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)(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 *inet_reqsk(struct sock *sk)
  73 {
  74         return (struct request_sock *)sk;
  75 }
  76
  77 static inline struct sock *req_to_sk(struct request_sock *req)
  78 {
  79         return (struct sock *)req;
  80 }
  81
  82 static inline struct request_sock *
  83 reqsk_alloc(const struct request_sock_ops *ops, struct sock *sk_listener)
  84 {
  85         struct request_sock *req = kmem_cache_alloc(ops->slab, GFP_ATOMIC);
  86
  87         if (req) {
  88                 req->rsk_ops = ops;
  89                 sock_hold(sk_listener);
  90                 req->rsk_listener = sk_listener;
  91                 req_to_sk(req)->sk_prot = sk_listener->sk_prot;
  92                 sk_node_init(&req_to_sk(req)->sk_node);
  93                 req->saved_syn = NULL;
  94                 /* Following is temporary. It is coupled with debugging
  95                  * helpers in reqsk_put() & reqsk_free()
  96                  */
  97                 atomic_set(&req->rsk_refcnt, 0);
  98         }
  99         return req;
 100 }
 101
 102 static inline void reqsk_free(struct request_sock *req)
 103 {
 104         /* temporary debugging */
 105         WARN_ON_ONCE(atomic_read(&req->rsk_refcnt) != 0);
 106
 107         req->rsk_ops->destructor(req);
 108         if (req->rsk_listener)
 109                 sock_put(req->rsk_listener);
 110         kfree(req->saved_syn);
 111         kmem_cache_free(req->rsk_ops->slab, req);
 112 }
 113
 114 static inline void reqsk_put(struct request_sock *req)
 115 {
 116         if (atomic_dec_and_test(&req->rsk_refcnt))
 117                 reqsk_free(req);
 118 }
 119
 120 extern int sysctl_max_syn_backlog;
 121
 122 /** struct listen_sock - listen state
 123  *
 124  * @max_qlen_log - log_2 of maximal queued SYNs/REQUESTs
 125  */
 126 struct listen_sock {
 127         u32                     max_qlen_log;
 128 };
 129
 130 /*
 131  * For a TCP Fast Open listener -
 132  *      lock - protects the access to all the reqsk, which is co-owned by
 133  *              the listener and the child socket.
 134  *      qlen - pending TFO requests (still in TCP_SYN_RECV).
 135  *      max_qlen - max TFO reqs allowed before TFO is disabled.
 136  *
 137  *      XXX (TFO) - ideally these fields can be made as part of "listen_sock"
 138  *      structure above. But there is some implementation difficulty due to
 139  *      listen_sock being part of request_sock_queue hence will be freed when
 140  *      a listener is stopped. But TFO related fields may continue to be
 141  *      accessed even after a listener is closed, until its sk_refcnt drops
 142  *      to 0 implying no more outstanding TFO reqs. One solution is to keep
 143  *      listen_opt around until sk_refcnt drops to 0. But there is some other
 144  *      complexity that needs to be resolved. E.g., a listener can be disabled
 145  *      temporarily through shutdown()->tcp_disconnect(), and re-enabled later.
 146  */
 147 struct fastopen_queue {
 148         struct request_sock     *rskq_rst_head; /* Keep track of past TFO */
 149         struct request_sock     *rskq_rst_tail; /* requests that caused RST.
 150                                                  * This is part of the defense
 151                                                  * against spoofing attack.
 152                                                  */
 153         spinlock_t      lock;
 154         int             qlen;           /* # of pending (TCP_SYN_RECV) reqs */
 155         int             max_qlen;       /* != 0 iff TFO is currently enabled */
 156 };
 157
 158 /** struct request_sock_queue - queue of request_socks
 159  *
 160  * @rskq_accept_head - FIFO head of established children
 161  * @rskq_accept_tail - FIFO tail of established children
 162  * @rskq_defer_accept - User waits for some data after accept()
 163  * @syn_wait_lock - serializer
 164  *
 165  * %syn_wait_lock is necessary only to avoid proc interface having to grab the main
 166  * lock sock while browsing the listening hash (otherwise it's deadlock prone).
 167  *
 168  */
 169 struct request_sock_queue {
 170         spinlock_t              rskq_lock;
 171         u8                      rskq_defer_accept;
 172         u32                     synflood_warned;
 173
 174         atomic_t                qlen;
 175         atomic_t                young;
 176
 177         struct request_sock     *rskq_accept_head;
 178         struct request_sock     *rskq_accept_tail;
 179         struct listen_sock      *listen_opt;
 180         struct fastopen_queue   fastopenq;  /* Check max_qlen != 0 to determine
 181                                              * if TFO is enabled.
 182                                              */
 183
 184         /* temporary alignment, our goal is to get rid of this lock */
 185         spinlock_t              syn_wait_lock ____cacheline_aligned_in_smp;
 186 };
 187
 188 int reqsk_queue_alloc(struct request_sock_queue *queue,
 189                       unsigned int nr_table_entries);
 190
 191 void __reqsk_queue_destroy(struct request_sock_queue *queue);
 192 void reqsk_queue_destroy(struct request_sock_queue *queue);
 193 void reqsk_fastopen_remove(struct sock *sk, struct request_sock *req,
 194                            bool reset);
 195
 196 static inline bool reqsk_queue_empty(const struct request_sock_queue *queue)
 197 {
 198         return queue->rskq_accept_head == NULL;
 199 }
 200
 201 static inline void reqsk_queue_add(struct request_sock_queue *queue,
 202                                    struct request_sock *req,
 203                                    struct sock *parent,
 204                                    struct sock *child)
 205 {
 206         spin_lock(&queue->rskq_lock);
 207         req->sk = child;
 208         sk_acceptq_added(parent);
 209
 210         if (queue->rskq_accept_head == NULL)
 211                 queue->rskq_accept_head = req;
 212         else
 213                 queue->rskq_accept_tail->dl_next = req;
 214
 215         queue->rskq_accept_tail = req;
 216         req->dl_next = NULL;
 217         spin_unlock(&queue->rskq_lock);
 218 }
 219
 220 static inline struct request_sock *reqsk_queue_remove(struct request_sock_queue *queue,
 221                                                       struct sock *parent)
 222 {
 223         struct request_sock *req;
 224
 225         spin_lock_bh(&queue->rskq_lock);
 226         req = queue->rskq_accept_head;
 227         if (req) {
 228                 sk_acceptq_removed(parent);
 229                 queue->rskq_accept_head = req->dl_next;
 230                 if (queue->rskq_accept_head == NULL)
 231                         queue->rskq_accept_tail = NULL;
 232         }
 233         spin_unlock_bh(&queue->rskq_lock);
 234         return req;
 235 }
 236
 237 static inline void reqsk_queue_removed(struct request_sock_queue *queue,
 238                                        const struct request_sock *req)
 239 {
 240         if (req->num_timeout == 0)
 241                 atomic_dec(&queue->young);
 242         atomic_dec(&queue->qlen);
 243 }
 244
 245 static inline void reqsk_queue_added(struct request_sock_queue *queue)
 246 {
 247         atomic_inc(&queue->young);
 248         atomic_inc(&queue->qlen);
 249 }
 250
 251 static inline int reqsk_queue_len(const struct request_sock_queue *queue)
 252 {
 253         return atomic_read(&queue->qlen);
 254 }
 255
 256 static inline int reqsk_queue_len_young(const struct request_sock_queue *queue)
 257 {
 258         return atomic_read(&queue->young);
 259 }
 260
 261 static inline int reqsk_queue_is_full(const struct request_sock_queue *queue)
 262 {
 263         return reqsk_queue_len(queue) >> queue->listen_opt->max_qlen_log;
 264 }
 265
 266 #endif /* _REQUEST_SOCK_H */