xref: /openbmc/linux/include/net/request_sock.h (revision d0bd7f2a)
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				secid;
68 	u32				peer_secid;
69 };
70 
71 static inline struct request_sock *
72 reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener)
73 {
74 	struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC);
75 
76 	if (req) {
77 		req->rsk_ops = ops;
78 		sock_hold(sk_listener);
79 		req->rsk_listener = sk_listener;
80 
81 		/* Following is temporary. It is coupled with debugging
82 		 * helpers in reqsk_put() & reqsk_free()
83 		 */
84 		atomic_set(&req->rsk_refcnt, 0);
85 	}
86 	return req;
87 }
88 
89 static inline struct request_sock *inet_reqsk(struct sock *sk)
90 {
91 	return (struct request_sock *)sk;
92 }
93 
94 static inline struct sock *req_to_sk(struct request_sock *req)
95 {
96 	return (struct sock *)req;
97 }
98 
99 static inline void reqsk_free(struct request_sock *req)
100 {
101 	/* temporary debugging */
102 	WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0);
103 
104 	req->rsk_ops->destructor(req);
105 	if (req->rsk_listener)
106 		sock_put(req->rsk_listener);
107 	kmem_cache_free(req->rsk_ops->slab, req);
108 }
109 
110 static inline void reqsk_put(struct request_sock *req)
111 {
112 	if (atomic_dec_and_test(&req->rsk_refcnt))
113 		reqsk_free(req);
114 }
115 
116 extern int sysctl_max_syn_backlog;
117 
118 /** struct listen_sock - listen state
119  *
120  * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
121  */
122 struct listen_sock {
123 	int			qlen_inc; /* protected by listener lock */
124 	int			young_inc;/* protected by listener lock */
125 
126 	/* following fields can be updated by timer */
127 	atomic_t		qlen_dec; /* qlen = qlen_inc - qlen_dec */
128 	atomic_t		young_dec;
129 
130 	u8			max_qlen_log ____cacheline_aligned_in_smp;
131 	u8			synflood_warned;
132 	/* 2 bytes hole, try to use */
133 	u32			hash_rnd;
134 	u32			nr_table_entries;
135 	struct request_sock	*syn_table[0];
136 };
137 
138 /*
139  * For a TCP Fast Open listener -
140  *	lock - protects the access to all the reqsk, which is co-owned by
141  *		the listener and the child socket.
142  *	qlen - pending TFO requests (still in TCP_SYN_RECV).
143  *	max_qlen - max TFO reqs allowed before TFO is disabled.
144  *
145  *	XXX (TFO) - ideally these fields can be made as part of "listen_sock"
146  *	structure above. But there is some implementation difficulty due to
147  *	listen_sock being part of request_sock_queue hence will be freed when
148  *	a listener is stopped. But TFO related fields may continue to be
149  *	accessed even after a listener is closed, until its sk_refcnt drops
150  *	to 0 implying no more outstanding TFO reqs. One solution is to keep
151  *	listen_opt around until	sk_refcnt drops to 0. But there is some other
152  *	complexity that needs to be resolved. E.g., a listener can be disabled
153  *	temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
154  */
155 struct fastopen_queue {
156 	struct request_sock	*rskq_rst_head; /* Keep track of past TFO */
157 	struct request_sock	*rskq_rst_tail; /* requests that caused RST.
158 						 * This is part of the defense
159 						 * against spoofing attack.
160 						 */
161 	spinlock_t	lock;
162 	int		qlen;		/* # of pending (TCP_SYN_RECV) reqs */
163 	int		max_qlen;	/* != 0 iff TFO is currently enabled */
164 };
165 
166 /** struct request_sock_queue - queue of request_socks
167  *
168  * @rskq_accept_head - FIFO head of established children
169  * @rskq_accept_tail - FIFO tail of established children
170  * @rskq_defer_accept - User waits for some data after accept()
171  * @syn_wait_lock - serializer
172  *
173  * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
174  * lock sock while browsing the listening hash (otherwise it's deadlock prone).
175  *
176  */
177 struct request_sock_queue {
178 	struct request_sock	*rskq_accept_head;
179 	struct request_sock	*rskq_accept_tail;
180 	u8			rskq_defer_accept;
181 	struct listen_sock	*listen_opt;
182 	struct fastopen_queue	*fastopenq; /* This is non-NULL iff TFO has been
183 					     * enabled on this listener. Check
184 					     * max_qlen != 0 in fastopen_queue
185 					     * to determine if TFO is enabled
186 					     * right at this moment.
187 					     */
188 
189 	/* temporary alignment, our goal is to get rid of this lock */
190 	spinlock_t		syn_wait_lock ____cacheline_aligned_in_smp;
191 };
192 
193 int reqsk_queue_alloc(struct request_sock_queue *queue,
194 		      unsigned int nr_table_entries);
195 
196 void __reqsk_queue_destroy(struct request_sock_queue *queue);
197 void reqsk_queue_destroy(struct request_sock_queue *queue);
198 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
199 			   bool reset);
200 
201 static inline struct request_sock *
202 	reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
203 {
204 	struct request_sock *req = queue->rskq_accept_head;
205 
206 	queue->rskq_accept_head = NULL;
207 	return req;
208 }
209 
210 static inline int reqsk_queue_empty(struct request_sock_queue *queue)
211 {
212 	return queue->rskq_accept_head == NULL;
213 }
214 
215 static inline void reqsk_queue_add(struct request_sock_queue *queue,
216 				   struct request_sock *req,
217 				   struct sock *parent,
218 				   struct sock *child)
219 {
220 	req->sk = child;
221 	sk_acceptq_added(parent);
222 
223 	if (queue->rskq_accept_head == NULL)
224 		queue->rskq_accept_head = req;
225 	else
226 		queue->rskq_accept_tail->dl_next = req;
227 
228 	queue->rskq_accept_tail = req;
229 	req->dl_next = NULL;
230 }
231 
232 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
233 {
234 	struct request_sock *req = queue->rskq_accept_head;
235 
236 	WARN_ON(req == NULL);
237 
238 	queue->rskq_accept_head = req->dl_next;
239 	if (queue->rskq_accept_head == NULL)
240 		queue->rskq_accept_tail = NULL;
241 
242 	return req;
243 }
244 
245 static inline void reqsk_queue_removed(struct request_sock_queue *queue,
246 				       const struct request_sock *req)
247 {
248 	struct listen_sock *lopt = queue->listen_opt;
249 
250 	if (req->num_timeout == 0)
251 		atomic_inc(&lopt->young_dec);
252 	atomic_inc(&lopt->qlen_dec);
253 }
254 
255 static inline void reqsk_queue_added(struct request_sock_queue *queue)
256 {
257 	struct listen_sock *lopt = queue->listen_opt;
258 
259 	lopt->young_inc++;
260 	lopt->qlen_inc++;
261 }
262 
263 static inline int listen_sock_qlen(const struct listen_sock *lopt)
264 {
265 	return lopt->qlen_inc - atomic_read(&lopt->qlen_dec);
266 }
267 
268 static inline int listen_sock_young(const struct listen_sock *lopt)
269 {
270 	return lopt->young_inc - atomic_read(&lopt->young_dec);
271 }
272 
273 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
274 {
275 	const struct listen_sock *lopt = queue->listen_opt;
276 
277 	return lopt ? listen_sock_qlen(lopt) : 0;
278 }
279 
280 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
281 {
282 	return listen_sock_young(queue->listen_opt);
283 }
284 
285 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
286 {
287 	return reqsk_queue_len(queue) >> queue->listen_opt->max_qlen_log;
288 }
289 
290 void reqsk_queue_hash_req(struct request_sock_queue *queue,
291 			  u32 hash, struct request_sock *req,
292 			  unsigned long timeout);
293 
294 #endif /* _REQUEST_SOCK_H */
295