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); |
72659ecc OP |
42 | void (*syn_ack_timeout)(struct sock *sk, |
43 | struct request_sock *req); | |
2e6599cb ACM |
44 | }; |
45 | ||
c0f4502a | 46 | int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req); |
e6c022a4 | 47 | |
60236fdd | 48 | /* struct request_sock - mini sock to represent a connection request |
2e6599cb | 49 | */ |
60236fdd | 50 | struct request_sock { |
634fb979 | 51 | struct sock_common __req_common; |
1e2e0117 ED |
52 | #define rsk_refcnt __req_common.skc_refcnt |
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; | |
64 | unsigned long expires; | |
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 | ||
60236fdd | 94 | static inline void reqsk_free(struct request_sock *req) |
2e6599cb | 95 | { |
13854e5a ED |
96 | /* temporary debugging */ |
97 | WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0); | |
98 | ||
60236fdd | 99 | req->rsk_ops->destructor(req); |
4e9a578e ED |
100 | if (req->rsk_listener) |
101 | sock_put(req->rsk_listener); | |
13854e5a | 102 | kmem_cache_free(req->rsk_ops->slab, req); |
2e6599cb ACM |
103 | } |
104 | ||
1e2e0117 ED |
105 | static inline void reqsk_put(struct request_sock *req) |
106 | { | |
13854e5a ED |
107 | /* temporary debugging, until req sock are put into ehash table */ |
108 | WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 1); | |
109 | ||
1e2e0117 ED |
110 | if (atomic_dec_and_test(&req->rsk_refcnt)) |
111 | reqsk_free(req); | |
112 | } | |
113 | ||
0e87506f ACM |
114 | extern int sysctl_max_syn_backlog; |
115 | ||
2ad69c55 | 116 | /** struct listen_sock - listen state |
0e87506f ACM |
117 | * |
118 | * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs | |
119 | */ | |
2ad69c55 | 120 | struct listen_sock { |
0e87506f | 121 | u8 max_qlen_log; |
946cedcc ED |
122 | u8 synflood_warned; |
123 | /* 2 bytes hole, try to use */ | |
0e87506f ACM |
124 | int qlen; |
125 | int qlen_young; | |
126 | int clock_hand; | |
127 | u32 hash_rnd; | |
83e3609e | 128 | u32 nr_table_entries; |
0e87506f ACM |
129 | struct request_sock *syn_table[0]; |
130 | }; | |
131 | ||
10467163 JC |
132 | /* |
133 | * For a TCP Fast Open listener - | |
134 | * lock - protects the access to all the reqsk, which is co-owned by | |
135 | * the listener and the child socket. | |
136 | * qlen - pending TFO requests (still in TCP_SYN_RECV). | |
137 | * max_qlen - max TFO reqs allowed before TFO is disabled. | |
138 | * | |
139 | * XXX (TFO) - ideally these fields can be made as part of "listen_sock" | |
140 | * structure above. But there is some implementation difficulty due to | |
141 | * listen_sock being part of request_sock_queue hence will be freed when | |
142 | * a listener is stopped. But TFO related fields may continue to be | |
143 | * accessed even after a listener is closed, until its sk_refcnt drops | |
144 | * to 0 implying no more outstanding TFO reqs. One solution is to keep | |
145 | * listen_opt around until sk_refcnt drops to 0. But there is some other | |
146 | * complexity that needs to be resolved. E.g., a listener can be disabled | |
147 | * temporarily through shutdown()->tcp_disconnect(), and re-enabled later. | |
148 | */ | |
149 | struct fastopen_queue { | |
150 | struct request_sock *rskq_rst_head; /* Keep track of past TFO */ | |
151 | struct request_sock *rskq_rst_tail; /* requests that caused RST. | |
152 | * This is part of the defense | |
153 | * against spoofing attack. | |
154 | */ | |
155 | spinlock_t lock; | |
156 | int qlen; /* # of pending (TCP_SYN_RECV) reqs */ | |
157 | int max_qlen; /* != 0 iff TFO is currently enabled */ | |
158 | }; | |
159 | ||
0e87506f ACM |
160 | /** struct request_sock_queue - queue of request_socks |
161 | * | |
162 | * @rskq_accept_head - FIFO head of established children | |
163 | * @rskq_accept_tail - FIFO tail of established children | |
295f7324 | 164 | * @rskq_defer_accept - User waits for some data after accept() |
0e87506f ACM |
165 | * @syn_wait_lock - serializer |
166 | * | |
167 | * %syn_wait_lock is necessary only to avoid proc interface having to grab the main | |
168 | * lock sock while browsing the listening hash (otherwise it's deadlock prone). | |
169 | * | |
170 | * This lock is acquired in read mode only from listening_get_next() seq_file | |
171 | * op and it's acquired in write mode _only_ from code that is actively | |
172 | * changing rskq_accept_head. All readers that are holding the master sock lock | |
173 | * don't need to grab this lock in read mode too as rskq_accept_head. writes | |
174 | * are always protected from the main sock lock. | |
175 | */ | |
176 | struct request_sock_queue { | |
177 | struct request_sock *rskq_accept_head; | |
178 | struct request_sock *rskq_accept_tail; | |
179 | rwlock_t syn_wait_lock; | |
ec0a1966 DM |
180 | u8 rskq_defer_accept; |
181 | /* 3 bytes hole, try to pack */ | |
2ad69c55 | 182 | struct listen_sock *listen_opt; |
10467163 JC |
183 | struct fastopen_queue *fastopenq; /* This is non-NULL iff TFO has been |
184 | * enabled on this listener. Check | |
185 | * max_qlen != 0 in fastopen_queue | |
186 | * to determine if TFO is enabled | |
187 | * right at this moment. | |
188 | */ | |
0e87506f ACM |
189 | }; |
190 | ||
c0f4502a JP |
191 | int reqsk_queue_alloc(struct request_sock_queue *queue, |
192 | unsigned int nr_table_entries); | |
0e87506f | 193 | |
c0f4502a JP |
194 | void __reqsk_queue_destroy(struct request_sock_queue *queue); |
195 | void reqsk_queue_destroy(struct request_sock_queue *queue); | |
196 | void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req, | |
197 | bool reset); | |
83e3609e | 198 | |
0e87506f ACM |
199 | static inline struct request_sock * |
200 | reqsk_queue_yank_acceptq(struct request_sock_queue *queue) | |
201 | { | |
202 | struct request_sock *req = queue->rskq_accept_head; | |
203 | ||
3eb4801d | 204 | queue->rskq_accept_head = NULL; |
0e87506f ACM |
205 | return req; |
206 | } | |
207 | ||
208 | static inline int reqsk_queue_empty(struct request_sock_queue *queue) | |
209 | { | |
210 | return queue->rskq_accept_head == NULL; | |
211 | } | |
212 | ||
213 | static inline void reqsk_queue_unlink(struct request_sock_queue *queue, | |
214 | struct request_sock *req, | |
215 | struct request_sock **prev_req) | |
216 | { | |
217 | write_lock(&queue->syn_wait_lock); | |
218 | *prev_req = req->dl_next; | |
219 | write_unlock(&queue->syn_wait_lock); | |
220 | } | |
221 | ||
222 | static inline void reqsk_queue_add(struct request_sock_queue *queue, | |
223 | struct request_sock *req, | |
224 | struct sock *parent, | |
225 | struct sock *child) | |
226 | { | |
227 | req->sk = child; | |
228 | sk_acceptq_added(parent); | |
229 | ||
230 | if (queue->rskq_accept_head == NULL) | |
231 | queue->rskq_accept_head = req; | |
232 | else | |
233 | queue->rskq_accept_tail->dl_next = req; | |
234 | ||
235 | queue->rskq_accept_tail = req; | |
236 | req->dl_next = NULL; | |
237 | } | |
238 | ||
239 | static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue) | |
240 | { | |
241 | struct request_sock *req = queue->rskq_accept_head; | |
242 | ||
547b792c | 243 | WARN_ON(req == NULL); |
0e87506f ACM |
244 | |
245 | queue->rskq_accept_head = req->dl_next; | |
246 | if (queue->rskq_accept_head == NULL) | |
247 | queue->rskq_accept_tail = NULL; | |
248 | ||
249 | return req; | |
250 | } | |
251 | ||
0e87506f ACM |
252 | static inline int reqsk_queue_removed(struct request_sock_queue *queue, |
253 | struct request_sock *req) | |
254 | { | |
2ad69c55 | 255 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f | 256 | |
e6c022a4 | 257 | if (req->num_timeout == 0) |
0e87506f ACM |
258 | --lopt->qlen_young; |
259 | ||
260 | return --lopt->qlen; | |
261 | } | |
262 | ||
263 | static inline int reqsk_queue_added(struct request_sock_queue *queue) | |
264 | { | |
2ad69c55 | 265 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f ACM |
266 | const int prev_qlen = lopt->qlen; |
267 | ||
268 | lopt->qlen_young++; | |
269 | lopt->qlen++; | |
270 | return prev_qlen; | |
271 | } | |
272 | ||
463c84b9 | 273 | static inline int reqsk_queue_len(const struct request_sock_queue *queue) |
0e87506f ACM |
274 | { |
275 | return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0; | |
276 | } | |
277 | ||
463c84b9 | 278 | static inline int reqsk_queue_len_young(const struct request_sock_queue *queue) |
0e87506f ACM |
279 | { |
280 | return queue->listen_opt->qlen_young; | |
281 | } | |
282 | ||
463c84b9 | 283 | static inline int reqsk_queue_is_full(const struct request_sock_queue *queue) |
0e87506f ACM |
284 | { |
285 | return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log; | |
286 | } | |
287 | ||
288 | static inline void reqsk_queue_hash_req(struct request_sock_queue *queue, | |
289 | u32 hash, struct request_sock *req, | |
8129765a | 290 | unsigned long timeout) |
0e87506f | 291 | { |
2ad69c55 | 292 | struct listen_sock *lopt = queue->listen_opt; |
0e87506f ACM |
293 | |
294 | req->expires = jiffies + timeout; | |
e6c022a4 ED |
295 | req->num_retrans = 0; |
296 | req->num_timeout = 0; | |
0e87506f ACM |
297 | req->sk = NULL; |
298 | req->dl_next = lopt->syn_table[hash]; | |
299 | ||
0470c8ca ED |
300 | /* before letting lookups find us, make sure all req fields |
301 | * are committed to memory and refcnt initialized. | |
302 | */ | |
303 | smp_wmb(); | |
304 | atomic_set(&req->rsk_refcnt, 1); | |
305 | ||
0e87506f ACM |
306 | write_lock(&queue->syn_wait_lock); |
307 | lopt->syn_table[hash] = req; | |
308 | write_unlock(&queue->syn_wait_lock); | |
309 | } | |
310 | ||
2e6599cb | 311 | #endif /* _REQUEST_SOCK_H */ |