xref: /linux/include/net/inet_connection_sock.h (revision ca220141fa8ebae09765a242076b2b77338106b0)
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 				      void (*opt_child_init)(struct sock *newsk,
47 							     const struct sock *sk));
48 	u16	    net_header_len;
49 	int	    (*setsockopt)(struct sock *sk, int level, int optname,
50 				  sockptr_t optval, unsigned int optlen);
51 	int	    (*getsockopt)(struct sock *sk, int level, int optname,
52 				  char __user *optval, int __user *optlen);
53 	void	    (*mtu_reduced)(struct sock *sk);
54 };
55 
56 /** inet_connection_sock - INET connection oriented sock
57  *
58  * @icsk_accept_queue:	   FIFO of established children
59  * @icsk_bind_hash:	   Bind node
60  * @icsk_bind2_hash:	   Bind node in the bhash2 table
61  * @icsk_delack_timer:     Delayed ACK timer
62  * @icsk_keepalive_timer:  Keepalive timer
63  * @mptcp_tout_timer: mptcp timer
64  * @icsk_rto:		   Retransmit timeout
65  * @icsk_pmtu_cookie	   Last pmtu seen by socket
66  * @icsk_ca_ops		   Pluggable congestion control hook
67  * @icsk_af_ops		   Operations which are AF_INET{4,6} specific
68  * @icsk_ulp_ops	   Pluggable ULP control hook
69  * @icsk_ulp_data	   ULP private data
70  * @icsk_ca_state:	   Congestion control state
71  * @icsk_retransmits:	   Number of unrecovered [RTO] timeouts
72  * @icsk_pending:	   Scheduled timer event
73  * @icsk_backoff:	   Backoff
74  * @icsk_syn_retries:      Number of allowed SYN (or equivalent) retries
75  * @icsk_probes_out:	   unanswered 0 window probes
76  * @icsk_ext_hdr_len:	   Network protocol overhead (IP/IPv6 options)
77  * @icsk_ack:		   Delayed ACK control data
78  * @icsk_mtup;		   MTU probing control data
79  * @icsk_probes_tstamp:    Probe timestamp (cleared by non-zero window ack)
80  * @icsk_user_timeout:	   TCP_USER_TIMEOUT value
81  */
82 struct inet_connection_sock {
83 	/* inet_sock has to be the first member! */
84 	struct inet_sock	  icsk_inet;
85 	struct request_sock_queue icsk_accept_queue;
86 	struct inet_bind_bucket	  *icsk_bind_hash;
87 	struct inet_bind2_bucket  *icsk_bind2_hash;
88 	struct timer_list	  icsk_delack_timer;
89 	union {
90 		struct timer_list icsk_keepalive_timer;
91 		struct timer_list mptcp_tout_timer;
92 	};
93 	__u32			  icsk_rto;
94 	__u32                     icsk_rto_min;
95 	u32			  icsk_rto_max;
96 	__u32                     icsk_delack_max;
97 	__u32			  icsk_pmtu_cookie;
98 	const struct tcp_congestion_ops *icsk_ca_ops;
99 	const struct inet_connection_sock_af_ops *icsk_af_ops;
100 	const struct tcp_ulp_ops  *icsk_ulp_ops;
101 	void __rcu		  *icsk_ulp_data;
102 	unsigned int		  (*icsk_sync_mss)(struct sock *sk, u32 pmtu);
103 	__u8			  icsk_ca_state:5,
104 				  icsk_ca_initialized:1,
105 				  icsk_ca_setsockopt:1,
106 				  icsk_ca_dst_locked:1;
107 	__u8			  icsk_retransmits;
108 	__u8			  icsk_pending;
109 	__u8			  icsk_backoff;
110 	__u8			  icsk_syn_retries;
111 	__u8			  icsk_probes_out;
112 	__u16			  icsk_ext_hdr_len;
113 	struct {
114 		__u8		  pending;	 /* ACK is pending			   */
115 		__u8		  quick;	 /* Scheduled number of quick acks	   */
116 		__u8		  pingpong;	 /* The session is interactive		   */
117 		__u8		  retry;	 /* Number of attempts			   */
118 		#define ATO_BITS 8
119 		__u32		  ato:ATO_BITS,	 /* Predicted tick of soft clock	   */
120 				  lrcv_flowlabel:20, /* last received ipv6 flowlabel	   */
121 				  dst_quick_ack:1, /* cache dst RTAX_QUICKACK		   */
122 				  unused:3;
123 		__u32		  lrcvtime;	 /* timestamp of last received data packet */
124 		__u16		  last_seg_size; /* Size of last incoming segment	   */
125 		__u16		  rcv_mss;	 /* MSS used for delayed ACK decisions	   */
126 	} icsk_ack;
127 	struct {
128 		/* Range of MTUs to search */
129 		int		  search_high;
130 		int		  search_low;
131 
132 		/* Information on the current probe. */
133 		u32		  probe_size:31,
134 		/* Is the MTUP feature enabled for this connection? */
135 				  enabled:1;
136 
137 		u32		  probe_timestamp;
138 	} icsk_mtup;
139 	u32			  icsk_probes_tstamp;
140 	u32			  icsk_user_timeout;
141 
142 	u64			  icsk_ca_priv[104 / sizeof(u64)];
143 #define ICSK_CA_PRIV_SIZE	  sizeof_field(struct inet_connection_sock, icsk_ca_priv)
144 };
145 
146 #define ICSK_TIME_RETRANS	1	/* Retransmit timer */
147 #define ICSK_TIME_DACK		2	/* Delayed ack timer */
148 #define ICSK_TIME_PROBE0	3	/* Zero window probe timer */
149 #define ICSK_TIME_LOSS_PROBE	5	/* Tail loss probe timer */
150 #define ICSK_TIME_REO_TIMEOUT	6	/* Reordering timer */
151 
152 #define inet_csk(ptr) container_of_const(ptr, struct inet_connection_sock, icsk_inet.sk)
153 
154 static inline void *inet_csk_ca(const struct sock *sk)
155 {
156 	return (void *)inet_csk(sk)->icsk_ca_priv;
157 }
158 
159 struct sock *inet_csk_clone_lock(const struct sock *sk,
160 				 const struct request_sock *req,
161 				 const gfp_t priority);
162 
163 enum inet_csk_ack_state_t {
164 	ICSK_ACK_SCHED	= 1,
165 	ICSK_ACK_TIMER  = 2,
166 	ICSK_ACK_PUSHED = 4,
167 	ICSK_ACK_PUSHED2 = 8,
168 	ICSK_ACK_NOW = 16,	/* Send the next ACK immediately (once) */
169 	ICSK_ACK_NOMEM = 32,
170 };
171 
172 void inet_csk_init_xmit_timers(struct sock *sk,
173 			       void (*retransmit_handler)(struct timer_list *),
174 			       void (*delack_handler)(struct timer_list *),
175 			       void (*keepalive_handler)(struct timer_list *));
176 void inet_csk_clear_xmit_timers(struct sock *sk);
177 void inet_csk_clear_xmit_timers_sync(struct sock *sk);
178 
179 static inline void inet_csk_schedule_ack(struct sock *sk)
180 {
181 	inet_csk(sk)->icsk_ack.pending |= ICSK_ACK_SCHED;
182 }
183 
184 static inline int inet_csk_ack_scheduled(const struct sock *sk)
185 {
186 	return inet_csk(sk)->icsk_ack.pending & ICSK_ACK_SCHED;
187 }
188 
189 static inline void inet_csk_delack_init(struct sock *sk)
190 {
191 	memset(&inet_csk(sk)->icsk_ack, 0, sizeof(inet_csk(sk)->icsk_ack));
192 }
193 
194 static inline unsigned long tcp_timeout_expires(const struct sock *sk)
195 {
196 	return READ_ONCE(sk->tcp_retransmit_timer.expires);
197 }
198 
199 static inline unsigned long
200 icsk_delack_timeout(const struct inet_connection_sock *icsk)
201 {
202 	return READ_ONCE(icsk->icsk_delack_timer.expires);
203 }
204 
205 static inline void inet_csk_clear_xmit_timer(struct sock *sk, const int what)
206 {
207 	struct inet_connection_sock *icsk = inet_csk(sk);
208 
209 	if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0) {
210 		smp_store_release(&icsk->icsk_pending, 0);
211 #ifdef INET_CSK_CLEAR_TIMERS
212 		sk_stop_timer(sk, &sk->tcp_retransmit_timer);
213 #endif
214 	} else if (what == ICSK_TIME_DACK) {
215 		smp_store_release(&icsk->icsk_ack.pending, 0);
216 		icsk->icsk_ack.retry = 0;
217 #ifdef INET_CSK_CLEAR_TIMERS
218 		sk_stop_timer(sk, &icsk->icsk_delack_timer);
219 #endif
220 	} else {
221 		pr_debug("inet_csk BUG: unknown timer value\n");
222 	}
223 }
224 
225 /*
226  *	Reset the retransmission timer
227  */
228 static inline void inet_csk_reset_xmit_timer(struct sock *sk, const int what,
229 					     unsigned long when,
230 					     const unsigned long max_when)
231 {
232 	struct inet_connection_sock *icsk = inet_csk(sk);
233 
234 	if (when > max_when) {
235 		pr_debug("reset_xmit_timer: sk=%p %d when=0x%lx, caller=%p\n",
236 			 sk, what, when, (void *)_THIS_IP_);
237 		when = max_when;
238 	}
239 
240 	when += jiffies;
241 	if (what == ICSK_TIME_RETRANS || what == ICSK_TIME_PROBE0 ||
242 	    what == ICSK_TIME_LOSS_PROBE || what == ICSK_TIME_REO_TIMEOUT) {
243 		smp_store_release(&icsk->icsk_pending, what);
244 		sk_reset_timer(sk, &sk->tcp_retransmit_timer, when);
245 	} else if (what == ICSK_TIME_DACK) {
246 		smp_store_release(&icsk->icsk_ack.pending,
247 				  icsk->icsk_ack.pending | ICSK_ACK_TIMER);
248 		sk_reset_timer(sk, &icsk->icsk_delack_timer, when);
249 	} else {
250 		pr_debug("inet_csk BUG: unknown timer value\n");
251 	}
252 }
253 
254 static inline unsigned long
255 inet_csk_rto_backoff(const struct inet_connection_sock *icsk,
256 		     unsigned long max_when)
257 {
258         u64 when = (u64)icsk->icsk_rto << icsk->icsk_backoff;
259 
260         return (unsigned long)min_t(u64, when, max_when);
261 }
262 
263 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg);
264 
265 int inet_csk_get_port(struct sock *sk, unsigned short snum);
266 
267 struct dst_entry *inet_csk_route_req(const struct sock *sk, struct flowi4 *fl4,
268 				     const struct request_sock *req);
269 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
270 					    struct sock *newsk,
271 					    const struct request_sock *req);
272 
273 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
274 				      struct request_sock *req,
275 				      struct sock *child);
276 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req);
277 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
278 					 struct request_sock *req,
279 					 bool own_req);
280 
281 static inline void inet_csk_reqsk_queue_added(struct sock *sk)
282 {
283 	reqsk_queue_added(&inet_csk(sk)->icsk_accept_queue);
284 }
285 
286 static inline int inet_csk_reqsk_queue_len(const struct sock *sk)
287 {
288 	return reqsk_queue_len(&inet_csk(sk)->icsk_accept_queue);
289 }
290 
291 static inline int inet_csk_reqsk_queue_is_full(const struct sock *sk)
292 {
293 	return inet_csk_reqsk_queue_len(sk) > READ_ONCE(sk->sk_max_ack_backlog);
294 }
295 
296 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req);
297 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req);
298 
299 void inet_csk_destroy_sock(struct sock *sk);
300 void inet_csk_prepare_for_destroy_sock(struct sock *sk);
301 void inet_csk_prepare_forced_close(struct sock *sk);
302 
303 /*
304  * LISTEN is a special case for poll..
305  */
306 static inline __poll_t inet_csk_listen_poll(const struct sock *sk)
307 {
308 	return !reqsk_queue_empty(&inet_csk(sk)->icsk_accept_queue) ?
309 			(EPOLLIN | EPOLLRDNORM) : 0;
310 }
311 
312 int inet_csk_listen_start(struct sock *sk);
313 void inet_csk_listen_stop(struct sock *sk);
314 
315 /* update the fast reuse flag when adding a socket */
316 void inet_csk_update_fastreuse(const struct sock *sk,
317 			       struct inet_bind_bucket *tb,
318 			       struct inet_bind2_bucket *tb2);
319 
320 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu);
321 
322 static inline void inet_csk_enter_pingpong_mode(struct sock *sk)
323 {
324 	inet_csk(sk)->icsk_ack.pingpong =
325 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
326 }
327 
328 static inline void inet_csk_exit_pingpong_mode(struct sock *sk)
329 {
330 	inet_csk(sk)->icsk_ack.pingpong = 0;
331 }
332 
333 static inline bool inet_csk_in_pingpong_mode(struct sock *sk)
334 {
335 	return inet_csk(sk)->icsk_ack.pingpong >=
336 	       READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_pingpong_thresh);
337 }
338 
339 static inline void inet_csk_inc_pingpong_cnt(struct sock *sk)
340 {
341 	struct inet_connection_sock *icsk = inet_csk(sk);
342 
343 	if (icsk->icsk_ack.pingpong < U8_MAX)
344 		icsk->icsk_ack.pingpong++;
345 }
346 
347 static inline bool inet_csk_has_ulp(const struct sock *sk)
348 {
349 	return inet_test_bit(IS_ICSK, sk) && !!inet_csk(sk)->icsk_ulp_ops;
350 }
351 
352 static inline void inet_init_csk_locks(struct sock *sk)
353 {
354 	struct inet_connection_sock *icsk = inet_csk(sk);
355 
356 	spin_lock_init(&icsk->icsk_accept_queue.rskq_lock);
357 	spin_lock_init(&icsk->icsk_accept_queue.fastopenq.lock);
358 }
359 
360 #endif /* _INET_CONNECTION_SOCK_H */
361