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 | ||
60236fdd | 30 | struct request_sock_ops { |
2e6599cb | 31 | int family; |
2e6599cb | 32 | int obj_size; |
e18b890b | 33 | struct kmem_cache *slab; |
7e56b5d6 | 34 | char *slab_name; |
2e6599cb | 35 | int (*rtx_syn_ack)(struct sock *sk, |
1a2c6181 | 36 | struct request_sock *req); |
6edafaaf | 37 | void (*send_ack)(struct sock *sk, struct sk_buff *skb, |
60236fdd | 38 | struct request_sock *req); |
cfb6eeb4 YH |
39 | void (*send_reset)(struct sock *sk, |
40 | struct sk_buff *skb); | |
60236fdd | 41 | void (*destructor)(struct request_sock *req); |
42cb80a2 | 42 | void (*syn_ack_timeout)(const struct request_sock *req); |
2e6599cb ACM |
43 | }; |
44 | ||
c0f4502a | 45 | int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req); |
e6c022a4 | 46 | |
60236fdd | 47 | /* struct request_sock - mini sock to represent a connection request |
2e6599cb | 48 | */ |
60236fdd | 49 | struct request_sock { |
634fb979 | 50 | struct sock_common __req_common; |
1e2e0117 | 51 | #define rsk_refcnt __req_common.skc_refcnt |
52452c54 | 52 | #define rsk_hash __req_common.skc_hash |
1e2e0117 | 53 | |
3fb62c5d | 54 | struct request_sock *dl_next; |
4e9a578e | 55 | struct sock *rsk_listener; |
2e6599cb | 56 | u16 mss; |
e6c022a4 ED |
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 */ | |
2e6599cb ACM |
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; | |
fa76ce73 | 64 | struct timer_list rsk_timer; |
72a3effa | 65 | const struct request_sock_ops *rsk_ops; |
2e6599cb | 66 | struct sock *sk; |
4237c75c | 67 | u32 secid; |
6b877699 | 68 | u32 peer_secid; |
2e6599cb ACM |
69 | }; |
70 | ||
4e9a578e ED |
71 | static inline struct request_sock * |
72 | reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener) | |
2e6599cb | 73 | { |
54e6ecb2 | 74 | struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC); |
2e6599cb | 75 | |
4e9a578e | 76 | if (req) { |
60236fdd | 77 | req->rsk_ops = ops; |
4e9a578e ED |
78 | sock_hold(sk_listener); |
79 | req->rsk_listener = sk_listener; | |
0470c8ca ED |
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); | |
4e9a578e | 85 | } |
2e6599cb ACM |
86 | return req; |
87 | } | |
88 | ||
41b822c5 ED |
89 | static inline struct request_sock *inet_reqsk(struct sock *sk) |
90 | { | |
91 | return (struct request_sock *)sk; | |
92 | } | |
93 | ||
08d2cc3b ED |
94 | static inline struct sock *req_to_sk(struct request_sock *req) |
95 | { | |
96 | return (struct sock *)req; | |
97 | } | |
98 | ||
60236fdd | 99 | static inline void reqsk_free(struct request_sock *req) |
2e6599cb | 100 | { |
13854e5a ED |
101 | /* temporary debugging */ |
102 | WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0); | |
103 | ||
60236fdd | 104 | req->rsk_ops->destructor(req); |
4e9a578e ED |
105 | if (req->rsk_listener) |
106 | sock_put(req->rsk_listener); | |
13854e5a | 107 | kmem_cache_free(req->rsk_ops->slab, req); |
2e6599cb ACM |
108 | } |
109 | ||
1e2e0117 ED |
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 | ||
0e87506f ACM |
116 | extern int sysctl_max_syn_backlog; |
117 | ||
2ad69c55 | 118 | /** struct listen_sock - listen state |
0e87506f ACM |
119 | * |
120 | * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs | |
121 | */ | |
2ad69c55 | 122 | struct listen_sock { |
fa76ce73 ED |
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; | |
946cedcc ED |
131 | u8 synflood_warned; |
132 | /* 2 bytes hole, try to use */ | |
0e87506f | 133 | u32 hash_rnd; |
83e3609e | 134 | u32 nr_table_entries; |
0e87506f ACM |
135 | struct request_sock *syn_table[0]; |
136 | }; | |
137 | ||
10467163 JC |
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 | ||
0e87506f ACM |
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 | |
295f7324 | 170 | * @rskq_defer_accept - User waits for some data after accept() |
0e87506f ACM |
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 | * | |
0e87506f ACM |
176 | */ |
177 | struct request_sock_queue { | |
178 | struct request_sock *rskq_accept_head; | |
179 | struct request_sock *rskq_accept_tail; | |
ec0a1966 | 180 | u8 rskq_defer_accept; |
2ad69c55 | 181 | struct listen_sock *listen_opt; |
10467163 JC |
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 | */ | |
fa76ce73 ED |
188 | |
189 | /* temporary alignment, our goal is to get rid of this lock */ | |
b2827053 | 190 | spinlock_t syn_wait_lock ____cacheline_aligned_in_smp; |
0e87506f ACM |
191 | }; |
192 | ||
c0f4502a JP |
193 | int reqsk_queue_alloc(struct request_sock_queue *queue, |
194 | unsigned int nr_table_entries); | |
0e87506f | 195 | |
c0f4502a JP |
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); | |
83e3609e | 200 | |
0e87506f ACM |
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 | ||
3eb4801d | 206 | queue->rskq_accept_head = NULL; |
0e87506f ACM |
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 | ||
0e87506f ACM |
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 | ||
547b792c | 236 | WARN_ON(req == NULL); |
0e87506f ACM |
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 | ||
fa76ce73 ED |
245 | static inline void reqsk_queue_removed(struct request_sock_queue *queue, |
246 | const struct request_sock *req) | |
0e87506f | 247 | { |
2ad69c55 | 248 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f | 249 | |
e6c022a4 | 250 | if (req->num_timeout == 0) |
fa76ce73 ED |
251 | atomic_inc(&lopt->young_dec); |
252 | atomic_inc(&lopt->qlen_dec); | |
0e87506f ACM |
253 | } |
254 | ||
fa76ce73 | 255 | static inline void reqsk_queue_added(struct request_sock_queue *queue) |
0e87506f | 256 | { |
2ad69c55 | 257 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f | 258 | |
fa76ce73 ED |
259 | lopt->young_inc++; |
260 | lopt->qlen_inc++; | |
0e87506f ACM |
261 | } |
262 | ||
fa76ce73 | 263 | static inline int listen_sock_qlen(const struct listen_sock *lopt) |
0e87506f | 264 | { |
fa76ce73 | 265 | return lopt->qlen_inc - atomic_read(&lopt->qlen_dec); |
0e87506f ACM |
266 | } |
267 | ||
fa76ce73 | 268 | static inline int listen_sock_young(const struct listen_sock *lopt) |
0e87506f | 269 | { |
fa76ce73 | 270 | return lopt->young_inc - atomic_read(&lopt->young_dec); |
0e87506f ACM |
271 | } |
272 | ||
fa76ce73 | 273 | static inline int reqsk_queue_len(const struct request_sock_queue *queue) |
0e87506f | 274 | { |
fa76ce73 ED |
275 | const struct listen_sock *lopt = queue->listen_opt; |
276 | ||
277 | return lopt ? listen_sock_qlen(lopt) : 0; | |
0e87506f ACM |
278 | } |
279 | ||
fa76ce73 | 280 | static inline int reqsk_queue_len_young(const struct request_sock_queue *queue) |
0e87506f | 281 | { |
fa76ce73 ED |
282 | return listen_sock_young(queue->listen_opt); |
283 | } | |
0470c8ca | 284 | |
fa76ce73 ED |
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; | |
0e87506f ACM |
288 | } |
289 | ||
fa76ce73 ED |
290 | void reqsk_queue_hash_req(struct request_sock_queue *queue, |
291 | u32 hash, struct request_sock *req, | |
292 | unsigned long timeout); | |
293 | ||
2e6599cb | 294 | #endif /* _REQUEST_SOCK_H */ |