xref: /linux/include/net/request_sock.h (revision b85d45947951d23cb22d90caecf4c1eb81342c96)
1 /*
2  * NET		Generic infrastructure for Network protocols.
3  *
4  *		Definitions for request_sock
5  *
6  * Authors:	Arnaldo Carvalho de Melo <acme@conectiva.com.br>
7  *
8  * 		From code originally in include/net/tcp.h
9  *
10  *		This program is free software; you can redistribute it and/or
11  *		modify it under the terms of the GNU General Public License
12  *		as published by the Free Software Foundation; either version
13  *		2 of the License, or (at your option) any later version.
14  */
15 #ifndef _REQUEST_SOCK_H
16 #define _REQUEST_SOCK_H
17 
18 #include <linux/slab.h>
19 #include <linux/spinlock.h>
20 #include <linux/types.h>
21 #include <linux/bug.h>
22 
23 #include <net/sock.h>
24 
25 struct request_sock;
26 struct sk_buff;
27 struct dst_entry;
28 struct proto;
29 
30 struct request_sock_ops {
31 	int		family;
32 	int		obj_size;
33 	struct kmem_cache	*slab;
34 	char		*slab_name;
35 	int		(*rtx_syn_ack)(struct sock *sk,
36 				       struct request_sock *req);
37 	void		(*send_ack)(struct sock *sk, struct sk_buff *skb,
38 				    struct request_sock *req);
39 	void		(*send_reset)(struct sock *sk,
40 				      struct sk_buff *skb);
41 	void		(*destructor)(struct request_sock *req);
42 	void		(*syn_ack_timeout)(const struct request_sock *req);
43 };
44 
45 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req);
46 
47 /* struct request_sock - mini sock to represent a connection request
48  */
49 struct request_sock {
50 	struct sock_common		__req_common;
51 #define rsk_refcnt			__req_common.skc_refcnt
52 #define rsk_hash			__req_common.skc_hash
53 
54 	struct request_sock		*dl_next;
55 	struct sock			*rsk_listener;
56 	u16				mss;
57 	u8				num_retrans; /* number of retransmits */
58 	u8				cookie_ts:1; /* syncookie: encode tcpopts in timestamp */
59 	u8				num_timeout:7; /* number of timeouts */
60 	/* The following two fields can be easily recomputed I think -AK */
61 	u32				window_clamp; /* window clamp at creation time */
62 	u32				rcv_wnd;	  /* rcv_wnd offered first time */
63 	u32				ts_recent;
64 	struct timer_list		rsk_timer;
65 	const struct request_sock_ops	*rsk_ops;
66 	struct sock			*sk;
67 	u32				*saved_syn;
68 	u32				secid;
69 	u32				peer_secid;
70 };
71 
72 static inline struct request_sock *
73 reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener)
74 {
75 	struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC);
76 
77 	if (req) {
78 		req->rsk_ops = ops;
79 		sock_hold(sk_listener);
80 		req->rsk_listener = sk_listener;
81 		req->saved_syn = NULL;
82 		/* Following is temporary. It is coupled with debugging
83 		 * helpers in reqsk_put() & reqsk_free()
84 		 */
85 		atomic_set(&req->rsk_refcnt, 0);
86 	}
87 	return req;
88 }
89 
90 static inline struct request_sock *inet_reqsk(struct sock *sk)
91 {
92 	return (struct request_sock *)sk;
93 }
94 
95 static inline struct sock *req_to_sk(struct request_sock *req)
96 {
97 	return (struct sock *)req;
98 }
99 
100 static inline void reqsk_free(struct request_sock *req)
101 {
102 	/* temporary debugging */
103 	WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0);
104 
105 	req->rsk_ops->destructor(req);
106 	if (req->rsk_listener)
107 		sock_put(req->rsk_listener);
108 	kfree(req->saved_syn);
109 	kmem_cache_free(req->rsk_ops->slab, req);
110 }
111 
112 static inline void reqsk_put(struct request_sock *req)
113 {
114 	if (atomic_dec_and_test(&req->rsk_refcnt))
115 		reqsk_free(req);
116 }
117 
118 extern int sysctl_max_syn_backlog;
119 
120 /** struct listen_sock - listen state
121  *
122  * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
123  */
124 struct listen_sock {
125 	int			qlen_inc; /* protected by listener lock */
126 	int			young_inc;/* protected by listener lock */
127 
128 	/* following fields can be updated by timer */
129 	atomic_t		qlen_dec; /* qlen = qlen_inc - qlen_dec */
130 	atomic_t		young_dec;
131 
132 	u8			max_qlen_log ____cacheline_aligned_in_smp;
133 	u8			synflood_warned;
134 	/* 2 bytes hole, try to use */
135 	u32			hash_rnd;
136 	u32			nr_table_entries;
137 	struct request_sock	*syn_table[0];
138 };
139 
140 /*
141  * For a TCP Fast Open listener -
142  *	lock - protects the access to all the reqsk, which is co-owned by
143  *		the listener and the child socket.
144  *	qlen - pending TFO requests (still in TCP_SYN_RECV).
145  *	max_qlen - max TFO reqs allowed before TFO is disabled.
146  *
147  *	XXX (TFO) - ideally these fields can be made as part of "listen_sock"
148  *	structure above. But there is some implementation difficulty due to
149  *	listen_sock being part of request_sock_queue hence will be freed when
150  *	a listener is stopped. But TFO related fields may continue to be
151  *	accessed even after a listener is closed, until its sk_refcnt drops
152  *	to 0 implying no more outstanding TFO reqs. One solution is to keep
153  *	listen_opt around until	sk_refcnt drops to 0. But there is some other
154  *	complexity that needs to be resolved. E.g., a listener can be disabled
155  *	temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
156  */
157 struct fastopen_queue {
158 	struct request_sock	*rskq_rst_head; /* Keep track of past TFO */
159 	struct request_sock	*rskq_rst_tail; /* requests that caused RST.
160 						 * This is part of the defense
161 						 * against spoofing attack.
162 						 */
163 	spinlock_t	lock;
164 	int		qlen;		/* # of pending (TCP_SYN_RECV) reqs */
165 	int		max_qlen;	/* != 0 iff TFO is currently enabled */
166 };
167 
168 /** struct request_sock_queue - queue of request_socks
169  *
170  * @rskq_accept_head - FIFO head of established children
171  * @rskq_accept_tail - FIFO tail of established children
172  * @rskq_defer_accept - User waits for some data after accept()
173  * @syn_wait_lock - serializer
174  *
175  * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
176  * lock sock while browsing the listening hash (otherwise it's deadlock prone).
177  *
178  */
179 struct request_sock_queue {
180 	struct request_sock	*rskq_accept_head;
181 	struct request_sock	*rskq_accept_tail;
182 	u8			rskq_defer_accept;
183 	struct listen_sock	*listen_opt;
184 	struct fastopen_queue	*fastopenq; /* This is non-NULL iff TFO has been
185 					     * enabled on this listener. Check
186 					     * max_qlen != 0 in fastopen_queue
187 					     * to determine if TFO is enabled
188 					     * right at this moment.
189 					     */
190 
191 	/* temporary alignment, our goal is to get rid of this lock */
192 	spinlock_t		syn_wait_lock ____cacheline_aligned_in_smp;
193 };
194 
195 int reqsk_queue_alloc(struct request_sock_queue *queue,
196 		      unsigned int nr_table_entries);
197 
198 void __reqsk_queue_destroy(struct request_sock_queue *queue);
199 void reqsk_queue_destroy(struct request_sock_queue *queue);
200 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
201 			   bool reset);
202 
203 static inline struct request_sock *
204 	reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
205 {
206 	struct request_sock *req = queue->rskq_accept_head;
207 
208 	queue->rskq_accept_head = NULL;
209 	return req;
210 }
211 
212 static inline int reqsk_queue_empty(struct request_sock_queue *queue)
213 {
214 	return queue->rskq_accept_head == NULL;
215 }
216 
217 static inline void reqsk_queue_add(struct request_sock_queue *queue,
218 				   struct request_sock *req,
219 				   struct sock *parent,
220 				   struct sock *child)
221 {
222 	req->sk = child;
223 	sk_acceptq_added(parent);
224 
225 	if (queue->rskq_accept_head == NULL)
226 		queue->rskq_accept_head = req;
227 	else
228 		queue->rskq_accept_tail->dl_next = req;
229 
230 	queue->rskq_accept_tail = req;
231 	req->dl_next = NULL;
232 }
233 
234 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
235 {
236 	struct request_sock *req = queue->rskq_accept_head;
237 
238 	WARN_ON(req == NULL);
239 
240 	queue->rskq_accept_head = req->dl_next;
241 	if (queue->rskq_accept_head == NULL)
242 		queue->rskq_accept_tail = NULL;
243 
244 	return req;
245 }
246 
247 static inline void reqsk_queue_removed(struct request_sock_queue *queue,
248 				       const struct request_sock *req)
249 {
250 	struct listen_sock *lopt = queue->listen_opt;
251 
252 	if (req->num_timeout == 0)
253 		atomic_inc(&lopt->young_dec);
254 	atomic_inc(&lopt->qlen_dec);
255 }
256 
257 static inline void reqsk_queue_added(struct request_sock_queue *queue)
258 {
259 	struct listen_sock *lopt = queue->listen_opt;
260 
261 	lopt->young_inc++;
262 	lopt->qlen_inc++;
263 }
264 
265 static inline int listen_sock_qlen(const struct listen_sock *lopt)
266 {
267 	return lopt->qlen_inc - atomic_read(&lopt->qlen_dec);
268 }
269 
270 static inline int listen_sock_young(const struct listen_sock *lopt)
271 {
272 	return lopt->young_inc - atomic_read(&lopt->young_dec);
273 }
274 
275 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
276 {
277 	const struct listen_sock *lopt = queue->listen_opt;
278 
279 	return lopt ? listen_sock_qlen(lopt) : 0;
280 }
281 
282 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
283 {
284 	return listen_sock_young(queue->listen_opt);
285 }
286 
287 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
288 {
289 	return reqsk_queue_len(queue) >> queue->listen_opt->max_qlen_log;
290 }
291 
292 void reqsk_queue_hash_req(struct request_sock_queue *queue,
293 			  u32 hash, struct request_sock *req,
294 			  unsigned long timeout);
295 
296 #endif /* _REQUEST_SOCK_H */
297