include/net/request_sock.h

   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)(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,
  40                                       struct sk_buff *skb);
  41         void            (*destructor)(struct request_sock *req);
  42         void            (*syn_ack_timeout)(struct sock *sk,
  43                                            struct request_sock *req);
  44 };
  45
  46 int inet_rtx_syn_ack(struct sock *parent, struct request_sock *req);
  47
  48 /* struct request_sock - mini sock to represent a connection request
  49  */
  50 struct request_sock {
  51         struct sock_common              __req_common;
  52 #define rsk_refcnt                      __req_common.skc_refcnt
  53 #define rsk_hash                        __req_common.skc_hash
  54
  55         struct request_sock             *dl_next;
  56         struct sock                     *rsk_listener;
  57         u16                             mss;
  58         u8                              num_retrans; /* number of retransmits */
  59         u8                              cookie_ts:1; /* syncookie: encode tcpopts in timestamp */
  60         u8                              num_timeout:7; /* number of timeouts */
  61         /* The following two fields can be easily recomputed I think -AK */
  62         u32                             window_clamp; /* window clamp at creation time */
  63         u32                             rcv_wnd;          /* rcv_wnd offered first time */
  64         u32                             ts_recent;
  65         struct timer_list               rsk_timer;
  66         const struct request_sock_ops   *rsk_ops;
  67         struct sock                     *sk;
  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
  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         kmem_cache_free(req->rsk_ops->slab, req);
 109 }
 110
 111 static inline void reqsk_put(struct request_sock *req)
 112 {
 113         if (atomic_dec_and_test(&req->rsk_refcnt))
 114                 reqsk_free(req);
 115 }
 116
 117 extern int sysctl_max_syn_backlog;
 118
 119 /** struct listen_sock - listen state
 120  *
 121  * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
 122  */
 123 struct listen_sock {
 124         int                     qlen_inc; /* protected by listener lock */
 125         int                     young_inc;/* protected by listener lock */
 126
 127         /* following fields can be updated by timer */
 128         atomic_t                qlen_dec; /* qlen = qlen_inc - qlen_dec */
 129         atomic_t                young_dec;
 130
 131         u8                      max_qlen_log ____cacheline_aligned_in_smp;
 132         u8                      synflood_warned;
 133         /* 2 bytes hole, try to use */
 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  * This lock is acquired in read mode only from listening_get_next() seq_file
 178  * op and it's acquired in write mode _only_ from code that is actively
 179  * changing rskq_accept_head. All readers that are holding the master sock lock
 180  * don't need to grab this lock in read mode too as rskq_accept_head. writes
 181  * are always protected from the main sock lock.
 182  */
 183 struct request_sock_queue {
 184         struct request_sock     *rskq_accept_head;
 185         struct request_sock     *rskq_accept_tail;
 186         u8                      rskq_defer_accept;
 187         struct listen_sock      *listen_opt;
 188         struct fastopen_queue   *fastopenq; /* This is non-NULL iff TFO has been
 189                                              * enabled on this listener. Check
 190                                              * max_qlen != 0 in fastopen_queue
 191                                              * to determine if TFO is enabled
 192                                              * right at this moment.
 193                                              */
 194
 195         /* temporary alignment, our goal is to get rid of this lock */
 196         rwlock_t                syn_wait_lock ____cacheline_aligned_in_smp;
 197 };
 198
 199 int reqsk_queue_alloc(struct request_sock_queue *queue,
 200                       unsigned int nr_table_entries);
 201
 202 void __reqsk_queue_destroy(struct request_sock_queue *queue);
 203 void reqsk_queue_destroy(struct request_sock_queue *queue);
 204 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
 205                            bool reset);
 206
 207 static inline struct request_sock *
 208         reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
 209 {
 210         struct request_sock *req = queue->rskq_accept_head;
 211
 212         queue->rskq_accept_head = NULL;
 213         return req;
 214 }
 215
 216 static inline int reqsk_queue_empty(struct request_sock_queue *queue)
 217 {
 218         return queue->rskq_accept_head == NULL;
 219 }
 220
 221 static inline void reqsk_queue_unlink(struct request_sock_queue *queue,
 222                                       struct request_sock *req)
 223 {
 224         struct listen_sock *lopt = queue->listen_opt;
 225         struct request_sock **prev;
 226
 227         write_lock(&queue->syn_wait_lock);
 228
 229         prev = &lopt->syn_table[req->rsk_hash];
 230         while (*prev != req)
 231                 prev = &(*prev)->dl_next;
 232         *prev = req->dl_next;
 233
 234         write_unlock(&queue->syn_wait_lock);
 235         if (del_timer(&req->rsk_timer))
 236                 reqsk_put(req);
 237 }
 238
 239 static inline void reqsk_queue_add(struct request_sock_queue *queue,
 240                                    struct request_sock *req,
 241                                    struct sock *parent,
 242                                    struct sock *child)
 243 {
 244         req->sk = child;
 245         sk_acceptq_added(parent);
 246
 247         if (queue->rskq_accept_head == NULL)
 248                 queue->rskq_accept_head = req;
 249         else
 250                 queue->rskq_accept_tail->dl_next = req;
 251
 252         queue->rskq_accept_tail = req;
 253         req->dl_next = NULL;
 254 }
 255
 256 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
 257 {
 258         struct request_sock *req = queue->rskq_accept_head;
 259
 260         WARN_ON(req == NULL);
 261
 262         queue->rskq_accept_head = req->dl_next;
 263         if (queue->rskq_accept_head == NULL)
 264                 queue->rskq_accept_tail = NULL;
 265
 266         return req;
 267 }
 268
 269 static inline void reqsk_queue_removed(struct request_sock_queue *queue,
 270                                        const struct request_sock *req)
 271 {
 272         struct listen_sock *lopt = queue->listen_opt;
 273
 274         if (req->num_timeout == 0)
 275                 atomic_inc(&lopt->young_dec);
 276         atomic_inc(&lopt->qlen_dec);
 277 }
 278
 279 static inline void reqsk_queue_added(struct request_sock_queue *queue)
 280 {
 281         struct listen_sock *lopt = queue->listen_opt;
 282
 283         lopt->young_inc++;
 284         lopt->qlen_inc++;
 285 }
 286
 287 static inline int listen_sock_qlen(const struct listen_sock *lopt)
 288 {
 289         return lopt->qlen_inc - atomic_read(&lopt->qlen_dec);
 290 }
 291
 292 static inline int listen_sock_young(const struct listen_sock *lopt)
 293 {
 294         return lopt->young_inc - atomic_read(&lopt->young_dec);
 295 }
 296
 297 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
 298 {
 299         const struct listen_sock *lopt = queue->listen_opt;
 300
 301         return lopt ? listen_sock_qlen(lopt) : 0;
 302 }
 303
 304 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
 305 {
 306         return listen_sock_young(queue->listen_opt);
 307 }
 308
 309 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
 310 {
 311         return reqsk_queue_len(queue) >> queue->listen_opt->max_qlen_log;
 312 }
 313
 314 void reqsk_queue_hash_req(struct request_sock_queue *queue,
 315                           u32 hash, struct request_sock *req,
 316                           unsigned long timeout);
 317
 318 #endif /* _REQUEST_SOCK_H */