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