xref: /linux/include/net/inet_connection_sock.h (revision b0249c0d41b306ddd79de58ca7fea543ab5e7a2e)
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 	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 	int	    (*setsockopt)(struct sock *sk, int level, int optname,
47 				  sockptr_t optval, unsigned int optlen);
48 	int	    (*getsockopt)(struct sock *sk, int level, int optname,
49 				  char __user *optval, int __user *optlen);
50 	void	    (*mtu_reduced)(struct sock *sk);
51 };
52 
53 /** inet_connection_sock - INET connection oriented sock
54  *
55  * @icsk_accept_queue:	   FIFO of established children
56  * @icsk_bind_hash:	   Bind node
57  * @icsk_bind2_hash:	   Bind node in the bhash2 table
58  * @icsk_delack_timer:     Delayed ACK timer
59  * @icsk_keepalive_timer:  Keepalive timer
60  * @mptcp_tout_timer: mptcp timer
61  * @icsk_rto:		   Retransmit timeout
62  * @icsk_pmtu_cookie	   Last pmtu seen by socket
63  * @icsk_ca_ops		   Pluggable congestion control hook
64  * @icsk_af_ops		   Operations which are AF_INET{4,6} specific
65  * @icsk_ulp_ops	   Pluggable ULP control hook
66  * @icsk_ulp_data	   ULP private data
67  * @icsk_ca_state:	   Congestion control state
68  * @icsk_retransmits:	   Number of unrecovered [RTO] timeouts
69  * @icsk_pending:	   Scheduled timer event
70  * @icsk_backoff:	   Backoff
71  * @icsk_syn_retries:      Number of allowed SYN (or equivalent) retries
72  * @icsk_probes_out:	   unanswered 0 window probes
73  * @icsk_ext_hdr_len:	   Network protocol overhead (IP/IPv6 options)
74  * @icsk_ack:		   Delayed ACK control data
75  * @icsk_mtup;		   MTU probing control data
76  * @icsk_probes_tstamp:    Probe timestamp (cleared by non-zero window ack)
77  * @icsk_user_timeout:	   TCP_USER_TIMEOUT value
78  */
79 struct inet_connection_sock {
80 	/* inet_sock has to be the first member! */
81 	struct inet_sock	  icsk_inet;
82 	struct request_sock_queue icsk_accept_queue;
83 	struct inet_bind_bucket	  *icsk_bind_hash;
84 	struct inet_bind2_bucket  *icsk_bind2_hash;
85 	struct timer_list	  icsk_delack_timer;
86 	union {
87 		struct timer_list icsk_keepalive_timer;
88 		struct timer_list mptcp_tout_timer;
89 	};
90 	__u32			  icsk_rto;
91 	__u32                     icsk_rto_min;
92 	u32			  icsk_rto_max;
93 	__u32                     icsk_delack_max;
94 	__u32			  icsk_pmtu_cookie;
95 	const struct tcp_congestion_ops *icsk_ca_ops;
96 	const struct inet_connection_sock_af_ops *icsk_af_ops;
97 	const struct tcp_ulp_ops  *icsk_ulp_ops;
98 	void __rcu		  *icsk_ulp_data;
99 	unsigned int		  (*icsk_sync_mss)(struct sock *sk, u32 pmtu);
100 	__u8			  icsk_ca_state:5,
101 				  icsk_ca_initialized:1,
102 				  icsk_ca_setsockopt:1,
103 				  icsk_ca_dst_locked:1;
104 	__u8			  icsk_retransmits;
105 	__u8			  icsk_pending;
106 	__u8			  icsk_backoff;
107 	__u8			  icsk_syn_retries;
108 	__u8			  icsk_probes_out;
109 	__u16			  icsk_ext_hdr_len;
110 	struct {
111 		__u8		  pending;	 /* ACK is pending			   */
112 		__u8		  quick;	 /* Scheduled number of quick acks	   */
113 		__u8		  pingpong;	 /* The session is interactive		   */
114 		__u8		  retry;	 /* Number of attempts			   */
115 		#define ATO_BITS 8
116 		__u32		  ato:ATO_BITS,	 /* Predicted tick of soft clock	   */
117 				  lrcv_flowlabel:20, /* last received ipv6 flowlabel	   */
118 				  dst_quick_ack:1, /* cache dst RTAX_QUICKACK		   */
119 				  unused:3;
120 		__u32		  lrcvtime;	 /* timestamp of last received data packet */
121 		__u16		  last_seg_size; /* Size of last incoming segment	   */
122 		__u16		  rcv_mss;	 /* MSS used for delayed ACK decisions	   */
123 	} icsk_ack;
124 	struct {
125 		/* Range of MTUs to search */
126 		int		  search_high;
127 		int		  search_low;
128 
129 		/* Information on the current probe. */
130 		u32		  probe_size:31,
131 		/* Is the MTUP feature enabled for this connection? */
132 				  enabled:1;
133 
134 		u32		  probe_timestamp;
135 	} icsk_mtup;
136 	u32			  icsk_probes_tstamp;
137 	u32			  icsk_user_timeout;
138 
139 	u64			  icsk_ca_priv[104 / sizeof(u64)];
140 #define ICSK_CA_PRIV_SIZE	  sizeof_field(struct inet_connection_sock, icsk_ca_priv)
141 };
142 
143 #define ICSK_TIME_RETRANS	1	/* Retransmit timer */
144 #define ICSK_TIME_DACK		2	/* Delayed ack timer */
145 #define ICSK_TIME_PROBE0	3	/* Zero window probe timer */
146 #define ICSK_TIME_LOSS_PROBE	5	/* Tail loss probe timer */
147 #define ICSK_TIME_REO_TIMEOUT	6	/* Reordering timer */
148 
149 #define inet_csk(ptr) container_of_const(ptr, struct inet_connection_sock, icsk_inet.sk)
150 
151 static inline void *inet_csk_ca(const struct sock *sk)
152 {
153 	return (void *)inet_csk(sk)->icsk_ca_priv;
154 }
155 
156 struct sock *inet_csk_clone_lock(const struct sock *sk,
157 				 const struct request_sock *req,
158 				 const gfp_t priority);
159 
160 enum inet_csk_ack_state_t {
161 	ICSK_ACK_SCHED	= 1,
162 	ICSK_ACK_TIMER  = 2,
163 	ICSK_ACK_PUSHED = 4,
164 	ICSK_ACK_PUSHED2 = 8,
165 	ICSK_ACK_NOW = 16,	/* Send the next ACK immediately (once) */
166 	ICSK_ACK_NOMEM = 32,
167 };
168 
169 void inet_csk_init_xmit_timers(struct sock *sk,
170 			       void (*retransmit_handler)(struct timer_list *),
171 			       void (*delack_handler)(struct timer_list *),
172 			       void (*keepalive_handler)(struct timer_list *));
173 void inet_csk_clear_xmit_timers(struct sock *sk);
174 void inet_csk_clear_xmit_timers_sync(struct sock *sk);
175 
176 static inline void inet_csk_schedule_ack(struct sock *sk)
177 {
178 	inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_SCHED;
179 }
180 
181 static inline int inet_csk_ack_scheduled(const struct sock *sk)
182 {
183 	return inet_csk(sk)->icsk_ack.pending & ICSK_ACK_SCHED;
184 }
185 
186 static inline void inet_csk_delack_init(struct sock *sk)
187 {
188 	memset(&inet_csk(sk)->icsk_ack, 0, sizeof(inet_csk(sk)->icsk_ack));
189 }
190 
191 static inline unsigned long tcp_timeout_expires(const struct sock *sk)
192 {
193 	return READ_ONCE(sk->tcp_retransmit_timer.expires);
194 }
195 
196 static inline unsigned long
197 icsk_delack_timeout(const struct inet_connection_sock *icsk)
198 {
199 	return READ_ONCE(icsk->icsk_delack_timer.expires);
200 }
201 
202 static inline void inet_csk_clear_xmit_timer(struct sock *sk, const int what)
203 {
204 	struct inet_connection_sock *icsk = inet_csk(sk);
205 
206 	if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
207 		smp_store_release(&icsk->icsk_pending, 0);
208 #ifdef INET_CSK_CLEAR_TIMERS
209 		sk_stop_timer(sk, &sk->tcp_retransmit_timer);
210 #endif
211 	} else if (what == ICSK_TIME_DACK) {
212 		smp_store_release(&icsk->icsk_ack.pending, 0);
213 		icsk->icsk_ack.retry = 0;
214 #ifdef INET_CSK_CLEAR_TIMERS
215 		sk_stop_timer(sk, &icsk->icsk_delack_timer);
216 #endif
217 	} else {
218 		pr_debug("inet_csk BUG: unknown timer value\n");
219 	}
220 }
221 
222 /*
223  *	Reset the retransmission timer
224  */
225 static inline void inet_csk_reset_xmit_timer(struct sock *sk, const int what,
226 					     unsigned long when,
227 					     const unsigned long max_when)
228 {
229 	struct inet_connection_sock *icsk = inet_csk(sk);
230 
231 	if (when > max_when) {
232 		pr_debug("reset_xmit_timer: sk=%p %d when=0x%lx, caller=%p\n",
233 			 sk, what, when, (void *)_THIS_IP_);
234 		when = max_when;
235 	}
236 
237 	when += jiffies;
238 	if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0 ||
239 	    what == ICSK_TIME_LOSS_PROBE || what == ICSK_TIME_REO_TIMEOUT) {
240 		smp_store_release(&icsk->icsk_pending, what);
241 		sk_reset_timer(sk, &sk->tcp_retransmit_timer, when);
242 	} else if (what == ICSK_TIME_DACK) {
243 		smp_store_release(&icsk->icsk_ack.pending,
244 				  icsk->icsk_ack.pending | ICSK_ACK_TIMER);
245 		sk_reset_timer(sk, &icsk->icsk_delack_timer, when);
246 	} else {
247 		pr_debug("inet_csk BUG: unknown timer value\n");
248 	}
249 }
250 
251 static inline unsigned long
252 inet_csk_rto_backoff(const struct inet_connection_sock *icsk,
253 		     unsigned long max_when)
254 {
255         u64 when = (u64)icsk->icsk_rto << icsk->icsk_backoff;
256 
257         return (unsigned long)min_t(u64, when, max_when);
258 }
259 
260 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg);
261 
262 int inet_csk_get_port(struct sock *sk, unsigned short snum);
263 
264 struct dst_entry *inet_csk_route_req(const struct sock *sk, struct flowi4 *fl4,
265 				     const struct request_sock *req);
266 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
267 					    struct sock *newsk,
268 					    const struct request_sock *req);
269 
270 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
271 				      struct request_sock *req,
272 				      struct sock *child);
273 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req);
274 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
275 					 struct request_sock *req,
276 					 bool own_req);
277 
278 static inline void inet_csk_reqsk_queue_added(struct sock *sk)
279 {
280 	reqsk_queue_added(&inet_csk(sk)->icsk_accept_queue);
281 }
282 
283 static inline int inet_csk_reqsk_queue_len(const struct sock *sk)
284 {
285 	return reqsk_queue_len(&inet_csk(sk)->icsk_accept_queue);
286 }
287 
288 static inline int inet_csk_reqsk_queue_is_full(const struct sock *sk)
289 {
290 	return inet_csk_reqsk_queue_len(sk) > READ_ONCE(sk->sk_max_ack_backlog);
291 }
292 
293 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req);
294 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req);
295 
296 void inet_csk_destroy_sock(struct sock *sk);
297 void inet_csk_prepare_for_destroy_sock(struct sock *sk);
298 void inet_csk_prepare_forced_close(struct sock *sk);
299 
300 /*
301  * LISTEN is a special case for poll..
302  */
303 static inline __poll_t inet_csk_listen_poll(const struct sock *sk)
304 {
305 	return !reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue) ?
306 			(EPOLLIN | EPOLLRDNORM) : 0;
307 }
308 
309 int inet_csk_listen_start(struct sock *sk);
310 void inet_csk_listen_stop(struct sock *sk);
311 
312 /* update the fast reuse flag when adding a socket */
313 void inet_csk_update_fastreuse(const struct sock *sk,
314 			       struct inet_bind_bucket *tb,
315 			       struct inet_bind2_bucket *tb2);
316 
317 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu);
318 
319 static inline void inet_csk_enter_pingpong_mode(struct sock *sk)
320 {
321 	inet_csk(sk)->icsk_ack.pingpong =
322 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
323 }
324 
325 static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
326 {
327 	inet_csk(sk)->icsk_ack.pingpong = 0;
328 }
329 
330 static inline bool inet_csk_in_pingpong_mode(struct sock *sk)
331 {
332 	return inet_csk(sk)->icsk_ack.pingpong >=
333 	       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
334 }
335 
336 static inline void inet_csk_inc_pingpong_cnt(struct sock *sk)
337 {
338 	struct inet_connection_sock *icsk = inet_csk(sk);
339 
340 	if (icsk->icsk_ack.pingpong < U8_MAX)
341 		icsk->icsk_ack.pingpong++;
342 }
343 
344 static inline bool inet_csk_has_ulp(const struct sock *sk)
345 {
346 	return inet_test_bit(IS_ICSK, sk) && !!inet_csk(sk)->icsk_ulp_ops;
347 }
348 
349 static inline void inet_init_csk_locks(struct sock *sk)
350 {
351 	struct inet_connection_sock *icsk = inet_csk(sk);
352 
353 	spin_lock_init(&icsk->icsk_accept_queue.rskq_lock);
354 	spin_lock_init(&icsk->icsk_accept_queue.fastopenq.lock);
355 }
356 
357 #endif /* _INET_CONNECTION_SOCK_H */
358