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 #include <linux/bug.h> 22 23 #include <net/sock.h> 24 25 struct request_sock; 26 struct sk_buff; 27 struct dst_entry; 28 struct proto; 29 30 struct request_sock_ops { 31 int family; 32 int obj_size; 33 struct kmem_cache *slab; 34 char *slab_name; 35 int (*rtx_syn_ack)(const struct sock *sk, 36 struct request_sock *req); 37 void (*send_ack)(const struct sock *sk, struct sk_buff *skb, 38 struct request_sock *req); 39 void (*send_reset)(const struct sock *sk, 40 struct sk_buff *skb); 41 void (*destructor)(struct request_sock *req); 42 void (*syn_ack_timeout)(const struct request_sock *req); 43 }; 44 45 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req); 46 47 /* struct request_sock - mini sock to represent a connection request 48 */ 49 struct request_sock { 50 struct sock_common __req_common; 51 #define rsk_refcnt __req_common.skc_refcnt 52 #define rsk_hash __req_common.skc_hash 53 54 struct request_sock *dl_next; 55 struct sock *rsk_listener; 56 u16 mss; 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 */ 63 u32 ts_recent; 64 struct timer_list rsk_timer; 65 const struct request_sock_ops *rsk_ops; 66 struct sock *sk; 67 u32 *saved_syn; 68 u32 secid; 69 u32 peer_secid; 70 }; 71 72 static inline struct request_sock * 73 reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener) 74 { 75 struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC); 76 77 if (req) { 78 req->rsk_ops = ops; 79 sock_hold(sk_listener); 80 req->rsk_listener = sk_listener; 81 req->saved_syn = NULL; 82 /* Following is temporary. It is coupled with debugging 83 * helpers in reqsk_put() & reqsk_free() 84 */ 85 atomic_set(&req->rsk_refcnt, 0); 86 } 87 return req; 88 } 89 90 static inline struct request_sock *inet_reqsk(struct sock *sk) 91 { 92 return (struct request_sock *)sk; 93 } 94 95 static inline struct sock *req_to_sk(struct request_sock *req) 96 { 97 return (struct sock *)req; 98 } 99 100 static inline void reqsk_free(struct request_sock *req) 101 { 102 /* temporary debugging */ 103 WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0); 104 105 req->rsk_ops->destructor(req); 106 if (req->rsk_listener) 107 sock_put(req->rsk_listener); 108 kfree(req->saved_syn); 109 kmem_cache_free(req->rsk_ops->slab, req); 110 } 111 112 static inline void reqsk_put(struct request_sock *req) 113 { 114 if (atomic_dec_and_test(&req->rsk_refcnt)) 115 reqsk_free(req); 116 } 117 118 extern int sysctl_max_syn_backlog; 119 120 /** struct listen_sock - listen state 121 * 122 * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs 123 */ 124 struct listen_sock { 125 int qlen_inc; /* protected by listener lock */ 126 int young_inc;/* protected by listener lock */ 127 128 /* following fields can be updated by timer */ 129 atomic_t qlen_dec; /* qlen = qlen_inc - qlen_dec */ 130 atomic_t young_dec; 131 132 u32 max_qlen_log ____cacheline_aligned_in_smp; 133 u32 synflood_warned; 134 u32 hash_rnd; 135 u32 nr_table_entries; 136 struct request_sock *syn_table[0]; 137 }; 138 139 /* 140 * For a TCP Fast Open listener - 141 * lock - protects the access to all the reqsk, which is co-owned by 142 * the listener and the child socket. 143 * qlen - pending TFO requests (still in TCP_SYN_RECV). 144 * max_qlen - max TFO reqs allowed before TFO is disabled. 145 * 146 * XXX (TFO) - ideally these fields can be made as part of "listen_sock" 147 * structure above. But there is some implementation difficulty due to 148 * listen_sock being part of request_sock_queue hence will be freed when 149 * a listener is stopped. But TFO related fields may continue to be 150 * accessed even after a listener is closed, until its sk_refcnt drops 151 * to 0 implying no more outstanding TFO reqs. One solution is to keep 152 * listen_opt around until sk_refcnt drops to 0. But there is some other 153 * complexity that needs to be resolved. E.g., a listener can be disabled 154 * temporarily through shutdown()->tcp_disconnect(), and re-enabled later. 155 */ 156 struct fastopen_queue { 157 struct request_sock *rskq_rst_head; /* Keep track of past TFO */ 158 struct request_sock *rskq_rst_tail; /* requests that caused RST. 159 * This is part of the defense 160 * against spoofing attack. 161 */ 162 spinlock_t lock; 163 int qlen; /* # of pending (TCP_SYN_RECV) reqs */ 164 int max_qlen; /* != 0 iff TFO is currently enabled */ 165 }; 166 167 /** struct request_sock_queue - queue of request_socks 168 * 169 * @rskq_accept_head - FIFO head of established children 170 * @rskq_accept_tail - FIFO tail of established children 171 * @rskq_defer_accept - User waits for some data after accept() 172 * @syn_wait_lock - serializer 173 * 174 * %syn_wait_lock is necessary only to avoid proc interface having to grab the main 175 * lock sock while browsing the listening hash (otherwise it's deadlock prone). 176 * 177 */ 178 struct request_sock_queue { 179 struct request_sock *rskq_accept_head; 180 struct request_sock *rskq_accept_tail; 181 u8 rskq_defer_accept; 182 struct listen_sock *listen_opt; 183 struct fastopen_queue fastopenq; /* Check max_qlen != 0 to determine 184 * if TFO is enabled. 185 */ 186 187 /* temporary alignment, our goal is to get rid of this lock */ 188 spinlock_t syn_wait_lock ____cacheline_aligned_in_smp; 189 }; 190 191 int reqsk_queue_alloc(struct request_sock_queue *queue, 192 unsigned int nr_table_entries); 193 194 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); 198 199 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; 203 204 queue->rskq_accept_head = NULL; 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_add(struct request_sock_queue *queue, 214 struct request_sock *req, 215 struct sock *parent, 216 struct sock *child) 217 { 218 req->sk = child; 219 sk_acceptq_added(parent); 220 221 if (queue->rskq_accept_head == NULL) 222 queue->rskq_accept_head = req; 223 else 224 queue->rskq_accept_tail->dl_next = req; 225 226 queue->rskq_accept_tail = req; 227 req->dl_next = NULL; 228 } 229 230 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue) 231 { 232 struct request_sock *req = queue->rskq_accept_head; 233 234 WARN_ON(req == NULL); 235 236 queue->rskq_accept_head = req->dl_next; 237 if (queue->rskq_accept_head == NULL) 238 queue->rskq_accept_tail = NULL; 239 240 return req; 241 } 242 243 static inline void reqsk_queue_removed(struct request_sock_queue *queue, 244 const struct request_sock *req) 245 { 246 struct listen_sock *lopt = queue->listen_opt; 247 248 if (req->num_timeout == 0) 249 atomic_inc(&lopt->young_dec); 250 atomic_inc(&lopt->qlen_dec); 251 } 252 253 static inline void reqsk_queue_added(struct request_sock_queue *queue) 254 { 255 struct listen_sock *lopt = queue->listen_opt; 256 257 lopt->young_inc++; 258 lopt->qlen_inc++; 259 } 260 261 static inline int listen_sock_qlen(const struct listen_sock *lopt) 262 { 263 return lopt->qlen_inc - atomic_read(&lopt->qlen_dec); 264 } 265 266 static inline int listen_sock_young(const struct listen_sock *lopt) 267 { 268 return lopt->young_inc - atomic_read(&lopt->young_dec); 269 } 270 271 static inline int reqsk_queue_len(const struct request_sock_queue *queue) 272 { 273 const struct listen_sock *lopt = queue->listen_opt; 274 275 return lopt ? listen_sock_qlen(lopt) : 0; 276 } 277 278 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue) 279 { 280 return listen_sock_young(queue->listen_opt); 281 } 282 283 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue) 284 { 285 return reqsk_queue_len(queue) >> queue->listen_opt->max_qlen_log; 286 } 287 288 void reqsk_queue_hash_req(struct request_sock_queue *queue, 289 u32 hash, struct request_sock *req, 290 unsigned long timeout); 291 292 #endif /* _REQUEST_SOCK_H */ 293