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