Commit | Line | Data |
---|---|---|
2e6599cb ACM |
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> | |
0e87506f | 19 | #include <linux/spinlock.h> |
2e6599cb | 20 | #include <linux/types.h> |
547b792c | 21 | #include <linux/bug.h> |
0e87506f | 22 | |
2e6599cb ACM |
23 | #include <net/sock.h> |
24 | ||
60236fdd | 25 | struct request_sock; |
2e6599cb ACM |
26 | struct sk_buff; |
27 | struct dst_entry; | |
28 | struct proto; | |
29 | ||
e6b4d113 WAS |
30 | /* empty to "strongly type" an otherwise void parameter. |
31 | */ | |
32 | struct request_values { | |
33 | }; | |
34 | ||
60236fdd | 35 | struct request_sock_ops { |
2e6599cb | 36 | int family; |
2e6599cb | 37 | int obj_size; |
e18b890b | 38 | struct kmem_cache *slab; |
7e56b5d6 | 39 | char *slab_name; |
2e6599cb | 40 | int (*rtx_syn_ack)(struct sock *sk, |
e6b4d113 WAS |
41 | struct request_sock *req, |
42 | struct request_values *rvp); | |
6edafaaf | 43 | void (*send_ack)(struct sock *sk, struct sk_buff *skb, |
60236fdd | 44 | struct request_sock *req); |
cfb6eeb4 YH |
45 | void (*send_reset)(struct sock *sk, |
46 | struct sk_buff *skb); | |
60236fdd | 47 | void (*destructor)(struct request_sock *req); |
72659ecc OP |
48 | void (*syn_ack_timeout)(struct sock *sk, |
49 | struct request_sock *req); | |
2e6599cb ACM |
50 | }; |
51 | ||
60236fdd | 52 | /* struct request_sock - mini sock to represent a connection request |
2e6599cb | 53 | */ |
60236fdd ACM |
54 | struct request_sock { |
55 | struct request_sock *dl_next; /* Must be first member! */ | |
2e6599cb ACM |
56 | u16 mss; |
57 | u8 retrans; | |
4dfc2817 | 58 | u8 cookie_ts; /* syncookie: encode tcpopts in timestamp */ |
2e6599cb ACM |
59 | /* The following two fields can be easily recomputed I think -AK */ |
60 | u32 window_clamp; /* window clamp at creation time */ | |
61 | u32 rcv_wnd; /* rcv_wnd offered first time */ | |
62 | u32 ts_recent; | |
63 | unsigned long expires; | |
72a3effa | 64 | const struct request_sock_ops *rsk_ops; |
2e6599cb | 65 | struct sock *sk; |
4237c75c | 66 | u32 secid; |
6b877699 | 67 | u32 peer_secid; |
2e6599cb ACM |
68 | }; |
69 | ||
72a3effa | 70 | static inline struct request_sock *reqsk_alloc(const struct request_sock_ops *ops) |
2e6599cb | 71 | { |
54e6ecb2 | 72 | struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC); |
2e6599cb ACM |
73 | |
74 | if (req != NULL) | |
60236fdd | 75 | req->rsk_ops = ops; |
2e6599cb ACM |
76 | |
77 | return req; | |
78 | } | |
79 | ||
60236fdd | 80 | static inline void __reqsk_free(struct request_sock *req) |
2e6599cb | 81 | { |
60236fdd | 82 | kmem_cache_free(req->rsk_ops->slab, req); |
2e6599cb ACM |
83 | } |
84 | ||
60236fdd | 85 | static inline void reqsk_free(struct request_sock *req) |
2e6599cb | 86 | { |
60236fdd ACM |
87 | req->rsk_ops->destructor(req); |
88 | __reqsk_free(req); | |
2e6599cb ACM |
89 | } |
90 | ||
0e87506f ACM |
91 | extern int sysctl_max_syn_backlog; |
92 | ||
2ad69c55 | 93 | /** struct listen_sock - listen state |
0e87506f ACM |
94 | * |
95 | * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs | |
96 | */ | |
2ad69c55 | 97 | struct listen_sock { |
0e87506f | 98 | u8 max_qlen_log; |
946cedcc ED |
99 | u8 synflood_warned; |
100 | /* 2 bytes hole, try to use */ | |
0e87506f ACM |
101 | int qlen; |
102 | int qlen_young; | |
103 | int clock_hand; | |
104 | u32 hash_rnd; | |
83e3609e | 105 | u32 nr_table_entries; |
0e87506f ACM |
106 | struct request_sock *syn_table[0]; |
107 | }; | |
108 | ||
10467163 JC |
109 | /* |
110 | * For a TCP Fast Open listener - | |
111 | * lock - protects the access to all the reqsk, which is co-owned by | |
112 | * the listener and the child socket. | |
113 | * qlen - pending TFO requests (still in TCP_SYN_RECV). | |
114 | * max_qlen - max TFO reqs allowed before TFO is disabled. | |
115 | * | |
116 | * XXX (TFO) - ideally these fields can be made as part of "listen_sock" | |
117 | * structure above. But there is some implementation difficulty due to | |
118 | * listen_sock being part of request_sock_queue hence will be freed when | |
119 | * a listener is stopped. But TFO related fields may continue to be | |
120 | * accessed even after a listener is closed, until its sk_refcnt drops | |
121 | * to 0 implying no more outstanding TFO reqs. One solution is to keep | |
122 | * listen_opt around until sk_refcnt drops to 0. But there is some other | |
123 | * complexity that needs to be resolved. E.g., a listener can be disabled | |
124 | * temporarily through shutdown()->tcp_disconnect(), and re-enabled later. | |
125 | */ | |
126 | struct fastopen_queue { | |
127 | struct request_sock *rskq_rst_head; /* Keep track of past TFO */ | |
128 | struct request_sock *rskq_rst_tail; /* requests that caused RST. | |
129 | * This is part of the defense | |
130 | * against spoofing attack. | |
131 | */ | |
132 | spinlock_t lock; | |
133 | int qlen; /* # of pending (TCP_SYN_RECV) reqs */ | |
134 | int max_qlen; /* != 0 iff TFO is currently enabled */ | |
135 | }; | |
136 | ||
0e87506f ACM |
137 | /** struct request_sock_queue - queue of request_socks |
138 | * | |
139 | * @rskq_accept_head - FIFO head of established children | |
140 | * @rskq_accept_tail - FIFO tail of established children | |
295f7324 | 141 | * @rskq_defer_accept - User waits for some data after accept() |
0e87506f ACM |
142 | * @syn_wait_lock - serializer |
143 | * | |
144 | * %syn_wait_lock is necessary only to avoid proc interface having to grab the main | |
145 | * lock sock while browsing the listening hash (otherwise it's deadlock prone). | |
146 | * | |
147 | * This lock is acquired in read mode only from listening_get_next() seq_file | |
148 | * op and it's acquired in write mode _only_ from code that is actively | |
149 | * changing rskq_accept_head. All readers that are holding the master sock lock | |
150 | * don't need to grab this lock in read mode too as rskq_accept_head. writes | |
151 | * are always protected from the main sock lock. | |
152 | */ | |
153 | struct request_sock_queue { | |
154 | struct request_sock *rskq_accept_head; | |
155 | struct request_sock *rskq_accept_tail; | |
156 | rwlock_t syn_wait_lock; | |
ec0a1966 DM |
157 | u8 rskq_defer_accept; |
158 | /* 3 bytes hole, try to pack */ | |
2ad69c55 | 159 | struct listen_sock *listen_opt; |
10467163 JC |
160 | struct fastopen_queue *fastopenq; /* This is non-NULL iff TFO has been |
161 | * enabled on this listener. Check | |
162 | * max_qlen != 0 in fastopen_queue | |
163 | * to determine if TFO is enabled | |
164 | * right at this moment. | |
165 | */ | |
0e87506f ACM |
166 | }; |
167 | ||
168 | extern int reqsk_queue_alloc(struct request_sock_queue *queue, | |
72a3effa | 169 | unsigned int nr_table_entries); |
0e87506f | 170 | |
dab6ba36 | 171 | extern void __reqsk_queue_destroy(struct request_sock_queue *queue); |
83e3609e | 172 | extern void reqsk_queue_destroy(struct request_sock_queue *queue); |
10467163 JC |
173 | extern void reqsk_fastopen_remove(struct sock *sk, |
174 | struct request_sock *req, bool reset); | |
83e3609e | 175 | |
0e87506f ACM |
176 | static inline struct request_sock * |
177 | reqsk_queue_yank_acceptq(struct request_sock_queue *queue) | |
178 | { | |
179 | struct request_sock *req = queue->rskq_accept_head; | |
180 | ||
3eb4801d | 181 | queue->rskq_accept_head = NULL; |
0e87506f ACM |
182 | return req; |
183 | } | |
184 | ||
185 | static inline int reqsk_queue_empty(struct request_sock_queue *queue) | |
186 | { | |
187 | return queue->rskq_accept_head == NULL; | |
188 | } | |
189 | ||
190 | static inline void reqsk_queue_unlink(struct request_sock_queue *queue, | |
191 | struct request_sock *req, | |
192 | struct request_sock **prev_req) | |
193 | { | |
194 | write_lock(&queue->syn_wait_lock); | |
195 | *prev_req = req->dl_next; | |
196 | write_unlock(&queue->syn_wait_lock); | |
197 | } | |
198 | ||
199 | static inline void reqsk_queue_add(struct request_sock_queue *queue, | |
200 | struct request_sock *req, | |
201 | struct sock *parent, | |
202 | struct sock *child) | |
203 | { | |
204 | req->sk = child; | |
205 | sk_acceptq_added(parent); | |
206 | ||
207 | if (queue->rskq_accept_head == NULL) | |
208 | queue->rskq_accept_head = req; | |
209 | else | |
210 | queue->rskq_accept_tail->dl_next = req; | |
211 | ||
212 | queue->rskq_accept_tail = req; | |
213 | req->dl_next = NULL; | |
214 | } | |
215 | ||
216 | static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue) | |
217 | { | |
218 | struct request_sock *req = queue->rskq_accept_head; | |
219 | ||
547b792c | 220 | WARN_ON(req == NULL); |
0e87506f ACM |
221 | |
222 | queue->rskq_accept_head = req->dl_next; | |
223 | if (queue->rskq_accept_head == NULL) | |
224 | queue->rskq_accept_tail = NULL; | |
225 | ||
226 | return req; | |
227 | } | |
228 | ||
0e87506f ACM |
229 | static inline int reqsk_queue_removed(struct request_sock_queue *queue, |
230 | struct request_sock *req) | |
231 | { | |
2ad69c55 | 232 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f ACM |
233 | |
234 | if (req->retrans == 0) | |
235 | --lopt->qlen_young; | |
236 | ||
237 | return --lopt->qlen; | |
238 | } | |
239 | ||
240 | static inline int reqsk_queue_added(struct request_sock_queue *queue) | |
241 | { | |
2ad69c55 | 242 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f ACM |
243 | const int prev_qlen = lopt->qlen; |
244 | ||
245 | lopt->qlen_young++; | |
246 | lopt->qlen++; | |
247 | return prev_qlen; | |
248 | } | |
249 | ||
463c84b9 | 250 | static inline int reqsk_queue_len(const struct request_sock_queue *queue) |
0e87506f ACM |
251 | { |
252 | return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0; | |
253 | } | |
254 | ||
463c84b9 | 255 | static inline int reqsk_queue_len_young(const struct request_sock_queue *queue) |
0e87506f ACM |
256 | { |
257 | return queue->listen_opt->qlen_young; | |
258 | } | |
259 | ||
463c84b9 | 260 | static inline int reqsk_queue_is_full(const struct request_sock_queue *queue) |
0e87506f ACM |
261 | { |
262 | return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log; | |
263 | } | |
264 | ||
265 | static inline void reqsk_queue_hash_req(struct request_sock_queue *queue, | |
266 | u32 hash, struct request_sock *req, | |
8129765a | 267 | unsigned long timeout) |
0e87506f | 268 | { |
2ad69c55 | 269 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f ACM |
270 | |
271 | req->expires = jiffies + timeout; | |
272 | req->retrans = 0; | |
273 | req->sk = NULL; | |
274 | req->dl_next = lopt->syn_table[hash]; | |
275 | ||
276 | write_lock(&queue->syn_wait_lock); | |
277 | lopt->syn_table[hash] = req; | |
278 | write_unlock(&queue->syn_wait_lock); | |
279 | } | |
280 | ||
2e6599cb | 281 | #endif /* _REQUEST_SOCK_H */ |