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
  54         struct request_sock             *dl_next;
  55         u16                             mss;
  56         u8                              num_retrans; /* number of retransmits */
  57         u8                              cookie_ts:1; /* syncookie: encode tcpopts in timestamp */
  58         u8                              num_timeout:7; /* number of timeouts */
  59         /* The following two fields can be easily recomputed I think -AK */
  60         u32                             window_clamp; /* window clamp at creation time */
  61         u32                             rcv_wnd;          /* rcv_wnd offered first time */
  62         u32                             ts_recent;
  63         unsigned long                   expires;
  64         const struct request_sock_ops   *rsk_ops;
  65         struct sock                     *sk;
  66         u32                             secid;
  67         u32                             peer_secid;
  68 };
  69
  70 static inline struct request_sock *reqsk_alloc(const struct request_sock_ops *ops)
  71 {
  72         struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC);
  73
  74         if (req != NULL)
  75                 req->rsk_ops = ops;
  76
  77         return req;
  78 }
  79
  80 static inline struct request_sock *inet_reqsk(struct sock *sk)
  81 {
  82         return (struct request_sock *)sk;
  83 }
  84
  85 static inline void reqsk_free(struct request_sock *req)
  86 {
  87         /* temporary debugging */
  88         WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0);
  89
  90         req->rsk_ops->destructor(req);
  91         kmem_cache_free(req->rsk_ops->slab, req);
  92 }
  93
  94 static inline void reqsk_put(struct request_sock *req)
  95 {
  96         /* temporary debugging, until req sock are put into ehash table */
  97         WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 1);
  98
  99         if (atomic_dec_and_test(&req->rsk_refcnt))
 100                 reqsk_free(req);
 101 }
 102
 103 extern int sysctl_max_syn_backlog;
 104
 105 /** struct listen_sock - listen state
 106  *
 107  * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
 108  */
 109 struct listen_sock {
 110         u8                      max_qlen_log;
 111         u8                      synflood_warned;
 112         /* 2 bytes hole, try to use */
 113         int                     qlen;
 114         int                     qlen_young;
 115         int                     clock_hand;
 116         u32                     hash_rnd;
 117         u32                     nr_table_entries;
 118         struct request_sock     *syn_table[0];
 119 };
 120
 121 /*
 122  * For a TCP Fast Open listener -
 123  *      lock - protects the access to all the reqsk, which is co-owned by
 124  *              the listener and the child socket.
 125  *      qlen - pending TFO requests (still in TCP_SYN_RECV).
 126  *      max_qlen - max TFO reqs allowed before TFO is disabled.
 127  *
 128  *      XXX (TFO) - ideally these fields can be made as part of "listen_sock"
 129  *      structure above. But there is some implementation difficulty due to
 130  *      listen_sock being part of request_sock_queue hence will be freed when
 131  *      a listener is stopped. But TFO related fields may continue to be
 132  *      accessed even after a listener is closed, until its sk_refcnt drops
 133  *      to 0 implying no more outstanding TFO reqs. One solution is to keep
 134  *      listen_opt around until sk_refcnt drops to 0. But there is some other
 135  *      complexity that needs to be resolved. E.g., a listener can be disabled
 136  *      temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
 137  */
 138 struct fastopen_queue {
 139         struct request_sock     *rskq_rst_head; /* Keep track of past TFO */
 140         struct request_sock     *rskq_rst_tail; /* requests that caused RST.
 141                                                  * This is part of the defense
 142                                                  * against spoofing attack.
 143                                                  */
 144         spinlock_t      lock;
 145         int             qlen;           /* # of pending (TCP_SYN_RECV) reqs */
 146         int             max_qlen;       /* != 0 iff TFO is currently enabled */
 147 };
 148
 149 /** struct request_sock_queue - queue of request_socks
 150  *
 151  * @rskq_accept_head - FIFO head of established children
 152  * @rskq_accept_tail - FIFO tail of established children
 153  * @rskq_defer_accept - User waits for some data after accept()
 154  * @syn_wait_lock - serializer
 155  *
 156  * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
 157  * lock sock while browsing the listening hash (otherwise it's deadlock prone).
 158  *
 159  * This lock is acquired in read mode only from listening_get_next() seq_file
 160  * op and it's acquired in write mode _only_ from code that is actively
 161  * changing rskq_accept_head. All readers that are holding the master sock lock
 162  * don't need to grab this lock in read mode too as rskq_accept_head. writes
 163  * are always protected from the main sock lock.
 164  */
 165 struct request_sock_queue {
 166         struct request_sock     *rskq_accept_head;
 167         struct request_sock     *rskq_accept_tail;
 168         rwlock_t                syn_wait_lock;
 169         u8                      rskq_defer_accept;
 170         /* 3 bytes hole, try to pack */
 171         struct listen_sock      *listen_opt;
 172         struct fastopen_queue   *fastopenq; /* This is non-NULL iff TFO has been
 173                                              * enabled on this listener. Check
 174                                              * max_qlen != 0 in fastopen_queue
 175                                              * to determine if TFO is enabled
 176                                              * right at this moment.
 177                                              */
 178 };
 179
 180 int reqsk_queue_alloc(struct request_sock_queue *queue,
 181                       unsigned int nr_table_entries);
 182
 183 void __reqsk_queue_destroy(struct request_sock_queue *queue);
 184 void reqsk_queue_destroy(struct request_sock_queue *queue);
 185 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
 186                            bool reset);
 187
 188 static inline struct request_sock *
 189         reqsk_queue_yank_acceptq(struct request_sock_queue *queue)
 190 {
 191         struct request_sock *req = queue->rskq_accept_head;
 192
 193         queue->rskq_accept_head = NULL;
 194         return req;
 195 }
 196
 197 static inline int reqsk_queue_empty(struct request_sock_queue *queue)
 198 {
 199         return queue->rskq_accept_head == NULL;
 200 }
 201
 202 static inline void reqsk_queue_unlink(struct request_sock_queue *queue,
 203                                       struct request_sock *req,
 204                                       struct request_sock **prev_req)
 205 {
 206         write_lock(&queue->syn_wait_lock);
 207         *prev_req = req->dl_next;
 208         write_unlock(&queue->syn_wait_lock);
 209 }
 210
 211 static inline void reqsk_queue_add(struct request_sock_queue *queue,
 212                                    struct request_sock *req,
 213                                    struct sock *parent,
 214                                    struct sock *child)
 215 {
 216         req->sk = child;
 217         sk_acceptq_added(parent);
 218
 219         if (queue->rskq_accept_head == NULL)
 220                 queue->rskq_accept_head = req;
 221         else
 222                 queue->rskq_accept_tail->dl_next = req;
 223
 224         queue->rskq_accept_tail = req;
 225         req->dl_next = NULL;
 226 }
 227
 228 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue)
 229 {
 230         struct request_sock *req = queue->rskq_accept_head;
 231
 232         WARN_ON(req == NULL);
 233
 234         queue->rskq_accept_head = req->dl_next;
 235         if (queue->rskq_accept_head == NULL)
 236                 queue->rskq_accept_tail = NULL;
 237
 238         return req;
 239 }
 240
 241 static inline int reqsk_queue_removed(struct request_sock_queue *queue,
 242                                       struct request_sock *req)
 243 {
 244         struct listen_sock *lopt = queue->listen_opt;
 245
 246         if (req->num_timeout == 0)
 247                 --lopt->qlen_young;
 248
 249         return --lopt->qlen;
 250 }
 251
 252 static inline int reqsk_queue_added(struct request_sock_queue *queue)
 253 {
 254         struct listen_sock *lopt = queue->listen_opt;
 255         const int prev_qlen = lopt->qlen;
 256
 257         lopt->qlen_young++;
 258         lopt->qlen++;
 259         return prev_qlen;
 260 }
 261
 262 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
 263 {
 264         return queue->listen_opt != NULL ? queue->listen_opt->qlen : 0;
 265 }
 266
 267 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
 268 {
 269         return queue->listen_opt->qlen_young;
 270 }
 271
 272 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
 273 {
 274         return queue->listen_opt->qlen >> queue->listen_opt->max_qlen_log;
 275 }
 276
 277 static inline void reqsk_queue_hash_req(struct request_sock_queue *queue,
 278                                         u32 hash, struct request_sock *req,
 279                                         unsigned long timeout)
 280 {
 281         struct listen_sock *lopt = queue->listen_opt;
 282
 283         req->expires = jiffies + timeout;
 284         req->num_retrans = 0;
 285         req->num_timeout = 0;
 286         req->sk = NULL;
 287         req->dl_next = lopt->syn_table[hash];
 288
 289         write_lock(&queue->syn_wait_lock);
 290         lopt->syn_table[hash] = req;
 291         write_unlock(&queue->syn_wait_lock);
 292 }
 293
 294 #endif /* _REQUEST_SOCK_H */