Loading...
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 */
83typedef 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
98struct sock;
99struct proto;
100struct net;
101
102typedef __u32 __bitwise __portpair;
103typedef __u64 __bitwise __addrpair;
104
105/**
106 * struct sock_common - minimal network layer representation of sockets
107 * @skc_daddr: Foreign IPv4 addr
108 * @skc_rcv_saddr: Bound local IPv4 addr
109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
110 * @skc_hash: hash value used with various protocol lookup tables
111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
112 * @skc_dport: placeholder for inet_dport/tw_dport
113 * @skc_num: placeholder for inet_num/tw_num
114 * @skc_portpair: __u32 union of @skc_dport & @skc_num
115 * @skc_family: network address family
116 * @skc_state: Connection state
117 * @skc_reuse: %SO_REUSEADDR setting
118 * @skc_reuseport: %SO_REUSEPORT setting
119 * @skc_ipv6only: socket is IPV6 only
120 * @skc_net_refcnt: socket is using net ref counting
121 * @skc_bound_dev_if: bound device index if != 0
122 * @skc_bind_node: bind hash linkage for various protocol lookup tables
123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
124 * @skc_prot: protocol handlers inside a network family
125 * @skc_net: reference to the network namespace of this socket
126 * @skc_v6_daddr: IPV6 destination address
127 * @skc_v6_rcv_saddr: IPV6 source address
128 * @skc_cookie: socket's cookie value
129 * @skc_node: main hash linkage for various protocol lookup tables
130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
131 * @skc_tx_queue_mapping: tx queue number for this connection
132 * @skc_rx_queue_mapping: rx queue number for this connection
133 * @skc_flags: place holder for sk_flags
134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
136 * @skc_listener: connection request listener socket (aka rsk_listener)
137 * [union with @skc_flags]
138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
139 * [union with @skc_flags]
140 * @skc_incoming_cpu: record/match cpu processing incoming packets
141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
142 * [union with @skc_incoming_cpu]
143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
144 * [union with @skc_incoming_cpu]
145 * @skc_refcnt: reference count
146 *
147 * This is the minimal network layer representation of sockets, the header
148 * for struct sock and struct inet_timewait_sock.
149 */
150struct sock_common {
151 union {
152 __addrpair skc_addrpair;
153 struct {
154 __be32 skc_daddr;
155 __be32 skc_rcv_saddr;
156 };
157 };
158 union {
159 unsigned int skc_hash;
160 __u16 skc_u16hashes[2];
161 };
162 /* skc_dport && skc_num must be grouped as well */
163 union {
164 __portpair skc_portpair;
165 struct {
166 __be16 skc_dport;
167 __u16 skc_num;
168 };
169 };
170
171 unsigned short skc_family;
172 volatile unsigned char skc_state;
173 unsigned char skc_reuse:4;
174 unsigned char skc_reuseport:1;
175 unsigned char skc_ipv6only:1;
176 unsigned char skc_net_refcnt:1;
177 int skc_bound_dev_if;
178 union {
179 struct hlist_node skc_bind_node;
180 struct hlist_node skc_portaddr_node;
181 };
182 struct proto *skc_prot;
183 possible_net_t skc_net;
184
185#if IS_ENABLED(CONFIG_IPV6)
186 struct in6_addr skc_v6_daddr;
187 struct in6_addr skc_v6_rcv_saddr;
188#endif
189
190 atomic64_t skc_cookie;
191
192 /* following fields are padding to force
193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
194 * assuming IPV6 is enabled. We use this padding differently
195 * for different kind of 'sockets'
196 */
197 union {
198 unsigned long skc_flags;
199 struct sock *skc_listener; /* request_sock */
200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
201 };
202 /*
203 * fields between dontcopy_begin/dontcopy_end
204 * are not copied in sock_copy()
205 */
206 /* private: */
207 int skc_dontcopy_begin[0];
208 /* public: */
209 union {
210 struct hlist_node skc_node;
211 struct hlist_nulls_node skc_nulls_node;
212 };
213 unsigned short skc_tx_queue_mapping;
214#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
215 unsigned short skc_rx_queue_mapping;
216#endif
217 union {
218 int skc_incoming_cpu;
219 u32 skc_rcv_wnd;
220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
221 };
222
223 refcount_t skc_refcnt;
224 /* private: */
225 int skc_dontcopy_end[0];
226 union {
227 u32 skc_rxhash;
228 u32 skc_window_clamp;
229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
230 };
231 /* public: */
232};
233
234struct bpf_local_storage;
235struct sk_filter;
236
237/**
238 * struct sock - network layer representation of sockets
239 * @__sk_common: shared layout with inet_timewait_sock
240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
242 * @sk_lock: synchronizer
243 * @sk_kern_sock: True if sock is using kernel lock classes
244 * @sk_rcvbuf: size of receive buffer in bytes
245 * @sk_wq: sock wait queue and async head
246 * @sk_rx_dst: receive input route used by early demux
247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
249 * @sk_dst_cache: destination cache
250 * @sk_dst_pending_confirm: need to confirm neighbour
251 * @sk_policy: flow policy
252 * @sk_receive_queue: incoming packets
253 * @sk_wmem_alloc: transmit queue bytes committed
254 * @sk_tsq_flags: TCP Small Queues flags
255 * @sk_write_queue: Packet sending queue
256 * @sk_omem_alloc: "o" is "option" or "other"
257 * @sk_wmem_queued: persistent queue size
258 * @sk_forward_alloc: space allocated forward
259 * @sk_reserved_mem: space reserved and non-reclaimable for the socket
260 * @sk_napi_id: id of the last napi context to receive data for sk
261 * @sk_ll_usec: usecs to busypoll when there is no data
262 * @sk_allocation: allocation mode
263 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
264 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
265 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
266 * @sk_sndbuf: size of send buffer in bytes
267 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
268 * @sk_no_check_rx: allow zero checksum in RX packets
269 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
270 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
271 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
272 * @sk_gso_max_size: Maximum GSO segment size to build
273 * @sk_gso_max_segs: Maximum number of GSO segments
274 * @sk_pacing_shift: scaling factor for TCP Small Queues
275 * @sk_lingertime: %SO_LINGER l_linger setting
276 * @sk_backlog: always used with the per-socket spinlock held
277 * @sk_callback_lock: used with the callbacks in the end of this struct
278 * @sk_error_queue: rarely used
279 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
280 * IPV6_ADDRFORM for instance)
281 * @sk_err: last error
282 * @sk_err_soft: errors that don't cause failure but are the cause of a
283 * persistent failure not just 'timed out'
284 * @sk_drops: raw/udp drops counter
285 * @sk_ack_backlog: current listen backlog
286 * @sk_max_ack_backlog: listen backlog set in listen()
287 * @sk_uid: user id of owner
288 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
289 * @sk_busy_poll_budget: napi processing budget when busypolling
290 * @sk_priority: %SO_PRIORITY setting
291 * @sk_type: socket type (%SOCK_STREAM, etc)
292 * @sk_protocol: which protocol this socket belongs in this network family
293 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
294 * @sk_peer_pid: &struct pid for this socket's peer
295 * @sk_peer_cred: %SO_PEERCRED setting
296 * @sk_rcvlowat: %SO_RCVLOWAT setting
297 * @sk_rcvtimeo: %SO_RCVTIMEO setting
298 * @sk_sndtimeo: %SO_SNDTIMEO setting
299 * @sk_txhash: computed flow hash for use on transmit
300 * @sk_txrehash: enable TX hash rethink
301 * @sk_filter: socket filtering instructions
302 * @sk_timer: sock cleanup timer
303 * @sk_stamp: time stamp of last packet received
304 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
305 * @sk_tsflags: SO_TIMESTAMPING flags
306 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
307 * Sockets that can be used under memory reclaim should
308 * set this to false.
309 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
310 * for timestamping
311 * @sk_tskey: counter to disambiguate concurrent tstamp requests
312 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
313 * @sk_socket: Identd and reporting IO signals
314 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
315 * @sk_frag: cached page frag
316 * @sk_peek_off: current peek_offset value
317 * @sk_send_head: front of stuff to transmit
318 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
319 * @sk_security: used by security modules
320 * @sk_mark: generic packet mark
321 * @sk_cgrp_data: cgroup data for this cgroup
322 * @sk_memcg: this socket's memory cgroup association
323 * @sk_write_pending: a write to stream socket waits to start
324 * @sk_disconnects: number of disconnect operations performed on this sock
325 * @sk_state_change: callback to indicate change in the state of the sock
326 * @sk_data_ready: callback to indicate there is data to be processed
327 * @sk_write_space: callback to indicate there is bf sending space available
328 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
329 * @sk_backlog_rcv: callback to process the backlog
330 * @sk_validate_xmit_skb: ptr to an optional validate function
331 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
332 * @sk_reuseport_cb: reuseport group container
333 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
334 * @sk_rcu: used during RCU grace period
335 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
336 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
337 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
338 * @sk_txtime_unused: unused txtime flags
339 * @ns_tracker: tracker for netns reference
340 * @sk_user_frags: xarray of pages the user is holding a reference on.
341 */
342struct sock {
343 /*
344 * Now struct inet_timewait_sock also uses sock_common, so please just
345 * don't add nothing before this first member (__sk_common) --acme
346 */
347 struct sock_common __sk_common;
348#define sk_node __sk_common.skc_node
349#define sk_nulls_node __sk_common.skc_nulls_node
350#define sk_refcnt __sk_common.skc_refcnt
351#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
352#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
353#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
354#endif
355
356#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
357#define sk_dontcopy_end __sk_common.skc_dontcopy_end
358#define sk_hash __sk_common.skc_hash
359#define sk_portpair __sk_common.skc_portpair
360#define sk_num __sk_common.skc_num
361#define sk_dport __sk_common.skc_dport
362#define sk_addrpair __sk_common.skc_addrpair
363#define sk_daddr __sk_common.skc_daddr
364#define sk_rcv_saddr __sk_common.skc_rcv_saddr
365#define sk_family __sk_common.skc_family
366#define sk_state __sk_common.skc_state
367#define sk_reuse __sk_common.skc_reuse
368#define sk_reuseport __sk_common.skc_reuseport
369#define sk_ipv6only __sk_common.skc_ipv6only
370#define sk_net_refcnt __sk_common.skc_net_refcnt
371#define sk_bound_dev_if __sk_common.skc_bound_dev_if
372#define sk_bind_node __sk_common.skc_bind_node
373#define sk_prot __sk_common.skc_prot
374#define sk_net __sk_common.skc_net
375#define sk_v6_daddr __sk_common.skc_v6_daddr
376#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
377#define sk_cookie __sk_common.skc_cookie
378#define sk_incoming_cpu __sk_common.skc_incoming_cpu
379#define sk_flags __sk_common.skc_flags
380#define sk_rxhash __sk_common.skc_rxhash
381
382 __cacheline_group_begin(sock_write_rx);
383
384 atomic_t sk_drops;
385 __s32 sk_peek_off;
386 struct sk_buff_head sk_error_queue;
387 struct sk_buff_head sk_receive_queue;
388 /*
389 * The backlog queue is special, it is always used with
390 * the per-socket spinlock held and requires low latency
391 * access. Therefore we special case it's implementation.
392 * Note : rmem_alloc is in this structure to fill a hole
393 * on 64bit arches, not because its logically part of
394 * backlog.
395 */
396 struct {
397 atomic_t rmem_alloc;
398 int len;
399 struct sk_buff *head;
400 struct sk_buff *tail;
401 } sk_backlog;
402#define sk_rmem_alloc sk_backlog.rmem_alloc
403
404 __cacheline_group_end(sock_write_rx);
405
406 __cacheline_group_begin(sock_read_rx);
407 /* early demux fields */
408 struct dst_entry __rcu *sk_rx_dst;
409 int sk_rx_dst_ifindex;
410 u32 sk_rx_dst_cookie;
411
412#ifdef CONFIG_NET_RX_BUSY_POLL
413 unsigned int sk_ll_usec;
414 unsigned int sk_napi_id;
415 u16 sk_busy_poll_budget;
416 u8 sk_prefer_busy_poll;
417#endif
418 u8 sk_userlocks;
419 int sk_rcvbuf;
420
421 struct sk_filter __rcu *sk_filter;
422 union {
423 struct socket_wq __rcu *sk_wq;
424 /* private: */
425 struct socket_wq *sk_wq_raw;
426 /* public: */
427 };
428
429 void (*sk_data_ready)(struct sock *sk);
430 long sk_rcvtimeo;
431 int sk_rcvlowat;
432 __cacheline_group_end(sock_read_rx);
433
434 __cacheline_group_begin(sock_read_rxtx);
435 int sk_err;
436 struct socket *sk_socket;
437 struct mem_cgroup *sk_memcg;
438#ifdef CONFIG_XFRM
439 struct xfrm_policy __rcu *sk_policy[2];
440#endif
441 __cacheline_group_end(sock_read_rxtx);
442
443 __cacheline_group_begin(sock_write_rxtx);
444 socket_lock_t sk_lock;
445 u32 sk_reserved_mem;
446 int sk_forward_alloc;
447 u32 sk_tsflags;
448 __cacheline_group_end(sock_write_rxtx);
449
450 __cacheline_group_begin(sock_write_tx);
451 int sk_write_pending;
452 atomic_t sk_omem_alloc;
453 int sk_sndbuf;
454
455 int sk_wmem_queued;
456 refcount_t sk_wmem_alloc;
457 unsigned long sk_tsq_flags;
458 union {
459 struct sk_buff *sk_send_head;
460 struct rb_root tcp_rtx_queue;
461 };
462 struct sk_buff_head sk_write_queue;
463 u32 sk_dst_pending_confirm;
464 u32 sk_pacing_status; /* see enum sk_pacing */
465 struct page_frag sk_frag;
466 struct timer_list sk_timer;
467
468 unsigned long sk_pacing_rate; /* bytes per second */
469 atomic_t sk_zckey;
470 atomic_t sk_tskey;
471 __cacheline_group_end(sock_write_tx);
472
473 __cacheline_group_begin(sock_read_tx);
474 unsigned long sk_max_pacing_rate;
475 long sk_sndtimeo;
476 u32 sk_priority;
477 u32 sk_mark;
478 struct dst_entry __rcu *sk_dst_cache;
479 netdev_features_t sk_route_caps;
480#ifdef CONFIG_SOCK_VALIDATE_XMIT
481 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
482 struct net_device *dev,
483 struct sk_buff *skb);
484#endif
485 u16 sk_gso_type;
486 u16 sk_gso_max_segs;
487 unsigned int sk_gso_max_size;
488 gfp_t sk_allocation;
489 u32 sk_txhash;
490 u8 sk_pacing_shift;
491 bool sk_use_task_frag;
492 __cacheline_group_end(sock_read_tx);
493
494 /*
495 * Because of non atomicity rules, all
496 * changes are protected by socket lock.
497 */
498 u8 sk_gso_disabled : 1,
499 sk_kern_sock : 1,
500 sk_no_check_tx : 1,
501 sk_no_check_rx : 1;
502 u8 sk_shutdown;
503 u16 sk_type;
504 u16 sk_protocol;
505 unsigned long sk_lingertime;
506 struct proto *sk_prot_creator;
507 rwlock_t sk_callback_lock;
508 int sk_err_soft;
509 u32 sk_ack_backlog;
510 u32 sk_max_ack_backlog;
511 kuid_t sk_uid;
512 spinlock_t sk_peer_lock;
513 int sk_bind_phc;
514 struct pid *sk_peer_pid;
515 const struct cred *sk_peer_cred;
516
517 ktime_t sk_stamp;
518#if BITS_PER_LONG==32
519 seqlock_t sk_stamp_seq;
520#endif
521 int sk_disconnects;
522
523 u8 sk_txrehash;
524 u8 sk_clockid;
525 u8 sk_txtime_deadline_mode : 1,
526 sk_txtime_report_errors : 1,
527 sk_txtime_unused : 6;
528
529 void *sk_user_data;
530#ifdef CONFIG_SECURITY
531 void *sk_security;
532#endif
533 struct sock_cgroup_data sk_cgrp_data;
534 void (*sk_state_change)(struct sock *sk);
535 void (*sk_write_space)(struct sock *sk);
536 void (*sk_error_report)(struct sock *sk);
537 int (*sk_backlog_rcv)(struct sock *sk,
538 struct sk_buff *skb);
539 void (*sk_destruct)(struct sock *sk);
540 struct sock_reuseport __rcu *sk_reuseport_cb;
541#ifdef CONFIG_BPF_SYSCALL
542 struct bpf_local_storage __rcu *sk_bpf_storage;
543#endif
544 struct rcu_head sk_rcu;
545 netns_tracker ns_tracker;
546 struct xarray sk_user_frags;
547};
548
549struct sock_bh_locked {
550 struct sock *sock;
551 local_lock_t bh_lock;
552};
553
554enum sk_pacing {
555 SK_PACING_NONE = 0,
556 SK_PACING_NEEDED = 1,
557 SK_PACING_FQ = 2,
558};
559
560/* flag bits in sk_user_data
561 *
562 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
563 * not be suitable for copying when cloning the socket. For instance,
564 * it can point to a reference counted object. sk_user_data bottom
565 * bit is set if pointer must not be copied.
566 *
567 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
568 * managed/owned by a BPF reuseport array. This bit should be set
569 * when sk_user_data's sk is added to the bpf's reuseport_array.
570 *
571 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
572 * sk_user_data points to psock type. This bit should be set
573 * when sk_user_data is assigned to a psock object.
574 */
575#define SK_USER_DATA_NOCOPY 1UL
576#define SK_USER_DATA_BPF 2UL
577#define SK_USER_DATA_PSOCK 4UL
578#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
579 SK_USER_DATA_PSOCK)
580
581/**
582 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
583 * @sk: socket
584 */
585static inline bool sk_user_data_is_nocopy(const struct sock *sk)
586{
587 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
588}
589
590#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
591
592/**
593 * __locked_read_sk_user_data_with_flags - return the pointer
594 * only if argument flags all has been set in sk_user_data. Otherwise
595 * return NULL
596 *
597 * @sk: socket
598 * @flags: flag bits
599 *
600 * The caller must be holding sk->sk_callback_lock.
601 */
602static inline void *
603__locked_read_sk_user_data_with_flags(const struct sock *sk,
604 uintptr_t flags)
605{
606 uintptr_t sk_user_data =
607 (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
608 lockdep_is_held(&sk->sk_callback_lock));
609
610 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
611
612 if ((sk_user_data & flags) == flags)
613 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
614 return NULL;
615}
616
617/**
618 * __rcu_dereference_sk_user_data_with_flags - return the pointer
619 * only if argument flags all has been set in sk_user_data. Otherwise
620 * return NULL
621 *
622 * @sk: socket
623 * @flags: flag bits
624 */
625static inline void *
626__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
627 uintptr_t flags)
628{
629 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
630
631 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
632
633 if ((sk_user_data & flags) == flags)
634 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
635 return NULL;
636}
637
638#define rcu_dereference_sk_user_data(sk) \
639 __rcu_dereference_sk_user_data_with_flags(sk, 0)
640#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
641({ \
642 uintptr_t __tmp1 = (uintptr_t)(ptr), \
643 __tmp2 = (uintptr_t)(flags); \
644 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
645 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
646 rcu_assign_pointer(__sk_user_data((sk)), \
647 __tmp1 | __tmp2); \
648})
649#define rcu_assign_sk_user_data(sk, ptr) \
650 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
651
652static inline
653struct net *sock_net(const struct sock *sk)
654{
655 return read_pnet(&sk->sk_net);
656}
657
658static inline
659void sock_net_set(struct sock *sk, struct net *net)
660{
661 write_pnet(&sk->sk_net, net);
662}
663
664/*
665 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
666 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
667 * on a socket means that the socket will reuse everybody else's port
668 * without looking at the other's sk_reuse value.
669 */
670
671#define SK_NO_REUSE 0
672#define SK_CAN_REUSE 1
673#define SK_FORCE_REUSE 2
674
675int sk_set_peek_off(struct sock *sk, int val);
676
677static inline int sk_peek_offset(const struct sock *sk, int flags)
678{
679 if (unlikely(flags & MSG_PEEK)) {
680 return READ_ONCE(sk->sk_peek_off);
681 }
682
683 return 0;
684}
685
686static inline void sk_peek_offset_bwd(struct sock *sk, int val)
687{
688 s32 off = READ_ONCE(sk->sk_peek_off);
689
690 if (unlikely(off >= 0)) {
691 off = max_t(s32, off - val, 0);
692 WRITE_ONCE(sk->sk_peek_off, off);
693 }
694}
695
696static inline void sk_peek_offset_fwd(struct sock *sk, int val)
697{
698 sk_peek_offset_bwd(sk, -val);
699}
700
701/*
702 * Hashed lists helper routines
703 */
704static inline struct sock *sk_entry(const struct hlist_node *node)
705{
706 return hlist_entry(node, struct sock, sk_node);
707}
708
709static inline struct sock *__sk_head(const struct hlist_head *head)
710{
711 return hlist_entry(head->first, struct sock, sk_node);
712}
713
714static inline struct sock *sk_head(const struct hlist_head *head)
715{
716 return hlist_empty(head) ? NULL : __sk_head(head);
717}
718
719static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
720{
721 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
722}
723
724static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
725{
726 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
727}
728
729static inline struct sock *sk_next(const struct sock *sk)
730{
731 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
732}
733
734static inline struct sock *sk_nulls_next(const struct sock *sk)
735{
736 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
737 hlist_nulls_entry(sk->sk_nulls_node.next,
738 struct sock, sk_nulls_node) :
739 NULL;
740}
741
742static inline bool sk_unhashed(const struct sock *sk)
743{
744 return hlist_unhashed(&sk->sk_node);
745}
746
747static inline bool sk_hashed(const struct sock *sk)
748{
749 return !sk_unhashed(sk);
750}
751
752static inline void sk_node_init(struct hlist_node *node)
753{
754 node->pprev = NULL;
755}
756
757static inline void __sk_del_node(struct sock *sk)
758{
759 __hlist_del(&sk->sk_node);
760}
761
762/* NB: equivalent to hlist_del_init_rcu */
763static inline bool __sk_del_node_init(struct sock *sk)
764{
765 if (sk_hashed(sk)) {
766 __sk_del_node(sk);
767 sk_node_init(&sk->sk_node);
768 return true;
769 }
770 return false;
771}
772
773/* Grab socket reference count. This operation is valid only
774 when sk is ALREADY grabbed f.e. it is found in hash table
775 or a list and the lookup is made under lock preventing hash table
776 modifications.
777 */
778
779static __always_inline void sock_hold(struct sock *sk)
780{
781 refcount_inc(&sk->sk_refcnt);
782}
783
784/* Ungrab socket in the context, which assumes that socket refcnt
785 cannot hit zero, f.e. it is true in context of any socketcall.
786 */
787static __always_inline void __sock_put(struct sock *sk)
788{
789 refcount_dec(&sk->sk_refcnt);
790}
791
792static inline bool sk_del_node_init(struct sock *sk)
793{
794 bool rc = __sk_del_node_init(sk);
795
796 if (rc) {
797 /* paranoid for a while -acme */
798 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
799 __sock_put(sk);
800 }
801 return rc;
802}
803#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
804
805static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
806{
807 if (sk_hashed(sk)) {
808 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
809 return true;
810 }
811 return false;
812}
813
814static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
815{
816 bool rc = __sk_nulls_del_node_init_rcu(sk);
817
818 if (rc) {
819 /* paranoid for a while -acme */
820 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
821 __sock_put(sk);
822 }
823 return rc;
824}
825
826static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
827{
828 hlist_add_head(&sk->sk_node, list);
829}
830
831static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
832{
833 sock_hold(sk);
834 __sk_add_node(sk, list);
835}
836
837static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
838{
839 sock_hold(sk);
840 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
841 sk->sk_family == AF_INET6)
842 hlist_add_tail_rcu(&sk->sk_node, list);
843 else
844 hlist_add_head_rcu(&sk->sk_node, list);
845}
846
847static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
848{
849 sock_hold(sk);
850 hlist_add_tail_rcu(&sk->sk_node, list);
851}
852
853static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
854{
855 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
856}
857
858static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
859{
860 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
861}
862
863static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
864{
865 sock_hold(sk);
866 __sk_nulls_add_node_rcu(sk, list);
867}
868
869static inline void __sk_del_bind_node(struct sock *sk)
870{
871 __hlist_del(&sk->sk_bind_node);
872}
873
874static inline void sk_add_bind_node(struct sock *sk,
875 struct hlist_head *list)
876{
877 hlist_add_head(&sk->sk_bind_node, list);
878}
879
880#define sk_for_each(__sk, list) \
881 hlist_for_each_entry(__sk, list, sk_node)
882#define sk_for_each_rcu(__sk, list) \
883 hlist_for_each_entry_rcu(__sk, list, sk_node)
884#define sk_nulls_for_each(__sk, node, list) \
885 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
886#define sk_nulls_for_each_rcu(__sk, node, list) \
887 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
888#define sk_for_each_from(__sk) \
889 hlist_for_each_entry_from(__sk, sk_node)
890#define sk_nulls_for_each_from(__sk, node) \
891 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
892 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
893#define sk_for_each_safe(__sk, tmp, list) \
894 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
895#define sk_for_each_bound(__sk, list) \
896 hlist_for_each_entry(__sk, list, sk_bind_node)
897#define sk_for_each_bound_safe(__sk, tmp, list) \
898 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node)
899
900/**
901 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
902 * @tpos: the type * to use as a loop cursor.
903 * @pos: the &struct hlist_node to use as a loop cursor.
904 * @head: the head for your list.
905 * @offset: offset of hlist_node within the struct.
906 *
907 */
908#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
909 for (pos = rcu_dereference(hlist_first_rcu(head)); \
910 pos != NULL && \
911 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
912 pos = rcu_dereference(hlist_next_rcu(pos)))
913
914static inline struct user_namespace *sk_user_ns(const struct sock *sk)
915{
916 /* Careful only use this in a context where these parameters
917 * can not change and must all be valid, such as recvmsg from
918 * userspace.
919 */
920 return sk->sk_socket->file->f_cred->user_ns;
921}
922
923/* Sock flags */
924enum sock_flags {
925 SOCK_DEAD,
926 SOCK_DONE,
927 SOCK_URGINLINE,
928 SOCK_KEEPOPEN,
929 SOCK_LINGER,
930 SOCK_DESTROY,
931 SOCK_BROADCAST,
932 SOCK_TIMESTAMP,
933 SOCK_ZAPPED,
934 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
935 SOCK_DBG, /* %SO_DEBUG setting */
936 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
937 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
938 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
939 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
940 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
941 SOCK_FASYNC, /* fasync() active */
942 SOCK_RXQ_OVFL,
943 SOCK_ZEROCOPY, /* buffers from userspace */
944 SOCK_WIFI_STATUS, /* push wifi status to userspace */
945 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
946 * Will use last 4 bytes of packet sent from
947 * user-space instead.
948 */
949 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
950 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
951 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
952 SOCK_TXTIME,
953 SOCK_XDP, /* XDP is attached */
954 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
955 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
956};
957
958#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
959/*
960 * The highest bit of sk_tsflags is reserved for kernel-internal
961 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
962 * SOF_TIMESTAMPING* values do not reach this reserved area
963 */
964#define SOCKCM_FLAG_TS_OPT_ID BIT(31)
965
966static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
967{
968 nsk->sk_flags = osk->sk_flags;
969}
970
971static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
972{
973 __set_bit(flag, &sk->sk_flags);
974}
975
976static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
977{
978 __clear_bit(flag, &sk->sk_flags);
979}
980
981static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
982 int valbool)
983{
984 if (valbool)
985 sock_set_flag(sk, bit);
986 else
987 sock_reset_flag(sk, bit);
988}
989
990static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
991{
992 return test_bit(flag, &sk->sk_flags);
993}
994
995#ifdef CONFIG_NET
996DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
997static inline int sk_memalloc_socks(void)
998{
999 return static_branch_unlikely(&memalloc_socks_key);
1000}
1001
1002void __receive_sock(struct file *file);
1003#else
1004
1005static inline int sk_memalloc_socks(void)
1006{
1007 return 0;
1008}
1009
1010static inline void __receive_sock(struct file *file)
1011{ }
1012#endif
1013
1014static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1015{
1016 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1017}
1018
1019static inline void sk_acceptq_removed(struct sock *sk)
1020{
1021 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1022}
1023
1024static inline void sk_acceptq_added(struct sock *sk)
1025{
1026 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1027}
1028
1029/* Note: If you think the test should be:
1030 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1031 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1032 */
1033static inline bool sk_acceptq_is_full(const struct sock *sk)
1034{
1035 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1036}
1037
1038/*
1039 * Compute minimal free write space needed to queue new packets.
1040 */
1041static inline int sk_stream_min_wspace(const struct sock *sk)
1042{
1043 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1044}
1045
1046static inline int sk_stream_wspace(const struct sock *sk)
1047{
1048 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1049}
1050
1051static inline void sk_wmem_queued_add(struct sock *sk, int val)
1052{
1053 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1054}
1055
1056static inline void sk_forward_alloc_add(struct sock *sk, int val)
1057{
1058 /* Paired with lockless reads of sk->sk_forward_alloc */
1059 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1060}
1061
1062void sk_stream_write_space(struct sock *sk);
1063
1064/* OOB backlog add */
1065static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1066{
1067 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1068 skb_dst_force(skb);
1069
1070 if (!sk->sk_backlog.tail)
1071 WRITE_ONCE(sk->sk_backlog.head, skb);
1072 else
1073 sk->sk_backlog.tail->next = skb;
1074
1075 WRITE_ONCE(sk->sk_backlog.tail, skb);
1076 skb->next = NULL;
1077}
1078
1079/*
1080 * Take into account size of receive queue and backlog queue
1081 * Do not take into account this skb truesize,
1082 * to allow even a single big packet to come.
1083 */
1084static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1085{
1086 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1087
1088 return qsize > limit;
1089}
1090
1091/* The per-socket spinlock must be held here. */
1092static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1093 unsigned int limit)
1094{
1095 if (sk_rcvqueues_full(sk, limit))
1096 return -ENOBUFS;
1097
1098 /*
1099 * If the skb was allocated from pfmemalloc reserves, only
1100 * allow SOCK_MEMALLOC sockets to use it as this socket is
1101 * helping free memory
1102 */
1103 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1104 return -ENOMEM;
1105
1106 __sk_add_backlog(sk, skb);
1107 sk->sk_backlog.len += skb->truesize;
1108 return 0;
1109}
1110
1111int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1112
1113INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1114INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1115
1116static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1117{
1118 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1119 return __sk_backlog_rcv(sk, skb);
1120
1121 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1122 tcp_v6_do_rcv,
1123 tcp_v4_do_rcv,
1124 sk, skb);
1125}
1126
1127static inline void sk_incoming_cpu_update(struct sock *sk)
1128{
1129 int cpu = raw_smp_processor_id();
1130
1131 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1132 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1133}
1134
1135
1136static inline void sock_rps_save_rxhash(struct sock *sk,
1137 const struct sk_buff *skb)
1138{
1139#ifdef CONFIG_RPS
1140 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1141 * here, and another one in sock_rps_record_flow().
1142 */
1143 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1144 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1145#endif
1146}
1147
1148static inline void sock_rps_reset_rxhash(struct sock *sk)
1149{
1150#ifdef CONFIG_RPS
1151 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1152 WRITE_ONCE(sk->sk_rxhash, 0);
1153#endif
1154}
1155
1156#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1157 ({ int __rc, __dis = __sk->sk_disconnects; \
1158 release_sock(__sk); \
1159 __rc = __condition; \
1160 if (!__rc) { \
1161 *(__timeo) = wait_woken(__wait, \
1162 TASK_INTERRUPTIBLE, \
1163 *(__timeo)); \
1164 } \
1165 sched_annotate_sleep(); \
1166 lock_sock(__sk); \
1167 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1168 __rc; \
1169 })
1170
1171int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1172int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1173void sk_stream_wait_close(struct sock *sk, long timeo_p);
1174int sk_stream_error(struct sock *sk, int flags, int err);
1175void sk_stream_kill_queues(struct sock *sk);
1176void sk_set_memalloc(struct sock *sk);
1177void sk_clear_memalloc(struct sock *sk);
1178
1179void __sk_flush_backlog(struct sock *sk);
1180
1181static inline bool sk_flush_backlog(struct sock *sk)
1182{
1183 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1184 __sk_flush_backlog(sk);
1185 return true;
1186 }
1187 return false;
1188}
1189
1190int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1191
1192struct request_sock_ops;
1193struct timewait_sock_ops;
1194struct inet_hashinfo;
1195struct raw_hashinfo;
1196struct smc_hashinfo;
1197struct module;
1198struct sk_psock;
1199
1200/*
1201 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1202 * un-modified. Special care is taken when initializing object to zero.
1203 */
1204static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1205{
1206 if (offsetof(struct sock, sk_node.next) != 0)
1207 memset(sk, 0, offsetof(struct sock, sk_node.next));
1208 memset(&sk->sk_node.pprev, 0,
1209 size - offsetof(struct sock, sk_node.pprev));
1210}
1211
1212struct proto_accept_arg {
1213 int flags;
1214 int err;
1215 int is_empty;
1216 bool kern;
1217};
1218
1219/* Networking protocol blocks we attach to sockets.
1220 * socket layer -> transport layer interface
1221 */
1222struct proto {
1223 void (*close)(struct sock *sk,
1224 long timeout);
1225 int (*pre_connect)(struct sock *sk,
1226 struct sockaddr *uaddr,
1227 int addr_len);
1228 int (*connect)(struct sock *sk,
1229 struct sockaddr *uaddr,
1230 int addr_len);
1231 int (*disconnect)(struct sock *sk, int flags);
1232
1233 struct sock * (*accept)(struct sock *sk,
1234 struct proto_accept_arg *arg);
1235
1236 int (*ioctl)(struct sock *sk, int cmd,
1237 int *karg);
1238 int (*init)(struct sock *sk);
1239 void (*destroy)(struct sock *sk);
1240 void (*shutdown)(struct sock *sk, int how);
1241 int (*setsockopt)(struct sock *sk, int level,
1242 int optname, sockptr_t optval,
1243 unsigned int optlen);
1244 int (*getsockopt)(struct sock *sk, int level,
1245 int optname, char __user *optval,
1246 int __user *option);
1247 void (*keepalive)(struct sock *sk, int valbool);
1248#ifdef CONFIG_COMPAT
1249 int (*compat_ioctl)(struct sock *sk,
1250 unsigned int cmd, unsigned long arg);
1251#endif
1252 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1253 size_t len);
1254 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1255 size_t len, int flags, int *addr_len);
1256 void (*splice_eof)(struct socket *sock);
1257 int (*bind)(struct sock *sk,
1258 struct sockaddr *addr, int addr_len);
1259 int (*bind_add)(struct sock *sk,
1260 struct sockaddr *addr, int addr_len);
1261
1262 int (*backlog_rcv) (struct sock *sk,
1263 struct sk_buff *skb);
1264 bool (*bpf_bypass_getsockopt)(int level,
1265 int optname);
1266
1267 void (*release_cb)(struct sock *sk);
1268
1269 /* Keeping track of sk's, looking them up, and port selection methods. */
1270 int (*hash)(struct sock *sk);
1271 void (*unhash)(struct sock *sk);
1272 void (*rehash)(struct sock *sk);
1273 int (*get_port)(struct sock *sk, unsigned short snum);
1274 void (*put_port)(struct sock *sk);
1275#ifdef CONFIG_BPF_SYSCALL
1276 int (*psock_update_sk_prot)(struct sock *sk,
1277 struct sk_psock *psock,
1278 bool restore);
1279#endif
1280
1281 /* Keeping track of sockets in use */
1282#ifdef CONFIG_PROC_FS
1283 unsigned int inuse_idx;
1284#endif
1285
1286#if IS_ENABLED(CONFIG_MPTCP)
1287 int (*forward_alloc_get)(const struct sock *sk);
1288#endif
1289
1290 bool (*stream_memory_free)(const struct sock *sk, int wake);
1291 bool (*sock_is_readable)(struct sock *sk);
1292 /* Memory pressure */
1293 void (*enter_memory_pressure)(struct sock *sk);
1294 void (*leave_memory_pressure)(struct sock *sk);
1295 atomic_long_t *memory_allocated; /* Current allocated memory. */
1296 int __percpu *per_cpu_fw_alloc;
1297 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1298
1299 /*
1300 * Pressure flag: try to collapse.
1301 * Technical note: it is used by multiple contexts non atomically.
1302 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1303 * All the __sk_mem_schedule() is of this nature: accounting
1304 * is strict, actions are advisory and have some latency.
1305 */
1306 unsigned long *memory_pressure;
1307 long *sysctl_mem;
1308
1309 int *sysctl_wmem;
1310 int *sysctl_rmem;
1311 u32 sysctl_wmem_offset;
1312 u32 sysctl_rmem_offset;
1313
1314 int max_header;
1315 bool no_autobind;
1316
1317 struct kmem_cache *slab;
1318 unsigned int obj_size;
1319 unsigned int ipv6_pinfo_offset;
1320 slab_flags_t slab_flags;
1321 unsigned int useroffset; /* Usercopy region offset */
1322 unsigned int usersize; /* Usercopy region size */
1323
1324 unsigned int __percpu *orphan_count;
1325
1326 struct request_sock_ops *rsk_prot;
1327 struct timewait_sock_ops *twsk_prot;
1328
1329 union {
1330 struct inet_hashinfo *hashinfo;
1331 struct udp_table *udp_table;
1332 struct raw_hashinfo *raw_hash;
1333 struct smc_hashinfo *smc_hash;
1334 } h;
1335
1336 struct module *owner;
1337
1338 char name[32];
1339
1340 struct list_head node;
1341 int (*diag_destroy)(struct sock *sk, int err);
1342} __randomize_layout;
1343
1344int proto_register(struct proto *prot, int alloc_slab);
1345void proto_unregister(struct proto *prot);
1346int sock_load_diag_module(int family, int protocol);
1347
1348INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1349
1350static inline int sk_forward_alloc_get(const struct sock *sk)
1351{
1352#if IS_ENABLED(CONFIG_MPTCP)
1353 if (sk->sk_prot->forward_alloc_get)
1354 return sk->sk_prot->forward_alloc_get(sk);
1355#endif
1356 return READ_ONCE(sk->sk_forward_alloc);
1357}
1358
1359static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1360{
1361 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1362 return false;
1363
1364 return sk->sk_prot->stream_memory_free ?
1365 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1366 tcp_stream_memory_free, sk, wake) : true;
1367}
1368
1369static inline bool sk_stream_memory_free(const struct sock *sk)
1370{
1371 return __sk_stream_memory_free(sk, 0);
1372}
1373
1374static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1375{
1376 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1377 __sk_stream_memory_free(sk, wake);
1378}
1379
1380static inline bool sk_stream_is_writeable(const struct sock *sk)
1381{
1382 return __sk_stream_is_writeable(sk, 0);
1383}
1384
1385static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1386 struct cgroup *ancestor)
1387{
1388#ifdef CONFIG_SOCK_CGROUP_DATA
1389 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1390 ancestor);
1391#else
1392 return -ENOTSUPP;
1393#endif
1394}
1395
1396#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1397
1398static inline void sk_sockets_allocated_dec(struct sock *sk)
1399{
1400 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1401 SK_ALLOC_PERCPU_COUNTER_BATCH);
1402}
1403
1404static inline void sk_sockets_allocated_inc(struct sock *sk)
1405{
1406 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1407 SK_ALLOC_PERCPU_COUNTER_BATCH);
1408}
1409
1410static inline u64
1411sk_sockets_allocated_read_positive(struct sock *sk)
1412{
1413 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1414}
1415
1416static inline int
1417proto_sockets_allocated_sum_positive(struct proto *prot)
1418{
1419 return percpu_counter_sum_positive(prot->sockets_allocated);
1420}
1421
1422#ifdef CONFIG_PROC_FS
1423#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1424struct prot_inuse {
1425 int all;
1426 int val[PROTO_INUSE_NR];
1427};
1428
1429static inline void sock_prot_inuse_add(const struct net *net,
1430 const struct proto *prot, int val)
1431{
1432 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1433}
1434
1435static inline void sock_inuse_add(const struct net *net, int val)
1436{
1437 this_cpu_add(net->core.prot_inuse->all, val);
1438}
1439
1440int sock_prot_inuse_get(struct net *net, struct proto *proto);
1441int sock_inuse_get(struct net *net);
1442#else
1443static inline void sock_prot_inuse_add(const struct net *net,
1444 const struct proto *prot, int val)
1445{
1446}
1447
1448static inline void sock_inuse_add(const struct net *net, int val)
1449{
1450}
1451#endif
1452
1453
1454/* With per-bucket locks this operation is not-atomic, so that
1455 * this version is not worse.
1456 */
1457static inline int __sk_prot_rehash(struct sock *sk)
1458{
1459 sk->sk_prot->unhash(sk);
1460 return sk->sk_prot->hash(sk);
1461}
1462
1463/* About 10 seconds */
1464#define SOCK_DESTROY_TIME (10*HZ)
1465
1466/* Sockets 0-1023 can't be bound to unless you are superuser */
1467#define PROT_SOCK 1024
1468
1469#define SHUTDOWN_MASK 3
1470#define RCV_SHUTDOWN 1
1471#define SEND_SHUTDOWN 2
1472
1473#define SOCK_BINDADDR_LOCK 4
1474#define SOCK_BINDPORT_LOCK 8
1475
1476struct socket_alloc {
1477 struct socket socket;
1478 struct inode vfs_inode;
1479};
1480
1481static inline struct socket *SOCKET_I(struct inode *inode)
1482{
1483 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1484}
1485
1486static inline struct inode *SOCK_INODE(struct socket *socket)
1487{
1488 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1489}
1490
1491/*
1492 * Functions for memory accounting
1493 */
1494int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1495int __sk_mem_schedule(struct sock *sk, int size, int kind);
1496void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1497void __sk_mem_reclaim(struct sock *sk, int amount);
1498
1499#define SK_MEM_SEND 0
1500#define SK_MEM_RECV 1
1501
1502/* sysctl_mem values are in pages */
1503static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1504{
1505 return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1506}
1507
1508static inline int sk_mem_pages(int amt)
1509{
1510 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1511}
1512
1513static inline bool sk_has_account(struct sock *sk)
1514{
1515 /* return true if protocol supports memory accounting */
1516 return !!sk->sk_prot->memory_allocated;
1517}
1518
1519static inline bool sk_wmem_schedule(struct sock *sk, int size)
1520{
1521 int delta;
1522
1523 if (!sk_has_account(sk))
1524 return true;
1525 delta = size - sk->sk_forward_alloc;
1526 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1527}
1528
1529static inline bool
1530__sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc)
1531{
1532 int delta;
1533
1534 if (!sk_has_account(sk))
1535 return true;
1536 delta = size - sk->sk_forward_alloc;
1537 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1538 pfmemalloc;
1539}
1540
1541static inline bool
1542sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1543{
1544 return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb));
1545}
1546
1547static inline int sk_unused_reserved_mem(const struct sock *sk)
1548{
1549 int unused_mem;
1550
1551 if (likely(!sk->sk_reserved_mem))
1552 return 0;
1553
1554 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1555 atomic_read(&sk->sk_rmem_alloc);
1556
1557 return unused_mem > 0 ? unused_mem : 0;
1558}
1559
1560static inline void sk_mem_reclaim(struct sock *sk)
1561{
1562 int reclaimable;
1563
1564 if (!sk_has_account(sk))
1565 return;
1566
1567 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1568
1569 if (reclaimable >= (int)PAGE_SIZE)
1570 __sk_mem_reclaim(sk, reclaimable);
1571}
1572
1573static inline void sk_mem_reclaim_final(struct sock *sk)
1574{
1575 sk->sk_reserved_mem = 0;
1576 sk_mem_reclaim(sk);
1577}
1578
1579static inline void sk_mem_charge(struct sock *sk, int size)
1580{
1581 if (!sk_has_account(sk))
1582 return;
1583 sk_forward_alloc_add(sk, -size);
1584}
1585
1586static inline void sk_mem_uncharge(struct sock *sk, int size)
1587{
1588 if (!sk_has_account(sk))
1589 return;
1590 sk_forward_alloc_add(sk, size);
1591 sk_mem_reclaim(sk);
1592}
1593
1594/*
1595 * Macro so as to not evaluate some arguments when
1596 * lockdep is not enabled.
1597 *
1598 * Mark both the sk_lock and the sk_lock.slock as a
1599 * per-address-family lock class.
1600 */
1601#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1602do { \
1603 sk->sk_lock.owned = 0; \
1604 init_waitqueue_head(&sk->sk_lock.wq); \
1605 spin_lock_init(&(sk)->sk_lock.slock); \
1606 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1607 sizeof((sk)->sk_lock)); \
1608 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1609 (skey), (sname)); \
1610 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1611} while (0)
1612
1613static inline bool lockdep_sock_is_held(const struct sock *sk)
1614{
1615 return lockdep_is_held(&sk->sk_lock) ||
1616 lockdep_is_held(&sk->sk_lock.slock);
1617}
1618
1619void lock_sock_nested(struct sock *sk, int subclass);
1620
1621static inline void lock_sock(struct sock *sk)
1622{
1623 lock_sock_nested(sk, 0);
1624}
1625
1626void __lock_sock(struct sock *sk);
1627void __release_sock(struct sock *sk);
1628void release_sock(struct sock *sk);
1629
1630/* BH context may only use the following locking interface. */
1631#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1632#define bh_lock_sock_nested(__sk) \
1633 spin_lock_nested(&((__sk)->sk_lock.slock), \
1634 SINGLE_DEPTH_NESTING)
1635#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1636
1637bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1638
1639/**
1640 * lock_sock_fast - fast version of lock_sock
1641 * @sk: socket
1642 *
1643 * This version should be used for very small section, where process won't block
1644 * return false if fast path is taken:
1645 *
1646 * sk_lock.slock locked, owned = 0, BH disabled
1647 *
1648 * return true if slow path is taken:
1649 *
1650 * sk_lock.slock unlocked, owned = 1, BH enabled
1651 */
1652static inline bool lock_sock_fast(struct sock *sk)
1653{
1654 /* The sk_lock has mutex_lock() semantics here. */
1655 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1656
1657 return __lock_sock_fast(sk);
1658}
1659
1660/* fast socket lock variant for caller already holding a [different] socket lock */
1661static inline bool lock_sock_fast_nested(struct sock *sk)
1662{
1663 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1664
1665 return __lock_sock_fast(sk);
1666}
1667
1668/**
1669 * unlock_sock_fast - complement of lock_sock_fast
1670 * @sk: socket
1671 * @slow: slow mode
1672 *
1673 * fast unlock socket for user context.
1674 * If slow mode is on, we call regular release_sock()
1675 */
1676static inline void unlock_sock_fast(struct sock *sk, bool slow)
1677 __releases(&sk->sk_lock.slock)
1678{
1679 if (slow) {
1680 release_sock(sk);
1681 __release(&sk->sk_lock.slock);
1682 } else {
1683 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1684 spin_unlock_bh(&sk->sk_lock.slock);
1685 }
1686}
1687
1688void sockopt_lock_sock(struct sock *sk);
1689void sockopt_release_sock(struct sock *sk);
1690bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1691bool sockopt_capable(int cap);
1692
1693/* Used by processes to "lock" a socket state, so that
1694 * interrupts and bottom half handlers won't change it
1695 * from under us. It essentially blocks any incoming
1696 * packets, so that we won't get any new data or any
1697 * packets that change the state of the socket.
1698 *
1699 * While locked, BH processing will add new packets to
1700 * the backlog queue. This queue is processed by the
1701 * owner of the socket lock right before it is released.
1702 *
1703 * Since ~2.3.5 it is also exclusive sleep lock serializing
1704 * accesses from user process context.
1705 */
1706
1707static inline void sock_owned_by_me(const struct sock *sk)
1708{
1709#ifdef CONFIG_LOCKDEP
1710 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1711#endif
1712}
1713
1714static inline void sock_not_owned_by_me(const struct sock *sk)
1715{
1716#ifdef CONFIG_LOCKDEP
1717 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1718#endif
1719}
1720
1721static inline bool sock_owned_by_user(const struct sock *sk)
1722{
1723 sock_owned_by_me(sk);
1724 return sk->sk_lock.owned;
1725}
1726
1727static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1728{
1729 return sk->sk_lock.owned;
1730}
1731
1732static inline void sock_release_ownership(struct sock *sk)
1733{
1734 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1735 sk->sk_lock.owned = 0;
1736
1737 /* The sk_lock has mutex_unlock() semantics: */
1738 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1739}
1740
1741/* no reclassification while locks are held */
1742static inline bool sock_allow_reclassification(const struct sock *csk)
1743{
1744 struct sock *sk = (struct sock *)csk;
1745
1746 return !sock_owned_by_user_nocheck(sk) &&
1747 !spin_is_locked(&sk->sk_lock.slock);
1748}
1749
1750struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1751 struct proto *prot, int kern);
1752void sk_free(struct sock *sk);
1753void sk_net_refcnt_upgrade(struct sock *sk);
1754void sk_destruct(struct sock *sk);
1755struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1756void sk_free_unlock_clone(struct sock *sk);
1757
1758struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1759 gfp_t priority);
1760void __sock_wfree(struct sk_buff *skb);
1761void sock_wfree(struct sk_buff *skb);
1762struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1763 gfp_t priority);
1764void skb_orphan_partial(struct sk_buff *skb);
1765void sock_rfree(struct sk_buff *skb);
1766void sock_efree(struct sk_buff *skb);
1767#ifdef CONFIG_INET
1768void sock_edemux(struct sk_buff *skb);
1769void sock_pfree(struct sk_buff *skb);
1770
1771static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk)
1772{
1773 skb_orphan(skb);
1774 if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1775 skb->sk = sk;
1776 skb->destructor = sock_edemux;
1777 }
1778}
1779#else
1780#define sock_edemux sock_efree
1781#endif
1782
1783int sk_setsockopt(struct sock *sk, int level, int optname,
1784 sockptr_t optval, unsigned int optlen);
1785int sock_setsockopt(struct socket *sock, int level, int op,
1786 sockptr_t optval, unsigned int optlen);
1787int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1788 int optname, sockptr_t optval, int optlen);
1789int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1790 int optname, sockptr_t optval, sockptr_t optlen);
1791
1792int sk_getsockopt(struct sock *sk, int level, int optname,
1793 sockptr_t optval, sockptr_t optlen);
1794int sock_gettstamp(struct socket *sock, void __user *userstamp,
1795 bool timeval, bool time32);
1796struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1797 unsigned long data_len, int noblock,
1798 int *errcode, int max_page_order);
1799
1800static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1801 unsigned long size,
1802 int noblock, int *errcode)
1803{
1804 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1805}
1806
1807void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1808void sock_kfree_s(struct sock *sk, void *mem, int size);
1809void sock_kzfree_s(struct sock *sk, void *mem, int size);
1810void sk_send_sigurg(struct sock *sk);
1811
1812static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1813{
1814 if (sk->sk_socket)
1815 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1816 WRITE_ONCE(sk->sk_prot, proto);
1817}
1818
1819struct sockcm_cookie {
1820 u64 transmit_time;
1821 u32 mark;
1822 u32 tsflags;
1823 u32 ts_opt_id;
1824};
1825
1826static inline void sockcm_init(struct sockcm_cookie *sockc,
1827 const struct sock *sk)
1828{
1829 *sockc = (struct sockcm_cookie) {
1830 .tsflags = READ_ONCE(sk->sk_tsflags)
1831 };
1832}
1833
1834int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1835 struct sockcm_cookie *sockc);
1836int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1837 struct sockcm_cookie *sockc);
1838
1839/*
1840 * Functions to fill in entries in struct proto_ops when a protocol
1841 * does not implement a particular function.
1842 */
1843int sock_no_bind(struct socket *, struct sockaddr *, int);
1844int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1845int sock_no_socketpair(struct socket *, struct socket *);
1846int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1847int sock_no_getname(struct socket *, struct sockaddr *, int);
1848int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1849int sock_no_listen(struct socket *, int);
1850int sock_no_shutdown(struct socket *, int);
1851int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1852int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1853int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1854int sock_no_mmap(struct file *file, struct socket *sock,
1855 struct vm_area_struct *vma);
1856
1857/*
1858 * Functions to fill in entries in struct proto_ops when a protocol
1859 * uses the inet style.
1860 */
1861int sock_common_getsockopt(struct socket *sock, int level, int optname,
1862 char __user *optval, int __user *optlen);
1863int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1864 int flags);
1865int sock_common_setsockopt(struct socket *sock, int level, int optname,
1866 sockptr_t optval, unsigned int optlen);
1867
1868void sk_common_release(struct sock *sk);
1869
1870/*
1871 * Default socket callbacks and setup code
1872 */
1873
1874/* Initialise core socket variables using an explicit uid. */
1875void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1876
1877/* Initialise core socket variables.
1878 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1879 */
1880void sock_init_data(struct socket *sock, struct sock *sk);
1881
1882/*
1883 * Socket reference counting postulates.
1884 *
1885 * * Each user of socket SHOULD hold a reference count.
1886 * * Each access point to socket (an hash table bucket, reference from a list,
1887 * running timer, skb in flight MUST hold a reference count.
1888 * * When reference count hits 0, it means it will never increase back.
1889 * * When reference count hits 0, it means that no references from
1890 * outside exist to this socket and current process on current CPU
1891 * is last user and may/should destroy this socket.
1892 * * sk_free is called from any context: process, BH, IRQ. When
1893 * it is called, socket has no references from outside -> sk_free
1894 * may release descendant resources allocated by the socket, but
1895 * to the time when it is called, socket is NOT referenced by any
1896 * hash tables, lists etc.
1897 * * Packets, delivered from outside (from network or from another process)
1898 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1899 * when they sit in queue. Otherwise, packets will leak to hole, when
1900 * socket is looked up by one cpu and unhasing is made by another CPU.
1901 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1902 * (leak to backlog). Packet socket does all the processing inside
1903 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1904 * use separate SMP lock, so that they are prone too.
1905 */
1906
1907/* Ungrab socket and destroy it, if it was the last reference. */
1908static inline void sock_put(struct sock *sk)
1909{
1910 if (refcount_dec_and_test(&sk->sk_refcnt))
1911 sk_free(sk);
1912}
1913/* Generic version of sock_put(), dealing with all sockets
1914 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1915 */
1916void sock_gen_put(struct sock *sk);
1917
1918int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1919 unsigned int trim_cap, bool refcounted);
1920static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1921 const int nested)
1922{
1923 return __sk_receive_skb(sk, skb, nested, 1, true);
1924}
1925
1926static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1927{
1928 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1929 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1930 return;
1931 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1932 * other WRITE_ONCE() because socket lock might be not held.
1933 */
1934 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1935}
1936
1937#define NO_QUEUE_MAPPING USHRT_MAX
1938
1939static inline void sk_tx_queue_clear(struct sock *sk)
1940{
1941 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1942 * other WRITE_ONCE() because socket lock might be not held.
1943 */
1944 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1945}
1946
1947static inline int sk_tx_queue_get(const struct sock *sk)
1948{
1949 if (sk) {
1950 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1951 * and sk_tx_queue_set().
1952 */
1953 int val = READ_ONCE(sk->sk_tx_queue_mapping);
1954
1955 if (val != NO_QUEUE_MAPPING)
1956 return val;
1957 }
1958 return -1;
1959}
1960
1961static inline void __sk_rx_queue_set(struct sock *sk,
1962 const struct sk_buff *skb,
1963 bool force_set)
1964{
1965#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1966 if (skb_rx_queue_recorded(skb)) {
1967 u16 rx_queue = skb_get_rx_queue(skb);
1968
1969 if (force_set ||
1970 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
1971 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
1972 }
1973#endif
1974}
1975
1976static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1977{
1978 __sk_rx_queue_set(sk, skb, true);
1979}
1980
1981static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
1982{
1983 __sk_rx_queue_set(sk, skb, false);
1984}
1985
1986static inline void sk_rx_queue_clear(struct sock *sk)
1987{
1988#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1989 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
1990#endif
1991}
1992
1993static inline int sk_rx_queue_get(const struct sock *sk)
1994{
1995#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1996 if (sk) {
1997 int res = READ_ONCE(sk->sk_rx_queue_mapping);
1998
1999 if (res != NO_QUEUE_MAPPING)
2000 return res;
2001 }
2002#endif
2003
2004 return -1;
2005}
2006
2007static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2008{
2009 sk->sk_socket = sock;
2010}
2011
2012static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2013{
2014 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2015 return &rcu_dereference_raw(sk->sk_wq)->wait;
2016}
2017/* Detach socket from process context.
2018 * Announce socket dead, detach it from wait queue and inode.
2019 * Note that parent inode held reference count on this struct sock,
2020 * we do not release it in this function, because protocol
2021 * probably wants some additional cleanups or even continuing
2022 * to work with this socket (TCP).
2023 */
2024static inline void sock_orphan(struct sock *sk)
2025{
2026 write_lock_bh(&sk->sk_callback_lock);
2027 sock_set_flag(sk, SOCK_DEAD);
2028 sk_set_socket(sk, NULL);
2029 sk->sk_wq = NULL;
2030 write_unlock_bh(&sk->sk_callback_lock);
2031}
2032
2033static inline void sock_graft(struct sock *sk, struct socket *parent)
2034{
2035 WARN_ON(parent->sk);
2036 write_lock_bh(&sk->sk_callback_lock);
2037 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2038 parent->sk = sk;
2039 sk_set_socket(sk, parent);
2040 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2041 security_sock_graft(sk, parent);
2042 write_unlock_bh(&sk->sk_callback_lock);
2043}
2044
2045kuid_t sock_i_uid(struct sock *sk);
2046unsigned long __sock_i_ino(struct sock *sk);
2047unsigned long sock_i_ino(struct sock *sk);
2048
2049static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2050{
2051 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2052}
2053
2054static inline u32 net_tx_rndhash(void)
2055{
2056 u32 v = get_random_u32();
2057
2058 return v ?: 1;
2059}
2060
2061static inline void sk_set_txhash(struct sock *sk)
2062{
2063 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2064 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2065}
2066
2067static inline bool sk_rethink_txhash(struct sock *sk)
2068{
2069 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2070 sk_set_txhash(sk);
2071 return true;
2072 }
2073 return false;
2074}
2075
2076static inline struct dst_entry *
2077__sk_dst_get(const struct sock *sk)
2078{
2079 return rcu_dereference_check(sk->sk_dst_cache,
2080 lockdep_sock_is_held(sk));
2081}
2082
2083static inline struct dst_entry *
2084sk_dst_get(const struct sock *sk)
2085{
2086 struct dst_entry *dst;
2087
2088 rcu_read_lock();
2089 dst = rcu_dereference(sk->sk_dst_cache);
2090 if (dst && !rcuref_get(&dst->__rcuref))
2091 dst = NULL;
2092 rcu_read_unlock();
2093 return dst;
2094}
2095
2096static inline void __dst_negative_advice(struct sock *sk)
2097{
2098 struct dst_entry *dst = __sk_dst_get(sk);
2099
2100 if (dst && dst->ops->negative_advice)
2101 dst->ops->negative_advice(sk, dst);
2102}
2103
2104static inline void dst_negative_advice(struct sock *sk)
2105{
2106 sk_rethink_txhash(sk);
2107 __dst_negative_advice(sk);
2108}
2109
2110static inline void
2111__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2112{
2113 struct dst_entry *old_dst;
2114
2115 sk_tx_queue_clear(sk);
2116 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2117 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2118 lockdep_sock_is_held(sk));
2119 rcu_assign_pointer(sk->sk_dst_cache, dst);
2120 dst_release(old_dst);
2121}
2122
2123static inline void
2124sk_dst_set(struct sock *sk, struct dst_entry *dst)
2125{
2126 struct dst_entry *old_dst;
2127
2128 sk_tx_queue_clear(sk);
2129 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2130 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2131 dst_release(old_dst);
2132}
2133
2134static inline void
2135__sk_dst_reset(struct sock *sk)
2136{
2137 __sk_dst_set(sk, NULL);
2138}
2139
2140static inline void
2141sk_dst_reset(struct sock *sk)
2142{
2143 sk_dst_set(sk, NULL);
2144}
2145
2146struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2147
2148struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2149
2150static inline void sk_dst_confirm(struct sock *sk)
2151{
2152 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2153 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2154}
2155
2156static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2157{
2158 if (skb_get_dst_pending_confirm(skb)) {
2159 struct sock *sk = skb->sk;
2160
2161 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2162 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2163 neigh_confirm(n);
2164 }
2165}
2166
2167bool sk_mc_loop(const struct sock *sk);
2168
2169static inline bool sk_can_gso(const struct sock *sk)
2170{
2171 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2172}
2173
2174void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2175
2176static inline void sk_gso_disable(struct sock *sk)
2177{
2178 sk->sk_gso_disabled = 1;
2179 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2180}
2181
2182static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2183 struct iov_iter *from, char *to,
2184 int copy, int offset)
2185{
2186 if (skb->ip_summed == CHECKSUM_NONE) {
2187 __wsum csum = 0;
2188 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2189 return -EFAULT;
2190 skb->csum = csum_block_add(skb->csum, csum, offset);
2191 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2192 if (!copy_from_iter_full_nocache(to, copy, from))
2193 return -EFAULT;
2194 } else if (!copy_from_iter_full(to, copy, from))
2195 return -EFAULT;
2196
2197 return 0;
2198}
2199
2200static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2201 struct iov_iter *from, int copy)
2202{
2203 int err, offset = skb->len;
2204
2205 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2206 copy, offset);
2207 if (err)
2208 __skb_trim(skb, offset);
2209
2210 return err;
2211}
2212
2213static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2214 struct sk_buff *skb,
2215 struct page *page,
2216 int off, int copy)
2217{
2218 int err;
2219
2220 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2221 copy, skb->len);
2222 if (err)
2223 return err;
2224
2225 skb_len_add(skb, copy);
2226 sk_wmem_queued_add(sk, copy);
2227 sk_mem_charge(sk, copy);
2228 return 0;
2229}
2230
2231/**
2232 * sk_wmem_alloc_get - returns write allocations
2233 * @sk: socket
2234 *
2235 * Return: sk_wmem_alloc minus initial offset of one
2236 */
2237static inline int sk_wmem_alloc_get(const struct sock *sk)
2238{
2239 return refcount_read(&sk->sk_wmem_alloc) - 1;
2240}
2241
2242/**
2243 * sk_rmem_alloc_get - returns read allocations
2244 * @sk: socket
2245 *
2246 * Return: sk_rmem_alloc
2247 */
2248static inline int sk_rmem_alloc_get(const struct sock *sk)
2249{
2250 return atomic_read(&sk->sk_rmem_alloc);
2251}
2252
2253/**
2254 * sk_has_allocations - check if allocations are outstanding
2255 * @sk: socket
2256 *
2257 * Return: true if socket has write or read allocations
2258 */
2259static inline bool sk_has_allocations(const struct sock *sk)
2260{
2261 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2262}
2263
2264/**
2265 * skwq_has_sleeper - check if there are any waiting processes
2266 * @wq: struct socket_wq
2267 *
2268 * Return: true if socket_wq has waiting processes
2269 *
2270 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2271 * barrier call. They were added due to the race found within the tcp code.
2272 *
2273 * Consider following tcp code paths::
2274 *
2275 * CPU1 CPU2
2276 * sys_select receive packet
2277 * ... ...
2278 * __add_wait_queue update tp->rcv_nxt
2279 * ... ...
2280 * tp->rcv_nxt check sock_def_readable
2281 * ... {
2282 * schedule rcu_read_lock();
2283 * wq = rcu_dereference(sk->sk_wq);
2284 * if (wq && waitqueue_active(&wq->wait))
2285 * wake_up_interruptible(&wq->wait)
2286 * ...
2287 * }
2288 *
2289 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2290 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2291 * could then endup calling schedule and sleep forever if there are no more
2292 * data on the socket.
2293 *
2294 */
2295static inline bool skwq_has_sleeper(struct socket_wq *wq)
2296{
2297 return wq && wq_has_sleeper(&wq->wait);
2298}
2299
2300/**
2301 * sock_poll_wait - wrapper for the poll_wait call.
2302 * @filp: file
2303 * @sock: socket to wait on
2304 * @p: poll_table
2305 *
2306 * See the comments in the wq_has_sleeper function.
2307 */
2308static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2309 poll_table *p)
2310{
2311 /* Provides a barrier we need to be sure we are in sync
2312 * with the socket flags modification.
2313 *
2314 * This memory barrier is paired in the wq_has_sleeper.
2315 */
2316 poll_wait(filp, &sock->wq.wait, p);
2317}
2318
2319static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2320{
2321 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2322 u32 txhash = READ_ONCE(sk->sk_txhash);
2323
2324 if (txhash) {
2325 skb->l4_hash = 1;
2326 skb->hash = txhash;
2327 }
2328}
2329
2330void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2331
2332/*
2333 * Queue a received datagram if it will fit. Stream and sequenced
2334 * protocols can't normally use this as they need to fit buffers in
2335 * and play with them.
2336 *
2337 * Inlined as it's very short and called for pretty much every
2338 * packet ever received.
2339 */
2340static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2341{
2342 skb_orphan(skb);
2343 skb->sk = sk;
2344 skb->destructor = sock_rfree;
2345 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2346 sk_mem_charge(sk, skb->truesize);
2347}
2348
2349static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2350{
2351 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2352 skb_orphan(skb);
2353 skb->destructor = sock_efree;
2354 skb->sk = sk;
2355 return true;
2356 }
2357 return false;
2358}
2359
2360static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2361{
2362 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2363 if (skb) {
2364 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2365 skb_set_owner_r(skb, sk);
2366 return skb;
2367 }
2368 __kfree_skb(skb);
2369 }
2370 return NULL;
2371}
2372
2373static inline void skb_prepare_for_gro(struct sk_buff *skb)
2374{
2375 if (skb->destructor != sock_wfree) {
2376 skb_orphan(skb);
2377 return;
2378 }
2379 skb->slow_gro = 1;
2380}
2381
2382void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2383 unsigned long expires);
2384
2385void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2386
2387void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2388
2389int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2390 struct sk_buff *skb, unsigned int flags,
2391 void (*destructor)(struct sock *sk,
2392 struct sk_buff *skb));
2393int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2394
2395int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2396 enum skb_drop_reason *reason);
2397
2398static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2399{
2400 return sock_queue_rcv_skb_reason(sk, skb, NULL);
2401}
2402
2403int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2404struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2405
2406/*
2407 * Recover an error report and clear atomically
2408 */
2409
2410static inline int sock_error(struct sock *sk)
2411{
2412 int err;
2413
2414 /* Avoid an atomic operation for the common case.
2415 * This is racy since another cpu/thread can change sk_err under us.
2416 */
2417 if (likely(data_race(!sk->sk_err)))
2418 return 0;
2419
2420 err = xchg(&sk->sk_err, 0);
2421 return -err;
2422}
2423
2424void sk_error_report(struct sock *sk);
2425
2426static inline unsigned long sock_wspace(struct sock *sk)
2427{
2428 int amt = 0;
2429
2430 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2431 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2432 if (amt < 0)
2433 amt = 0;
2434 }
2435 return amt;
2436}
2437
2438/* Note:
2439 * We use sk->sk_wq_raw, from contexts knowing this
2440 * pointer is not NULL and cannot disappear/change.
2441 */
2442static inline void sk_set_bit(int nr, struct sock *sk)
2443{
2444 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2445 !sock_flag(sk, SOCK_FASYNC))
2446 return;
2447
2448 set_bit(nr, &sk->sk_wq_raw->flags);
2449}
2450
2451static inline void sk_clear_bit(int nr, struct sock *sk)
2452{
2453 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2454 !sock_flag(sk, SOCK_FASYNC))
2455 return;
2456
2457 clear_bit(nr, &sk->sk_wq_raw->flags);
2458}
2459
2460static inline void sk_wake_async(const struct sock *sk, int how, int band)
2461{
2462 if (sock_flag(sk, SOCK_FASYNC)) {
2463 rcu_read_lock();
2464 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2465 rcu_read_unlock();
2466 }
2467}
2468
2469static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2470{
2471 if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2472 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2473}
2474
2475/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2476 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2477 * Note: for send buffers, TCP works better if we can build two skbs at
2478 * minimum.
2479 */
2480#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2481
2482#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2483#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2484
2485static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2486{
2487 u32 val;
2488
2489 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2490 return;
2491
2492 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2493 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2494
2495 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2496}
2497
2498/**
2499 * sk_page_frag - return an appropriate page_frag
2500 * @sk: socket
2501 *
2502 * Use the per task page_frag instead of the per socket one for
2503 * optimization when we know that we're in process context and own
2504 * everything that's associated with %current.
2505 *
2506 * Both direct reclaim and page faults can nest inside other
2507 * socket operations and end up recursing into sk_page_frag()
2508 * while it's already in use: explicitly avoid task page_frag
2509 * when users disable sk_use_task_frag.
2510 *
2511 * Return: a per task page_frag if context allows that,
2512 * otherwise a per socket one.
2513 */
2514static inline struct page_frag *sk_page_frag(struct sock *sk)
2515{
2516 if (sk->sk_use_task_frag)
2517 return ¤t->task_frag;
2518
2519 return &sk->sk_frag;
2520}
2521
2522bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2523
2524/*
2525 * Default write policy as shown to user space via poll/select/SIGIO
2526 */
2527static inline bool sock_writeable(const struct sock *sk)
2528{
2529 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2530}
2531
2532static inline gfp_t gfp_any(void)
2533{
2534 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2535}
2536
2537static inline gfp_t gfp_memcg_charge(void)
2538{
2539 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2540}
2541
2542static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2543{
2544 return noblock ? 0 : sk->sk_rcvtimeo;
2545}
2546
2547static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2548{
2549 return noblock ? 0 : sk->sk_sndtimeo;
2550}
2551
2552static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2553{
2554 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2555
2556 return v ?: 1;
2557}
2558
2559/* Alas, with timeout socket operations are not restartable.
2560 * Compare this to poll().
2561 */
2562static inline int sock_intr_errno(long timeo)
2563{
2564 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2565}
2566
2567struct sock_skb_cb {
2568 u32 dropcount;
2569};
2570
2571/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2572 * using skb->cb[] would keep using it directly and utilize its
2573 * alignment guarantee.
2574 */
2575#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2576 sizeof(struct sock_skb_cb)))
2577
2578#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2579 SOCK_SKB_CB_OFFSET))
2580
2581#define sock_skb_cb_check_size(size) \
2582 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2583
2584static inline void
2585sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2586{
2587 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2588 atomic_read(&sk->sk_drops) : 0;
2589}
2590
2591static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2592{
2593 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2594
2595 atomic_add(segs, &sk->sk_drops);
2596}
2597
2598static inline ktime_t sock_read_timestamp(struct sock *sk)
2599{
2600#if BITS_PER_LONG==32
2601 unsigned int seq;
2602 ktime_t kt;
2603
2604 do {
2605 seq = read_seqbegin(&sk->sk_stamp_seq);
2606 kt = sk->sk_stamp;
2607 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2608
2609 return kt;
2610#else
2611 return READ_ONCE(sk->sk_stamp);
2612#endif
2613}
2614
2615static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2616{
2617#if BITS_PER_LONG==32
2618 write_seqlock(&sk->sk_stamp_seq);
2619 sk->sk_stamp = kt;
2620 write_sequnlock(&sk->sk_stamp_seq);
2621#else
2622 WRITE_ONCE(sk->sk_stamp, kt);
2623#endif
2624}
2625
2626void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2627 struct sk_buff *skb);
2628void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2629 struct sk_buff *skb);
2630
2631static inline void
2632sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2633{
2634 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2635 u32 tsflags = READ_ONCE(sk->sk_tsflags);
2636 ktime_t kt = skb->tstamp;
2637 /*
2638 * generate control messages if
2639 * - receive time stamping in software requested
2640 * - software time stamp available and wanted
2641 * - hardware time stamps available and wanted
2642 */
2643 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2644 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2645 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2646 (hwtstamps->hwtstamp &&
2647 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2648 __sock_recv_timestamp(msg, sk, skb);
2649 else
2650 sock_write_timestamp(sk, kt);
2651
2652 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2653 __sock_recv_wifi_status(msg, sk, skb);
2654}
2655
2656void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2657 struct sk_buff *skb);
2658
2659#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2660static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2661 struct sk_buff *skb)
2662{
2663#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2664 (1UL << SOCK_RCVTSTAMP) | \
2665 (1UL << SOCK_RCVMARK))
2666#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2667 SOF_TIMESTAMPING_RAW_HARDWARE)
2668
2669 if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2670 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2671 __sock_recv_cmsgs(msg, sk, skb);
2672 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2673 sock_write_timestamp(sk, skb->tstamp);
2674 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2675 sock_write_timestamp(sk, 0);
2676}
2677
2678void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2679
2680/**
2681 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2682 * @sk: socket sending this packet
2683 * @sockc: pointer to socket cmsg cookie to get timestamping info
2684 * @tx_flags: completed with instructions for time stamping
2685 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2686 *
2687 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2688 */
2689static inline void _sock_tx_timestamp(struct sock *sk,
2690 const struct sockcm_cookie *sockc,
2691 __u8 *tx_flags, __u32 *tskey)
2692{
2693 __u32 tsflags = sockc->tsflags;
2694
2695 if (unlikely(tsflags)) {
2696 __sock_tx_timestamp(tsflags, tx_flags);
2697 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2698 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2699 if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2700 *tskey = sockc->ts_opt_id;
2701 else
2702 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2703 }
2704 }
2705 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2706 *tx_flags |= SKBTX_WIFI_STATUS;
2707}
2708
2709static inline void sock_tx_timestamp(struct sock *sk,
2710 const struct sockcm_cookie *sockc,
2711 __u8 *tx_flags)
2712{
2713 _sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2714}
2715
2716static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2717 const struct sockcm_cookie *sockc)
2718{
2719 _sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags,
2720 &skb_shinfo(skb)->tskey);
2721}
2722
2723static inline bool sk_is_inet(const struct sock *sk)
2724{
2725 int family = READ_ONCE(sk->sk_family);
2726
2727 return family == AF_INET || family == AF_INET6;
2728}
2729
2730static inline bool sk_is_tcp(const struct sock *sk)
2731{
2732 return sk_is_inet(sk) &&
2733 sk->sk_type == SOCK_STREAM &&
2734 sk->sk_protocol == IPPROTO_TCP;
2735}
2736
2737static inline bool sk_is_udp(const struct sock *sk)
2738{
2739 return sk_is_inet(sk) &&
2740 sk->sk_type == SOCK_DGRAM &&
2741 sk->sk_protocol == IPPROTO_UDP;
2742}
2743
2744static inline bool sk_is_stream_unix(const struct sock *sk)
2745{
2746 return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM;
2747}
2748
2749static inline bool sk_is_vsock(const struct sock *sk)
2750{
2751 return sk->sk_family == AF_VSOCK;
2752}
2753
2754/**
2755 * sk_eat_skb - Release a skb if it is no longer needed
2756 * @sk: socket to eat this skb from
2757 * @skb: socket buffer to eat
2758 *
2759 * This routine must be called with interrupts disabled or with the socket
2760 * locked so that the sk_buff queue operation is ok.
2761*/
2762static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2763{
2764 __skb_unlink(skb, &sk->sk_receive_queue);
2765 __kfree_skb(skb);
2766}
2767
2768static inline bool
2769skb_sk_is_prefetched(struct sk_buff *skb)
2770{
2771#ifdef CONFIG_INET
2772 return skb->destructor == sock_pfree;
2773#else
2774 return false;
2775#endif /* CONFIG_INET */
2776}
2777
2778/* This helper checks if a socket is a full socket,
2779 * ie _not_ a timewait or request socket.
2780 */
2781static inline bool sk_fullsock(const struct sock *sk)
2782{
2783 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2784}
2785
2786static inline bool
2787sk_is_refcounted(struct sock *sk)
2788{
2789 /* Only full sockets have sk->sk_flags. */
2790 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2791}
2792
2793/* Checks if this SKB belongs to an HW offloaded socket
2794 * and whether any SW fallbacks are required based on dev.
2795 * Check decrypted mark in case skb_orphan() cleared socket.
2796 */
2797static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2798 struct net_device *dev)
2799{
2800#ifdef CONFIG_SOCK_VALIDATE_XMIT
2801 struct sock *sk = skb->sk;
2802
2803 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2804 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2805 } else if (unlikely(skb_is_decrypted(skb))) {
2806 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2807 kfree_skb(skb);
2808 skb = NULL;
2809 }
2810#endif
2811
2812 return skb;
2813}
2814
2815/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2816 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2817 */
2818static inline bool sk_listener(const struct sock *sk)
2819{
2820 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2821}
2822
2823/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT
2824 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE)
2825 * TCP RST and ACK can be attached to TIME_WAIT.
2826 */
2827static inline bool sk_listener_or_tw(const struct sock *sk)
2828{
2829 return (1 << READ_ONCE(sk->sk_state)) &
2830 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT);
2831}
2832
2833void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2834int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2835 int type);
2836
2837bool sk_ns_capable(const struct sock *sk,
2838 struct user_namespace *user_ns, int cap);
2839bool sk_capable(const struct sock *sk, int cap);
2840bool sk_net_capable(const struct sock *sk, int cap);
2841
2842void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2843
2844/* Take into consideration the size of the struct sk_buff overhead in the
2845 * determination of these values, since that is non-constant across
2846 * platforms. This makes socket queueing behavior and performance
2847 * not depend upon such differences.
2848 */
2849#define _SK_MEM_PACKETS 256
2850#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2851#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2852#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2853
2854extern __u32 sysctl_wmem_max;
2855extern __u32 sysctl_rmem_max;
2856
2857extern __u32 sysctl_wmem_default;
2858extern __u32 sysctl_rmem_default;
2859
2860#define SKB_FRAG_PAGE_ORDER get_order(32768)
2861DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2862
2863static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2864{
2865 /* Does this proto have per netns sysctl_wmem ? */
2866 if (proto->sysctl_wmem_offset)
2867 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2868
2869 return READ_ONCE(*proto->sysctl_wmem);
2870}
2871
2872static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2873{
2874 /* Does this proto have per netns sysctl_rmem ? */
2875 if (proto->sysctl_rmem_offset)
2876 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2877
2878 return READ_ONCE(*proto->sysctl_rmem);
2879}
2880
2881/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2882 * Some wifi drivers need to tweak it to get more chunks.
2883 * They can use this helper from their ndo_start_xmit()
2884 */
2885static inline void sk_pacing_shift_update(struct sock *sk, int val)
2886{
2887 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2888 return;
2889 WRITE_ONCE(sk->sk_pacing_shift, val);
2890}
2891
2892/* if a socket is bound to a device, check that the given device
2893 * index is either the same or that the socket is bound to an L3
2894 * master device and the given device index is also enslaved to
2895 * that L3 master
2896 */
2897static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2898{
2899 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2900 int mdif;
2901
2902 if (!bound_dev_if || bound_dev_if == dif)
2903 return true;
2904
2905 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2906 if (mdif && mdif == bound_dev_if)
2907 return true;
2908
2909 return false;
2910}
2911
2912void sock_def_readable(struct sock *sk);
2913
2914int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2915void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2916int sock_set_timestamping(struct sock *sk, int optname,
2917 struct so_timestamping timestamping);
2918
2919void sock_enable_timestamps(struct sock *sk);
2920void sock_no_linger(struct sock *sk);
2921void sock_set_keepalive(struct sock *sk);
2922void sock_set_priority(struct sock *sk, u32 priority);
2923void sock_set_rcvbuf(struct sock *sk, int val);
2924void sock_set_mark(struct sock *sk, u32 val);
2925void sock_set_reuseaddr(struct sock *sk);
2926void sock_set_reuseport(struct sock *sk);
2927void sock_set_sndtimeo(struct sock *sk, s64 secs);
2928
2929int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2930
2931int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2932int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2933 sockptr_t optval, int optlen, bool old_timeval);
2934
2935int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
2936 void __user *arg, void *karg, size_t size);
2937int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
2938static inline bool sk_is_readable(struct sock *sk)
2939{
2940 if (sk->sk_prot->sock_is_readable)
2941 return sk->sk_prot->sock_is_readable(sk);
2942 return false;
2943}
2944#endif /* _SOCK_H */
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the AF_INET socket handler.
7 *
8 * Version: @(#)sock.h 1.0.4 05/13/93
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
14 *
15 * Fixes:
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
19 * than the reverse.
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 * David S. Miller : New socket lookup architecture.
26 * Steve Whitehouse: Default routines for sock_ops
27 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
28 * protinfo be just a void pointer, as the
29 * protocol specific parts were moved to
30 * respective headers and ipv4/v6, etc now
31 * use private slabcaches for its socks
32 * Pedro Hortas : New flags field for socket options
33 *
34 *
35 * This program is free software; you can redistribute it and/or
36 * modify it under the terms of the GNU General Public License
37 * as published by the Free Software Foundation; either version
38 * 2 of the License, or (at your option) any later version.
39 */
40#ifndef _SOCK_H
41#define _SOCK_H
42
43#include <linux/hardirq.h>
44#include <linux/kernel.h>
45#include <linux/list.h>
46#include <linux/list_nulls.h>
47#include <linux/timer.h>
48#include <linux/cache.h>
49#include <linux/bitops.h>
50#include <linux/lockdep.h>
51#include <linux/netdevice.h>
52#include <linux/skbuff.h> /* struct sk_buff */
53#include <linux/mm.h>
54#include <linux/security.h>
55#include <linux/slab.h>
56#include <linux/uaccess.h>
57#include <linux/page_counter.h>
58#include <linux/memcontrol.h>
59#include <linux/static_key.h>
60#include <linux/sched.h>
61#include <linux/wait.h>
62#include <linux/cgroup-defs.h>
63#include <linux/rbtree.h>
64#include <linux/filter.h>
65#include <linux/rculist_nulls.h>
66#include <linux/poll.h>
67
68#include <linux/atomic.h>
69#include <linux/refcount.h>
70#include <net/dst.h>
71#include <net/checksum.h>
72#include <net/tcp_states.h>
73#include <linux/net_tstamp.h>
74#include <net/smc.h>
75#include <net/l3mdev.h>
76
77/*
78 * This structure really needs to be cleaned up.
79 * Most of it is for TCP, and not used by any of
80 * the other protocols.
81 */
82
83/* Define this to get the SOCK_DBG debugging facility. */
84#define SOCK_DEBUGGING
85#ifdef SOCK_DEBUGGING
86#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
87 printk(KERN_DEBUG msg); } while (0)
88#else
89/* Validate arguments and do nothing */
90static inline __printf(2, 3)
91void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
92{
93}
94#endif
95
96/* This is the per-socket lock. The spinlock provides a synchronization
97 * between user contexts and software interrupt processing, whereas the
98 * mini-semaphore synchronizes multiple users amongst themselves.
99 */
100typedef struct {
101 spinlock_t slock;
102 int owned;
103 wait_queue_head_t wq;
104 /*
105 * We express the mutex-alike socket_lock semantics
106 * to the lock validator by explicitly managing
107 * the slock as a lock variant (in addition to
108 * the slock itself):
109 */
110#ifdef CONFIG_DEBUG_LOCK_ALLOC
111 struct lockdep_map dep_map;
112#endif
113} socket_lock_t;
114
115struct sock;
116struct proto;
117struct net;
118
119typedef __u32 __bitwise __portpair;
120typedef __u64 __bitwise __addrpair;
121
122/**
123 * struct sock_common - minimal network layer representation of sockets
124 * @skc_daddr: Foreign IPv4 addr
125 * @skc_rcv_saddr: Bound local IPv4 addr
126 * @skc_hash: hash value used with various protocol lookup tables
127 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
128 * @skc_dport: placeholder for inet_dport/tw_dport
129 * @skc_num: placeholder for inet_num/tw_num
130 * @skc_family: network address family
131 * @skc_state: Connection state
132 * @skc_reuse: %SO_REUSEADDR setting
133 * @skc_reuseport: %SO_REUSEPORT setting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_node: main hash linkage for various protocol lookup tables
140 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
141 * @skc_tx_queue_mapping: tx queue number for this connection
142 * @skc_flags: place holder for sk_flags
143 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
144 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
145 * @skc_incoming_cpu: record/match cpu processing incoming packets
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 */
151struct sock_common {
152 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
153 * address on 64bit arches : cf INET_MATCH()
154 */
155 union {
156 __addrpair skc_addrpair;
157 struct {
158 __be32 skc_daddr;
159 __be32 skc_rcv_saddr;
160 };
161 };
162 union {
163 unsigned int skc_hash;
164 __u16 skc_u16hashes[2];
165 };
166 /* skc_dport && skc_num must be grouped as well */
167 union {
168 __portpair skc_portpair;
169 struct {
170 __be16 skc_dport;
171 __u16 skc_num;
172 };
173 };
174
175 unsigned short skc_family;
176 volatile unsigned char skc_state;
177 unsigned char skc_reuse:4;
178 unsigned char skc_reuseport:1;
179 unsigned char skc_ipv6only:1;
180 unsigned char skc_net_refcnt:1;
181 int skc_bound_dev_if;
182 union {
183 struct hlist_node skc_bind_node;
184 struct hlist_node skc_portaddr_node;
185 };
186 struct proto *skc_prot;
187 possible_net_t skc_net;
188
189#if IS_ENABLED(CONFIG_IPV6)
190 struct in6_addr skc_v6_daddr;
191 struct in6_addr skc_v6_rcv_saddr;
192#endif
193
194 atomic64_t skc_cookie;
195
196 /* following fields are padding to force
197 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
198 * assuming IPV6 is enabled. We use this padding differently
199 * for different kind of 'sockets'
200 */
201 union {
202 unsigned long skc_flags;
203 struct sock *skc_listener; /* request_sock */
204 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
205 };
206 /*
207 * fields between dontcopy_begin/dontcopy_end
208 * are not copied in sock_copy()
209 */
210 /* private: */
211 int skc_dontcopy_begin[0];
212 /* public: */
213 union {
214 struct hlist_node skc_node;
215 struct hlist_nulls_node skc_nulls_node;
216 };
217 int skc_tx_queue_mapping;
218 union {
219 int skc_incoming_cpu;
220 u32 skc_rcv_wnd;
221 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
222 };
223
224 refcount_t skc_refcnt;
225 /* private: */
226 int skc_dontcopy_end[0];
227 union {
228 u32 skc_rxhash;
229 u32 skc_window_clamp;
230 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
231 };
232 /* public: */
233};
234
235/**
236 * struct sock - network layer representation of sockets
237 * @__sk_common: shared layout with inet_timewait_sock
238 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
239 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
240 * @sk_lock: synchronizer
241 * @sk_kern_sock: True if sock is using kernel lock classes
242 * @sk_rcvbuf: size of receive buffer in bytes
243 * @sk_wq: sock wait queue and async head
244 * @sk_rx_dst: receive input route used by early demux
245 * @sk_dst_cache: destination cache
246 * @sk_dst_pending_confirm: need to confirm neighbour
247 * @sk_policy: flow policy
248 * @sk_receive_queue: incoming packets
249 * @sk_wmem_alloc: transmit queue bytes committed
250 * @sk_tsq_flags: TCP Small Queues flags
251 * @sk_write_queue: Packet sending queue
252 * @sk_omem_alloc: "o" is "option" or "other"
253 * @sk_wmem_queued: persistent queue size
254 * @sk_forward_alloc: space allocated forward
255 * @sk_napi_id: id of the last napi context to receive data for sk
256 * @sk_ll_usec: usecs to busypoll when there is no data
257 * @sk_allocation: allocation mode
258 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
259 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
260 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
261 * @sk_sndbuf: size of send buffer in bytes
262 * @__sk_flags_offset: empty field used to determine location of bitfield
263 * @sk_padding: unused element for alignment
264 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
265 * @sk_no_check_rx: allow zero checksum in RX packets
266 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
267 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
268 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
269 * @sk_gso_max_size: Maximum GSO segment size to build
270 * @sk_gso_max_segs: Maximum number of GSO segments
271 * @sk_pacing_shift: scaling factor for TCP Small Queues
272 * @sk_lingertime: %SO_LINGER l_linger setting
273 * @sk_backlog: always used with the per-socket spinlock held
274 * @sk_callback_lock: used with the callbacks in the end of this struct
275 * @sk_error_queue: rarely used
276 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
277 * IPV6_ADDRFORM for instance)
278 * @sk_err: last error
279 * @sk_err_soft: errors that don't cause failure but are the cause of a
280 * persistent failure not just 'timed out'
281 * @sk_drops: raw/udp drops counter
282 * @sk_ack_backlog: current listen backlog
283 * @sk_max_ack_backlog: listen backlog set in listen()
284 * @sk_uid: user id of owner
285 * @sk_priority: %SO_PRIORITY setting
286 * @sk_type: socket type (%SOCK_STREAM, etc)
287 * @sk_protocol: which protocol this socket belongs in this network family
288 * @sk_peer_pid: &struct pid for this socket's peer
289 * @sk_peer_cred: %SO_PEERCRED setting
290 * @sk_rcvlowat: %SO_RCVLOWAT setting
291 * @sk_rcvtimeo: %SO_RCVTIMEO setting
292 * @sk_sndtimeo: %SO_SNDTIMEO setting
293 * @sk_txhash: computed flow hash for use on transmit
294 * @sk_filter: socket filtering instructions
295 * @sk_timer: sock cleanup timer
296 * @sk_stamp: time stamp of last packet received
297 * @sk_tsflags: SO_TIMESTAMPING socket options
298 * @sk_tskey: counter to disambiguate concurrent tstamp requests
299 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
300 * @sk_socket: Identd and reporting IO signals
301 * @sk_user_data: RPC layer private data
302 * @sk_frag: cached page frag
303 * @sk_peek_off: current peek_offset value
304 * @sk_send_head: front of stuff to transmit
305 * @sk_security: used by security modules
306 * @sk_mark: generic packet mark
307 * @sk_cgrp_data: cgroup data for this cgroup
308 * @sk_memcg: this socket's memory cgroup association
309 * @sk_write_pending: a write to stream socket waits to start
310 * @sk_state_change: callback to indicate change in the state of the sock
311 * @sk_data_ready: callback to indicate there is data to be processed
312 * @sk_write_space: callback to indicate there is bf sending space available
313 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
314 * @sk_backlog_rcv: callback to process the backlog
315 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
316 * @sk_reuseport_cb: reuseport group container
317 * @sk_rcu: used during RCU grace period
318 */
319struct sock {
320 /*
321 * Now struct inet_timewait_sock also uses sock_common, so please just
322 * don't add nothing before this first member (__sk_common) --acme
323 */
324 struct sock_common __sk_common;
325#define sk_node __sk_common.skc_node
326#define sk_nulls_node __sk_common.skc_nulls_node
327#define sk_refcnt __sk_common.skc_refcnt
328#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
329
330#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
331#define sk_dontcopy_end __sk_common.skc_dontcopy_end
332#define sk_hash __sk_common.skc_hash
333#define sk_portpair __sk_common.skc_portpair
334#define sk_num __sk_common.skc_num
335#define sk_dport __sk_common.skc_dport
336#define sk_addrpair __sk_common.skc_addrpair
337#define sk_daddr __sk_common.skc_daddr
338#define sk_rcv_saddr __sk_common.skc_rcv_saddr
339#define sk_family __sk_common.skc_family
340#define sk_state __sk_common.skc_state
341#define sk_reuse __sk_common.skc_reuse
342#define sk_reuseport __sk_common.skc_reuseport
343#define sk_ipv6only __sk_common.skc_ipv6only
344#define sk_net_refcnt __sk_common.skc_net_refcnt
345#define sk_bound_dev_if __sk_common.skc_bound_dev_if
346#define sk_bind_node __sk_common.skc_bind_node
347#define sk_prot __sk_common.skc_prot
348#define sk_net __sk_common.skc_net
349#define sk_v6_daddr __sk_common.skc_v6_daddr
350#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
351#define sk_cookie __sk_common.skc_cookie
352#define sk_incoming_cpu __sk_common.skc_incoming_cpu
353#define sk_flags __sk_common.skc_flags
354#define sk_rxhash __sk_common.skc_rxhash
355
356 socket_lock_t sk_lock;
357 atomic_t sk_drops;
358 int sk_rcvlowat;
359 struct sk_buff_head sk_error_queue;
360 struct sk_buff_head sk_receive_queue;
361 /*
362 * The backlog queue is special, it is always used with
363 * the per-socket spinlock held and requires low latency
364 * access. Therefore we special case it's implementation.
365 * Note : rmem_alloc is in this structure to fill a hole
366 * on 64bit arches, not because its logically part of
367 * backlog.
368 */
369 struct {
370 atomic_t rmem_alloc;
371 int len;
372 struct sk_buff *head;
373 struct sk_buff *tail;
374 } sk_backlog;
375#define sk_rmem_alloc sk_backlog.rmem_alloc
376
377 int sk_forward_alloc;
378#ifdef CONFIG_NET_RX_BUSY_POLL
379 unsigned int sk_ll_usec;
380 /* ===== mostly read cache line ===== */
381 unsigned int sk_napi_id;
382#endif
383 int sk_rcvbuf;
384
385 struct sk_filter __rcu *sk_filter;
386 union {
387 struct socket_wq __rcu *sk_wq;
388 struct socket_wq *sk_wq_raw;
389 };
390#ifdef CONFIG_XFRM
391 struct xfrm_policy __rcu *sk_policy[2];
392#endif
393 struct dst_entry *sk_rx_dst;
394 struct dst_entry __rcu *sk_dst_cache;
395 atomic_t sk_omem_alloc;
396 int sk_sndbuf;
397
398 /* ===== cache line for TX ===== */
399 int sk_wmem_queued;
400 refcount_t sk_wmem_alloc;
401 unsigned long sk_tsq_flags;
402 union {
403 struct sk_buff *sk_send_head;
404 struct rb_root tcp_rtx_queue;
405 };
406 struct sk_buff_head sk_write_queue;
407 __s32 sk_peek_off;
408 int sk_write_pending;
409 __u32 sk_dst_pending_confirm;
410 u32 sk_pacing_status; /* see enum sk_pacing */
411 long sk_sndtimeo;
412 struct timer_list sk_timer;
413 __u32 sk_priority;
414 __u32 sk_mark;
415 u32 sk_pacing_rate; /* bytes per second */
416 u32 sk_max_pacing_rate;
417 struct page_frag sk_frag;
418 netdev_features_t sk_route_caps;
419 netdev_features_t sk_route_nocaps;
420 netdev_features_t sk_route_forced_caps;
421 int sk_gso_type;
422 unsigned int sk_gso_max_size;
423 gfp_t sk_allocation;
424 __u32 sk_txhash;
425
426 /*
427 * Because of non atomicity rules, all
428 * changes are protected by socket lock.
429 */
430 unsigned int __sk_flags_offset[0];
431#ifdef __BIG_ENDIAN_BITFIELD
432#define SK_FL_PROTO_SHIFT 16
433#define SK_FL_PROTO_MASK 0x00ff0000
434
435#define SK_FL_TYPE_SHIFT 0
436#define SK_FL_TYPE_MASK 0x0000ffff
437#else
438#define SK_FL_PROTO_SHIFT 8
439#define SK_FL_PROTO_MASK 0x0000ff00
440
441#define SK_FL_TYPE_SHIFT 16
442#define SK_FL_TYPE_MASK 0xffff0000
443#endif
444
445 unsigned int sk_padding : 1,
446 sk_kern_sock : 1,
447 sk_no_check_tx : 1,
448 sk_no_check_rx : 1,
449 sk_userlocks : 4,
450 sk_protocol : 8,
451 sk_type : 16;
452#define SK_PROTOCOL_MAX U8_MAX
453 u16 sk_gso_max_segs;
454 u8 sk_pacing_shift;
455 unsigned long sk_lingertime;
456 struct proto *sk_prot_creator;
457 rwlock_t sk_callback_lock;
458 int sk_err,
459 sk_err_soft;
460 u32 sk_ack_backlog;
461 u32 sk_max_ack_backlog;
462 kuid_t sk_uid;
463 struct pid *sk_peer_pid;
464 const struct cred *sk_peer_cred;
465 long sk_rcvtimeo;
466 ktime_t sk_stamp;
467 u16 sk_tsflags;
468 u8 sk_shutdown;
469 u32 sk_tskey;
470 atomic_t sk_zckey;
471 struct socket *sk_socket;
472 void *sk_user_data;
473#ifdef CONFIG_SECURITY
474 void *sk_security;
475#endif
476 struct sock_cgroup_data sk_cgrp_data;
477 struct mem_cgroup *sk_memcg;
478 void (*sk_state_change)(struct sock *sk);
479 void (*sk_data_ready)(struct sock *sk);
480 void (*sk_write_space)(struct sock *sk);
481 void (*sk_error_report)(struct sock *sk);
482 int (*sk_backlog_rcv)(struct sock *sk,
483 struct sk_buff *skb);
484 void (*sk_destruct)(struct sock *sk);
485 struct sock_reuseport __rcu *sk_reuseport_cb;
486 struct rcu_head sk_rcu;
487};
488
489enum sk_pacing {
490 SK_PACING_NONE = 0,
491 SK_PACING_NEEDED = 1,
492 SK_PACING_FQ = 2,
493};
494
495#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
496
497#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
498#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
499
500/*
501 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
502 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
503 * on a socket means that the socket will reuse everybody else's port
504 * without looking at the other's sk_reuse value.
505 */
506
507#define SK_NO_REUSE 0
508#define SK_CAN_REUSE 1
509#define SK_FORCE_REUSE 2
510
511int sk_set_peek_off(struct sock *sk, int val);
512
513static inline int sk_peek_offset(struct sock *sk, int flags)
514{
515 if (unlikely(flags & MSG_PEEK)) {
516 return READ_ONCE(sk->sk_peek_off);
517 }
518
519 return 0;
520}
521
522static inline void sk_peek_offset_bwd(struct sock *sk, int val)
523{
524 s32 off = READ_ONCE(sk->sk_peek_off);
525
526 if (unlikely(off >= 0)) {
527 off = max_t(s32, off - val, 0);
528 WRITE_ONCE(sk->sk_peek_off, off);
529 }
530}
531
532static inline void sk_peek_offset_fwd(struct sock *sk, int val)
533{
534 sk_peek_offset_bwd(sk, -val);
535}
536
537/*
538 * Hashed lists helper routines
539 */
540static inline struct sock *sk_entry(const struct hlist_node *node)
541{
542 return hlist_entry(node, struct sock, sk_node);
543}
544
545static inline struct sock *__sk_head(const struct hlist_head *head)
546{
547 return hlist_entry(head->first, struct sock, sk_node);
548}
549
550static inline struct sock *sk_head(const struct hlist_head *head)
551{
552 return hlist_empty(head) ? NULL : __sk_head(head);
553}
554
555static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
556{
557 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
558}
559
560static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
561{
562 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
563}
564
565static inline struct sock *sk_next(const struct sock *sk)
566{
567 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
568}
569
570static inline struct sock *sk_nulls_next(const struct sock *sk)
571{
572 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
573 hlist_nulls_entry(sk->sk_nulls_node.next,
574 struct sock, sk_nulls_node) :
575 NULL;
576}
577
578static inline bool sk_unhashed(const struct sock *sk)
579{
580 return hlist_unhashed(&sk->sk_node);
581}
582
583static inline bool sk_hashed(const struct sock *sk)
584{
585 return !sk_unhashed(sk);
586}
587
588static inline void sk_node_init(struct hlist_node *node)
589{
590 node->pprev = NULL;
591}
592
593static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
594{
595 node->pprev = NULL;
596}
597
598static inline void __sk_del_node(struct sock *sk)
599{
600 __hlist_del(&sk->sk_node);
601}
602
603/* NB: equivalent to hlist_del_init_rcu */
604static inline bool __sk_del_node_init(struct sock *sk)
605{
606 if (sk_hashed(sk)) {
607 __sk_del_node(sk);
608 sk_node_init(&sk->sk_node);
609 return true;
610 }
611 return false;
612}
613
614/* Grab socket reference count. This operation is valid only
615 when sk is ALREADY grabbed f.e. it is found in hash table
616 or a list and the lookup is made under lock preventing hash table
617 modifications.
618 */
619
620static __always_inline void sock_hold(struct sock *sk)
621{
622 refcount_inc(&sk->sk_refcnt);
623}
624
625/* Ungrab socket in the context, which assumes that socket refcnt
626 cannot hit zero, f.e. it is true in context of any socketcall.
627 */
628static __always_inline void __sock_put(struct sock *sk)
629{
630 refcount_dec(&sk->sk_refcnt);
631}
632
633static inline bool sk_del_node_init(struct sock *sk)
634{
635 bool rc = __sk_del_node_init(sk);
636
637 if (rc) {
638 /* paranoid for a while -acme */
639 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
640 __sock_put(sk);
641 }
642 return rc;
643}
644#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
645
646static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
647{
648 if (sk_hashed(sk)) {
649 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
650 return true;
651 }
652 return false;
653}
654
655static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
656{
657 bool rc = __sk_nulls_del_node_init_rcu(sk);
658
659 if (rc) {
660 /* paranoid for a while -acme */
661 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
662 __sock_put(sk);
663 }
664 return rc;
665}
666
667static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
668{
669 hlist_add_head(&sk->sk_node, list);
670}
671
672static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
673{
674 sock_hold(sk);
675 __sk_add_node(sk, list);
676}
677
678static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
679{
680 sock_hold(sk);
681 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
682 sk->sk_family == AF_INET6)
683 hlist_add_tail_rcu(&sk->sk_node, list);
684 else
685 hlist_add_head_rcu(&sk->sk_node, list);
686}
687
688static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
689{
690 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
691}
692
693static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
694{
695 sock_hold(sk);
696 __sk_nulls_add_node_rcu(sk, list);
697}
698
699static inline void __sk_del_bind_node(struct sock *sk)
700{
701 __hlist_del(&sk->sk_bind_node);
702}
703
704static inline void sk_add_bind_node(struct sock *sk,
705 struct hlist_head *list)
706{
707 hlist_add_head(&sk->sk_bind_node, list);
708}
709
710#define sk_for_each(__sk, list) \
711 hlist_for_each_entry(__sk, list, sk_node)
712#define sk_for_each_rcu(__sk, list) \
713 hlist_for_each_entry_rcu(__sk, list, sk_node)
714#define sk_nulls_for_each(__sk, node, list) \
715 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
716#define sk_nulls_for_each_rcu(__sk, node, list) \
717 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
718#define sk_for_each_from(__sk) \
719 hlist_for_each_entry_from(__sk, sk_node)
720#define sk_nulls_for_each_from(__sk, node) \
721 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
722 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
723#define sk_for_each_safe(__sk, tmp, list) \
724 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
725#define sk_for_each_bound(__sk, list) \
726 hlist_for_each_entry(__sk, list, sk_bind_node)
727
728/**
729 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
730 * @tpos: the type * to use as a loop cursor.
731 * @pos: the &struct hlist_node to use as a loop cursor.
732 * @head: the head for your list.
733 * @offset: offset of hlist_node within the struct.
734 *
735 */
736#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
737 for (pos = rcu_dereference(hlist_first_rcu(head)); \
738 pos != NULL && \
739 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
740 pos = rcu_dereference(hlist_next_rcu(pos)))
741
742static inline struct user_namespace *sk_user_ns(struct sock *sk)
743{
744 /* Careful only use this in a context where these parameters
745 * can not change and must all be valid, such as recvmsg from
746 * userspace.
747 */
748 return sk->sk_socket->file->f_cred->user_ns;
749}
750
751/* Sock flags */
752enum sock_flags {
753 SOCK_DEAD,
754 SOCK_DONE,
755 SOCK_URGINLINE,
756 SOCK_KEEPOPEN,
757 SOCK_LINGER,
758 SOCK_DESTROY,
759 SOCK_BROADCAST,
760 SOCK_TIMESTAMP,
761 SOCK_ZAPPED,
762 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
763 SOCK_DBG, /* %SO_DEBUG setting */
764 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
765 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
766 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
767 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
768 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
769 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
770 SOCK_FASYNC, /* fasync() active */
771 SOCK_RXQ_OVFL,
772 SOCK_ZEROCOPY, /* buffers from userspace */
773 SOCK_WIFI_STATUS, /* push wifi status to userspace */
774 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
775 * Will use last 4 bytes of packet sent from
776 * user-space instead.
777 */
778 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
779 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
780 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
781};
782
783#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
784
785static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
786{
787 nsk->sk_flags = osk->sk_flags;
788}
789
790static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
791{
792 __set_bit(flag, &sk->sk_flags);
793}
794
795static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
796{
797 __clear_bit(flag, &sk->sk_flags);
798}
799
800static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
801{
802 return test_bit(flag, &sk->sk_flags);
803}
804
805#ifdef CONFIG_NET
806extern struct static_key memalloc_socks;
807static inline int sk_memalloc_socks(void)
808{
809 return static_key_false(&memalloc_socks);
810}
811#else
812
813static inline int sk_memalloc_socks(void)
814{
815 return 0;
816}
817
818#endif
819
820static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
821{
822 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
823}
824
825static inline void sk_acceptq_removed(struct sock *sk)
826{
827 sk->sk_ack_backlog--;
828}
829
830static inline void sk_acceptq_added(struct sock *sk)
831{
832 sk->sk_ack_backlog++;
833}
834
835static inline bool sk_acceptq_is_full(const struct sock *sk)
836{
837 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
838}
839
840/*
841 * Compute minimal free write space needed to queue new packets.
842 */
843static inline int sk_stream_min_wspace(const struct sock *sk)
844{
845 return sk->sk_wmem_queued >> 1;
846}
847
848static inline int sk_stream_wspace(const struct sock *sk)
849{
850 return sk->sk_sndbuf - sk->sk_wmem_queued;
851}
852
853void sk_stream_write_space(struct sock *sk);
854
855/* OOB backlog add */
856static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
857{
858 /* dont let skb dst not refcounted, we are going to leave rcu lock */
859 skb_dst_force(skb);
860
861 if (!sk->sk_backlog.tail)
862 sk->sk_backlog.head = skb;
863 else
864 sk->sk_backlog.tail->next = skb;
865
866 sk->sk_backlog.tail = skb;
867 skb->next = NULL;
868}
869
870/*
871 * Take into account size of receive queue and backlog queue
872 * Do not take into account this skb truesize,
873 * to allow even a single big packet to come.
874 */
875static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
876{
877 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
878
879 return qsize > limit;
880}
881
882/* The per-socket spinlock must be held here. */
883static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
884 unsigned int limit)
885{
886 if (sk_rcvqueues_full(sk, limit))
887 return -ENOBUFS;
888
889 /*
890 * If the skb was allocated from pfmemalloc reserves, only
891 * allow SOCK_MEMALLOC sockets to use it as this socket is
892 * helping free memory
893 */
894 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
895 return -ENOMEM;
896
897 __sk_add_backlog(sk, skb);
898 sk->sk_backlog.len += skb->truesize;
899 return 0;
900}
901
902int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
903
904static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
905{
906 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
907 return __sk_backlog_rcv(sk, skb);
908
909 return sk->sk_backlog_rcv(sk, skb);
910}
911
912static inline void sk_incoming_cpu_update(struct sock *sk)
913{
914 int cpu = raw_smp_processor_id();
915
916 if (unlikely(sk->sk_incoming_cpu != cpu))
917 sk->sk_incoming_cpu = cpu;
918}
919
920static inline void sock_rps_record_flow_hash(__u32 hash)
921{
922#ifdef CONFIG_RPS
923 struct rps_sock_flow_table *sock_flow_table;
924
925 rcu_read_lock();
926 sock_flow_table = rcu_dereference(rps_sock_flow_table);
927 rps_record_sock_flow(sock_flow_table, hash);
928 rcu_read_unlock();
929#endif
930}
931
932static inline void sock_rps_record_flow(const struct sock *sk)
933{
934#ifdef CONFIG_RPS
935 if (static_key_false(&rfs_needed)) {
936 /* Reading sk->sk_rxhash might incur an expensive cache line
937 * miss.
938 *
939 * TCP_ESTABLISHED does cover almost all states where RFS
940 * might be useful, and is cheaper [1] than testing :
941 * IPv4: inet_sk(sk)->inet_daddr
942 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
943 * OR an additional socket flag
944 * [1] : sk_state and sk_prot are in the same cache line.
945 */
946 if (sk->sk_state == TCP_ESTABLISHED)
947 sock_rps_record_flow_hash(sk->sk_rxhash);
948 }
949#endif
950}
951
952static inline void sock_rps_save_rxhash(struct sock *sk,
953 const struct sk_buff *skb)
954{
955#ifdef CONFIG_RPS
956 if (unlikely(sk->sk_rxhash != skb->hash))
957 sk->sk_rxhash = skb->hash;
958#endif
959}
960
961static inline void sock_rps_reset_rxhash(struct sock *sk)
962{
963#ifdef CONFIG_RPS
964 sk->sk_rxhash = 0;
965#endif
966}
967
968#define sk_wait_event(__sk, __timeo, __condition, __wait) \
969 ({ int __rc; \
970 release_sock(__sk); \
971 __rc = __condition; \
972 if (!__rc) { \
973 *(__timeo) = wait_woken(__wait, \
974 TASK_INTERRUPTIBLE, \
975 *(__timeo)); \
976 } \
977 sched_annotate_sleep(); \
978 lock_sock(__sk); \
979 __rc = __condition; \
980 __rc; \
981 })
982
983int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
984int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
985void sk_stream_wait_close(struct sock *sk, long timeo_p);
986int sk_stream_error(struct sock *sk, int flags, int err);
987void sk_stream_kill_queues(struct sock *sk);
988void sk_set_memalloc(struct sock *sk);
989void sk_clear_memalloc(struct sock *sk);
990
991void __sk_flush_backlog(struct sock *sk);
992
993static inline bool sk_flush_backlog(struct sock *sk)
994{
995 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
996 __sk_flush_backlog(sk);
997 return true;
998 }
999 return false;
1000}
1001
1002int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1003
1004struct request_sock_ops;
1005struct timewait_sock_ops;
1006struct inet_hashinfo;
1007struct raw_hashinfo;
1008struct smc_hashinfo;
1009struct module;
1010
1011/*
1012 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1013 * un-modified. Special care is taken when initializing object to zero.
1014 */
1015static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1016{
1017 if (offsetof(struct sock, sk_node.next) != 0)
1018 memset(sk, 0, offsetof(struct sock, sk_node.next));
1019 memset(&sk->sk_node.pprev, 0,
1020 size - offsetof(struct sock, sk_node.pprev));
1021}
1022
1023/* Networking protocol blocks we attach to sockets.
1024 * socket layer -> transport layer interface
1025 */
1026struct proto {
1027 void (*close)(struct sock *sk,
1028 long timeout);
1029 int (*pre_connect)(struct sock *sk,
1030 struct sockaddr *uaddr,
1031 int addr_len);
1032 int (*connect)(struct sock *sk,
1033 struct sockaddr *uaddr,
1034 int addr_len);
1035 int (*disconnect)(struct sock *sk, int flags);
1036
1037 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1038 bool kern);
1039
1040 int (*ioctl)(struct sock *sk, int cmd,
1041 unsigned long arg);
1042 int (*init)(struct sock *sk);
1043 void (*destroy)(struct sock *sk);
1044 void (*shutdown)(struct sock *sk, int how);
1045 int (*setsockopt)(struct sock *sk, int level,
1046 int optname, char __user *optval,
1047 unsigned int optlen);
1048 int (*getsockopt)(struct sock *sk, int level,
1049 int optname, char __user *optval,
1050 int __user *option);
1051 void (*keepalive)(struct sock *sk, int valbool);
1052#ifdef CONFIG_COMPAT
1053 int (*compat_setsockopt)(struct sock *sk,
1054 int level,
1055 int optname, char __user *optval,
1056 unsigned int optlen);
1057 int (*compat_getsockopt)(struct sock *sk,
1058 int level,
1059 int optname, char __user *optval,
1060 int __user *option);
1061 int (*compat_ioctl)(struct sock *sk,
1062 unsigned int cmd, unsigned long arg);
1063#endif
1064 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1065 size_t len);
1066 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1067 size_t len, int noblock, int flags,
1068 int *addr_len);
1069 int (*sendpage)(struct sock *sk, struct page *page,
1070 int offset, size_t size, int flags);
1071 int (*bind)(struct sock *sk,
1072 struct sockaddr *uaddr, int addr_len);
1073
1074 int (*backlog_rcv) (struct sock *sk,
1075 struct sk_buff *skb);
1076
1077 void (*release_cb)(struct sock *sk);
1078
1079 /* Keeping track of sk's, looking them up, and port selection methods. */
1080 int (*hash)(struct sock *sk);
1081 void (*unhash)(struct sock *sk);
1082 void (*rehash)(struct sock *sk);
1083 int (*get_port)(struct sock *sk, unsigned short snum);
1084
1085 /* Keeping track of sockets in use */
1086#ifdef CONFIG_PROC_FS
1087 unsigned int inuse_idx;
1088#endif
1089
1090 bool (*stream_memory_free)(const struct sock *sk);
1091 bool (*stream_memory_read)(const struct sock *sk);
1092 /* Memory pressure */
1093 void (*enter_memory_pressure)(struct sock *sk);
1094 void (*leave_memory_pressure)(struct sock *sk);
1095 atomic_long_t *memory_allocated; /* Current allocated memory. */
1096 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1097 /*
1098 * Pressure flag: try to collapse.
1099 * Technical note: it is used by multiple contexts non atomically.
1100 * All the __sk_mem_schedule() is of this nature: accounting
1101 * is strict, actions are advisory and have some latency.
1102 */
1103 unsigned long *memory_pressure;
1104 long *sysctl_mem;
1105
1106 int *sysctl_wmem;
1107 int *sysctl_rmem;
1108 u32 sysctl_wmem_offset;
1109 u32 sysctl_rmem_offset;
1110
1111 int max_header;
1112 bool no_autobind;
1113
1114 struct kmem_cache *slab;
1115 unsigned int obj_size;
1116 slab_flags_t slab_flags;
1117 unsigned int useroffset; /* Usercopy region offset */
1118 unsigned int usersize; /* Usercopy region size */
1119
1120 struct percpu_counter *orphan_count;
1121
1122 struct request_sock_ops *rsk_prot;
1123 struct timewait_sock_ops *twsk_prot;
1124
1125 union {
1126 struct inet_hashinfo *hashinfo;
1127 struct udp_table *udp_table;
1128 struct raw_hashinfo *raw_hash;
1129 struct smc_hashinfo *smc_hash;
1130 } h;
1131
1132 struct module *owner;
1133
1134 char name[32];
1135
1136 struct list_head node;
1137#ifdef SOCK_REFCNT_DEBUG
1138 atomic_t socks;
1139#endif
1140 int (*diag_destroy)(struct sock *sk, int err);
1141} __randomize_layout;
1142
1143int proto_register(struct proto *prot, int alloc_slab);
1144void proto_unregister(struct proto *prot);
1145int sock_load_diag_module(int family, int protocol);
1146
1147#ifdef SOCK_REFCNT_DEBUG
1148static inline void sk_refcnt_debug_inc(struct sock *sk)
1149{
1150 atomic_inc(&sk->sk_prot->socks);
1151}
1152
1153static inline void sk_refcnt_debug_dec(struct sock *sk)
1154{
1155 atomic_dec(&sk->sk_prot->socks);
1156 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1157 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1158}
1159
1160static inline void sk_refcnt_debug_release(const struct sock *sk)
1161{
1162 if (refcount_read(&sk->sk_refcnt) != 1)
1163 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1164 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1165}
1166#else /* SOCK_REFCNT_DEBUG */
1167#define sk_refcnt_debug_inc(sk) do { } while (0)
1168#define sk_refcnt_debug_dec(sk) do { } while (0)
1169#define sk_refcnt_debug_release(sk) do { } while (0)
1170#endif /* SOCK_REFCNT_DEBUG */
1171
1172static inline bool sk_stream_memory_free(const struct sock *sk)
1173{
1174 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1175 return false;
1176
1177 return sk->sk_prot->stream_memory_free ?
1178 sk->sk_prot->stream_memory_free(sk) : true;
1179}
1180
1181static inline bool sk_stream_is_writeable(const struct sock *sk)
1182{
1183 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1184 sk_stream_memory_free(sk);
1185}
1186
1187static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1188 struct cgroup *ancestor)
1189{
1190#ifdef CONFIG_SOCK_CGROUP_DATA
1191 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1192 ancestor);
1193#else
1194 return -ENOTSUPP;
1195#endif
1196}
1197
1198static inline bool sk_has_memory_pressure(const struct sock *sk)
1199{
1200 return sk->sk_prot->memory_pressure != NULL;
1201}
1202
1203static inline bool sk_under_memory_pressure(const struct sock *sk)
1204{
1205 if (!sk->sk_prot->memory_pressure)
1206 return false;
1207
1208 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1209 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1210 return true;
1211
1212 return !!*sk->sk_prot->memory_pressure;
1213}
1214
1215static inline long
1216sk_memory_allocated(const struct sock *sk)
1217{
1218 return atomic_long_read(sk->sk_prot->memory_allocated);
1219}
1220
1221static inline long
1222sk_memory_allocated_add(struct sock *sk, int amt)
1223{
1224 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1225}
1226
1227static inline void
1228sk_memory_allocated_sub(struct sock *sk, int amt)
1229{
1230 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1231}
1232
1233static inline void sk_sockets_allocated_dec(struct sock *sk)
1234{
1235 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1236}
1237
1238static inline void sk_sockets_allocated_inc(struct sock *sk)
1239{
1240 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1241}
1242
1243static inline int
1244sk_sockets_allocated_read_positive(struct sock *sk)
1245{
1246 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1247}
1248
1249static inline int
1250proto_sockets_allocated_sum_positive(struct proto *prot)
1251{
1252 return percpu_counter_sum_positive(prot->sockets_allocated);
1253}
1254
1255static inline long
1256proto_memory_allocated(struct proto *prot)
1257{
1258 return atomic_long_read(prot->memory_allocated);
1259}
1260
1261static inline bool
1262proto_memory_pressure(struct proto *prot)
1263{
1264 if (!prot->memory_pressure)
1265 return false;
1266 return !!*prot->memory_pressure;
1267}
1268
1269
1270#ifdef CONFIG_PROC_FS
1271/* Called with local bh disabled */
1272void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1273int sock_prot_inuse_get(struct net *net, struct proto *proto);
1274int sock_inuse_get(struct net *net);
1275#else
1276static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1277 int inc)
1278{
1279}
1280#endif
1281
1282
1283/* With per-bucket locks this operation is not-atomic, so that
1284 * this version is not worse.
1285 */
1286static inline int __sk_prot_rehash(struct sock *sk)
1287{
1288 sk->sk_prot->unhash(sk);
1289 return sk->sk_prot->hash(sk);
1290}
1291
1292/* About 10 seconds */
1293#define SOCK_DESTROY_TIME (10*HZ)
1294
1295/* Sockets 0-1023 can't be bound to unless you are superuser */
1296#define PROT_SOCK 1024
1297
1298#define SHUTDOWN_MASK 3
1299#define RCV_SHUTDOWN 1
1300#define SEND_SHUTDOWN 2
1301
1302#define SOCK_SNDBUF_LOCK 1
1303#define SOCK_RCVBUF_LOCK 2
1304#define SOCK_BINDADDR_LOCK 4
1305#define SOCK_BINDPORT_LOCK 8
1306
1307struct socket_alloc {
1308 struct socket socket;
1309 struct inode vfs_inode;
1310};
1311
1312static inline struct socket *SOCKET_I(struct inode *inode)
1313{
1314 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1315}
1316
1317static inline struct inode *SOCK_INODE(struct socket *socket)
1318{
1319 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1320}
1321
1322/*
1323 * Functions for memory accounting
1324 */
1325int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1326int __sk_mem_schedule(struct sock *sk, int size, int kind);
1327void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1328void __sk_mem_reclaim(struct sock *sk, int amount);
1329
1330/* We used to have PAGE_SIZE here, but systems with 64KB pages
1331 * do not necessarily have 16x time more memory than 4KB ones.
1332 */
1333#define SK_MEM_QUANTUM 4096
1334#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1335#define SK_MEM_SEND 0
1336#define SK_MEM_RECV 1
1337
1338/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1339static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1340{
1341 long val = sk->sk_prot->sysctl_mem[index];
1342
1343#if PAGE_SIZE > SK_MEM_QUANTUM
1344 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1345#elif PAGE_SIZE < SK_MEM_QUANTUM
1346 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1347#endif
1348 return val;
1349}
1350
1351static inline int sk_mem_pages(int amt)
1352{
1353 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1354}
1355
1356static inline bool sk_has_account(struct sock *sk)
1357{
1358 /* return true if protocol supports memory accounting */
1359 return !!sk->sk_prot->memory_allocated;
1360}
1361
1362static inline bool sk_wmem_schedule(struct sock *sk, int size)
1363{
1364 if (!sk_has_account(sk))
1365 return true;
1366 return size <= sk->sk_forward_alloc ||
1367 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1368}
1369
1370static inline bool
1371sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1372{
1373 if (!sk_has_account(sk))
1374 return true;
1375 return size<= sk->sk_forward_alloc ||
1376 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1377 skb_pfmemalloc(skb);
1378}
1379
1380static inline void sk_mem_reclaim(struct sock *sk)
1381{
1382 if (!sk_has_account(sk))
1383 return;
1384 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1385 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1386}
1387
1388static inline void sk_mem_reclaim_partial(struct sock *sk)
1389{
1390 if (!sk_has_account(sk))
1391 return;
1392 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1393 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1394}
1395
1396static inline void sk_mem_charge(struct sock *sk, int size)
1397{
1398 if (!sk_has_account(sk))
1399 return;
1400 sk->sk_forward_alloc -= size;
1401}
1402
1403static inline void sk_mem_uncharge(struct sock *sk, int size)
1404{
1405 if (!sk_has_account(sk))
1406 return;
1407 sk->sk_forward_alloc += size;
1408
1409 /* Avoid a possible overflow.
1410 * TCP send queues can make this happen, if sk_mem_reclaim()
1411 * is not called and more than 2 GBytes are released at once.
1412 *
1413 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1414 * no need to hold that much forward allocation anyway.
1415 */
1416 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1417 __sk_mem_reclaim(sk, 1 << 20);
1418}
1419
1420static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1421{
1422 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1423 sk->sk_wmem_queued -= skb->truesize;
1424 sk_mem_uncharge(sk, skb->truesize);
1425 __kfree_skb(skb);
1426}
1427
1428static inline void sock_release_ownership(struct sock *sk)
1429{
1430 if (sk->sk_lock.owned) {
1431 sk->sk_lock.owned = 0;
1432
1433 /* The sk_lock has mutex_unlock() semantics: */
1434 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1435 }
1436}
1437
1438/*
1439 * Macro so as to not evaluate some arguments when
1440 * lockdep is not enabled.
1441 *
1442 * Mark both the sk_lock and the sk_lock.slock as a
1443 * per-address-family lock class.
1444 */
1445#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1446do { \
1447 sk->sk_lock.owned = 0; \
1448 init_waitqueue_head(&sk->sk_lock.wq); \
1449 spin_lock_init(&(sk)->sk_lock.slock); \
1450 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1451 sizeof((sk)->sk_lock)); \
1452 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1453 (skey), (sname)); \
1454 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1455} while (0)
1456
1457#ifdef CONFIG_LOCKDEP
1458static inline bool lockdep_sock_is_held(const struct sock *sk)
1459{
1460 return lockdep_is_held(&sk->sk_lock) ||
1461 lockdep_is_held(&sk->sk_lock.slock);
1462}
1463#endif
1464
1465void lock_sock_nested(struct sock *sk, int subclass);
1466
1467static inline void lock_sock(struct sock *sk)
1468{
1469 lock_sock_nested(sk, 0);
1470}
1471
1472void release_sock(struct sock *sk);
1473
1474/* BH context may only use the following locking interface. */
1475#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1476#define bh_lock_sock_nested(__sk) \
1477 spin_lock_nested(&((__sk)->sk_lock.slock), \
1478 SINGLE_DEPTH_NESTING)
1479#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1480
1481bool lock_sock_fast(struct sock *sk);
1482/**
1483 * unlock_sock_fast - complement of lock_sock_fast
1484 * @sk: socket
1485 * @slow: slow mode
1486 *
1487 * fast unlock socket for user context.
1488 * If slow mode is on, we call regular release_sock()
1489 */
1490static inline void unlock_sock_fast(struct sock *sk, bool slow)
1491{
1492 if (slow)
1493 release_sock(sk);
1494 else
1495 spin_unlock_bh(&sk->sk_lock.slock);
1496}
1497
1498/* Used by processes to "lock" a socket state, so that
1499 * interrupts and bottom half handlers won't change it
1500 * from under us. It essentially blocks any incoming
1501 * packets, so that we won't get any new data or any
1502 * packets that change the state of the socket.
1503 *
1504 * While locked, BH processing will add new packets to
1505 * the backlog queue. This queue is processed by the
1506 * owner of the socket lock right before it is released.
1507 *
1508 * Since ~2.3.5 it is also exclusive sleep lock serializing
1509 * accesses from user process context.
1510 */
1511
1512static inline void sock_owned_by_me(const struct sock *sk)
1513{
1514#ifdef CONFIG_LOCKDEP
1515 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1516#endif
1517}
1518
1519static inline bool sock_owned_by_user(const struct sock *sk)
1520{
1521 sock_owned_by_me(sk);
1522 return sk->sk_lock.owned;
1523}
1524
1525static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1526{
1527 return sk->sk_lock.owned;
1528}
1529
1530/* no reclassification while locks are held */
1531static inline bool sock_allow_reclassification(const struct sock *csk)
1532{
1533 struct sock *sk = (struct sock *)csk;
1534
1535 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1536}
1537
1538struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1539 struct proto *prot, int kern);
1540void sk_free(struct sock *sk);
1541void sk_destruct(struct sock *sk);
1542struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1543void sk_free_unlock_clone(struct sock *sk);
1544
1545struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1546 gfp_t priority);
1547void __sock_wfree(struct sk_buff *skb);
1548void sock_wfree(struct sk_buff *skb);
1549struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1550 gfp_t priority);
1551void skb_orphan_partial(struct sk_buff *skb);
1552void sock_rfree(struct sk_buff *skb);
1553void sock_efree(struct sk_buff *skb);
1554#ifdef CONFIG_INET
1555void sock_edemux(struct sk_buff *skb);
1556#else
1557#define sock_edemux sock_efree
1558#endif
1559
1560int sock_setsockopt(struct socket *sock, int level, int op,
1561 char __user *optval, unsigned int optlen);
1562
1563int sock_getsockopt(struct socket *sock, int level, int op,
1564 char __user *optval, int __user *optlen);
1565struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1566 int noblock, int *errcode);
1567struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1568 unsigned long data_len, int noblock,
1569 int *errcode, int max_page_order);
1570void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1571void sock_kfree_s(struct sock *sk, void *mem, int size);
1572void sock_kzfree_s(struct sock *sk, void *mem, int size);
1573void sk_send_sigurg(struct sock *sk);
1574
1575struct sockcm_cookie {
1576 u32 mark;
1577 u16 tsflags;
1578};
1579
1580int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1581 struct sockcm_cookie *sockc);
1582int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1583 struct sockcm_cookie *sockc);
1584
1585/*
1586 * Functions to fill in entries in struct proto_ops when a protocol
1587 * does not implement a particular function.
1588 */
1589int sock_no_bind(struct socket *, struct sockaddr *, int);
1590int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1591int sock_no_socketpair(struct socket *, struct socket *);
1592int sock_no_accept(struct socket *, struct socket *, int, bool);
1593int sock_no_getname(struct socket *, struct sockaddr *, int);
1594__poll_t sock_no_poll(struct file *, struct socket *,
1595 struct poll_table_struct *);
1596int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1597int sock_no_listen(struct socket *, int);
1598int sock_no_shutdown(struct socket *, int);
1599int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1600int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1601int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1602int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1603int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1604int sock_no_mmap(struct file *file, struct socket *sock,
1605 struct vm_area_struct *vma);
1606ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1607 size_t size, int flags);
1608ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1609 int offset, size_t size, int flags);
1610
1611/*
1612 * Functions to fill in entries in struct proto_ops when a protocol
1613 * uses the inet style.
1614 */
1615int sock_common_getsockopt(struct socket *sock, int level, int optname,
1616 char __user *optval, int __user *optlen);
1617int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1618 int flags);
1619int sock_common_setsockopt(struct socket *sock, int level, int optname,
1620 char __user *optval, unsigned int optlen);
1621int compat_sock_common_getsockopt(struct socket *sock, int level,
1622 int optname, char __user *optval, int __user *optlen);
1623int compat_sock_common_setsockopt(struct socket *sock, int level,
1624 int optname, char __user *optval, unsigned int optlen);
1625
1626void sk_common_release(struct sock *sk);
1627
1628/*
1629 * Default socket callbacks and setup code
1630 */
1631
1632/* Initialise core socket variables */
1633void sock_init_data(struct socket *sock, struct sock *sk);
1634
1635/*
1636 * Socket reference counting postulates.
1637 *
1638 * * Each user of socket SHOULD hold a reference count.
1639 * * Each access point to socket (an hash table bucket, reference from a list,
1640 * running timer, skb in flight MUST hold a reference count.
1641 * * When reference count hits 0, it means it will never increase back.
1642 * * When reference count hits 0, it means that no references from
1643 * outside exist to this socket and current process on current CPU
1644 * is last user and may/should destroy this socket.
1645 * * sk_free is called from any context: process, BH, IRQ. When
1646 * it is called, socket has no references from outside -> sk_free
1647 * may release descendant resources allocated by the socket, but
1648 * to the time when it is called, socket is NOT referenced by any
1649 * hash tables, lists etc.
1650 * * Packets, delivered from outside (from network or from another process)
1651 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1652 * when they sit in queue. Otherwise, packets will leak to hole, when
1653 * socket is looked up by one cpu and unhasing is made by another CPU.
1654 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1655 * (leak to backlog). Packet socket does all the processing inside
1656 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1657 * use separate SMP lock, so that they are prone too.
1658 */
1659
1660/* Ungrab socket and destroy it, if it was the last reference. */
1661static inline void sock_put(struct sock *sk)
1662{
1663 if (refcount_dec_and_test(&sk->sk_refcnt))
1664 sk_free(sk);
1665}
1666/* Generic version of sock_put(), dealing with all sockets
1667 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1668 */
1669void sock_gen_put(struct sock *sk);
1670
1671int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1672 unsigned int trim_cap, bool refcounted);
1673static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1674 const int nested)
1675{
1676 return __sk_receive_skb(sk, skb, nested, 1, true);
1677}
1678
1679static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1680{
1681 sk->sk_tx_queue_mapping = tx_queue;
1682}
1683
1684static inline void sk_tx_queue_clear(struct sock *sk)
1685{
1686 sk->sk_tx_queue_mapping = -1;
1687}
1688
1689static inline int sk_tx_queue_get(const struct sock *sk)
1690{
1691 return sk ? sk->sk_tx_queue_mapping : -1;
1692}
1693
1694static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1695{
1696 sk_tx_queue_clear(sk);
1697 sk->sk_socket = sock;
1698}
1699
1700static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1701{
1702 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1703 return &rcu_dereference_raw(sk->sk_wq)->wait;
1704}
1705/* Detach socket from process context.
1706 * Announce socket dead, detach it from wait queue and inode.
1707 * Note that parent inode held reference count on this struct sock,
1708 * we do not release it in this function, because protocol
1709 * probably wants some additional cleanups or even continuing
1710 * to work with this socket (TCP).
1711 */
1712static inline void sock_orphan(struct sock *sk)
1713{
1714 write_lock_bh(&sk->sk_callback_lock);
1715 sock_set_flag(sk, SOCK_DEAD);
1716 sk_set_socket(sk, NULL);
1717 sk->sk_wq = NULL;
1718 write_unlock_bh(&sk->sk_callback_lock);
1719}
1720
1721static inline void sock_graft(struct sock *sk, struct socket *parent)
1722{
1723 WARN_ON(parent->sk);
1724 write_lock_bh(&sk->sk_callback_lock);
1725 sk->sk_wq = parent->wq;
1726 parent->sk = sk;
1727 sk_set_socket(sk, parent);
1728 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1729 security_sock_graft(sk, parent);
1730 write_unlock_bh(&sk->sk_callback_lock);
1731}
1732
1733kuid_t sock_i_uid(struct sock *sk);
1734unsigned long sock_i_ino(struct sock *sk);
1735
1736static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1737{
1738 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1739}
1740
1741static inline u32 net_tx_rndhash(void)
1742{
1743 u32 v = prandom_u32();
1744
1745 return v ?: 1;
1746}
1747
1748static inline void sk_set_txhash(struct sock *sk)
1749{
1750 sk->sk_txhash = net_tx_rndhash();
1751}
1752
1753static inline void sk_rethink_txhash(struct sock *sk)
1754{
1755 if (sk->sk_txhash)
1756 sk_set_txhash(sk);
1757}
1758
1759static inline struct dst_entry *
1760__sk_dst_get(struct sock *sk)
1761{
1762 return rcu_dereference_check(sk->sk_dst_cache,
1763 lockdep_sock_is_held(sk));
1764}
1765
1766static inline struct dst_entry *
1767sk_dst_get(struct sock *sk)
1768{
1769 struct dst_entry *dst;
1770
1771 rcu_read_lock();
1772 dst = rcu_dereference(sk->sk_dst_cache);
1773 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1774 dst = NULL;
1775 rcu_read_unlock();
1776 return dst;
1777}
1778
1779static inline void dst_negative_advice(struct sock *sk)
1780{
1781 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1782
1783 sk_rethink_txhash(sk);
1784
1785 if (dst && dst->ops->negative_advice) {
1786 ndst = dst->ops->negative_advice(dst);
1787
1788 if (ndst != dst) {
1789 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1790 sk_tx_queue_clear(sk);
1791 sk->sk_dst_pending_confirm = 0;
1792 }
1793 }
1794}
1795
1796static inline void
1797__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1798{
1799 struct dst_entry *old_dst;
1800
1801 sk_tx_queue_clear(sk);
1802 sk->sk_dst_pending_confirm = 0;
1803 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1804 lockdep_sock_is_held(sk));
1805 rcu_assign_pointer(sk->sk_dst_cache, dst);
1806 dst_release(old_dst);
1807}
1808
1809static inline void
1810sk_dst_set(struct sock *sk, struct dst_entry *dst)
1811{
1812 struct dst_entry *old_dst;
1813
1814 sk_tx_queue_clear(sk);
1815 sk->sk_dst_pending_confirm = 0;
1816 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1817 dst_release(old_dst);
1818}
1819
1820static inline void
1821__sk_dst_reset(struct sock *sk)
1822{
1823 __sk_dst_set(sk, NULL);
1824}
1825
1826static inline void
1827sk_dst_reset(struct sock *sk)
1828{
1829 sk_dst_set(sk, NULL);
1830}
1831
1832struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1833
1834struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1835
1836static inline void sk_dst_confirm(struct sock *sk)
1837{
1838 if (!sk->sk_dst_pending_confirm)
1839 sk->sk_dst_pending_confirm = 1;
1840}
1841
1842static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1843{
1844 if (skb_get_dst_pending_confirm(skb)) {
1845 struct sock *sk = skb->sk;
1846 unsigned long now = jiffies;
1847
1848 /* avoid dirtying neighbour */
1849 if (n->confirmed != now)
1850 n->confirmed = now;
1851 if (sk && sk->sk_dst_pending_confirm)
1852 sk->sk_dst_pending_confirm = 0;
1853 }
1854}
1855
1856bool sk_mc_loop(struct sock *sk);
1857
1858static inline bool sk_can_gso(const struct sock *sk)
1859{
1860 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1861}
1862
1863void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1864
1865static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1866{
1867 sk->sk_route_nocaps |= flags;
1868 sk->sk_route_caps &= ~flags;
1869}
1870
1871static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1872 struct iov_iter *from, char *to,
1873 int copy, int offset)
1874{
1875 if (skb->ip_summed == CHECKSUM_NONE) {
1876 __wsum csum = 0;
1877 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1878 return -EFAULT;
1879 skb->csum = csum_block_add(skb->csum, csum, offset);
1880 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1881 if (!copy_from_iter_full_nocache(to, copy, from))
1882 return -EFAULT;
1883 } else if (!copy_from_iter_full(to, copy, from))
1884 return -EFAULT;
1885
1886 return 0;
1887}
1888
1889static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1890 struct iov_iter *from, int copy)
1891{
1892 int err, offset = skb->len;
1893
1894 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1895 copy, offset);
1896 if (err)
1897 __skb_trim(skb, offset);
1898
1899 return err;
1900}
1901
1902static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1903 struct sk_buff *skb,
1904 struct page *page,
1905 int off, int copy)
1906{
1907 int err;
1908
1909 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1910 copy, skb->len);
1911 if (err)
1912 return err;
1913
1914 skb->len += copy;
1915 skb->data_len += copy;
1916 skb->truesize += copy;
1917 sk->sk_wmem_queued += copy;
1918 sk_mem_charge(sk, copy);
1919 return 0;
1920}
1921
1922/**
1923 * sk_wmem_alloc_get - returns write allocations
1924 * @sk: socket
1925 *
1926 * Returns sk_wmem_alloc minus initial offset of one
1927 */
1928static inline int sk_wmem_alloc_get(const struct sock *sk)
1929{
1930 return refcount_read(&sk->sk_wmem_alloc) - 1;
1931}
1932
1933/**
1934 * sk_rmem_alloc_get - returns read allocations
1935 * @sk: socket
1936 *
1937 * Returns sk_rmem_alloc
1938 */
1939static inline int sk_rmem_alloc_get(const struct sock *sk)
1940{
1941 return atomic_read(&sk->sk_rmem_alloc);
1942}
1943
1944/**
1945 * sk_has_allocations - check if allocations are outstanding
1946 * @sk: socket
1947 *
1948 * Returns true if socket has write or read allocations
1949 */
1950static inline bool sk_has_allocations(const struct sock *sk)
1951{
1952 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1953}
1954
1955/**
1956 * skwq_has_sleeper - check if there are any waiting processes
1957 * @wq: struct socket_wq
1958 *
1959 * Returns true if socket_wq has waiting processes
1960 *
1961 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1962 * barrier call. They were added due to the race found within the tcp code.
1963 *
1964 * Consider following tcp code paths::
1965 *
1966 * CPU1 CPU2
1967 * sys_select receive packet
1968 * ... ...
1969 * __add_wait_queue update tp->rcv_nxt
1970 * ... ...
1971 * tp->rcv_nxt check sock_def_readable
1972 * ... {
1973 * schedule rcu_read_lock();
1974 * wq = rcu_dereference(sk->sk_wq);
1975 * if (wq && waitqueue_active(&wq->wait))
1976 * wake_up_interruptible(&wq->wait)
1977 * ...
1978 * }
1979 *
1980 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1981 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1982 * could then endup calling schedule and sleep forever if there are no more
1983 * data on the socket.
1984 *
1985 */
1986static inline bool skwq_has_sleeper(struct socket_wq *wq)
1987{
1988 return wq && wq_has_sleeper(&wq->wait);
1989}
1990
1991/**
1992 * sock_poll_wait - place memory barrier behind the poll_wait call.
1993 * @filp: file
1994 * @wait_address: socket wait queue
1995 * @p: poll_table
1996 *
1997 * See the comments in the wq_has_sleeper function.
1998 */
1999static inline void sock_poll_wait(struct file *filp,
2000 wait_queue_head_t *wait_address, poll_table *p)
2001{
2002 if (!poll_does_not_wait(p) && wait_address) {
2003 poll_wait(filp, wait_address, p);
2004 /* We need to be sure we are in sync with the
2005 * socket flags modification.
2006 *
2007 * This memory barrier is paired in the wq_has_sleeper.
2008 */
2009 smp_mb();
2010 }
2011}
2012
2013static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2014{
2015 if (sk->sk_txhash) {
2016 skb->l4_hash = 1;
2017 skb->hash = sk->sk_txhash;
2018 }
2019}
2020
2021void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2022
2023/*
2024 * Queue a received datagram if it will fit. Stream and sequenced
2025 * protocols can't normally use this as they need to fit buffers in
2026 * and play with them.
2027 *
2028 * Inlined as it's very short and called for pretty much every
2029 * packet ever received.
2030 */
2031static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2032{
2033 skb_orphan(skb);
2034 skb->sk = sk;
2035 skb->destructor = sock_rfree;
2036 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2037 sk_mem_charge(sk, skb->truesize);
2038}
2039
2040void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2041 unsigned long expires);
2042
2043void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2044
2045int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2046 struct sk_buff *skb, unsigned int flags,
2047 void (*destructor)(struct sock *sk,
2048 struct sk_buff *skb));
2049int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2050int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2051
2052int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2053struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2054
2055/*
2056 * Recover an error report and clear atomically
2057 */
2058
2059static inline int sock_error(struct sock *sk)
2060{
2061 int err;
2062 if (likely(!sk->sk_err))
2063 return 0;
2064 err = xchg(&sk->sk_err, 0);
2065 return -err;
2066}
2067
2068static inline unsigned long sock_wspace(struct sock *sk)
2069{
2070 int amt = 0;
2071
2072 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2073 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2074 if (amt < 0)
2075 amt = 0;
2076 }
2077 return amt;
2078}
2079
2080/* Note:
2081 * We use sk->sk_wq_raw, from contexts knowing this
2082 * pointer is not NULL and cannot disappear/change.
2083 */
2084static inline void sk_set_bit(int nr, struct sock *sk)
2085{
2086 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2087 !sock_flag(sk, SOCK_FASYNC))
2088 return;
2089
2090 set_bit(nr, &sk->sk_wq_raw->flags);
2091}
2092
2093static inline void sk_clear_bit(int nr, struct sock *sk)
2094{
2095 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2096 !sock_flag(sk, SOCK_FASYNC))
2097 return;
2098
2099 clear_bit(nr, &sk->sk_wq_raw->flags);
2100}
2101
2102static inline void sk_wake_async(const struct sock *sk, int how, int band)
2103{
2104 if (sock_flag(sk, SOCK_FASYNC)) {
2105 rcu_read_lock();
2106 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2107 rcu_read_unlock();
2108 }
2109}
2110
2111/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2112 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2113 * Note: for send buffers, TCP works better if we can build two skbs at
2114 * minimum.
2115 */
2116#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2117
2118#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2119#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2120
2121static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2122{
2123 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2124 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2125 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2126 }
2127}
2128
2129struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2130 bool force_schedule);
2131
2132/**
2133 * sk_page_frag - return an appropriate page_frag
2134 * @sk: socket
2135 *
2136 * If socket allocation mode allows current thread to sleep, it means its
2137 * safe to use the per task page_frag instead of the per socket one.
2138 */
2139static inline struct page_frag *sk_page_frag(struct sock *sk)
2140{
2141 if (gfpflags_allow_blocking(sk->sk_allocation))
2142 return ¤t->task_frag;
2143
2144 return &sk->sk_frag;
2145}
2146
2147bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2148
2149int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2150 int sg_start, int *sg_curr, unsigned int *sg_size,
2151 int first_coalesce);
2152
2153/*
2154 * Default write policy as shown to user space via poll/select/SIGIO
2155 */
2156static inline bool sock_writeable(const struct sock *sk)
2157{
2158 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2159}
2160
2161static inline gfp_t gfp_any(void)
2162{
2163 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2164}
2165
2166static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2167{
2168 return noblock ? 0 : sk->sk_rcvtimeo;
2169}
2170
2171static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2172{
2173 return noblock ? 0 : sk->sk_sndtimeo;
2174}
2175
2176static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2177{
2178 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2179}
2180
2181/* Alas, with timeout socket operations are not restartable.
2182 * Compare this to poll().
2183 */
2184static inline int sock_intr_errno(long timeo)
2185{
2186 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2187}
2188
2189struct sock_skb_cb {
2190 u32 dropcount;
2191};
2192
2193/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2194 * using skb->cb[] would keep using it directly and utilize its
2195 * alignement guarantee.
2196 */
2197#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2198 sizeof(struct sock_skb_cb)))
2199
2200#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2201 SOCK_SKB_CB_OFFSET))
2202
2203#define sock_skb_cb_check_size(size) \
2204 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2205
2206static inline void
2207sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2208{
2209 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2210 atomic_read(&sk->sk_drops) : 0;
2211}
2212
2213static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2214{
2215 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2216
2217 atomic_add(segs, &sk->sk_drops);
2218}
2219
2220void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2221 struct sk_buff *skb);
2222void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2223 struct sk_buff *skb);
2224
2225static inline void
2226sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2227{
2228 ktime_t kt = skb->tstamp;
2229 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2230
2231 /*
2232 * generate control messages if
2233 * - receive time stamping in software requested
2234 * - software time stamp available and wanted
2235 * - hardware time stamps available and wanted
2236 */
2237 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2238 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2239 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2240 (hwtstamps->hwtstamp &&
2241 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2242 __sock_recv_timestamp(msg, sk, skb);
2243 else
2244 sk->sk_stamp = kt;
2245
2246 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2247 __sock_recv_wifi_status(msg, sk, skb);
2248}
2249
2250void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2251 struct sk_buff *skb);
2252
2253#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2254static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2255 struct sk_buff *skb)
2256{
2257#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2258 (1UL << SOCK_RCVTSTAMP))
2259#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2260 SOF_TIMESTAMPING_RAW_HARDWARE)
2261
2262 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2263 __sock_recv_ts_and_drops(msg, sk, skb);
2264 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2265 sk->sk_stamp = skb->tstamp;
2266 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2267 sk->sk_stamp = 0;
2268}
2269
2270void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2271
2272/**
2273 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2274 * @sk: socket sending this packet
2275 * @tsflags: timestamping flags to use
2276 * @tx_flags: completed with instructions for time stamping
2277 *
2278 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2279 */
2280static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2281 __u8 *tx_flags)
2282{
2283 if (unlikely(tsflags))
2284 __sock_tx_timestamp(tsflags, tx_flags);
2285 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2286 *tx_flags |= SKBTX_WIFI_STATUS;
2287}
2288
2289/**
2290 * sk_eat_skb - Release a skb if it is no longer needed
2291 * @sk: socket to eat this skb from
2292 * @skb: socket buffer to eat
2293 *
2294 * This routine must be called with interrupts disabled or with the socket
2295 * locked so that the sk_buff queue operation is ok.
2296*/
2297static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2298{
2299 __skb_unlink(skb, &sk->sk_receive_queue);
2300 __kfree_skb(skb);
2301}
2302
2303static inline
2304struct net *sock_net(const struct sock *sk)
2305{
2306 return read_pnet(&sk->sk_net);
2307}
2308
2309static inline
2310void sock_net_set(struct sock *sk, struct net *net)
2311{
2312 write_pnet(&sk->sk_net, net);
2313}
2314
2315static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2316{
2317 if (skb->sk) {
2318 struct sock *sk = skb->sk;
2319
2320 skb->destructor = NULL;
2321 skb->sk = NULL;
2322 return sk;
2323 }
2324 return NULL;
2325}
2326
2327/* This helper checks if a socket is a full socket,
2328 * ie _not_ a timewait or request socket.
2329 */
2330static inline bool sk_fullsock(const struct sock *sk)
2331{
2332 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2333}
2334
2335/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2336 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2337 */
2338static inline bool sk_listener(const struct sock *sk)
2339{
2340 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2341}
2342
2343void sock_enable_timestamp(struct sock *sk, int flag);
2344int sock_get_timestamp(struct sock *, struct timeval __user *);
2345int sock_get_timestampns(struct sock *, struct timespec __user *);
2346int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2347 int type);
2348
2349bool sk_ns_capable(const struct sock *sk,
2350 struct user_namespace *user_ns, int cap);
2351bool sk_capable(const struct sock *sk, int cap);
2352bool sk_net_capable(const struct sock *sk, int cap);
2353
2354void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2355
2356/* Take into consideration the size of the struct sk_buff overhead in the
2357 * determination of these values, since that is non-constant across
2358 * platforms. This makes socket queueing behavior and performance
2359 * not depend upon such differences.
2360 */
2361#define _SK_MEM_PACKETS 256
2362#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2363#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2364#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2365
2366extern __u32 sysctl_wmem_max;
2367extern __u32 sysctl_rmem_max;
2368
2369extern int sysctl_tstamp_allow_data;
2370extern int sysctl_optmem_max;
2371
2372extern __u32 sysctl_wmem_default;
2373extern __u32 sysctl_rmem_default;
2374
2375static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2376{
2377 /* Does this proto have per netns sysctl_wmem ? */
2378 if (proto->sysctl_wmem_offset)
2379 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2380
2381 return *proto->sysctl_wmem;
2382}
2383
2384static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2385{
2386 /* Does this proto have per netns sysctl_rmem ? */
2387 if (proto->sysctl_rmem_offset)
2388 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2389
2390 return *proto->sysctl_rmem;
2391}
2392
2393/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2394 * Some wifi drivers need to tweak it to get more chunks.
2395 * They can use this helper from their ndo_start_xmit()
2396 */
2397static inline void sk_pacing_shift_update(struct sock *sk, int val)
2398{
2399 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2400 return;
2401 sk->sk_pacing_shift = val;
2402}
2403
2404/* if a socket is bound to a device, check that the given device
2405 * index is either the same or that the socket is bound to an L3
2406 * master device and the given device index is also enslaved to
2407 * that L3 master
2408 */
2409static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2410{
2411 int mdif;
2412
2413 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2414 return true;
2415
2416 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2417 if (mdif && mdif == sk->sk_bound_dev_if)
2418 return true;
2419
2420 return false;
2421}
2422
2423#endif /* _SOCK_H */