sk_wq_raw, from contexts knowing this 2531 * pointer is not NULL and cannot disappear/change. 2532 */ 2533 static inline void sk_set_bit(int nr, struct sock *sk) 2534 { 2535 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2536 !sock_flag(sk, SOCK_FASYNC)) 2537 return; 2538 2539 set_bit(nr, &sk->sk_wq_raw->flags); 2540 } 2541 2542 static inline void sk_clear_bit(int nr, struct sock *sk) 2543 { 2544 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) && 2545 !sock_flag(sk, SOCK_FASYNC)) 2546 return; 2547 2548 clear_bit(nr, &sk->sk_wq_raw->flags); 2549 } 2550 2551 static inline void sk_wake_async(const struct sock *sk, int how, int band) 2552 { 2553 if (sock_flag(sk, SOCK_FASYNC)) { 2554 rcu_read_lock(); 2555 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2556 rcu_read_unlock(); 2557 } 2558 } 2559 2560 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band) 2561 { 2562 if (unlikely(sock_flag(sk, SOCK_FASYNC))) 2563 sock_wake_async(rcu_dereference(sk->sk_wq), how, band); 2564 } 2565 2566 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might 2567 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak. 2568 * Note: for send buffers, TCP works better if we can build two skbs at 2569 * minimum. 2570 */ 2571 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff))) 2572 2573 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2) 2574 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE 2575 2576 static inline void sk_stream_moderate_sndbuf(struct sock *sk) 2577 { 2578 u32 val; 2579 2580 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK) 2581 return; 2582 2583 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1); 2584 val = max_t(u32, val, sk_unused_reserved_mem(sk)); 2585 2586 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF)); 2587 } 2588 2589 /** 2590 * sk_page_frag - return an appropriate page_frag 2591 * @sk: socket 2592 * 2593 * Use the per task page_frag instead of the per socket one for 2594 * optimization when we know that we're in process context and own 2595 * everything that's associated with %current. 2596 * 2597 * Both direct reclaim and page faults can nest inside other 2598 * socket operations and end up recursing into sk_page_frag() 2599 * while it's already in use: explicitly avoid task page_frag 2600 * when users disable sk_use_task_frag. 2601 * 2602 * Return: a per task page_frag if context allows that, 2603 * otherwise a per socket one. 2604 */ 2605 static inline struct page_frag *sk_page_frag(struct sock *sk) 2606 { 2607 if (sk->sk_use_task_frag) 2608 return ¤t->task_frag; 2609 2610 return &sk->sk_frag; 2611 } 2612 2613 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag); 2614 2615 static inline bool __sock_writeable(const struct sock *sk, int wmem_alloc) 2616 { 2617 return wmem_alloc < (READ_ONCE(sk->sk_sndbuf) >> 1); 2618 } 2619 /* 2620 * Default write policy as shown to user space via poll/select/SIGIO 2621 */ 2622 static inline bool sock_writeable(const struct sock *sk) 2623 { 2624 return __sock_writeable(sk, refcount_read(&sk->sk_wmem_alloc)); 2625 } 2626 2627 static inline gfp_t gfp_any(void) 2628 { 2629 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2630 } 2631 2632 static inline gfp_t gfp_memcg_charge(void) 2633 { 2634 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL; 2635 } 2636 2637 #ifdef CONFIG_MEMCG 2638 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk) 2639 { 2640 return sk->sk_memcg; 2641 } 2642 2643 static inline bool mem_cgroup_sk_enabled(const struct sock *sk) 2644 { 2645 return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk); 2646 } 2647 2648 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk) 2649 { 2650 struct mem_cgroup *memcg = mem_cgroup_from_sk(sk); 2651 2652 #ifdef CONFIG_MEMCG_V1 2653 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) 2654 return !!memcg->tcpmem_pressure; 2655 #endif /* CONFIG_MEMCG_V1 */ 2656 2657 do { 2658 if (time_before64(get_jiffies_64(), mem_cgroup_get_socket_pressure(memcg))) 2659 return true; 2660 } while ((memcg = parent_mem_cgroup(memcg))); 2661 2662 return false; 2663 } 2664 #else 2665 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk) 2666 { 2667 return NULL; 2668 } 2669 2670 static inline bool mem_cgroup_sk_enabled(const struct sock *sk) 2671 { 2672 return false; 2673 } 2674 2675 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk) 2676 { 2677 return false; 2678 } 2679 #endif 2680 2681 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock) 2682 { 2683 return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo); 2684 } 2685 2686 static inline long sock_sndtimeo(const struct sock *sk, bool noblock) 2687 { 2688 return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo); 2689 } 2690 2691 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len) 2692 { 2693 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len); 2694 2695 return v ?: 1; 2696 } 2697 2698 /* Alas, with timeout socket operations are not restartable. 2699 * Compare this to poll(). 2700 */ 2701 static inline int sock_intr_errno(long timeo) 2702 { 2703 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR; 2704 } 2705 2706 struct sock_skb_cb { 2707 u32 dropcount; 2708 }; 2709 2710 /* Store sock_skb_cb at the end of skb->cb[] so protocol families 2711 * using skb->cb[] would keep using it directly and utilize its 2712 * alignment guarantee. 2713 */ 2714 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \ 2715 sizeof(struct sock_skb_cb)) 2716 2717 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \ 2718 SOCK_SKB_CB_OFFSET)) 2719 2720 #define sock_skb_cb_check_size(size) \ 2721 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET) 2722 2723 static inline void sk_drops_add(struct sock *sk, int segs) 2724 { 2725 struct numa_drop_counters *ndc = sk->sk_drop_counters; 2726 2727 if (ndc) 2728 numa_drop_add(ndc, segs); 2729 else 2730 atomic_add(segs, &sk->sk_drops); 2731 } 2732 2733 static inline void sk_drops_inc(struct sock *sk) 2734 { 2735 sk_drops_add(sk, 1); 2736 } 2737 2738 static inline int sk_drops_read(const struct sock *sk) 2739 { 2740 const struct numa_drop_counters *ndc = sk->sk_drop_counters; 2741 2742 if (ndc) { 2743 DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops)); 2744 return numa_drop_read(ndc); 2745 } 2746 return atomic_read(&sk->sk_drops); 2747 } 2748 2749 static inline void sk_drops_reset(struct sock *sk) 2750 { 2751 struct numa_drop_counters *ndc = sk->sk_drop_counters; 2752 2753 if (ndc) 2754 numa_drop_reset(ndc); 2755 atomic_set(&sk->sk_drops, 0); 2756 } 2757 2758 static inline void 2759 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb) 2760 { 2761 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ? 2762 sk_drops_read(sk) : 0; 2763 } 2764 2765 static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb) 2766 { 2767 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs); 2768 2769 sk_drops_add(sk, segs); 2770 } 2771 2772 static inline ktime_t sock_read_timestamp(struct sock *sk) 2773 { 2774 #if BITS_PER_LONG==32 2775 unsigned int seq; 2776 ktime_t kt; 2777 2778 do { 2779 seq = read_seqbegin(&sk->sk_stamp_seq); 2780 kt = sk->sk_stamp; 2781 } while (read_seqretry(&sk->sk_stamp_seq, seq)); 2782 2783 return kt; 2784 #else 2785 return READ_ONCE(sk->sk_stamp); 2786 #endif 2787 } 2788 2789 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt) 2790 { 2791 #if BITS_PER_LONG==32 2792 write_seqlock(&sk->sk_stamp_seq); 2793 sk->sk_stamp = kt; 2794 write_sequnlock(&sk->sk_stamp_seq); 2795 #else 2796 WRITE_ONCE(sk->sk_stamp, kt); 2797 #endif 2798 } 2799 2800 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk, 2801 struct sk_buff *skb); 2802 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk, 2803 struct sk_buff *skb); 2804 2805 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk); 2806 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk, 2807 struct timespec64 *ts); 2808 2809 static inline void 2810 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb) 2811 { 2812 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb); 2813 u32 tsflags = READ_ONCE(sk->sk_tsflags); 2814 ktime_t kt = skb->tstamp; 2815 /* 2816 * generate control messages if 2817 * - receive time stamping in software requested 2818 * - software time stamp available and wanted 2819 * - hardware time stamps available and wanted 2820 */ 2821 if (sock_flag(sk, SOCK_RCVTSTAMP) || 2822 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) || 2823 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) || 2824 (hwtstamps->hwtstamp && 2825 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE))) 2826 __sock_recv_timestamp(msg, sk, skb); 2827 else 2828 sock_write_timestamp(sk, kt); 2829 2830 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb)) 2831 __sock_recv_wifi_status(msg, sk, skb); 2832 } 2833 2834 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2835 struct sk_buff *skb); 2836 2837 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC) 2838 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk, 2839 struct sk_buff *skb) 2840 { 2841 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \ 2842 (1UL << SOCK_RCVTSTAMP) | \ 2843 (1UL << SOCK_RCVMARK) | \ 2844 (1UL << SOCK_RCVPRIORITY) | \ 2845 (1UL << SOCK_TIMESTAMPING_ANY)) 2846 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \ 2847 SOF_TIMESTAMPING_RAW_HARDWARE) 2848 2849 if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS) 2850 __sock_recv_cmsgs(msg, sk, skb); 2851 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP))) 2852 sock_write_timestamp(sk, skb->tstamp); 2853 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP)) 2854 sock_write_timestamp(sk, 0); 2855 } 2856 2857 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags); 2858 2859 /** 2860 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped 2861 * @sk: socket sending this packet 2862 * @sockc: pointer to socket cmsg cookie to get timestamping info 2863 * @tx_flags: completed with instructions for time stamping 2864 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno) 2865 * 2866 * Note: callers should take care of initial ``*tx_flags`` value (usually 0) 2867 */ 2868 static inline void _sock_tx_timestamp(struct sock *sk, 2869 const struct sockcm_cookie *sockc, 2870 __u8 *tx_flags, __u32 *tskey) 2871 { 2872 __u32 tsflags = sockc->tsflags; 2873 2874 if (unlikely(tsflags)) { 2875 __sock_tx_timestamp(tsflags, tx_flags); 2876 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey && 2877 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) { 2878 if (tsflags & SOCKCM_FLAG_TS_OPT_ID) 2879 *tskey = sockc->ts_opt_id; 2880 else 2881 *tskey = atomic_inc_return(&sk->sk_tskey) - 1; 2882 } 2883 } 2884 } 2885 2886 static inline void sock_tx_timestamp(struct sock *sk, 2887 const struct sockcm_cookie *sockc, 2888 __u8 *tx_flags) 2889 { 2890 _sock_tx_timestamp(sk, sockc, tx_flags, NULL); 2891 } 2892 2893 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, 2894 const struct sockcm_cookie *sockc) 2895 { 2896 _sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags, 2897 &skb_shinfo(skb)->tskey); 2898 } 2899 2900 static inline bool sk_is_inet(const struct sock *sk) 2901 { 2902 int family = READ_ONCE(sk->sk_family); 2903 2904 return family == AF_INET || family == AF_INET6; 2905 } 2906 2907 static inline bool sk_is_tcp(const struct sock *sk) 2908 { 2909 return sk_is_inet(sk) && 2910 sk->sk_type == SOCK_STREAM && 2911 sk->sk_protocol == IPPROTO_TCP; 2912 } 2913 2914 static inline bool sk_is_udp(const struct sock *sk) 2915 { 2916 return sk_is_inet(sk) && 2917 sk->sk_type == SOCK_DGRAM && 2918 sk->sk_protocol == IPPROTO_UDP; 2919 } 2920 2921 static inline bool sk_is_unix(const struct sock *sk) 2922 { 2923 return sk->sk_family == AF_UNIX; 2924 } 2925 2926 static inline bool sk_is_stream_unix(const struct sock *sk) 2927 { 2928 return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM; 2929 } 2930 2931 static inline bool sk_is_vsock(const struct sock *sk) 2932 { 2933 return sk->sk_family == AF_VSOCK; 2934 } 2935 2936 static inline bool sk_may_scm_recv(const struct sock *sk) 2937 { 2938 return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) || 2939 sk->sk_family == AF_NETLINK || 2940 (IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH); 2941 } 2942 2943 /** 2944 * sk_eat_skb - Release a skb if it is no longer needed 2945 * @sk: socket to eat this skb from 2946 * @skb: socket buffer to eat 2947 * 2948 * This routine must be called with interrupts disabled or with the socket 2949 * locked so that the sk_buff queue operation is ok. 2950 */ 2951 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb) 2952 { 2953 __skb_unlink(skb, &sk->sk_receive_queue); 2954 __kfree_skb(skb); 2955 } 2956 2957 static inline bool 2958 skb_sk_is_prefetched(struct sk_buff *skb) 2959 { 2960 #ifdef CONFIG_INET 2961 return skb->destructor == sock_pfree; 2962 #else 2963 return false; 2964 #endif /* CONFIG_INET */ 2965 } 2966 2967 /* This helper checks if a socket is a full socket, 2968 * ie _not_ a timewait or request socket. 2969 */ 2970 static inline bool sk_fullsock(const struct sock *sk) 2971 { 2972 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV); 2973 } 2974 2975 static inline bool 2976 sk_is_refcounted(struct sock *sk) 2977 { 2978 /* Only full sockets have sk->sk_flags. */ 2979 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE); 2980 } 2981 2982 static inline bool 2983 sk_requests_wifi_status(struct sock *sk) 2984 { 2985 return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS); 2986 } 2987 2988 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV 2989 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE) 2990 */ 2991 static inline bool sk_listener(const struct sock *sk) 2992 { 2993 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV); 2994 } 2995 2996 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT 2997 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE) 2998 * TCP RST and ACK can be attached to TIME_WAIT. 2999 */ 3000 static inline bool sk_listener_or_tw(const struct sock *sk) 3001 { 3002 return (1 << READ_ONCE(sk->sk_state)) & 3003 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT); 3004 } 3005 3006 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag); 3007 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level, 3008 int type); 3009 3010 bool sk_ns_capable(const struct sock *sk, 3011 struct user_namespace *user_ns, int cap); 3012 bool sk_capable(const struct sock *sk, int cap); 3013 bool sk_net_capable(const struct sock *sk, int cap); 3014 3015 void sk_get_meminfo(const struct sock *sk, u32 *meminfo); 3016 3017 /* Take into consideration the size of the struct sk_buff overhead in the 3018 * determination of these values, since that is non-constant across 3019 * platforms. This makes socket queueing behavior and performance 3020 * not depend upon such differences. 3021 */ 3022 #define _SK_MEM_PACKETS 256 3023 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256) 3024 #define SK_WMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 3025 #define SK_RMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS) 3026 3027 extern __u32 sysctl_wmem_max; 3028 extern __u32 sysctl_rmem_max; 3029 3030 extern __u32 sysctl_wmem_default; 3031 extern __u32 sysctl_rmem_default; 3032 3033 #define SKB_FRAG_PAGE_ORDER get_order(32768) 3034 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key); 3035 3036 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto) 3037 { 3038 /* Does this proto have per netns sysctl_wmem ? */ 3039 if (proto->sysctl_wmem_offset) 3040 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset)); 3041 3042 return READ_ONCE(*proto->sysctl_wmem); 3043 } 3044 3045 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto) 3046 { 3047 /* Does this proto have per netns sysctl_rmem ? */ 3048 if (proto->sysctl_rmem_offset) 3049 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset)); 3050 3051 return READ_ONCE(*proto->sysctl_rmem); 3052 } 3053 3054 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10) 3055 * Some wifi drivers need to tweak it to get more chunks. 3056 * They can use this helper from their ndo_start_xmit() 3057 */ 3058 static inline void sk_pacing_shift_update(struct sock *sk, int val) 3059 { 3060 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val) 3061 return; 3062 WRITE_ONCE(sk->sk_pacing_shift, val); 3063 } 3064 3065 /* if a socket is bound to a device, check that the given device 3066 * index is either the same or that the socket is bound to an L3 3067 * master device and the given device index is also enslaved to 3068 * that L3 master 3069 */ 3070 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif) 3071 { 3072 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); 3073 int mdif; 3074 3075 if (!bound_dev_if || bound_dev_if == dif) 3076 return true; 3077 3078 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif); 3079 if (mdif && mdif == bound_dev_if) 3080 return true; 3081 3082 return false; 3083 } 3084 3085 void sock_def_readable(struct sock *sk); 3086 3087 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk); 3088 void sock_set_timestamp(struct sock *sk, int optname, bool valbool); 3089 int sock_set_timestamping(struct sock *sk, int optname, 3090 struct so_timestamping timestamping); 3091 3092 #if defined(CONFIG_CGROUP_BPF) 3093 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op); 3094 #else 3095 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op) 3096 { 3097 } 3098 #endif 3099 void sock_no_linger(struct sock *sk); 3100 void sock_set_keepalive(struct sock *sk); 3101 void sock_set_priority(struct sock *sk, u32 priority); 3102 void sock_set_rcvbuf(struct sock *sk, int val); 3103 void sock_set_mark(struct sock *sk, u32 val); 3104 void sock_set_reuseaddr(struct sock *sk); 3105 void sock_set_reuseport(struct sock *sk); 3106 void sock_set_sndtimeo(struct sock *sk, s64 secs); 3107 3108 int sock_bind_add(struct sock *sk, struct sockaddr_unsized *addr, int addr_len); 3109 3110 int sock_get_timeout(long timeo, void *optval, bool old_timeval); 3111 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv, 3112 sockptr_t optval, int optlen, bool old_timeval); 3113 3114 int sock_ioctl_inout(struct sock *sk, unsigned int cmd, 3115 void __user *arg, void *karg, size_t size); 3116 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg); 3117 static inline bool sk_is_readable(struct sock *sk) 3118 { 3119 const struct proto *prot = READ_ONCE(sk->sk_prot); 3120 3121 if (prot->sock_is_readable) 3122 return prot->sock_is_readable(sk); 3123 3124 return false; 3125 } 3126 #endif /* _SOCK_H */ 3127