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_rx_queue_mapping: rx queue number for this connection
143 * @skc_flags: place holder for sk_flags
144 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
145 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
146 * @skc_incoming_cpu: record/match cpu processing incoming packets
147 * @skc_refcnt: reference count
148 *
149 * This is the minimal network layer representation of sockets, the header
150 * for struct sock and struct inet_timewait_sock.
151 */
152struct sock_common {
153 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
154 * address on 64bit arches : cf INET_MATCH()
155 */
156 union {
157 __addrpair skc_addrpair;
158 struct {
159 __be32 skc_daddr;
160 __be32 skc_rcv_saddr;
161 };
162 };
163 union {
164 unsigned int skc_hash;
165 __u16 skc_u16hashes[2];
166 };
167 /* skc_dport && skc_num must be grouped as well */
168 union {
169 __portpair skc_portpair;
170 struct {
171 __be16 skc_dport;
172 __u16 skc_num;
173 };
174 };
175
176 unsigned short skc_family;
177 volatile unsigned char skc_state;
178 unsigned char skc_reuse:4;
179 unsigned char skc_reuseport:1;
180 unsigned char skc_ipv6only:1;
181 unsigned char skc_net_refcnt:1;
182 int skc_bound_dev_if;
183 union {
184 struct hlist_node skc_bind_node;
185 struct hlist_node skc_portaddr_node;
186 };
187 struct proto *skc_prot;
188 possible_net_t skc_net;
189
190#if IS_ENABLED(CONFIG_IPV6)
191 struct in6_addr skc_v6_daddr;
192 struct in6_addr skc_v6_rcv_saddr;
193#endif
194
195 atomic64_t skc_cookie;
196
197 /* following fields are padding to force
198 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
199 * assuming IPV6 is enabled. We use this padding differently
200 * for different kind of 'sockets'
201 */
202 union {
203 unsigned long skc_flags;
204 struct sock *skc_listener; /* request_sock */
205 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
206 };
207 /*
208 * fields between dontcopy_begin/dontcopy_end
209 * are not copied in sock_copy()
210 */
211 /* private: */
212 int skc_dontcopy_begin[0];
213 /* public: */
214 union {
215 struct hlist_node skc_node;
216 struct hlist_nulls_node skc_nulls_node;
217 };
218 unsigned short skc_tx_queue_mapping;
219#ifdef CONFIG_XPS
220 unsigned short skc_rx_queue_mapping;
221#endif
222 union {
223 int skc_incoming_cpu;
224 u32 skc_rcv_wnd;
225 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
226 };
227
228 refcount_t skc_refcnt;
229 /* private: */
230 int skc_dontcopy_end[0];
231 union {
232 u32 skc_rxhash;
233 u32 skc_window_clamp;
234 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
235 };
236 /* public: */
237};
238
239/**
240 * struct sock - network layer representation of sockets
241 * @__sk_common: shared layout with inet_timewait_sock
242 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
243 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
244 * @sk_lock: synchronizer
245 * @sk_kern_sock: True if sock is using kernel lock classes
246 * @sk_rcvbuf: size of receive buffer in bytes
247 * @sk_wq: sock wait queue and async head
248 * @sk_rx_dst: receive input route used by early demux
249 * @sk_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_napi_id: id of the last napi context to receive data for sk
260 * @sk_ll_usec: usecs to busypoll when there is no data
261 * @sk_allocation: allocation mode
262 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
263 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
264 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
265 * @sk_sndbuf: size of send buffer in bytes
266 * @__sk_flags_offset: empty field used to determine location of bitfield
267 * @sk_padding: unused element for alignment
268 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
269 * @sk_no_check_rx: allow zero checksum in RX packets
270 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
271 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
272 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
273 * @sk_gso_max_size: Maximum GSO segment size to build
274 * @sk_gso_max_segs: Maximum number of GSO segments
275 * @sk_pacing_shift: scaling factor for TCP Small Queues
276 * @sk_lingertime: %SO_LINGER l_linger setting
277 * @sk_backlog: always used with the per-socket spinlock held
278 * @sk_callback_lock: used with the callbacks in the end of this struct
279 * @sk_error_queue: rarely used
280 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
281 * IPV6_ADDRFORM for instance)
282 * @sk_err: last error
283 * @sk_err_soft: errors that don't cause failure but are the cause of a
284 * persistent failure not just 'timed out'
285 * @sk_drops: raw/udp drops counter
286 * @sk_ack_backlog: current listen backlog
287 * @sk_max_ack_backlog: listen backlog set in listen()
288 * @sk_uid: user id of owner
289 * @sk_priority: %SO_PRIORITY setting
290 * @sk_type: socket type (%SOCK_STREAM, etc)
291 * @sk_protocol: which protocol this socket belongs in this network family
292 * @sk_peer_pid: &struct pid for this socket's peer
293 * @sk_peer_cred: %SO_PEERCRED setting
294 * @sk_rcvlowat: %SO_RCVLOWAT setting
295 * @sk_rcvtimeo: %SO_RCVTIMEO setting
296 * @sk_sndtimeo: %SO_SNDTIMEO setting
297 * @sk_txhash: computed flow hash for use on transmit
298 * @sk_filter: socket filtering instructions
299 * @sk_timer: sock cleanup timer
300 * @sk_stamp: time stamp of last packet received
301 * @sk_tsflags: SO_TIMESTAMPING socket options
302 * @sk_tskey: counter to disambiguate concurrent tstamp requests
303 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
304 * @sk_socket: Identd and reporting IO signals
305 * @sk_user_data: RPC layer private data
306 * @sk_frag: cached page frag
307 * @sk_peek_off: current peek_offset value
308 * @sk_send_head: front of stuff to transmit
309 * @sk_security: used by security modules
310 * @sk_mark: generic packet mark
311 * @sk_cgrp_data: cgroup data for this cgroup
312 * @sk_memcg: this socket's memory cgroup association
313 * @sk_write_pending: a write to stream socket waits to start
314 * @sk_state_change: callback to indicate change in the state of the sock
315 * @sk_data_ready: callback to indicate there is data to be processed
316 * @sk_write_space: callback to indicate there is bf sending space available
317 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
318 * @sk_backlog_rcv: callback to process the backlog
319 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
320 * @sk_reuseport_cb: reuseport group container
321 * @sk_rcu: used during RCU grace period
322 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
323 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
324 * @sk_txtime_unused: unused txtime flags
325 */
326struct sock {
327 /*
328 * Now struct inet_timewait_sock also uses sock_common, so please just
329 * don't add nothing before this first member (__sk_common) --acme
330 */
331 struct sock_common __sk_common;
332#define sk_node __sk_common.skc_node
333#define sk_nulls_node __sk_common.skc_nulls_node
334#define sk_refcnt __sk_common.skc_refcnt
335#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
336#ifdef CONFIG_XPS
337#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
338#endif
339
340#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
341#define sk_dontcopy_end __sk_common.skc_dontcopy_end
342#define sk_hash __sk_common.skc_hash
343#define sk_portpair __sk_common.skc_portpair
344#define sk_num __sk_common.skc_num
345#define sk_dport __sk_common.skc_dport
346#define sk_addrpair __sk_common.skc_addrpair
347#define sk_daddr __sk_common.skc_daddr
348#define sk_rcv_saddr __sk_common.skc_rcv_saddr
349#define sk_family __sk_common.skc_family
350#define sk_state __sk_common.skc_state
351#define sk_reuse __sk_common.skc_reuse
352#define sk_reuseport __sk_common.skc_reuseport
353#define sk_ipv6only __sk_common.skc_ipv6only
354#define sk_net_refcnt __sk_common.skc_net_refcnt
355#define sk_bound_dev_if __sk_common.skc_bound_dev_if
356#define sk_bind_node __sk_common.skc_bind_node
357#define sk_prot __sk_common.skc_prot
358#define sk_net __sk_common.skc_net
359#define sk_v6_daddr __sk_common.skc_v6_daddr
360#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
361#define sk_cookie __sk_common.skc_cookie
362#define sk_incoming_cpu __sk_common.skc_incoming_cpu
363#define sk_flags __sk_common.skc_flags
364#define sk_rxhash __sk_common.skc_rxhash
365
366 socket_lock_t sk_lock;
367 atomic_t sk_drops;
368 int sk_rcvlowat;
369 struct sk_buff_head sk_error_queue;
370 struct sk_buff_head sk_receive_queue;
371 /*
372 * The backlog queue is special, it is always used with
373 * the per-socket spinlock held and requires low latency
374 * access. Therefore we special case it's implementation.
375 * Note : rmem_alloc is in this structure to fill a hole
376 * on 64bit arches, not because its logically part of
377 * backlog.
378 */
379 struct {
380 atomic_t rmem_alloc;
381 int len;
382 struct sk_buff *head;
383 struct sk_buff *tail;
384 } sk_backlog;
385#define sk_rmem_alloc sk_backlog.rmem_alloc
386
387 int sk_forward_alloc;
388#ifdef CONFIG_NET_RX_BUSY_POLL
389 unsigned int sk_ll_usec;
390 /* ===== mostly read cache line ===== */
391 unsigned int sk_napi_id;
392#endif
393 int sk_rcvbuf;
394
395 struct sk_filter __rcu *sk_filter;
396 union {
397 struct socket_wq __rcu *sk_wq;
398 struct socket_wq *sk_wq_raw;
399 };
400#ifdef CONFIG_XFRM
401 struct xfrm_policy __rcu *sk_policy[2];
402#endif
403 struct dst_entry *sk_rx_dst;
404 struct dst_entry __rcu *sk_dst_cache;
405 atomic_t sk_omem_alloc;
406 int sk_sndbuf;
407
408 /* ===== cache line for TX ===== */
409 int sk_wmem_queued;
410 refcount_t sk_wmem_alloc;
411 unsigned long sk_tsq_flags;
412 union {
413 struct sk_buff *sk_send_head;
414 struct rb_root tcp_rtx_queue;
415 };
416 struct sk_buff_head sk_write_queue;
417 __s32 sk_peek_off;
418 int sk_write_pending;
419 __u32 sk_dst_pending_confirm;
420 u32 sk_pacing_status; /* see enum sk_pacing */
421 long sk_sndtimeo;
422 struct timer_list sk_timer;
423 __u32 sk_priority;
424 __u32 sk_mark;
425 u32 sk_pacing_rate; /* bytes per second */
426 u32 sk_max_pacing_rate;
427 struct page_frag sk_frag;
428 netdev_features_t sk_route_caps;
429 netdev_features_t sk_route_nocaps;
430 netdev_features_t sk_route_forced_caps;
431 int sk_gso_type;
432 unsigned int sk_gso_max_size;
433 gfp_t sk_allocation;
434 __u32 sk_txhash;
435
436 /*
437 * Because of non atomicity rules, all
438 * changes are protected by socket lock.
439 */
440 unsigned int __sk_flags_offset[0];
441#ifdef __BIG_ENDIAN_BITFIELD
442#define SK_FL_PROTO_SHIFT 16
443#define SK_FL_PROTO_MASK 0x00ff0000
444
445#define SK_FL_TYPE_SHIFT 0
446#define SK_FL_TYPE_MASK 0x0000ffff
447#else
448#define SK_FL_PROTO_SHIFT 8
449#define SK_FL_PROTO_MASK 0x0000ff00
450
451#define SK_FL_TYPE_SHIFT 16
452#define SK_FL_TYPE_MASK 0xffff0000
453#endif
454
455 unsigned int sk_padding : 1,
456 sk_kern_sock : 1,
457 sk_no_check_tx : 1,
458 sk_no_check_rx : 1,
459 sk_userlocks : 4,
460 sk_protocol : 8,
461 sk_type : 16;
462#define SK_PROTOCOL_MAX U8_MAX
463 u16 sk_gso_max_segs;
464 u8 sk_pacing_shift;
465 unsigned long sk_lingertime;
466 struct proto *sk_prot_creator;
467 rwlock_t sk_callback_lock;
468 int sk_err,
469 sk_err_soft;
470 u32 sk_ack_backlog;
471 u32 sk_max_ack_backlog;
472 kuid_t sk_uid;
473 struct pid *sk_peer_pid;
474 const struct cred *sk_peer_cred;
475 long sk_rcvtimeo;
476 ktime_t sk_stamp;
477 u16 sk_tsflags;
478 u8 sk_shutdown;
479 u32 sk_tskey;
480 atomic_t sk_zckey;
481
482 u8 sk_clockid;
483 u8 sk_txtime_deadline_mode : 1,
484 sk_txtime_report_errors : 1,
485 sk_txtime_unused : 6;
486
487 struct socket *sk_socket;
488 void *sk_user_data;
489#ifdef CONFIG_SECURITY
490 void *sk_security;
491#endif
492 struct sock_cgroup_data sk_cgrp_data;
493 struct mem_cgroup *sk_memcg;
494 void (*sk_state_change)(struct sock *sk);
495 void (*sk_data_ready)(struct sock *sk);
496 void (*sk_write_space)(struct sock *sk);
497 void (*sk_error_report)(struct sock *sk);
498 int (*sk_backlog_rcv)(struct sock *sk,
499 struct sk_buff *skb);
500#ifdef CONFIG_SOCK_VALIDATE_XMIT
501 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
502 struct net_device *dev,
503 struct sk_buff *skb);
504#endif
505 void (*sk_destruct)(struct sock *sk);
506 struct sock_reuseport __rcu *sk_reuseport_cb;
507 struct rcu_head sk_rcu;
508};
509
510enum sk_pacing {
511 SK_PACING_NONE = 0,
512 SK_PACING_NEEDED = 1,
513 SK_PACING_FQ = 2,
514};
515
516#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
517
518#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
519#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
520
521/*
522 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
523 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
524 * on a socket means that the socket will reuse everybody else's port
525 * without looking at the other's sk_reuse value.
526 */
527
528#define SK_NO_REUSE 0
529#define SK_CAN_REUSE 1
530#define SK_FORCE_REUSE 2
531
532int sk_set_peek_off(struct sock *sk, int val);
533
534static inline int sk_peek_offset(struct sock *sk, int flags)
535{
536 if (unlikely(flags & MSG_PEEK)) {
537 return READ_ONCE(sk->sk_peek_off);
538 }
539
540 return 0;
541}
542
543static inline void sk_peek_offset_bwd(struct sock *sk, int val)
544{
545 s32 off = READ_ONCE(sk->sk_peek_off);
546
547 if (unlikely(off >= 0)) {
548 off = max_t(s32, off - val, 0);
549 WRITE_ONCE(sk->sk_peek_off, off);
550 }
551}
552
553static inline void sk_peek_offset_fwd(struct sock *sk, int val)
554{
555 sk_peek_offset_bwd(sk, -val);
556}
557
558/*
559 * Hashed lists helper routines
560 */
561static inline struct sock *sk_entry(const struct hlist_node *node)
562{
563 return hlist_entry(node, struct sock, sk_node);
564}
565
566static inline struct sock *__sk_head(const struct hlist_head *head)
567{
568 return hlist_entry(head->first, struct sock, sk_node);
569}
570
571static inline struct sock *sk_head(const struct hlist_head *head)
572{
573 return hlist_empty(head) ? NULL : __sk_head(head);
574}
575
576static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
577{
578 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
579}
580
581static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
582{
583 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
584}
585
586static inline struct sock *sk_next(const struct sock *sk)
587{
588 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
589}
590
591static inline struct sock *sk_nulls_next(const struct sock *sk)
592{
593 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
594 hlist_nulls_entry(sk->sk_nulls_node.next,
595 struct sock, sk_nulls_node) :
596 NULL;
597}
598
599static inline bool sk_unhashed(const struct sock *sk)
600{
601 return hlist_unhashed(&sk->sk_node);
602}
603
604static inline bool sk_hashed(const struct sock *sk)
605{
606 return !sk_unhashed(sk);
607}
608
609static inline void sk_node_init(struct hlist_node *node)
610{
611 node->pprev = NULL;
612}
613
614static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
615{
616 node->pprev = NULL;
617}
618
619static inline void __sk_del_node(struct sock *sk)
620{
621 __hlist_del(&sk->sk_node);
622}
623
624/* NB: equivalent to hlist_del_init_rcu */
625static inline bool __sk_del_node_init(struct sock *sk)
626{
627 if (sk_hashed(sk)) {
628 __sk_del_node(sk);
629 sk_node_init(&sk->sk_node);
630 return true;
631 }
632 return false;
633}
634
635/* Grab socket reference count. This operation is valid only
636 when sk is ALREADY grabbed f.e. it is found in hash table
637 or a list and the lookup is made under lock preventing hash table
638 modifications.
639 */
640
641static __always_inline void sock_hold(struct sock *sk)
642{
643 refcount_inc(&sk->sk_refcnt);
644}
645
646/* Ungrab socket in the context, which assumes that socket refcnt
647 cannot hit zero, f.e. it is true in context of any socketcall.
648 */
649static __always_inline void __sock_put(struct sock *sk)
650{
651 refcount_dec(&sk->sk_refcnt);
652}
653
654static inline bool sk_del_node_init(struct sock *sk)
655{
656 bool rc = __sk_del_node_init(sk);
657
658 if (rc) {
659 /* paranoid for a while -acme */
660 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
661 __sock_put(sk);
662 }
663 return rc;
664}
665#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
666
667static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
668{
669 if (sk_hashed(sk)) {
670 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
671 return true;
672 }
673 return false;
674}
675
676static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
677{
678 bool rc = __sk_nulls_del_node_init_rcu(sk);
679
680 if (rc) {
681 /* paranoid for a while -acme */
682 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
683 __sock_put(sk);
684 }
685 return rc;
686}
687
688static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
689{
690 hlist_add_head(&sk->sk_node, list);
691}
692
693static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
694{
695 sock_hold(sk);
696 __sk_add_node(sk, list);
697}
698
699static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
700{
701 sock_hold(sk);
702 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
703 sk->sk_family == AF_INET6)
704 hlist_add_tail_rcu(&sk->sk_node, list);
705 else
706 hlist_add_head_rcu(&sk->sk_node, list);
707}
708
709static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
710{
711 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
712}
713
714static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
715{
716 sock_hold(sk);
717 __sk_nulls_add_node_rcu(sk, list);
718}
719
720static inline void __sk_del_bind_node(struct sock *sk)
721{
722 __hlist_del(&sk->sk_bind_node);
723}
724
725static inline void sk_add_bind_node(struct sock *sk,
726 struct hlist_head *list)
727{
728 hlist_add_head(&sk->sk_bind_node, list);
729}
730
731#define sk_for_each(__sk, list) \
732 hlist_for_each_entry(__sk, list, sk_node)
733#define sk_for_each_rcu(__sk, list) \
734 hlist_for_each_entry_rcu(__sk, list, sk_node)
735#define sk_nulls_for_each(__sk, node, list) \
736 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
737#define sk_nulls_for_each_rcu(__sk, node, list) \
738 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
739#define sk_for_each_from(__sk) \
740 hlist_for_each_entry_from(__sk, sk_node)
741#define sk_nulls_for_each_from(__sk, node) \
742 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
743 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
744#define sk_for_each_safe(__sk, tmp, list) \
745 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
746#define sk_for_each_bound(__sk, list) \
747 hlist_for_each_entry(__sk, list, sk_bind_node)
748
749/**
750 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
751 * @tpos: the type * to use as a loop cursor.
752 * @pos: the &struct hlist_node to use as a loop cursor.
753 * @head: the head for your list.
754 * @offset: offset of hlist_node within the struct.
755 *
756 */
757#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
758 for (pos = rcu_dereference(hlist_first_rcu(head)); \
759 pos != NULL && \
760 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
761 pos = rcu_dereference(hlist_next_rcu(pos)))
762
763static inline struct user_namespace *sk_user_ns(struct sock *sk)
764{
765 /* Careful only use this in a context where these parameters
766 * can not change and must all be valid, such as recvmsg from
767 * userspace.
768 */
769 return sk->sk_socket->file->f_cred->user_ns;
770}
771
772/* Sock flags */
773enum sock_flags {
774 SOCK_DEAD,
775 SOCK_DONE,
776 SOCK_URGINLINE,
777 SOCK_KEEPOPEN,
778 SOCK_LINGER,
779 SOCK_DESTROY,
780 SOCK_BROADCAST,
781 SOCK_TIMESTAMP,
782 SOCK_ZAPPED,
783 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
784 SOCK_DBG, /* %SO_DEBUG setting */
785 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
786 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
787 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
788 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
789 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
790 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
791 SOCK_FASYNC, /* fasync() active */
792 SOCK_RXQ_OVFL,
793 SOCK_ZEROCOPY, /* buffers from userspace */
794 SOCK_WIFI_STATUS, /* push wifi status to userspace */
795 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
796 * Will use last 4 bytes of packet sent from
797 * user-space instead.
798 */
799 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
800 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
801 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
802 SOCK_TXTIME,
803};
804
805#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
806
807static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
808{
809 nsk->sk_flags = osk->sk_flags;
810}
811
812static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
813{
814 __set_bit(flag, &sk->sk_flags);
815}
816
817static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
818{
819 __clear_bit(flag, &sk->sk_flags);
820}
821
822static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
823{
824 return test_bit(flag, &sk->sk_flags);
825}
826
827#ifdef CONFIG_NET
828DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
829static inline int sk_memalloc_socks(void)
830{
831 return static_branch_unlikely(&memalloc_socks_key);
832}
833#else
834
835static inline int sk_memalloc_socks(void)
836{
837 return 0;
838}
839
840#endif
841
842static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
843{
844 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
845}
846
847static inline void sk_acceptq_removed(struct sock *sk)
848{
849 sk->sk_ack_backlog--;
850}
851
852static inline void sk_acceptq_added(struct sock *sk)
853{
854 sk->sk_ack_backlog++;
855}
856
857static inline bool sk_acceptq_is_full(const struct sock *sk)
858{
859 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
860}
861
862/*
863 * Compute minimal free write space needed to queue new packets.
864 */
865static inline int sk_stream_min_wspace(const struct sock *sk)
866{
867 return sk->sk_wmem_queued >> 1;
868}
869
870static inline int sk_stream_wspace(const struct sock *sk)
871{
872 return sk->sk_sndbuf - sk->sk_wmem_queued;
873}
874
875void sk_stream_write_space(struct sock *sk);
876
877/* OOB backlog add */
878static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
879{
880 /* dont let skb dst not refcounted, we are going to leave rcu lock */
881 skb_dst_force(skb);
882
883 if (!sk->sk_backlog.tail)
884 sk->sk_backlog.head = skb;
885 else
886 sk->sk_backlog.tail->next = skb;
887
888 sk->sk_backlog.tail = skb;
889 skb->next = NULL;
890}
891
892/*
893 * Take into account size of receive queue and backlog queue
894 * Do not take into account this skb truesize,
895 * to allow even a single big packet to come.
896 */
897static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
898{
899 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
900
901 return qsize > limit;
902}
903
904/* The per-socket spinlock must be held here. */
905static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
906 unsigned int limit)
907{
908 if (sk_rcvqueues_full(sk, limit))
909 return -ENOBUFS;
910
911 /*
912 * If the skb was allocated from pfmemalloc reserves, only
913 * allow SOCK_MEMALLOC sockets to use it as this socket is
914 * helping free memory
915 */
916 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
917 return -ENOMEM;
918
919 __sk_add_backlog(sk, skb);
920 sk->sk_backlog.len += skb->truesize;
921 return 0;
922}
923
924int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
925
926static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
927{
928 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
929 return __sk_backlog_rcv(sk, skb);
930
931 return sk->sk_backlog_rcv(sk, skb);
932}
933
934static inline void sk_incoming_cpu_update(struct sock *sk)
935{
936 int cpu = raw_smp_processor_id();
937
938 if (unlikely(sk->sk_incoming_cpu != cpu))
939 sk->sk_incoming_cpu = cpu;
940}
941
942static inline void sock_rps_record_flow_hash(__u32 hash)
943{
944#ifdef CONFIG_RPS
945 struct rps_sock_flow_table *sock_flow_table;
946
947 rcu_read_lock();
948 sock_flow_table = rcu_dereference(rps_sock_flow_table);
949 rps_record_sock_flow(sock_flow_table, hash);
950 rcu_read_unlock();
951#endif
952}
953
954static inline void sock_rps_record_flow(const struct sock *sk)
955{
956#ifdef CONFIG_RPS
957 if (static_key_false(&rfs_needed)) {
958 /* Reading sk->sk_rxhash might incur an expensive cache line
959 * miss.
960 *
961 * TCP_ESTABLISHED does cover almost all states where RFS
962 * might be useful, and is cheaper [1] than testing :
963 * IPv4: inet_sk(sk)->inet_daddr
964 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
965 * OR an additional socket flag
966 * [1] : sk_state and sk_prot are in the same cache line.
967 */
968 if (sk->sk_state == TCP_ESTABLISHED)
969 sock_rps_record_flow_hash(sk->sk_rxhash);
970 }
971#endif
972}
973
974static inline void sock_rps_save_rxhash(struct sock *sk,
975 const struct sk_buff *skb)
976{
977#ifdef CONFIG_RPS
978 if (unlikely(sk->sk_rxhash != skb->hash))
979 sk->sk_rxhash = skb->hash;
980#endif
981}
982
983static inline void sock_rps_reset_rxhash(struct sock *sk)
984{
985#ifdef CONFIG_RPS
986 sk->sk_rxhash = 0;
987#endif
988}
989
990#define sk_wait_event(__sk, __timeo, __condition, __wait) \
991 ({ int __rc; \
992 release_sock(__sk); \
993 __rc = __condition; \
994 if (!__rc) { \
995 *(__timeo) = wait_woken(__wait, \
996 TASK_INTERRUPTIBLE, \
997 *(__timeo)); \
998 } \
999 sched_annotate_sleep(); \
1000 lock_sock(__sk); \
1001 __rc = __condition; \
1002 __rc; \
1003 })
1004
1005int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1006int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1007void sk_stream_wait_close(struct sock *sk, long timeo_p);
1008int sk_stream_error(struct sock *sk, int flags, int err);
1009void sk_stream_kill_queues(struct sock *sk);
1010void sk_set_memalloc(struct sock *sk);
1011void sk_clear_memalloc(struct sock *sk);
1012
1013void __sk_flush_backlog(struct sock *sk);
1014
1015static inline bool sk_flush_backlog(struct sock *sk)
1016{
1017 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1018 __sk_flush_backlog(sk);
1019 return true;
1020 }
1021 return false;
1022}
1023
1024int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1025
1026struct request_sock_ops;
1027struct timewait_sock_ops;
1028struct inet_hashinfo;
1029struct raw_hashinfo;
1030struct smc_hashinfo;
1031struct module;
1032
1033/*
1034 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1035 * un-modified. Special care is taken when initializing object to zero.
1036 */
1037static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1038{
1039 if (offsetof(struct sock, sk_node.next) != 0)
1040 memset(sk, 0, offsetof(struct sock, sk_node.next));
1041 memset(&sk->sk_node.pprev, 0,
1042 size - offsetof(struct sock, sk_node.pprev));
1043}
1044
1045/* Networking protocol blocks we attach to sockets.
1046 * socket layer -> transport layer interface
1047 */
1048struct proto {
1049 void (*close)(struct sock *sk,
1050 long timeout);
1051 int (*pre_connect)(struct sock *sk,
1052 struct sockaddr *uaddr,
1053 int addr_len);
1054 int (*connect)(struct sock *sk,
1055 struct sockaddr *uaddr,
1056 int addr_len);
1057 int (*disconnect)(struct sock *sk, int flags);
1058
1059 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1060 bool kern);
1061
1062 int (*ioctl)(struct sock *sk, int cmd,
1063 unsigned long arg);
1064 int (*init)(struct sock *sk);
1065 void (*destroy)(struct sock *sk);
1066 void (*shutdown)(struct sock *sk, int how);
1067 int (*setsockopt)(struct sock *sk, int level,
1068 int optname, char __user *optval,
1069 unsigned int optlen);
1070 int (*getsockopt)(struct sock *sk, int level,
1071 int optname, char __user *optval,
1072 int __user *option);
1073 void (*keepalive)(struct sock *sk, int valbool);
1074#ifdef CONFIG_COMPAT
1075 int (*compat_setsockopt)(struct sock *sk,
1076 int level,
1077 int optname, char __user *optval,
1078 unsigned int optlen);
1079 int (*compat_getsockopt)(struct sock *sk,
1080 int level,
1081 int optname, char __user *optval,
1082 int __user *option);
1083 int (*compat_ioctl)(struct sock *sk,
1084 unsigned int cmd, unsigned long arg);
1085#endif
1086 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1087 size_t len);
1088 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1089 size_t len, int noblock, int flags,
1090 int *addr_len);
1091 int (*sendpage)(struct sock *sk, struct page *page,
1092 int offset, size_t size, int flags);
1093 int (*bind)(struct sock *sk,
1094 struct sockaddr *uaddr, int addr_len);
1095
1096 int (*backlog_rcv) (struct sock *sk,
1097 struct sk_buff *skb);
1098
1099 void (*release_cb)(struct sock *sk);
1100
1101 /* Keeping track of sk's, looking them up, and port selection methods. */
1102 int (*hash)(struct sock *sk);
1103 void (*unhash)(struct sock *sk);
1104 void (*rehash)(struct sock *sk);
1105 int (*get_port)(struct sock *sk, unsigned short snum);
1106
1107 /* Keeping track of sockets in use */
1108#ifdef CONFIG_PROC_FS
1109 unsigned int inuse_idx;
1110#endif
1111
1112 bool (*stream_memory_free)(const struct sock *sk);
1113 bool (*stream_memory_read)(const struct sock *sk);
1114 /* Memory pressure */
1115 void (*enter_memory_pressure)(struct sock *sk);
1116 void (*leave_memory_pressure)(struct sock *sk);
1117 atomic_long_t *memory_allocated; /* Current allocated memory. */
1118 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1119 /*
1120 * Pressure flag: try to collapse.
1121 * Technical note: it is used by multiple contexts non atomically.
1122 * All the __sk_mem_schedule() is of this nature: accounting
1123 * is strict, actions are advisory and have some latency.
1124 */
1125 unsigned long *memory_pressure;
1126 long *sysctl_mem;
1127
1128 int *sysctl_wmem;
1129 int *sysctl_rmem;
1130 u32 sysctl_wmem_offset;
1131 u32 sysctl_rmem_offset;
1132
1133 int max_header;
1134 bool no_autobind;
1135
1136 struct kmem_cache *slab;
1137 unsigned int obj_size;
1138 slab_flags_t slab_flags;
1139 unsigned int useroffset; /* Usercopy region offset */
1140 unsigned int usersize; /* Usercopy region size */
1141
1142 struct percpu_counter *orphan_count;
1143
1144 struct request_sock_ops *rsk_prot;
1145 struct timewait_sock_ops *twsk_prot;
1146
1147 union {
1148 struct inet_hashinfo *hashinfo;
1149 struct udp_table *udp_table;
1150 struct raw_hashinfo *raw_hash;
1151 struct smc_hashinfo *smc_hash;
1152 } h;
1153
1154 struct module *owner;
1155
1156 char name[32];
1157
1158 struct list_head node;
1159#ifdef SOCK_REFCNT_DEBUG
1160 atomic_t socks;
1161#endif
1162 int (*diag_destroy)(struct sock *sk, int err);
1163} __randomize_layout;
1164
1165int proto_register(struct proto *prot, int alloc_slab);
1166void proto_unregister(struct proto *prot);
1167int sock_load_diag_module(int family, int protocol);
1168
1169#ifdef SOCK_REFCNT_DEBUG
1170static inline void sk_refcnt_debug_inc(struct sock *sk)
1171{
1172 atomic_inc(&sk->sk_prot->socks);
1173}
1174
1175static inline void sk_refcnt_debug_dec(struct sock *sk)
1176{
1177 atomic_dec(&sk->sk_prot->socks);
1178 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1179 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1180}
1181
1182static inline void sk_refcnt_debug_release(const struct sock *sk)
1183{
1184 if (refcount_read(&sk->sk_refcnt) != 1)
1185 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1186 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1187}
1188#else /* SOCK_REFCNT_DEBUG */
1189#define sk_refcnt_debug_inc(sk) do { } while (0)
1190#define sk_refcnt_debug_dec(sk) do { } while (0)
1191#define sk_refcnt_debug_release(sk) do { } while (0)
1192#endif /* SOCK_REFCNT_DEBUG */
1193
1194static inline bool sk_stream_memory_free(const struct sock *sk)
1195{
1196 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1197 return false;
1198
1199 return sk->sk_prot->stream_memory_free ?
1200 sk->sk_prot->stream_memory_free(sk) : true;
1201}
1202
1203static inline bool sk_stream_is_writeable(const struct sock *sk)
1204{
1205 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1206 sk_stream_memory_free(sk);
1207}
1208
1209static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1210 struct cgroup *ancestor)
1211{
1212#ifdef CONFIG_SOCK_CGROUP_DATA
1213 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1214 ancestor);
1215#else
1216 return -ENOTSUPP;
1217#endif
1218}
1219
1220static inline bool sk_has_memory_pressure(const struct sock *sk)
1221{
1222 return sk->sk_prot->memory_pressure != NULL;
1223}
1224
1225static inline bool sk_under_memory_pressure(const struct sock *sk)
1226{
1227 if (!sk->sk_prot->memory_pressure)
1228 return false;
1229
1230 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1231 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1232 return true;
1233
1234 return !!*sk->sk_prot->memory_pressure;
1235}
1236
1237static inline long
1238sk_memory_allocated(const struct sock *sk)
1239{
1240 return atomic_long_read(sk->sk_prot->memory_allocated);
1241}
1242
1243static inline long
1244sk_memory_allocated_add(struct sock *sk, int amt)
1245{
1246 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1247}
1248
1249static inline void
1250sk_memory_allocated_sub(struct sock *sk, int amt)
1251{
1252 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1253}
1254
1255static inline void sk_sockets_allocated_dec(struct sock *sk)
1256{
1257 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1258}
1259
1260static inline void sk_sockets_allocated_inc(struct sock *sk)
1261{
1262 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1263}
1264
1265static inline int
1266sk_sockets_allocated_read_positive(struct sock *sk)
1267{
1268 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1269}
1270
1271static inline int
1272proto_sockets_allocated_sum_positive(struct proto *prot)
1273{
1274 return percpu_counter_sum_positive(prot->sockets_allocated);
1275}
1276
1277static inline long
1278proto_memory_allocated(struct proto *prot)
1279{
1280 return atomic_long_read(prot->memory_allocated);
1281}
1282
1283static inline bool
1284proto_memory_pressure(struct proto *prot)
1285{
1286 if (!prot->memory_pressure)
1287 return false;
1288 return !!*prot->memory_pressure;
1289}
1290
1291
1292#ifdef CONFIG_PROC_FS
1293/* Called with local bh disabled */
1294void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1295int sock_prot_inuse_get(struct net *net, struct proto *proto);
1296int sock_inuse_get(struct net *net);
1297#else
1298static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1299 int inc)
1300{
1301}
1302#endif
1303
1304
1305/* With per-bucket locks this operation is not-atomic, so that
1306 * this version is not worse.
1307 */
1308static inline int __sk_prot_rehash(struct sock *sk)
1309{
1310 sk->sk_prot->unhash(sk);
1311 return sk->sk_prot->hash(sk);
1312}
1313
1314/* About 10 seconds */
1315#define SOCK_DESTROY_TIME (10*HZ)
1316
1317/* Sockets 0-1023 can't be bound to unless you are superuser */
1318#define PROT_SOCK 1024
1319
1320#define SHUTDOWN_MASK 3
1321#define RCV_SHUTDOWN 1
1322#define SEND_SHUTDOWN 2
1323
1324#define SOCK_SNDBUF_LOCK 1
1325#define SOCK_RCVBUF_LOCK 2
1326#define SOCK_BINDADDR_LOCK 4
1327#define SOCK_BINDPORT_LOCK 8
1328
1329struct socket_alloc {
1330 struct socket socket;
1331 struct inode vfs_inode;
1332};
1333
1334static inline struct socket *SOCKET_I(struct inode *inode)
1335{
1336 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1337}
1338
1339static inline struct inode *SOCK_INODE(struct socket *socket)
1340{
1341 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1342}
1343
1344/*
1345 * Functions for memory accounting
1346 */
1347int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1348int __sk_mem_schedule(struct sock *sk, int size, int kind);
1349void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1350void __sk_mem_reclaim(struct sock *sk, int amount);
1351
1352/* We used to have PAGE_SIZE here, but systems with 64KB pages
1353 * do not necessarily have 16x time more memory than 4KB ones.
1354 */
1355#define SK_MEM_QUANTUM 4096
1356#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1357#define SK_MEM_SEND 0
1358#define SK_MEM_RECV 1
1359
1360/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1361static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1362{
1363 long val = sk->sk_prot->sysctl_mem[index];
1364
1365#if PAGE_SIZE > SK_MEM_QUANTUM
1366 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1367#elif PAGE_SIZE < SK_MEM_QUANTUM
1368 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1369#endif
1370 return val;
1371}
1372
1373static inline int sk_mem_pages(int amt)
1374{
1375 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1376}
1377
1378static inline bool sk_has_account(struct sock *sk)
1379{
1380 /* return true if protocol supports memory accounting */
1381 return !!sk->sk_prot->memory_allocated;
1382}
1383
1384static inline bool sk_wmem_schedule(struct sock *sk, int size)
1385{
1386 if (!sk_has_account(sk))
1387 return true;
1388 return size <= sk->sk_forward_alloc ||
1389 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1390}
1391
1392static inline bool
1393sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1394{
1395 if (!sk_has_account(sk))
1396 return true;
1397 return size<= sk->sk_forward_alloc ||
1398 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1399 skb_pfmemalloc(skb);
1400}
1401
1402static inline void sk_mem_reclaim(struct sock *sk)
1403{
1404 if (!sk_has_account(sk))
1405 return;
1406 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1407 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1408}
1409
1410static inline void sk_mem_reclaim_partial(struct sock *sk)
1411{
1412 if (!sk_has_account(sk))
1413 return;
1414 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1415 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1416}
1417
1418static inline void sk_mem_charge(struct sock *sk, int size)
1419{
1420 if (!sk_has_account(sk))
1421 return;
1422 sk->sk_forward_alloc -= size;
1423}
1424
1425static inline void sk_mem_uncharge(struct sock *sk, int size)
1426{
1427 if (!sk_has_account(sk))
1428 return;
1429 sk->sk_forward_alloc += size;
1430
1431 /* Avoid a possible overflow.
1432 * TCP send queues can make this happen, if sk_mem_reclaim()
1433 * is not called and more than 2 GBytes are released at once.
1434 *
1435 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1436 * no need to hold that much forward allocation anyway.
1437 */
1438 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1439 __sk_mem_reclaim(sk, 1 << 20);
1440}
1441
1442static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1443{
1444 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1445 sk->sk_wmem_queued -= skb->truesize;
1446 sk_mem_uncharge(sk, skb->truesize);
1447 __kfree_skb(skb);
1448}
1449
1450static inline void sock_release_ownership(struct sock *sk)
1451{
1452 if (sk->sk_lock.owned) {
1453 sk->sk_lock.owned = 0;
1454
1455 /* The sk_lock has mutex_unlock() semantics: */
1456 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1457 }
1458}
1459
1460/*
1461 * Macro so as to not evaluate some arguments when
1462 * lockdep is not enabled.
1463 *
1464 * Mark both the sk_lock and the sk_lock.slock as a
1465 * per-address-family lock class.
1466 */
1467#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1468do { \
1469 sk->sk_lock.owned = 0; \
1470 init_waitqueue_head(&sk->sk_lock.wq); \
1471 spin_lock_init(&(sk)->sk_lock.slock); \
1472 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1473 sizeof((sk)->sk_lock)); \
1474 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1475 (skey), (sname)); \
1476 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1477} while (0)
1478
1479#ifdef CONFIG_LOCKDEP
1480static inline bool lockdep_sock_is_held(const struct sock *sk)
1481{
1482 return lockdep_is_held(&sk->sk_lock) ||
1483 lockdep_is_held(&sk->sk_lock.slock);
1484}
1485#endif
1486
1487void lock_sock_nested(struct sock *sk, int subclass);
1488
1489static inline void lock_sock(struct sock *sk)
1490{
1491 lock_sock_nested(sk, 0);
1492}
1493
1494void release_sock(struct sock *sk);
1495
1496/* BH context may only use the following locking interface. */
1497#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1498#define bh_lock_sock_nested(__sk) \
1499 spin_lock_nested(&((__sk)->sk_lock.slock), \
1500 SINGLE_DEPTH_NESTING)
1501#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1502
1503bool lock_sock_fast(struct sock *sk);
1504/**
1505 * unlock_sock_fast - complement of lock_sock_fast
1506 * @sk: socket
1507 * @slow: slow mode
1508 *
1509 * fast unlock socket for user context.
1510 * If slow mode is on, we call regular release_sock()
1511 */
1512static inline void unlock_sock_fast(struct sock *sk, bool slow)
1513{
1514 if (slow)
1515 release_sock(sk);
1516 else
1517 spin_unlock_bh(&sk->sk_lock.slock);
1518}
1519
1520/* Used by processes to "lock" a socket state, so that
1521 * interrupts and bottom half handlers won't change it
1522 * from under us. It essentially blocks any incoming
1523 * packets, so that we won't get any new data or any
1524 * packets that change the state of the socket.
1525 *
1526 * While locked, BH processing will add new packets to
1527 * the backlog queue. This queue is processed by the
1528 * owner of the socket lock right before it is released.
1529 *
1530 * Since ~2.3.5 it is also exclusive sleep lock serializing
1531 * accesses from user process context.
1532 */
1533
1534static inline void sock_owned_by_me(const struct sock *sk)
1535{
1536#ifdef CONFIG_LOCKDEP
1537 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1538#endif
1539}
1540
1541static inline bool sock_owned_by_user(const struct sock *sk)
1542{
1543 sock_owned_by_me(sk);
1544 return sk->sk_lock.owned;
1545}
1546
1547static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1548{
1549 return sk->sk_lock.owned;
1550}
1551
1552/* no reclassification while locks are held */
1553static inline bool sock_allow_reclassification(const struct sock *csk)
1554{
1555 struct sock *sk = (struct sock *)csk;
1556
1557 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1558}
1559
1560struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1561 struct proto *prot, int kern);
1562void sk_free(struct sock *sk);
1563void sk_destruct(struct sock *sk);
1564struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1565void sk_free_unlock_clone(struct sock *sk);
1566
1567struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1568 gfp_t priority);
1569void __sock_wfree(struct sk_buff *skb);
1570void sock_wfree(struct sk_buff *skb);
1571struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1572 gfp_t priority);
1573void skb_orphan_partial(struct sk_buff *skb);
1574void sock_rfree(struct sk_buff *skb);
1575void sock_efree(struct sk_buff *skb);
1576#ifdef CONFIG_INET
1577void sock_edemux(struct sk_buff *skb);
1578#else
1579#define sock_edemux sock_efree
1580#endif
1581
1582int sock_setsockopt(struct socket *sock, int level, int op,
1583 char __user *optval, unsigned int optlen);
1584
1585int sock_getsockopt(struct socket *sock, int level, int op,
1586 char __user *optval, int __user *optlen);
1587struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1588 int noblock, int *errcode);
1589struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1590 unsigned long data_len, int noblock,
1591 int *errcode, int max_page_order);
1592void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1593void sock_kfree_s(struct sock *sk, void *mem, int size);
1594void sock_kzfree_s(struct sock *sk, void *mem, int size);
1595void sk_send_sigurg(struct sock *sk);
1596
1597struct sockcm_cookie {
1598 u64 transmit_time;
1599 u32 mark;
1600 u16 tsflags;
1601};
1602
1603static inline void sockcm_init(struct sockcm_cookie *sockc,
1604 const struct sock *sk)
1605{
1606 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1607}
1608
1609int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1610 struct sockcm_cookie *sockc);
1611int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1612 struct sockcm_cookie *sockc);
1613
1614/*
1615 * Functions to fill in entries in struct proto_ops when a protocol
1616 * does not implement a particular function.
1617 */
1618int sock_no_bind(struct socket *, struct sockaddr *, int);
1619int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1620int sock_no_socketpair(struct socket *, struct socket *);
1621int sock_no_accept(struct socket *, struct socket *, int, bool);
1622int sock_no_getname(struct socket *, struct sockaddr *, int);
1623int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1624int sock_no_listen(struct socket *, int);
1625int sock_no_shutdown(struct socket *, int);
1626int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1627int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1628int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1629int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1630int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1631int sock_no_mmap(struct file *file, struct socket *sock,
1632 struct vm_area_struct *vma);
1633ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1634 size_t size, int flags);
1635ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1636 int offset, size_t size, int flags);
1637
1638/*
1639 * Functions to fill in entries in struct proto_ops when a protocol
1640 * uses the inet style.
1641 */
1642int sock_common_getsockopt(struct socket *sock, int level, int optname,
1643 char __user *optval, int __user *optlen);
1644int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1645 int flags);
1646int sock_common_setsockopt(struct socket *sock, int level, int optname,
1647 char __user *optval, unsigned int optlen);
1648int compat_sock_common_getsockopt(struct socket *sock, int level,
1649 int optname, char __user *optval, int __user *optlen);
1650int compat_sock_common_setsockopt(struct socket *sock, int level,
1651 int optname, char __user *optval, unsigned int optlen);
1652
1653void sk_common_release(struct sock *sk);
1654
1655/*
1656 * Default socket callbacks and setup code
1657 */
1658
1659/* Initialise core socket variables */
1660void sock_init_data(struct socket *sock, struct sock *sk);
1661
1662/*
1663 * Socket reference counting postulates.
1664 *
1665 * * Each user of socket SHOULD hold a reference count.
1666 * * Each access point to socket (an hash table bucket, reference from a list,
1667 * running timer, skb in flight MUST hold a reference count.
1668 * * When reference count hits 0, it means it will never increase back.
1669 * * When reference count hits 0, it means that no references from
1670 * outside exist to this socket and current process on current CPU
1671 * is last user and may/should destroy this socket.
1672 * * sk_free is called from any context: process, BH, IRQ. When
1673 * it is called, socket has no references from outside -> sk_free
1674 * may release descendant resources allocated by the socket, but
1675 * to the time when it is called, socket is NOT referenced by any
1676 * hash tables, lists etc.
1677 * * Packets, delivered from outside (from network or from another process)
1678 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1679 * when they sit in queue. Otherwise, packets will leak to hole, when
1680 * socket is looked up by one cpu and unhasing is made by another CPU.
1681 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1682 * (leak to backlog). Packet socket does all the processing inside
1683 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1684 * use separate SMP lock, so that they are prone too.
1685 */
1686
1687/* Ungrab socket and destroy it, if it was the last reference. */
1688static inline void sock_put(struct sock *sk)
1689{
1690 if (refcount_dec_and_test(&sk->sk_refcnt))
1691 sk_free(sk);
1692}
1693/* Generic version of sock_put(), dealing with all sockets
1694 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1695 */
1696void sock_gen_put(struct sock *sk);
1697
1698int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1699 unsigned int trim_cap, bool refcounted);
1700static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1701 const int nested)
1702{
1703 return __sk_receive_skb(sk, skb, nested, 1, true);
1704}
1705
1706static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1707{
1708 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1709 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1710 return;
1711 sk->sk_tx_queue_mapping = tx_queue;
1712}
1713
1714#define NO_QUEUE_MAPPING USHRT_MAX
1715
1716static inline void sk_tx_queue_clear(struct sock *sk)
1717{
1718 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1719}
1720
1721static inline int sk_tx_queue_get(const struct sock *sk)
1722{
1723 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1724 return sk->sk_tx_queue_mapping;
1725
1726 return -1;
1727}
1728
1729static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1730{
1731#ifdef CONFIG_XPS
1732 if (skb_rx_queue_recorded(skb)) {
1733 u16 rx_queue = skb_get_rx_queue(skb);
1734
1735 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1736 return;
1737
1738 sk->sk_rx_queue_mapping = rx_queue;
1739 }
1740#endif
1741}
1742
1743static inline void sk_rx_queue_clear(struct sock *sk)
1744{
1745#ifdef CONFIG_XPS
1746 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1747#endif
1748}
1749
1750#ifdef CONFIG_XPS
1751static inline int sk_rx_queue_get(const struct sock *sk)
1752{
1753 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1754 return sk->sk_rx_queue_mapping;
1755
1756 return -1;
1757}
1758#endif
1759
1760static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1761{
1762 sk_tx_queue_clear(sk);
1763 sk->sk_socket = sock;
1764}
1765
1766static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1767{
1768 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1769 return &rcu_dereference_raw(sk->sk_wq)->wait;
1770}
1771/* Detach socket from process context.
1772 * Announce socket dead, detach it from wait queue and inode.
1773 * Note that parent inode held reference count on this struct sock,
1774 * we do not release it in this function, because protocol
1775 * probably wants some additional cleanups or even continuing
1776 * to work with this socket (TCP).
1777 */
1778static inline void sock_orphan(struct sock *sk)
1779{
1780 write_lock_bh(&sk->sk_callback_lock);
1781 sock_set_flag(sk, SOCK_DEAD);
1782 sk_set_socket(sk, NULL);
1783 sk->sk_wq = NULL;
1784 write_unlock_bh(&sk->sk_callback_lock);
1785}
1786
1787static inline void sock_graft(struct sock *sk, struct socket *parent)
1788{
1789 WARN_ON(parent->sk);
1790 write_lock_bh(&sk->sk_callback_lock);
1791 rcu_assign_pointer(sk->sk_wq, parent->wq);
1792 parent->sk = sk;
1793 sk_set_socket(sk, parent);
1794 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1795 security_sock_graft(sk, parent);
1796 write_unlock_bh(&sk->sk_callback_lock);
1797}
1798
1799kuid_t sock_i_uid(struct sock *sk);
1800unsigned long sock_i_ino(struct sock *sk);
1801
1802static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1803{
1804 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1805}
1806
1807static inline u32 net_tx_rndhash(void)
1808{
1809 u32 v = prandom_u32();
1810
1811 return v ?: 1;
1812}
1813
1814static inline void sk_set_txhash(struct sock *sk)
1815{
1816 sk->sk_txhash = net_tx_rndhash();
1817}
1818
1819static inline void sk_rethink_txhash(struct sock *sk)
1820{
1821 if (sk->sk_txhash)
1822 sk_set_txhash(sk);
1823}
1824
1825static inline struct dst_entry *
1826__sk_dst_get(struct sock *sk)
1827{
1828 return rcu_dereference_check(sk->sk_dst_cache,
1829 lockdep_sock_is_held(sk));
1830}
1831
1832static inline struct dst_entry *
1833sk_dst_get(struct sock *sk)
1834{
1835 struct dst_entry *dst;
1836
1837 rcu_read_lock();
1838 dst = rcu_dereference(sk->sk_dst_cache);
1839 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1840 dst = NULL;
1841 rcu_read_unlock();
1842 return dst;
1843}
1844
1845static inline void dst_negative_advice(struct sock *sk)
1846{
1847 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1848
1849 sk_rethink_txhash(sk);
1850
1851 if (dst && dst->ops->negative_advice) {
1852 ndst = dst->ops->negative_advice(dst);
1853
1854 if (ndst != dst) {
1855 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1856 sk_tx_queue_clear(sk);
1857 sk->sk_dst_pending_confirm = 0;
1858 }
1859 }
1860}
1861
1862static inline void
1863__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1864{
1865 struct dst_entry *old_dst;
1866
1867 sk_tx_queue_clear(sk);
1868 sk->sk_dst_pending_confirm = 0;
1869 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1870 lockdep_sock_is_held(sk));
1871 rcu_assign_pointer(sk->sk_dst_cache, dst);
1872 dst_release(old_dst);
1873}
1874
1875static inline void
1876sk_dst_set(struct sock *sk, struct dst_entry *dst)
1877{
1878 struct dst_entry *old_dst;
1879
1880 sk_tx_queue_clear(sk);
1881 sk->sk_dst_pending_confirm = 0;
1882 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1883 dst_release(old_dst);
1884}
1885
1886static inline void
1887__sk_dst_reset(struct sock *sk)
1888{
1889 __sk_dst_set(sk, NULL);
1890}
1891
1892static inline void
1893sk_dst_reset(struct sock *sk)
1894{
1895 sk_dst_set(sk, NULL);
1896}
1897
1898struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1899
1900struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1901
1902static inline void sk_dst_confirm(struct sock *sk)
1903{
1904 if (!sk->sk_dst_pending_confirm)
1905 sk->sk_dst_pending_confirm = 1;
1906}
1907
1908static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1909{
1910 if (skb_get_dst_pending_confirm(skb)) {
1911 struct sock *sk = skb->sk;
1912 unsigned long now = jiffies;
1913
1914 /* avoid dirtying neighbour */
1915 if (n->confirmed != now)
1916 n->confirmed = now;
1917 if (sk && sk->sk_dst_pending_confirm)
1918 sk->sk_dst_pending_confirm = 0;
1919 }
1920}
1921
1922bool sk_mc_loop(struct sock *sk);
1923
1924static inline bool sk_can_gso(const struct sock *sk)
1925{
1926 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1927}
1928
1929void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1930
1931static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1932{
1933 sk->sk_route_nocaps |= flags;
1934 sk->sk_route_caps &= ~flags;
1935}
1936
1937static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1938 struct iov_iter *from, char *to,
1939 int copy, int offset)
1940{
1941 if (skb->ip_summed == CHECKSUM_NONE) {
1942 __wsum csum = 0;
1943 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1944 return -EFAULT;
1945 skb->csum = csum_block_add(skb->csum, csum, offset);
1946 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1947 if (!copy_from_iter_full_nocache(to, copy, from))
1948 return -EFAULT;
1949 } else if (!copy_from_iter_full(to, copy, from))
1950 return -EFAULT;
1951
1952 return 0;
1953}
1954
1955static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1956 struct iov_iter *from, int copy)
1957{
1958 int err, offset = skb->len;
1959
1960 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1961 copy, offset);
1962 if (err)
1963 __skb_trim(skb, offset);
1964
1965 return err;
1966}
1967
1968static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1969 struct sk_buff *skb,
1970 struct page *page,
1971 int off, int copy)
1972{
1973 int err;
1974
1975 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1976 copy, skb->len);
1977 if (err)
1978 return err;
1979
1980 skb->len += copy;
1981 skb->data_len += copy;
1982 skb->truesize += copy;
1983 sk->sk_wmem_queued += copy;
1984 sk_mem_charge(sk, copy);
1985 return 0;
1986}
1987
1988/**
1989 * sk_wmem_alloc_get - returns write allocations
1990 * @sk: socket
1991 *
1992 * Returns sk_wmem_alloc minus initial offset of one
1993 */
1994static inline int sk_wmem_alloc_get(const struct sock *sk)
1995{
1996 return refcount_read(&sk->sk_wmem_alloc) - 1;
1997}
1998
1999/**
2000 * sk_rmem_alloc_get - returns read allocations
2001 * @sk: socket
2002 *
2003 * Returns sk_rmem_alloc
2004 */
2005static inline int sk_rmem_alloc_get(const struct sock *sk)
2006{
2007 return atomic_read(&sk->sk_rmem_alloc);
2008}
2009
2010/**
2011 * sk_has_allocations - check if allocations are outstanding
2012 * @sk: socket
2013 *
2014 * Returns true if socket has write or read allocations
2015 */
2016static inline bool sk_has_allocations(const struct sock *sk)
2017{
2018 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2019}
2020
2021/**
2022 * skwq_has_sleeper - check if there are any waiting processes
2023 * @wq: struct socket_wq
2024 *
2025 * Returns true if socket_wq has waiting processes
2026 *
2027 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2028 * barrier call. They were added due to the race found within the tcp code.
2029 *
2030 * Consider following tcp code paths::
2031 *
2032 * CPU1 CPU2
2033 * sys_select receive packet
2034 * ... ...
2035 * __add_wait_queue update tp->rcv_nxt
2036 * ... ...
2037 * tp->rcv_nxt check sock_def_readable
2038 * ... {
2039 * schedule rcu_read_lock();
2040 * wq = rcu_dereference(sk->sk_wq);
2041 * if (wq && waitqueue_active(&wq->wait))
2042 * wake_up_interruptible(&wq->wait)
2043 * ...
2044 * }
2045 *
2046 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2047 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2048 * could then endup calling schedule and sleep forever if there are no more
2049 * data on the socket.
2050 *
2051 */
2052static inline bool skwq_has_sleeper(struct socket_wq *wq)
2053{
2054 return wq && wq_has_sleeper(&wq->wait);
2055}
2056
2057/**
2058 * sock_poll_wait - place memory barrier behind the poll_wait call.
2059 * @filp: file
2060 * @p: poll_table
2061 *
2062 * See the comments in the wq_has_sleeper function.
2063 */
2064static inline void sock_poll_wait(struct file *filp, poll_table *p)
2065{
2066 struct socket *sock = filp->private_data;
2067
2068 if (!poll_does_not_wait(p)) {
2069 poll_wait(filp, &sock->wq->wait, p);
2070 /* We need to be sure we are in sync with the
2071 * socket flags modification.
2072 *
2073 * This memory barrier is paired in the wq_has_sleeper.
2074 */
2075 smp_mb();
2076 }
2077}
2078
2079static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2080{
2081 if (sk->sk_txhash) {
2082 skb->l4_hash = 1;
2083 skb->hash = sk->sk_txhash;
2084 }
2085}
2086
2087void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2088
2089/*
2090 * Queue a received datagram if it will fit. Stream and sequenced
2091 * protocols can't normally use this as they need to fit buffers in
2092 * and play with them.
2093 *
2094 * Inlined as it's very short and called for pretty much every
2095 * packet ever received.
2096 */
2097static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2098{
2099 skb_orphan(skb);
2100 skb->sk = sk;
2101 skb->destructor = sock_rfree;
2102 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2103 sk_mem_charge(sk, skb->truesize);
2104}
2105
2106void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2107 unsigned long expires);
2108
2109void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2110
2111int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2112 struct sk_buff *skb, unsigned int flags,
2113 void (*destructor)(struct sock *sk,
2114 struct sk_buff *skb));
2115int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2116int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2117
2118int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2119struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2120
2121/*
2122 * Recover an error report and clear atomically
2123 */
2124
2125static inline int sock_error(struct sock *sk)
2126{
2127 int err;
2128 if (likely(!sk->sk_err))
2129 return 0;
2130 err = xchg(&sk->sk_err, 0);
2131 return -err;
2132}
2133
2134static inline unsigned long sock_wspace(struct sock *sk)
2135{
2136 int amt = 0;
2137
2138 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2139 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2140 if (amt < 0)
2141 amt = 0;
2142 }
2143 return amt;
2144}
2145
2146/* Note:
2147 * We use sk->sk_wq_raw, from contexts knowing this
2148 * pointer is not NULL and cannot disappear/change.
2149 */
2150static inline void sk_set_bit(int nr, struct sock *sk)
2151{
2152 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2153 !sock_flag(sk, SOCK_FASYNC))
2154 return;
2155
2156 set_bit(nr, &sk->sk_wq_raw->flags);
2157}
2158
2159static inline void sk_clear_bit(int nr, struct sock *sk)
2160{
2161 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2162 !sock_flag(sk, SOCK_FASYNC))
2163 return;
2164
2165 clear_bit(nr, &sk->sk_wq_raw->flags);
2166}
2167
2168static inline void sk_wake_async(const struct sock *sk, int how, int band)
2169{
2170 if (sock_flag(sk, SOCK_FASYNC)) {
2171 rcu_read_lock();
2172 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2173 rcu_read_unlock();
2174 }
2175}
2176
2177/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2178 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2179 * Note: for send buffers, TCP works better if we can build two skbs at
2180 * minimum.
2181 */
2182#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2183
2184#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2185#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2186
2187static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2188{
2189 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2190 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2191 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2192 }
2193}
2194
2195struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2196 bool force_schedule);
2197
2198/**
2199 * sk_page_frag - return an appropriate page_frag
2200 * @sk: socket
2201 *
2202 * If socket allocation mode allows current thread to sleep, it means its
2203 * safe to use the per task page_frag instead of the per socket one.
2204 */
2205static inline struct page_frag *sk_page_frag(struct sock *sk)
2206{
2207 if (gfpflags_allow_blocking(sk->sk_allocation))
2208 return &current->task_frag;
2209
2210 return &sk->sk_frag;
2211}
2212
2213bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2214
2215int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2216 int sg_start, int *sg_curr, unsigned int *sg_size,
2217 int first_coalesce);
2218
2219/*
2220 * Default write policy as shown to user space via poll/select/SIGIO
2221 */
2222static inline bool sock_writeable(const struct sock *sk)
2223{
2224 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2225}
2226
2227static inline gfp_t gfp_any(void)
2228{
2229 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2230}
2231
2232static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2233{
2234 return noblock ? 0 : sk->sk_rcvtimeo;
2235}
2236
2237static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2238{
2239 return noblock ? 0 : sk->sk_sndtimeo;
2240}
2241
2242static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2243{
2244 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2245}
2246
2247/* Alas, with timeout socket operations are not restartable.
2248 * Compare this to poll().
2249 */
2250static inline int sock_intr_errno(long timeo)
2251{
2252 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2253}
2254
2255struct sock_skb_cb {
2256 u32 dropcount;
2257};
2258
2259/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2260 * using skb->cb[] would keep using it directly and utilize its
2261 * alignement guarantee.
2262 */
2263#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2264 sizeof(struct sock_skb_cb)))
2265
2266#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2267 SOCK_SKB_CB_OFFSET))
2268
2269#define sock_skb_cb_check_size(size) \
2270 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2271
2272static inline void
2273sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2274{
2275 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2276 atomic_read(&sk->sk_drops) : 0;
2277}
2278
2279static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2280{
2281 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2282
2283 atomic_add(segs, &sk->sk_drops);
2284}
2285
2286void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2287 struct sk_buff *skb);
2288void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2289 struct sk_buff *skb);
2290
2291static inline void
2292sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2293{
2294 ktime_t kt = skb->tstamp;
2295 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2296
2297 /*
2298 * generate control messages if
2299 * - receive time stamping in software requested
2300 * - software time stamp available and wanted
2301 * - hardware time stamps available and wanted
2302 */
2303 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2304 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2305 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2306 (hwtstamps->hwtstamp &&
2307 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2308 __sock_recv_timestamp(msg, sk, skb);
2309 else
2310 sk->sk_stamp = kt;
2311
2312 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2313 __sock_recv_wifi_status(msg, sk, skb);
2314}
2315
2316void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2317 struct sk_buff *skb);
2318
2319#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2320static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2321 struct sk_buff *skb)
2322{
2323#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2324 (1UL << SOCK_RCVTSTAMP))
2325#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2326 SOF_TIMESTAMPING_RAW_HARDWARE)
2327
2328 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2329 __sock_recv_ts_and_drops(msg, sk, skb);
2330 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2331 sk->sk_stamp = skb->tstamp;
2332 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2333 sk->sk_stamp = 0;
2334}
2335
2336void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2337
2338/**
2339 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2340 * @sk: socket sending this packet
2341 * @tsflags: timestamping flags to use
2342 * @tx_flags: completed with instructions for time stamping
2343 *
2344 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2345 */
2346static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2347 __u8 *tx_flags)
2348{
2349 if (unlikely(tsflags))
2350 __sock_tx_timestamp(tsflags, tx_flags);
2351 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2352 *tx_flags |= SKBTX_WIFI_STATUS;
2353}
2354
2355/**
2356 * sk_eat_skb - Release a skb if it is no longer needed
2357 * @sk: socket to eat this skb from
2358 * @skb: socket buffer to eat
2359 *
2360 * This routine must be called with interrupts disabled or with the socket
2361 * locked so that the sk_buff queue operation is ok.
2362*/
2363static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2364{
2365 __skb_unlink(skb, &sk->sk_receive_queue);
2366 __kfree_skb(skb);
2367}
2368
2369static inline
2370struct net *sock_net(const struct sock *sk)
2371{
2372 return read_pnet(&sk->sk_net);
2373}
2374
2375static inline
2376void sock_net_set(struct sock *sk, struct net *net)
2377{
2378 write_pnet(&sk->sk_net, net);
2379}
2380
2381static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2382{
2383 if (skb->sk) {
2384 struct sock *sk = skb->sk;
2385
2386 skb->destructor = NULL;
2387 skb->sk = NULL;
2388 return sk;
2389 }
2390 return NULL;
2391}
2392
2393/* This helper checks if a socket is a full socket,
2394 * ie _not_ a timewait or request socket.
2395 */
2396static inline bool sk_fullsock(const struct sock *sk)
2397{
2398 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2399}
2400
2401/* Checks if this SKB belongs to an HW offloaded socket
2402 * and whether any SW fallbacks are required based on dev.
2403 */
2404static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2405 struct net_device *dev)
2406{
2407#ifdef CONFIG_SOCK_VALIDATE_XMIT
2408 struct sock *sk = skb->sk;
2409
2410 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb)
2411 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2412#endif
2413
2414 return skb;
2415}
2416
2417/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2418 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2419 */
2420static inline bool sk_listener(const struct sock *sk)
2421{
2422 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2423}
2424
2425void sock_enable_timestamp(struct sock *sk, int flag);
2426int sock_get_timestamp(struct sock *, struct timeval __user *);
2427int sock_get_timestampns(struct sock *, struct timespec __user *);
2428int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2429 int type);
2430
2431bool sk_ns_capable(const struct sock *sk,
2432 struct user_namespace *user_ns, int cap);
2433bool sk_capable(const struct sock *sk, int cap);
2434bool sk_net_capable(const struct sock *sk, int cap);
2435
2436void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2437
2438/* Take into consideration the size of the struct sk_buff overhead in the
2439 * determination of these values, since that is non-constant across
2440 * platforms. This makes socket queueing behavior and performance
2441 * not depend upon such differences.
2442 */
2443#define _SK_MEM_PACKETS 256
2444#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2445#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2446#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2447
2448extern __u32 sysctl_wmem_max;
2449extern __u32 sysctl_rmem_max;
2450
2451extern int sysctl_tstamp_allow_data;
2452extern int sysctl_optmem_max;
2453
2454extern __u32 sysctl_wmem_default;
2455extern __u32 sysctl_rmem_default;
2456
2457static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2458{
2459 /* Does this proto have per netns sysctl_wmem ? */
2460 if (proto->sysctl_wmem_offset)
2461 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2462
2463 return *proto->sysctl_wmem;
2464}
2465
2466static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2467{
2468 /* Does this proto have per netns sysctl_rmem ? */
2469 if (proto->sysctl_rmem_offset)
2470 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2471
2472 return *proto->sysctl_rmem;
2473}
2474
2475/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2476 * Some wifi drivers need to tweak it to get more chunks.
2477 * They can use this helper from their ndo_start_xmit()
2478 */
2479static inline void sk_pacing_shift_update(struct sock *sk, int val)
2480{
2481 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2482 return;
2483 sk->sk_pacing_shift = val;
2484}
2485
2486/* if a socket is bound to a device, check that the given device
2487 * index is either the same or that the socket is bound to an L3
2488 * master device and the given device index is also enslaved to
2489 * that L3 master
2490 */
2491static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2492{
2493 int mdif;
2494
2495 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2496 return true;
2497
2498 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2499 if (mdif && mdif == sk->sk_bound_dev_if)
2500 return true;
2501
2502 return false;
2503}
2504
2505#endif /* _SOCK_H */
2506