[MLSXFRM]: Auto-labeling of child sockets
[linux-2.6.git] / include / net / request_sock.h
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
22 #include <net/sock.h>
23
24 struct request_sock;
25 struct sk_buff;
26 struct dst_entry;
27 struct proto;
28
29 struct request_sock_ops {
30         int             family;
31         kmem_cache_t    *slab;
32         int             obj_size;
33         int             (*rtx_syn_ack)(struct sock *sk,
34                                        struct request_sock *req,
35                                        struct dst_entry *dst);
36         void            (*send_ack)(struct sk_buff *skb,
37                                     struct request_sock *req);
38         void            (*send_reset)(struct sk_buff *skb);
39         void            (*destructor)(struct request_sock *req);
40 };
41
42 /* struct request_sock - mini sock to represent a connection request
43  */
44 struct request_sock {
45         struct request_sock             *dl_next; /* Must be first member! */
46         u16                             mss;
47         u8                              retrans;
48         u8                              __pad;
49         /* The following two fields can be easily recomputed I think -AK */
50         u32                             window_clamp; /* window clamp at creation time */
51         u32                             rcv_wnd;          /* rcv_wnd offered first time */
52         u32                             ts_recent;
53         unsigned long                   expires;
54         struct request_sock_ops         *rsk_ops;
55         struct sock                     *sk;
56         u32                             secid;
57 };
58
59 static inline struct request_sock *reqsk_alloc(struct request_sock_ops *ops)
60 {
61         struct request_sock *req = kmem_cache_alloc(ops->slab, SLAB_ATOMIC);
62
63         if (req != NULL)
64                 req->rsk_ops = ops;
65
66         return req;
67 }
68
69 static inline void __reqsk_free(struct request_sock *req)
70 {
71         kmem_cache_free(req->rsk_ops->slab, req);
72 }
73
74 static inline void reqsk_free(struct request_sock *req)
75 {
76         req->rsk_ops->destructor(req);
77         __reqsk_free(req);
78 }
79
80 extern int sysctl_max_syn_backlog;
81
82 /** struct listen_sock - listen state
83  *
84  * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
85  */
86 struct listen_sock {
87         u8                      max_qlen_log;
88         /* 3 bytes hole, try to use */
89         int                     qlen;
90         int                     qlen_young;
91         int                     clock_hand;
92         u32                     hash_rnd;
93         u32                     nr_table_entries;
94         struct request_sock     *syn_table[0];
95 };
96
97 /** struct request_sock_queue - queue of request_socks
98  *
99  * @rskq_accept_head - FIFO head of established children
100  * @rskq_accept_tail - FIFO tail of established children
101  * @rskq_defer_accept - User waits for some data after accept()
102  * @syn_wait_lock - serializer
103  *
104  * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
105  * lock sock while browsing the listening hash (otherwise it's deadlock prone).
106  *
107  * This lock is acquired in read mode only from listening_get_next() seq_file
108  * op and it's acquired in write mode _only_ from code that is actively
109  * changing rskq_accept_head. All readers that are holding the master sock lock
110  * don't need to grab this lock in read mode too as rskq_accept_head. writes
111  * are always protected from the main sock lock.
112  */
113 struct request_sock_queue {
114         struct request_sock     *rskq_accept_head;
115         struct request_sock     *rskq_accept_tail;
116         rwlock_t                syn_wait_lock;
117         u8                      rskq_defer_accept;
118         /* 3 bytes hole, try to pack */
119         struct listen_sock      *listen_opt;
120 };
121
122 extern int reqsk_queue_alloc(struct request_sock_queue *queue,
123                              const int nr_table_entries);
124
125 static inline struct listen_sock *reqsk_queue_yank_listen_sk(struct request_sock_queue *queue)
126 {
127         struct listen_sock *lopt;
128
129         write_lock_bh(&queue->syn_wait_lock);
130         lopt = queue->listen_opt;
131         queue->listen_opt = NULL;
132         write_unlock_bh(&queue->syn_wait_lock);
133
134         return lopt;
135 }
136
137 static inline void __reqsk_queue_destroy(struct request_sock_queue *queue)
138 {
139         kfree(reqsk_queue_yank_listen_sk(queue));
140 }
141
142 extern void reqsk_queue_destroy(struct request_sock_queue *queue);
143
144 static inline struct request_sock *
145         reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
146 {
147         struct request_sock *req = queue->rskq_accept_head;
148
149         queue->rskq_accept_head = NULL;
150         return req;
151 }
152
153 static inline int reqsk_queue_empty(struct request_sock_queue *queue)
154 {
155         return queue->rskq_accept_head == NULL;
156 }
157
158 static inline void reqsk_queue_unlink(struct request_sock_queue *queue,
159                                       struct request_sock *req,
160                                       struct request_sock **prev_req)
161 {
162         write_lock(&queue->syn_wait_lock);
163         *prev_req = req->dl_next;
164         write_unlock(&queue->syn_wait_lock);
165 }
166
167 static inline void reqsk_queue_add(struct request_sock_queue *queue,
168                                    struct request_sock *req,
169                                    struct sock *parent,
170                                    struct sock *child)
171 {
172         req->sk = child;
173         sk_acceptq_added(parent);
174
175         if (queue->rskq_accept_head == NULL)
176                 queue->rskq_accept_head = req;
177         else
178                 queue->rskq_accept_tail->dl_next = req;
179
180         queue->rskq_accept_tail = req;
181         req->dl_next = NULL;
182 }
183
184 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
185 {
186         struct request_sock *req = queue->rskq_accept_head;
187
188         BUG_TRAP(req != NULL);
189
190         queue->rskq_accept_head = req->dl_next;
191         if (queue->rskq_accept_head == NULL)
192                 queue->rskq_accept_tail = NULL;
193
194         return req;
195 }
196
197 static inline struct sock *reqsk_queue_get_child(struct request_sock_queue *queue,
198                                                  struct sock *parent)
199 {
200         struct request_sock *req = reqsk_queue_remove(queue);
201         struct sock *child = req->sk;
202
203         BUG_TRAP(child != NULL);
204
205         sk_acceptq_removed(parent);
206         __reqsk_free(req);
207         return child;
208 }
209
210 static inline int reqsk_queue_removed(struct request_sock_queue *queue,
211                                       struct request_sock *req)
212 {
213         struct listen_sock *lopt = queue->listen_opt;
214
215         if (req->retrans == 0)
216                 --lopt->qlen_young;
217
218         return --lopt->qlen;
219 }
220
221 static inline int reqsk_queue_added(struct request_sock_queue *queue)
222 {
223         struct listen_sock *lopt = queue->listen_opt;
224         const int prev_qlen = lopt->qlen;
225
226         lopt->qlen_young++;
227         lopt->qlen++;
228         return prev_qlen;
229 }
230
231 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
232 {
233         return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0;
234 }
235
236 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
237 {
238         return queue->listen_opt->qlen_young;
239 }
240
241 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
242 {
243         return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log;
244 }
245
246 static inline void reqsk_queue_hash_req(struct request_sock_queue *queue,
247                                         u32 hash, struct request_sock *req,
248                                         unsigned long timeout)
249 {
250         struct listen_sock *lopt = queue->listen_opt;
251
252         req->expires = jiffies + timeout;
253         req->retrans = 0;
254         req->sk = NULL;
255         req->dl_next = lopt->syn_table[hash];
256
257         write_lock(&queue->syn_wait_lock);
258         lopt->syn_table[hash] = req;
259         write_unlock(&queue->syn_wait_lock);
260 }
261
262 #endif /* _REQUEST_SOCK_H */