projects/linux·blob master · refs

359 lines · 11.3 KiB

1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * NET Generic infrastructure for INET connection oriented protocols.
4 *
5 * Definitions for inet_connection_sock
6 *
7 * Authors: Many people, see the TCP sources
8 *
9 * From code originally in TCP
10 */
11#ifndef _INET_CONNECTION_SOCK_H
12#define _INET_CONNECTION_SOCK_H
13
14#include <linux/compiler.h>
15#include <linux/string.h>
16#include <linux/timer.h>
17#include <linux/poll.h>
18#include <linux/kernel.h>
19#include <linux/sockptr.h>
20
21#include <net/inet_sock.h>
22#include <net/request_sock.h>
23
24/* Cancel timers, when they are not required. */
25#undef INET_CSK_CLEAR_TIMERS
26
27struct inet_bind_bucket;
28struct inet_bind2_bucket;
29struct tcp_congestion_ops;
30
31/*
32 * Pointers to address related TCP functions
33 * (i.e. things that depend on the address family)
34 */
35struct inet_connection_sock_af_ops {
36 int (*queue_xmit)(struct sock *sk, struct sk_buff *skb, struct flowi *fl);
37 u16 net_header_len;
38 int (*rebuild_header)(struct sock *sk);
39 void (*sk_rx_dst_set)(struct sock *sk, const struct sk_buff *skb);
40 int (*conn_request)(struct sock *sk, struct sk_buff *skb);
41 struct sock *(*syn_recv_sock)(const struct sock *sk, struct sk_buff *skb,
42 struct request_sock *req,
43 struct dst_entry *dst,
44 struct request_sock *req_unhash,
45 bool *own_req,
46 void (*opt_child_init)(struct sock *newsk,
47 const struct sock *sk));
48 int (*setsockopt)(struct sock *sk, int level, int optname,
49 sockptr_t optval, unsigned int optlen);
50 int (*getsockopt)(struct sock *sk, int level, int optname,
51 char __user *optval, int __user *optlen);
52 void (*mtu_reduced)(struct sock *sk);
53};
54
55/** inet_connection_sock - INET connection oriented sock
56 *
57 * @icsk_accept_queue: FIFO of established children
58 * @icsk_bind_hash: Bind node
59 * @icsk_bind2_hash: Bind node in the bhash2 table
60 * @icsk_delack_timer: Delayed ACK timer
61 * @icsk_keepalive_timer: Keepalive timer
62 * @mptcp_tout_timer: mptcp timer
63 * @icsk_rto: Retransmit timeout
64 * @icsk_pmtu_cookie Last pmtu seen by socket
65 * @icsk_ca_ops Pluggable congestion control hook
66 * @icsk_af_ops Operations which are AF_INET{4,6} specific
67 * @icsk_ulp_ops Pluggable ULP control hook
68 * @icsk_ulp_data ULP private data
69 * @icsk_ca_state: Congestion control state
70 * @icsk_retransmits: Number of unrecovered [RTO] timeouts
71 * @icsk_pending: Scheduled timer event
72 * @icsk_backoff: Backoff
73 * @icsk_syn_retries: Number of allowed SYN (or equivalent) retries
74 * @icsk_probes_out: unanswered 0 window probes
75 * @icsk_ext_hdr_len: Network protocol overhead (IP/IPv6 options)
76 * @icsk_ack: Delayed ACK control data
77 * @icsk_mtup; MTU probing control data
78 * @icsk_probes_tstamp: Probe timestamp (cleared by non-zero window ack)
79 * @icsk_user_timeout: TCP_USER_TIMEOUT value
80 */
81struct inet_connection_sock {
82 /* inet_sock has to be the first member! */
83 struct inet_sock icsk_inet;
84 struct request_sock_queue icsk_accept_queue;
85 struct inet_bind_bucket *icsk_bind_hash;
86 struct inet_bind2_bucket *icsk_bind2_hash;
87 struct timer_list icsk_delack_timer;
88 union {
89 struct timer_list icsk_keepalive_timer;
90 struct timer_list mptcp_tout_timer;
91 };
92 __u32 icsk_rto;
93 __u32 icsk_rto_min;
94 u32 icsk_rto_max;
95 __u32 icsk_delack_max;
96 __u32 icsk_pmtu_cookie;
97 const struct tcp_congestion_ops *icsk_ca_ops;
98 const struct inet_connection_sock_af_ops *icsk_af_ops;
99 const struct tcp_ulp_ops *icsk_ulp_ops;
100 void __rcu *icsk_ulp_data;
101 unsigned int (*icsk_sync_mss)(struct sock *sk, u32 pmtu);
102 __u8 icsk_ca_state:5,
103 icsk_ca_initialized:1,
104 icsk_ca_setsockopt:1,
105 icsk_ca_dst_locked:1;
106 __u8 icsk_retransmits;
107 __u8 icsk_pending;
108 __u8 icsk_backoff;
109 __u8 icsk_syn_retries;
110 __u8 icsk_probes_out;
111 __u16 icsk_ext_hdr_len;
112 struct {
113 __u8 pending; /* ACK is pending */
114 __u8 quick; /* Scheduled number of quick acks */
115 __u8 pingpong; /* The session is interactive */
116 __u8 retry; /* Number of attempts */
117 #define ATO_BITS 8
118 __u32 ato:ATO_BITS, /* Predicted tick of soft clock */
119 lrcv_flowlabel:20, /* last received ipv6 flowlabel */
120 dst_quick_ack:1, /* cache dst RTAX_QUICKACK */
121 unused:3;
122 __u32 lrcvtime; /* timestamp of last received data packet */
123 __u16 last_seg_size; /* Size of last incoming segment */
124 __u16 rcv_mss; /* MSS used for delayed ACK decisions */
125 } icsk_ack;
126 struct {
127 /* Range of MTUs to search */
128 int search_high;
129 int search_low;
130
131 /* Information on the current probe. */
132 u32 probe_size:31,
133 /* Is the MTUP feature enabled for this connection? */
134 enabled:1;
135
136 u32 probe_timestamp;
137 } icsk_mtup;
138 u32 icsk_probes_tstamp;
139 u32 icsk_user_timeout;
140
141 u64 icsk_ca_priv[104 / sizeof(u64)];
142#define ICSK_CA_PRIV_SIZE sizeof_field(struct inet_connection_sock, icsk_ca_priv)
143};
144
145#define ICSK_TIME_RETRANS 1 /* Retransmit timer */
146#define ICSK_TIME_DACK 2 /* Delayed ack timer */
147#define ICSK_TIME_PROBE0 3 /* Zero window probe timer */
148#define ICSK_TIME_LOSS_PROBE 5 /* Tail loss probe timer */
149#define ICSK_TIME_REO_TIMEOUT 6 /* Reordering timer */
150
151#define inet_csk(ptr) container_of_const(ptr, struct inet_connection_sock, icsk_inet.sk)
152
153static inline void *inet_csk_ca(const struct sock *sk)
154{
155 return (void *)inet_csk(sk)->icsk_ca_priv;
156}
157
158struct sock *inet_csk_clone_lock(const struct sock *sk,
159 const struct request_sock *req,
160 const gfp_t priority);
161
162enum inet_csk_ack_state_t {
163 ICSK_ACK_SCHED = 1,
164 ICSK_ACK_TIMER = 2,
165 ICSK_ACK_PUSHED = 4,
166 ICSK_ACK_PUSHED2 = 8,
167 ICSK_ACK_NOW = 16, /* Send the next ACK immediately (once) */
168 ICSK_ACK_NOMEM = 32,
169};
170
171void inet_csk_init_xmit_timers(struct sock *sk,
172 void (*retransmit_handler)(struct timer_list *),
173 void (*delack_handler)(struct timer_list *),
174 void (*keepalive_handler)(struct timer_list *));
175void inet_csk_clear_xmit_timers(struct sock *sk);
176void inet_csk_clear_xmit_timers_sync(struct sock *sk);
177
178static inline void inet_csk_schedule_ack(struct sock *sk)
179{
180 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_SCHED;
181}
182
183static inline int inet_csk_ack_scheduled(const struct sock *sk)
184{
185 return inet_csk(sk)->icsk_ack.pending & ICSK_ACK_SCHED;
186}
187
188static inline void inet_csk_delack_init(struct sock *sk)
189{
190 memset(&inet_csk(sk)->icsk_ack, 0, sizeof(inet_csk(sk)->icsk_ack));
191}
192
193static inline unsigned long tcp_timeout_expires(const struct sock *sk)
194{
195 return READ_ONCE(sk->tcp_retransmit_timer.expires);
196}
197
198static inline unsigned long
199icsk_delack_timeout(const struct inet_connection_sock *icsk)
200{
201 return READ_ONCE(icsk->icsk_delack_timer.expires);
202}
203
204static inline void inet_csk_clear_xmit_timer(struct sock *sk, const int what)
205{
206 struct inet_connection_sock *icsk = inet_csk(sk);
207
208 if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
209 smp_store_release(&icsk->icsk_pending, 0);
210#ifdef INET_CSK_CLEAR_TIMERS
211 sk_stop_timer(sk, &sk->tcp_retransmit_timer);
212#endif
213 } else if (what == ICSK_TIME_DACK) {
214 smp_store_release(&icsk->icsk_ack.pending, 0);
215 icsk->icsk_ack.retry = 0;
216#ifdef INET_CSK_CLEAR_TIMERS
217 sk_stop_timer(sk, &icsk->icsk_delack_timer);
218#endif
219 } else {
220 pr_debug("inet_csk BUG: unknown timer value\n");
221 }
222}
223
224/*
225 * Reset the retransmission timer
226 */
227static inline void inet_csk_reset_xmit_timer(struct sock *sk, const int what,
228 unsigned long when,
229 const unsigned long max_when)
230{
231 struct inet_connection_sock *icsk = inet_csk(sk);
232
233 if (when > max_when) {
234 pr_debug("reset_xmit_timer: sk=%p %d when=0x%lx, caller=%p\n",
235 sk, what, when, (void *)_THIS_IP_);
236 when = max_when;
237 }
238
239 when += jiffies;
240 if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0 ||
241 what == ICSK_TIME_LOSS_PROBE || what == ICSK_TIME_REO_TIMEOUT) {
242 smp_store_release(&icsk->icsk_pending, what);
243 sk_reset_timer(sk, &sk->tcp_retransmit_timer, when);
244 } else if (what == ICSK_TIME_DACK) {
245 smp_store_release(&icsk->icsk_ack.pending,
246 icsk->icsk_ack.pending | ICSK_ACK_TIMER);
247 sk_reset_timer(sk, &icsk->icsk_delack_timer, when);
248 } else {
249 pr_debug("inet_csk BUG: unknown timer value\n");
250 }
251}
252
253static inline unsigned long
254inet_csk_rto_backoff(const struct inet_connection_sock *icsk,
255 unsigned long max_when)
256{
257 u64 when = (u64)icsk->icsk_rto << icsk->icsk_backoff;
258
259 return (unsigned long)min_t(u64, when, max_when);
260}
261
262struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg);
263
264int inet_csk_get_port(struct sock *sk, unsigned short snum);
265
266struct dst_entry *inet_csk_route_req(const struct sock *sk, struct flowi4 *fl4,
267 const struct request_sock *req);
268struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
269 struct sock *newsk,
270 const struct request_sock *req);
271
272struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
273 struct request_sock *req,
274 struct sock *child);
275bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req);
276struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
277 struct request_sock *req,
278 bool own_req);
279
280static inline void inet_csk_reqsk_queue_added(struct sock *sk)
281{
282 reqsk_queue_added(&inet_csk(sk)->icsk_accept_queue);
283}
284
285static inline int inet_csk_reqsk_queue_len(const struct sock *sk)
286{
287 return reqsk_queue_len(&inet_csk(sk)->icsk_accept_queue);
288}
289
290static inline int inet_csk_reqsk_queue_is_full(const struct sock *sk)
291{
292 return inet_csk_reqsk_queue_len(sk) > READ_ONCE(sk->sk_max_ack_backlog);
293}
294
295bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req);
296void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req);
297
298void inet_csk_destroy_sock(struct sock *sk);
299void inet_csk_prepare_for_destroy_sock(struct sock *sk);
300void inet_csk_prepare_forced_close(struct sock *sk);
301
302/*
303 * LISTEN is a special case for poll..
304 */
305static inline __poll_t inet_csk_listen_poll(const struct sock *sk)
306{
307 return !reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue) ?
308 (EPOLLIN | EPOLLRDNORM) : 0;
309}
310
311int inet_csk_listen_start(struct sock *sk);
312void inet_csk_listen_stop(struct sock *sk);
313
314/* update the fast reuse flag when adding a socket */
315void inet_csk_update_fastreuse(const struct sock *sk,
316 struct inet_bind_bucket *tb,
317 struct inet_bind2_bucket *tb2);
318
319struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu);
320
321static inline void inet_csk_enter_pingpong_mode(struct sock *sk)
322{
323 inet_csk(sk)->icsk_ack.pingpong =
324 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
325}
326
327static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
328{
329 inet_csk(sk)->icsk_ack.pingpong = 0;
330}
331
332static inline bool inet_csk_in_pingpong_mode(struct sock *sk)
333{
334 return inet_csk(sk)->icsk_ack.pingpong >=
335 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
336}
337
338static inline void inet_csk_inc_pingpong_cnt(struct sock *sk)
339{
340 struct inet_connection_sock *icsk = inet_csk(sk);
341
342 if (icsk->icsk_ack.pingpong < U8_MAX)
343 icsk->icsk_ack.pingpong++;
344}
345
346static inline bool inet_csk_has_ulp(const struct sock *sk)
347{
348 return inet_test_bit(IS_ICSK, sk) && !!inet_csk(sk)->icsk_ulp_ops;
349}
350
351static inline void inet_init_csk_locks(struct sock *sk)
352{
353 struct inet_connection_sock *icsk = inet_csk(sk);
354
355 spin_lock_init(&icsk->icsk_accept_queue.rskq_lock);
356 spin_lock_init(&icsk->icsk_accept_queue.fastopenq.lock);
357}
358
359#endif /* _INET_CONNECTION_SOCK_H */