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
27 struct inet_bind_bucket;
28 struct inet_bind2_bucket;
29 struct tcp_congestion_ops;
30
31 /*
32 * Pointers to address related TCP functions
33 * (i.e. things that depend on the address family)
34 */
35 struct inet_connection_sock_af_ops {
36 int (*queue_xmit)(struct sock *sk, struct sk_buff *skb, struct flowi *fl);
37 void (*send_check)(struct sock *sk, struct sk_buff *skb);
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 u16 net_header_len;
47 int (*setsockopt)(struct sock *sk, int level, int optname,
48 sockptr_t optval, unsigned int optlen);
49 int (*getsockopt)(struct sock *sk, int level, int optname,
50 char __user *optval, int __user *optlen);
51 void (*mtu_reduced)(struct sock *sk);
52 };
53
54 /** inet_connection_sock - INET connection oriented sock
55 *
56 * @icsk_accept_queue: FIFO of established children
57 * @icsk_bind_hash: Bind node
58 * @icsk_bind2_hash: Bind node in the bhash2 table
59 * @icsk_retransmit_timer: Resend (no ack)
60 * @icsk_rto: Retransmit timeout
61 * @icsk_pmtu_cookie Last pmtu seen by socket
62 * @icsk_ca_ops Pluggable congestion control hook
63 * @icsk_af_ops Operations which are AF_INET{4,6} specific
64 * @icsk_ulp_ops Pluggable ULP control hook
65 * @icsk_ulp_data ULP private data
66 * @icsk_ca_state: Congestion control state
67 * @icsk_retransmits: Number of unrecovered [RTO] timeouts
68 * @icsk_pending: Scheduled timer event
69 * @icsk_backoff: Backoff
70 * @icsk_syn_retries: Number of allowed SYN (or equivalent) retries
71 * @icsk_probes_out: unanswered 0 window probes
72 * @icsk_ext_hdr_len: Network protocol overhead (IP/IPv6 options)
73 * @icsk_ack: Delayed ACK control data
74 * @icsk_mtup; MTU probing control data
75 * @icsk_probes_tstamp: Probe timestamp (cleared by non-zero window ack)
76 * @icsk_user_timeout: TCP_USER_TIMEOUT value
77 */
78 struct inet_connection_sock {
79 /* inet_sock has to be the first member! */
80 struct inet_sock icsk_inet;
81 struct request_sock_queue icsk_accept_queue;
82 struct inet_bind_bucket *icsk_bind_hash;
83 struct inet_bind2_bucket *icsk_bind2_hash;
84 struct timer_list icsk_retransmit_timer;
85 struct timer_list icsk_delack_timer;
86 __u32 icsk_rto;
87 __u32 icsk_rto_min;
88 u32 icsk_rto_max;
89 __u32 icsk_delack_max;
90 __u32 icsk_pmtu_cookie;
91 const struct tcp_congestion_ops *icsk_ca_ops;
92 const struct inet_connection_sock_af_ops *icsk_af_ops;
93 const struct tcp_ulp_ops *icsk_ulp_ops;
94 void __rcu *icsk_ulp_data;
95 unsigned int (*icsk_sync_mss)(struct sock *sk, u32 pmtu);
96 __u8 icsk_ca_state:5,
97 icsk_ca_initialized:1,
98 icsk_ca_setsockopt:1,
99 icsk_ca_dst_locked:1;
100 __u8 icsk_retransmits;
101 __u8 icsk_pending;
102 __u8 icsk_backoff;
103 __u8 icsk_syn_retries;
104 __u8 icsk_probes_out;
105 __u16 icsk_ext_hdr_len;
106 struct {
107 __u8 pending; /* ACK is pending */
108 __u8 quick; /* Scheduled number of quick acks */
109 __u8 pingpong; /* The session is interactive */
110 __u8 retry; /* Number of attempts */
111 #define ATO_BITS 8
112 __u32 ato:ATO_BITS, /* Predicted tick of soft clock */
113 lrcv_flowlabel:20, /* last received ipv6 flowlabel */
114 dst_quick_ack:1, /* cache dst RTAX_QUICKACK */
115 unused:3;
116 __u32 lrcvtime; /* timestamp of last received data packet */
117 __u16 last_seg_size; /* Size of last incoming segment */
118 __u16 rcv_mss; /* MSS used for delayed ACK decisions */
119 } icsk_ack;
120 struct {
121 /* Range of MTUs to search */
122 int search_high;
123 int search_low;
124
125 /* Information on the current probe. */
126 u32 probe_size:31,
127 /* Is the MTUP feature enabled for this connection? */
128 enabled:1;
129
130 u32 probe_timestamp;
131 } icsk_mtup;
132 u32 icsk_probes_tstamp;
133 u32 icsk_user_timeout;
134
135 u64 icsk_ca_priv[104 / sizeof(u64)];
136 #define ICSK_CA_PRIV_SIZE sizeof_field(struct inet_connection_sock, icsk_ca_priv)
137 };
138
139 #define ICSK_TIME_RETRANS 1 /* Retransmit timer */
140 #define ICSK_TIME_DACK 2 /* Delayed ack timer */
141 #define ICSK_TIME_PROBE0 3 /* Zero window probe timer */
142 #define ICSK_TIME_LOSS_PROBE 5 /* Tail loss probe timer */
143 #define ICSK_TIME_REO_TIMEOUT 6 /* Reordering timer */
144
145 #define inet_csk(ptr) container_of_const(ptr, struct inet_connection_sock, icsk_inet.sk)
146
inet_csk_ca(const struct sock * sk)147 static inline void *inet_csk_ca(const struct sock *sk)
148 {
149 return (void *)inet_csk(sk)->icsk_ca_priv;
150 }
151
152 struct sock *inet_csk_clone_lock(const struct sock *sk,
153 const struct request_sock *req,
154 const gfp_t priority);
155
156 enum inet_csk_ack_state_t {
157 ICSK_ACK_SCHED = 1,
158 ICSK_ACK_TIMER = 2,
159 ICSK_ACK_PUSHED = 4,
160 ICSK_ACK_PUSHED2 = 8,
161 ICSK_ACK_NOW = 16, /* Send the next ACK immediately (once) */
162 ICSK_ACK_NOMEM = 32,
163 };
164
165 void inet_csk_init_xmit_timers(struct sock *sk,
166 void (*retransmit_handler)(struct timer_list *),
167 void (*delack_handler)(struct timer_list *),
168 void (*keepalive_handler)(struct timer_list *));
169 void inet_csk_clear_xmit_timers(struct sock *sk);
170 void inet_csk_clear_xmit_timers_sync(struct sock *sk);
171
inet_csk_schedule_ack(struct sock * sk)172 static inline void inet_csk_schedule_ack(struct sock *sk)
173 {
174 inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_SCHED;
175 }
176
inet_csk_ack_scheduled(const struct sock * sk)177 static inline int inet_csk_ack_scheduled(const struct sock *sk)
178 {
179 return inet_csk(sk)->icsk_ack.pending & ICSK_ACK_SCHED;
180 }
181
inet_csk_delack_init(struct sock * sk)182 static inline void inet_csk_delack_init(struct sock *sk)
183 {
184 memset(&inet_csk(sk)->icsk_ack, 0, sizeof(inet_csk(sk)->icsk_ack));
185 }
186
187 static inline unsigned long
icsk_timeout(const struct inet_connection_sock * icsk)188 icsk_timeout(const struct inet_connection_sock *icsk)
189 {
190 return READ_ONCE(icsk->icsk_retransmit_timer.expires);
191 }
192
193 static inline unsigned long
icsk_delack_timeout(const struct inet_connection_sock * icsk)194 icsk_delack_timeout(const struct inet_connection_sock *icsk)
195 {
196 return READ_ONCE(icsk->icsk_delack_timer.expires);
197 }
198
inet_csk_clear_xmit_timer(struct sock * sk,const int what)199 static inline void inet_csk_clear_xmit_timer(struct sock *sk, const int what)
200 {
201 struct inet_connection_sock *icsk = inet_csk(sk);
202
203 if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
204 smp_store_release(&icsk->icsk_pending, 0);
205 #ifdef INET_CSK_CLEAR_TIMERS
206 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
207 #endif
208 } else if (what == ICSK_TIME_DACK) {
209 smp_store_release(&icsk->icsk_ack.pending, 0);
210 icsk->icsk_ack.retry = 0;
211 #ifdef INET_CSK_CLEAR_TIMERS
212 sk_stop_timer(sk, &icsk->icsk_delack_timer);
213 #endif
214 } else {
215 pr_debug("inet_csk BUG: unknown timer value\n");
216 }
217 }
218
219 /*
220 * Reset the retransmission timer
221 */
inet_csk_reset_xmit_timer(struct sock * sk,const int what,unsigned long when,const unsigned long max_when)222 static inline void inet_csk_reset_xmit_timer(struct sock *sk, const int what,
223 unsigned long when,
224 const unsigned long max_when)
225 {
226 struct inet_connection_sock *icsk = inet_csk(sk);
227
228 if (when > max_when) {
229 pr_debug("reset_xmit_timer: sk=%p %d when=0x%lx, caller=%p\n",
230 sk, what, when, (void *)_THIS_IP_);
231 when = max_when;
232 }
233
234 when += jiffies;
235 if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0 ||
236 what == ICSK_TIME_LOSS_PROBE || what == ICSK_TIME_REO_TIMEOUT) {
237 smp_store_release(&icsk->icsk_pending, what);
238 sk_reset_timer(sk, &icsk->icsk_retransmit_timer, when);
239 } else if (what == ICSK_TIME_DACK) {
240 smp_store_release(&icsk->icsk_ack.pending,
241 icsk->icsk_ack.pending | ICSK_ACK_TIMER);
242 sk_reset_timer(sk, &icsk->icsk_delack_timer, when);
243 } else {
244 pr_debug("inet_csk BUG: unknown timer value\n");
245 }
246 }
247
248 static inline unsigned long
inet_csk_rto_backoff(const struct inet_connection_sock * icsk,unsigned long max_when)249 inet_csk_rto_backoff(const struct inet_connection_sock *icsk,
250 unsigned long max_when)
251 {
252 u64 when = (u64)icsk->icsk_rto << icsk->icsk_backoff;
253
254 return (unsigned long)min_t(u64, when, max_when);
255 }
256
257 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg);
258
259 int inet_csk_get_port(struct sock *sk, unsigned short snum);
260
261 struct dst_entry *inet_csk_route_req(const struct sock *sk, struct flowi4 *fl4,
262 const struct request_sock *req);
263 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
264 struct sock *newsk,
265 const struct request_sock *req);
266
267 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
268 struct request_sock *req,
269 struct sock *child);
270 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
271 unsigned long timeout);
272 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
273 struct request_sock *req,
274 bool own_req);
275
inet_csk_reqsk_queue_added(struct sock * sk)276 static inline void inet_csk_reqsk_queue_added(struct sock *sk)
277 {
278 reqsk_queue_added(&inet_csk(sk)->icsk_accept_queue);
279 }
280
inet_csk_reqsk_queue_len(const struct sock * sk)281 static inline int inet_csk_reqsk_queue_len(const struct sock *sk)
282 {
283 return reqsk_queue_len(&inet_csk(sk)->icsk_accept_queue);
284 }
285
inet_csk_reqsk_queue_is_full(const struct sock * sk)286 static inline int inet_csk_reqsk_queue_is_full(const struct sock *sk)
287 {
288 return inet_csk_reqsk_queue_len(sk) > READ_ONCE(sk->sk_max_ack_backlog);
289 }
290
291 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req);
292 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req);
293
294 static inline unsigned long
reqsk_timeout(struct request_sock * req,unsigned long max_timeout)295 reqsk_timeout(struct request_sock *req, unsigned long max_timeout)
296 {
297 u64 timeout = (u64)req->timeout << req->num_timeout;
298
299 return (unsigned long)min_t(u64, timeout, max_timeout);
300 }
301
inet_csk_prepare_for_destroy_sock(struct sock * sk)302 static inline void inet_csk_prepare_for_destroy_sock(struct sock *sk)
303 {
304 /* The below has to be done to allow calling inet_csk_destroy_sock */
305 sock_set_flag(sk, SOCK_DEAD);
306 this_cpu_inc(*sk->sk_prot->orphan_count);
307 }
308
309 void inet_csk_destroy_sock(struct sock *sk);
310 void inet_csk_prepare_forced_close(struct sock *sk);
311
312 /*
313 * LISTEN is a special case for poll..
314 */
inet_csk_listen_poll(const struct sock * sk)315 static inline __poll_t inet_csk_listen_poll(const struct sock *sk)
316 {
317 return !reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue) ?
318 (EPOLLIN | EPOLLRDNORM) : 0;
319 }
320
321 int inet_csk_listen_start(struct sock *sk);
322 void inet_csk_listen_stop(struct sock *sk);
323
324 /* update the fast reuse flag when adding a socket */
325 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
326 struct sock *sk);
327
328 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu);
329
inet_csk_enter_pingpong_mode(struct sock * sk)330 static inline void inet_csk_enter_pingpong_mode(struct sock *sk)
331 {
332 inet_csk(sk)->icsk_ack.pingpong =
333 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
334 }
335
inet_csk_exit_pingpong_mode(struct sock * sk)336 static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
337 {
338 inet_csk(sk)->icsk_ack.pingpong = 0;
339 }
340
inet_csk_in_pingpong_mode(struct sock * sk)341 static inline bool inet_csk_in_pingpong_mode(struct sock *sk)
342 {
343 return inet_csk(sk)->icsk_ack.pingpong >=
344 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
345 }
346
inet_csk_inc_pingpong_cnt(struct sock * sk)347 static inline void inet_csk_inc_pingpong_cnt(struct sock *sk)
348 {
349 struct inet_connection_sock *icsk = inet_csk(sk);
350
351 if (icsk->icsk_ack.pingpong < U8_MAX)
352 icsk->icsk_ack.pingpong++;
353 }
354
inet_csk_has_ulp(const struct sock * sk)355 static inline bool inet_csk_has_ulp(const struct sock *sk)
356 {
357 return inet_test_bit(IS_ICSK, sk) && !!inet_csk(sk)->icsk_ulp_ops;
358 }
359
inet_init_csk_locks(struct sock * sk)360 static inline void inet_init_csk_locks(struct sock *sk)
361 {
362 struct inet_connection_sock *icsk = inet_csk(sk);
363
364 spin_lock_init(&icsk->icsk_accept_queue.rskq_lock);
365 spin_lock_init(&icsk->icsk_accept_queue.fastopenq.lock);
366 }
367
368 #endif /* _INET_CONNECTION_SOCK_H */
369