2 * NET Generic infrastructure for Network protocols.
4 * Definitions for request_sock
6 * Authors: Arnaldo Carvalho de Melo <acme@conectiva.com.br>
8 * From code originally in include/net/tcp.h
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.
15 #ifndef _REQUEST_SOCK_H
16 #define _REQUEST_SOCK_H
18 #include <linux/slab.h>
19 #include <linux/spinlock.h>
20 #include <linux/types.h>
21 #include <linux/bug.h>
30 struct request_sock_ops {
33 struct kmem_cache *slab;
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,
41 void (*destructor)(struct request_sock *req);
42 void (*syn_ack_timeout)(struct sock *sk,
43 struct request_sock *req);
46 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req);
48 /* struct request_sock - mini sock to represent a connection request
51 struct sock_common __req_common;
52 #define rsk_refcnt __req_common.skc_refcnt
54 struct request_sock *dl_next;
55 struct sock *rsk_listener;
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 */
64 unsigned long expires;
65 const struct request_sock_ops *rsk_ops;
71 static inline struct request_sock *
72 reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener)
74 struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC);
78 sock_hold(sk_listener);
79 req->rsk_listener = sk_listener;
81 /* Following is temporary. It is coupled with debugging
82 * helpers in reqsk_put() & reqsk_free()
84 atomic_set(&req->rsk_refcnt, 0);
89 static inline struct request_sock *inet_reqsk(struct sock *sk)
91 return (struct request_sock *)sk;
94 static inline void reqsk_free(struct request_sock *req)
96 /* temporary debugging */
97 WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0);
99 req->rsk_ops->destructor(req);
100 if (req->rsk_listener)
101 sock_put(req->rsk_listener);
102 kmem_cache_free(req->rsk_ops->slab, req);
105 static inline void reqsk_put(struct request_sock *req)
107 /* temporary debugging, until req sock are put into ehash table */
108 WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 1);
110 if (atomic_dec_and_test(&req->rsk_refcnt))
114 extern int sysctl_max_syn_backlog;
116 /** struct listen_sock - listen state
118 * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
123 /* 2 bytes hole, try to use */
128 u32 nr_table_entries;
129 struct request_sock *syn_table[0];
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.
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.
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.
156 int qlen; /* # of pending (TCP_SYN_RECV) reqs */
157 int max_qlen; /* != 0 iff TFO is currently enabled */
160 /** struct request_sock_queue - queue of request_socks
162 * @rskq_accept_head - FIFO head of established children
163 * @rskq_accept_tail - FIFO tail of established children
164 * @rskq_defer_accept - User waits for some data after accept()
165 * @syn_wait_lock - serializer
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).
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.
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;
180 u8 rskq_defer_accept;
181 /* 3 bytes hole, try to pack */
182 struct listen_sock *listen_opt;
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.
191 int reqsk_queue_alloc(struct request_sock_queue *queue,
192 unsigned int nr_table_entries);
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,
199 static inline struct request_sock *
200 reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
202 struct request_sock *req = queue->rskq_accept_head;
204 queue->rskq_accept_head = NULL;
208 static inline int reqsk_queue_empty(struct request_sock_queue *queue)
210 return queue->rskq_accept_head == NULL;
213 static inline void reqsk_queue_unlink(struct request_sock_queue *queue,
214 struct request_sock *req,
215 struct request_sock **prev_req)
217 write_lock(&queue->syn_wait_lock);
218 *prev_req = req->dl_next;
219 write_unlock(&queue->syn_wait_lock);
222 static inline void reqsk_queue_add(struct request_sock_queue *queue,
223 struct request_sock *req,
228 sk_acceptq_added(parent);
230 if (queue->rskq_accept_head == NULL)
231 queue->rskq_accept_head = req;
233 queue->rskq_accept_tail->dl_next = req;
235 queue->rskq_accept_tail = req;
239 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
241 struct request_sock *req = queue->rskq_accept_head;
243 WARN_ON(req == NULL);
245 queue->rskq_accept_head = req->dl_next;
246 if (queue->rskq_accept_head == NULL)
247 queue->rskq_accept_tail = NULL;
252 static inline int reqsk_queue_removed(struct request_sock_queue *queue,
253 struct request_sock *req)
255 struct listen_sock *lopt = queue->listen_opt;
257 if (req->num_timeout == 0)
263 static inline int reqsk_queue_added(struct request_sock_queue *queue)
265 struct listen_sock *lopt = queue->listen_opt;
266 const int prev_qlen = lopt->qlen;
273 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
275 return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0;
278 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
280 return queue->listen_opt->qlen_young;
283 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
285 return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log;
288 static inline void reqsk_queue_hash_req(struct request_sock_queue *queue,
289 u32 hash, struct request_sock *req,
290 unsigned long timeout)
292 struct listen_sock *lopt = queue->listen_opt;
294 req->expires = jiffies + timeout;
295 req->num_retrans = 0;
296 req->num_timeout = 0;
298 req->dl_next = lopt->syn_table[hash];
300 /* before letting lookups find us, make sure all req fields
301 * are committed to memory and refcnt initialized.
304 atomic_set(&req->rsk_refcnt, 1);
306 write_lock(&queue->syn_wait_lock);
307 lopt->syn_table[hash] = req;
308 write_unlock(&queue->syn_wait_lock);
311 #endif /* _REQUEST_SOCK_H */