xref: /linux/include/net/inet_connection_sock.h (revision 6dfafbd0299a60bfb5d5e277fdf100037c7ded07)
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_delack_timer:     Delayed ACK timer
60  * @icsk_keepalive_timer:  Keepalive timer
61  * @mptcp_tout_timer: mptcp timer
62  * @icsk_rto:		   Retransmit timeout
63  * @icsk_pmtu_cookie	   Last pmtu seen by socket
64  * @icsk_ca_ops		   Pluggable congestion control hook
65  * @icsk_af_ops		   Operations which are AF_INET{4,6} specific
66  * @icsk_ulp_ops	   Pluggable ULP control hook
67  * @icsk_ulp_data	   ULP private data
68  * @icsk_ca_state:	   Congestion control state
69  * @icsk_retransmits:	   Number of unrecovered [RTO] timeouts
70  * @icsk_pending:	   Scheduled timer event
71  * @icsk_backoff:	   Backoff
72  * @icsk_syn_retries:      Number of allowed SYN (or equivalent) retries
73  * @icsk_probes_out:	   unanswered 0 window probes
74  * @icsk_ext_hdr_len:	   Network protocol overhead (IP/IPv6 options)
75  * @icsk_ack:		   Delayed ACK control data
76  * @icsk_mtup;		   MTU probing control data
77  * @icsk_probes_tstamp:    Probe timestamp (cleared by non-zero window ack)
78  * @icsk_user_timeout:	   TCP_USER_TIMEOUT value
79  */
80 struct inet_connection_sock {
81 	/* inet_sock has to be the first member! */
82 	struct inet_sock	  icsk_inet;
83 	struct request_sock_queue icsk_accept_queue;
84 	struct inet_bind_bucket	  *icsk_bind_hash;
85 	struct inet_bind2_bucket  *icsk_bind2_hash;
86 	struct timer_list	  icsk_delack_timer;
87 	union {
88 		struct timer_list icsk_keepalive_timer;
89 		struct timer_list mptcp_tout_timer;
90 	};
91 	__u32			  icsk_rto;
92 	__u32                     icsk_rto_min;
93 	u32			  icsk_rto_max;
94 	__u32                     icsk_delack_max;
95 	__u32			  icsk_pmtu_cookie;
96 	const struct tcp_congestion_ops *icsk_ca_ops;
97 	const struct inet_connection_sock_af_ops *icsk_af_ops;
98 	const struct tcp_ulp_ops  *icsk_ulp_ops;
99 	void __rcu		  *icsk_ulp_data;
100 	unsigned int		  (*icsk_sync_mss)(struct sock *sk, u32 pmtu);
101 	__u8			  icsk_ca_state:5,
102 				  icsk_ca_initialized:1,
103 				  icsk_ca_setsockopt:1,
104 				  icsk_ca_dst_locked:1;
105 	__u8			  icsk_retransmits;
106 	__u8			  icsk_pending;
107 	__u8			  icsk_backoff;
108 	__u8			  icsk_syn_retries;
109 	__u8			  icsk_probes_out;
110 	__u16			  icsk_ext_hdr_len;
111 	struct {
112 		__u8		  pending;	 /* ACK is pending			   */
113 		__u8		  quick;	 /* Scheduled number of quick acks	   */
114 		__u8		  pingpong;	 /* The session is interactive		   */
115 		__u8		  retry;	 /* Number of attempts			   */
116 		#define ATO_BITS 8
117 		__u32		  ato:ATO_BITS,	 /* Predicted tick of soft clock	   */
118 				  lrcv_flowlabel:20, /* last received ipv6 flowlabel	   */
119 				  dst_quick_ack:1, /* cache dst RTAX_QUICKACK		   */
120 				  unused:3;
121 		__u32		  lrcvtime;	 /* timestamp of last received data packet */
122 		__u16		  last_seg_size; /* Size of last incoming segment	   */
123 		__u16		  rcv_mss;	 /* MSS used for delayed ACK decisions	   */
124 	} icsk_ack;
125 	struct {
126 		/* Range of MTUs to search */
127 		int		  search_high;
128 		int		  search_low;
129 
130 		/* Information on the current probe. */
131 		u32		  probe_size:31,
132 		/* Is the MTUP feature enabled for this connection? */
133 				  enabled:1;
134 
135 		u32		  probe_timestamp;
136 	} icsk_mtup;
137 	u32			  icsk_probes_tstamp;
138 	u32			  icsk_user_timeout;
139 
140 	u64			  icsk_ca_priv[104 / sizeof(u64)];
141 #define ICSK_CA_PRIV_SIZE	  sizeof_field(struct inet_connection_sock, icsk_ca_priv)
142 };
143 
144 #define ICSK_TIME_RETRANS	1	/* Retransmit timer */
145 #define ICSK_TIME_DACK		2	/* Delayed ack timer */
146 #define ICSK_TIME_PROBE0	3	/* Zero window probe timer */
147 #define ICSK_TIME_LOSS_PROBE	5	/* Tail loss probe timer */
148 #define ICSK_TIME_REO_TIMEOUT	6	/* Reordering timer */
149 
150 #define inet_csk(ptr) container_of_const(ptr, struct inet_connection_sock, icsk_inet.sk)
151 
152 static inline void *inet_csk_ca(const struct sock *sk)
153 {
154 	return (void *)inet_csk(sk)->icsk_ca_priv;
155 }
156 
157 struct sock *inet_csk_clone_lock(const struct sock *sk,
158 				 const struct request_sock *req,
159 				 const gfp_t priority);
160 
161 enum inet_csk_ack_state_t {
162 	ICSK_ACK_SCHED	= 1,
163 	ICSK_ACK_TIMER  = 2,
164 	ICSK_ACK_PUSHED = 4,
165 	ICSK_ACK_PUSHED2 = 8,
166 	ICSK_ACK_NOW = 16,	/* Send the next ACK immediately (once) */
167 	ICSK_ACK_NOMEM = 32,
168 };
169 
170 void inet_csk_init_xmit_timers(struct sock *sk,
171 			       void (*retransmit_handler)(struct timer_list *),
172 			       void (*delack_handler)(struct timer_list *),
173 			       void (*keepalive_handler)(struct timer_list *));
174 void inet_csk_clear_xmit_timers(struct sock *sk);
175 void inet_csk_clear_xmit_timers_sync(struct sock *sk);
176 
177 static inline void inet_csk_schedule_ack(struct sock *sk)
178 {
179 	inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_SCHED;
180 }
181 
182 static inline int inet_csk_ack_scheduled(const struct sock *sk)
183 {
184 	return inet_csk(sk)->icsk_ack.pending & ICSK_ACK_SCHED;
185 }
186 
187 static inline void inet_csk_delack_init(struct sock *sk)
188 {
189 	memset(&inet_csk(sk)->icsk_ack, 0, sizeof(inet_csk(sk)->icsk_ack));
190 }
191 
192 static inline unsigned long tcp_timeout_expires(const struct sock *sk)
193 {
194 	return READ_ONCE(sk->tcp_retransmit_timer.expires);
195 }
196 
197 static inline unsigned long
198 icsk_delack_timeout(const struct inet_connection_sock *icsk)
199 {
200 	return READ_ONCE(icsk->icsk_delack_timer.expires);
201 }
202 
203 static inline void inet_csk_clear_xmit_timer(struct sock *sk, const int what)
204 {
205 	struct inet_connection_sock *icsk = inet_csk(sk);
206 
207 	if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
208 		smp_store_release(&icsk->icsk_pending, 0);
209 #ifdef INET_CSK_CLEAR_TIMERS
210 		sk_stop_timer(sk, &sk->tcp_retransmit_timer);
211 #endif
212 	} else if (what == ICSK_TIME_DACK) {
213 		smp_store_release(&icsk->icsk_ack.pending, 0);
214 		icsk->icsk_ack.retry = 0;
215 #ifdef INET_CSK_CLEAR_TIMERS
216 		sk_stop_timer(sk, &icsk->icsk_delack_timer);
217 #endif
218 	} else {
219 		pr_debug("inet_csk BUG: unknown timer value\n");
220 	}
221 }
222 
223 /*
224  *	Reset the retransmission timer
225  */
226 static inline void inet_csk_reset_xmit_timer(struct sock *sk, const int what,
227 					     unsigned long when,
228 					     const unsigned long max_when)
229 {
230 	struct inet_connection_sock *icsk = inet_csk(sk);
231 
232 	if (when > max_when) {
233 		pr_debug("reset_xmit_timer: sk=%p %d when=0x%lx, caller=%p\n",
234 			 sk, what, when, (void *)_THIS_IP_);
235 		when = max_when;
236 	}
237 
238 	when += jiffies;
239 	if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0 ||
240 	    what == ICSK_TIME_LOSS_PROBE || what == ICSK_TIME_REO_TIMEOUT) {
241 		smp_store_release(&icsk->icsk_pending, what);
242 		sk_reset_timer(sk, &sk->tcp_retransmit_timer, when);
243 	} else if (what == ICSK_TIME_DACK) {
244 		smp_store_release(&icsk->icsk_ack.pending,
245 				  icsk->icsk_ack.pending | ICSK_ACK_TIMER);
246 		sk_reset_timer(sk, &icsk->icsk_delack_timer, when);
247 	} else {
248 		pr_debug("inet_csk BUG: unknown timer value\n");
249 	}
250 }
251 
252 static inline unsigned long
253 inet_csk_rto_backoff(const struct inet_connection_sock *icsk,
254 		     unsigned long max_when)
255 {
256         u64 when = (u64)icsk->icsk_rto << icsk->icsk_backoff;
257 
258         return (unsigned long)min_t(u64, when, max_when);
259 }
260 
261 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg);
262 
263 int inet_csk_get_port(struct sock *sk, unsigned short snum);
264 
265 struct dst_entry *inet_csk_route_req(const struct sock *sk, struct flowi4 *fl4,
266 				     const struct request_sock *req);
267 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
268 					    struct sock *newsk,
269 					    const struct request_sock *req);
270 
271 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
272 				      struct request_sock *req,
273 				      struct sock *child);
274 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req);
275 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
276 					 struct request_sock *req,
277 					 bool own_req);
278 
279 static inline void inet_csk_reqsk_queue_added(struct sock *sk)
280 {
281 	reqsk_queue_added(&inet_csk(sk)->icsk_accept_queue);
282 }
283 
284 static inline int inet_csk_reqsk_queue_len(const struct sock *sk)
285 {
286 	return reqsk_queue_len(&inet_csk(sk)->icsk_accept_queue);
287 }
288 
289 static inline int inet_csk_reqsk_queue_is_full(const struct sock *sk)
290 {
291 	return inet_csk_reqsk_queue_len(sk) > READ_ONCE(sk->sk_max_ack_backlog);
292 }
293 
294 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req);
295 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req);
296 
297 void inet_csk_destroy_sock(struct sock *sk);
298 void inet_csk_prepare_for_destroy_sock(struct sock *sk);
299 void inet_csk_prepare_forced_close(struct sock *sk);
300 
301 /*
302  * LISTEN is a special case for poll..
303  */
304 static inline __poll_t inet_csk_listen_poll(const struct sock *sk)
305 {
306 	return !reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue) ?
307 			(EPOLLIN | EPOLLRDNORM) : 0;
308 }
309 
310 int inet_csk_listen_start(struct sock *sk);
311 void inet_csk_listen_stop(struct sock *sk);
312 
313 /* update the fast reuse flag when adding a socket */
314 void inet_csk_update_fastreuse(const struct sock *sk,
315 			       struct inet_bind_bucket *tb,
316 			       struct inet_bind2_bucket *tb2);
317 
318 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu);
319 
320 static inline void inet_csk_enter_pingpong_mode(struct sock *sk)
321 {
322 	inet_csk(sk)->icsk_ack.pingpong =
323 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
324 }
325 
326 static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
327 {
328 	inet_csk(sk)->icsk_ack.pingpong = 0;
329 }
330 
331 static inline bool inet_csk_in_pingpong_mode(struct sock *sk)
332 {
333 	return inet_csk(sk)->icsk_ack.pingpong >=
334 	       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
335 }
336 
337 static inline void inet_csk_inc_pingpong_cnt(struct sock *sk)
338 {
339 	struct inet_connection_sock *icsk = inet_csk(sk);
340 
341 	if (icsk->icsk_ack.pingpong < U8_MAX)
342 		icsk->icsk_ack.pingpong++;
343 }
344 
345 static inline bool inet_csk_has_ulp(const struct sock *sk)
346 {
347 	return inet_test_bit(IS_ICSK, sk) && !!inet_csk(sk)->icsk_ulp_ops;
348 }
349 
350 static inline void inet_init_csk_locks(struct sock *sk)
351 {
352 	struct inet_connection_sock *icsk = inet_csk(sk);
353 
354 	spin_lock_init(&icsk->icsk_accept_queue.rskq_lock);
355 	spin_lock_init(&icsk->icsk_accept_queue.fastopenq.lock);
356 }
357 
358 #endif /* _INET_CONNECTION_SOCK_H */
359