net/dccp/timer.c

   1 /*
   2  *  net/dccp/timer.c
   3  *
   4  *  An implementation of the DCCP protocol
   5  *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
   6  *
   7  *      This program is free software; you can redistribute it and/or
   8  *      modify it under the terms of the GNU General Public License
   9  *      as published by the Free Software Foundation; either version
  10  *      2 of the License, or (at your option) any later version.
  11  */
  12
  13 #include <linux/dccp.h>
  14 #include <linux/skbuff.h>
  15
  16 #include "dccp.h"
  17
  18 /* sysctl variables governing numbers of retransmission attempts */
  19 int  sysctl_dccp_request_retries        __read_mostly = TCP_SYN_RETRIES;
  20 int  sysctl_dccp_retries1               __read_mostly = TCP_RETR1;
  21 int  sysctl_dccp_retries2               __read_mostly = TCP_RETR2;
  22
  23 static void dccp_write_err(struct sock *sk)
  24 {
  25         sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
  26         sk->sk_error_report(sk);
  27
  28         dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
  29         dccp_done(sk);
  30         DCCP_INC_STATS_BH(DCCP_MIB_ABORTONTIMEOUT);
  31 }
  32
  33 /* A write timeout has occurred. Process the after effects. */
  34 static int dccp_write_timeout(struct sock *sk)
  35 {
  36         const struct inet_connection_sock *icsk = inet_csk(sk);
  37         int retry_until;
  38
  39         if (sk->sk_state == DCCP_REQUESTING || sk->sk_state == DCCP_PARTOPEN) {
  40                 if (icsk->icsk_retransmits != 0)
  41                         dst_negative_advice(sk);
  42                 retry_until = icsk->icsk_syn_retries ?
  43                             : sysctl_dccp_request_retries;
  44         } else {
  45                 if (icsk->icsk_retransmits >= sysctl_dccp_retries1) {
  46                         /* NOTE. draft-ietf-tcpimpl-pmtud-01.txt requires pmtu
  47                            black hole detection. :-(
  48
  49                            It is place to make it. It is not made. I do not want
  50                            to make it. It is disguisting. It does not work in any
  51                            case. Let me to cite the same draft, which requires for
  52                            us to implement this:
  53
  54    "The one security concern raised by this memo is that ICMP black holes
  55    are often caused by over-zealous security administrators who block
  56    all ICMP messages.  It is vitally important that those who design and
  57    deploy security systems understand the impact of strict filtering on
  58    upper-layer protocols.  The safest web site in the world is worthless
  59    if most TCP implementations cannot transfer data from it.  It would
  60    be far nicer to have all of the black holes fixed rather than fixing
  61    all of the TCP implementations."
  62
  63                            Golden words :-).
  64                    */
  65
  66                         dst_negative_advice(sk);
  67                 }
  68
  69                 retry_until = sysctl_dccp_retries2;
  70                 /*
  71                  * FIXME: see tcp_write_timout and tcp_out_of_resources
  72                  */
  73         }
  74
  75         if (icsk->icsk_retransmits >= retry_until) {
  76                 /* Has it gone just too far? */
  77                 dccp_write_err(sk);
  78                 return 1;
  79         }
  80         return 0;
  81 }
  82
  83 /*
  84  *      The DCCP retransmit timer.
  85  */
  86 static void dccp_retransmit_timer(struct sock *sk)
  87 {
  88         struct inet_connection_sock *icsk = inet_csk(sk);
  89
  90         /*
  91          * More than than 4MSL (8 minutes) has passed, a RESET(aborted) was
  92          * sent, no need to retransmit, this sock is dead.
  93          */
  94         if (dccp_write_timeout(sk))
  95                 return;
  96
  97         /*
  98          * We want to know the number of packets retransmitted, not the
  99          * total number of retransmissions of clones of original packets.
 100          */
 101         if (icsk->icsk_retransmits == 0)
 102                 DCCP_INC_STATS_BH(DCCP_MIB_TIMEOUTS);
 103
 104         if (dccp_retransmit_skb(sk) != 0) {
 105                 /*
 106                  * Retransmission failed because of local congestion,
 107                  * do not backoff.
 108                  */
 109                 if (--icsk->icsk_retransmits == 0)
 110                         icsk->icsk_retransmits = 1;
 111                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 112                                           min(icsk->icsk_rto,
 113                                               TCP_RESOURCE_PROBE_INTERVAL),
 114                                           DCCP_RTO_MAX);
 115                 return;
 116         }
 117
 118         icsk->icsk_backoff++;
 119
 120         icsk->icsk_rto = min(icsk->icsk_rto << 1, DCCP_RTO_MAX);
 121         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto,
 122                                   DCCP_RTO_MAX);
 123         if (icsk->icsk_retransmits > sysctl_dccp_retries1)
 124                 __sk_dst_reset(sk);
 125 }
 126
 127 static void dccp_write_timer(unsigned long data)
 128 {
 129         struct sock *sk = (struct sock *)data;
 130         struct inet_connection_sock *icsk = inet_csk(sk);
 131         int event = 0;
 132
 133         bh_lock_sock(sk);
 134         if (sock_owned_by_user(sk)) {
 135                 /* Try again later */
 136                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
 137                                jiffies + (HZ / 20));
 138                 goto out;
 139         }
 140
 141         if (sk->sk_state == DCCP_CLOSED || !icsk->icsk_pending)
 142                 goto out;
 143
 144         if (time_after(icsk->icsk_timeout, jiffies)) {
 145                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
 146                                icsk->icsk_timeout);
 147                 goto out;
 148         }
 149
 150         event = icsk->icsk_pending;
 151         icsk->icsk_pending = 0;
 152
 153         switch (event) {
 154         case ICSK_TIME_RETRANS:
 155                 dccp_retransmit_timer(sk);
 156                 break;
 157         }
 158 out:
 159         bh_unlock_sock(sk);
 160         sock_put(sk);
 161 }
 162
 163 /*
 164  *      Timer for listening sockets
 165  */
 166 static void dccp_response_timer(struct sock *sk)
 167 {
 168         inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, DCCP_TIMEOUT_INIT,
 169                                    DCCP_RTO_MAX);
 170 }
 171
 172 static void dccp_keepalive_timer(unsigned long data)
 173 {
 174         struct sock *sk = (struct sock *)data;
 175
 176         /* Only process if socket is not in use. */
 177         bh_lock_sock(sk);
 178         if (sock_owned_by_user(sk)) {
 179                 /* Try again later. */
 180                 inet_csk_reset_keepalive_timer(sk, HZ / 20);
 181                 goto out;
 182         }
 183
 184         if (sk->sk_state == DCCP_LISTEN) {
 185                 dccp_response_timer(sk);
 186                 goto out;
 187         }
 188 out:
 189         bh_unlock_sock(sk);
 190         sock_put(sk);
 191 }
 192
 193 /* This is the same as tcp_delack_timer, sans prequeue & mem_reclaim stuff */
 194 static void dccp_delack_timer(unsigned long data)
 195 {
 196         struct sock *sk = (struct sock *)data;
 197         struct inet_connection_sock *icsk = inet_csk(sk);
 198
 199         bh_lock_sock(sk);
 200         if (sock_owned_by_user(sk)) {
 201                 /* Try again later. */
 202                 icsk->icsk_ack.blocked = 1;
 203                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
 204                 sk_reset_timer(sk, &icsk->icsk_delack_timer,
 205                                jiffies + TCP_DELACK_MIN);
 206                 goto out;
 207         }
 208
 209         if (sk->sk_state == DCCP_CLOSED ||
 210             !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
 211                 goto out;
 212         if (time_after(icsk->icsk_ack.timeout, jiffies)) {
 213                 sk_reset_timer(sk, &icsk->icsk_delack_timer,
 214                                icsk->icsk_ack.timeout);
 215                 goto out;
 216         }
 217
 218         icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
 219
 220         if (inet_csk_ack_scheduled(sk)) {
 221                 if (!icsk->icsk_ack.pingpong) {
 222                         /* Delayed ACK missed: inflate ATO. */
 223                         icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1,
 224                                                  icsk->icsk_rto);
 225                 } else {
 226                         /* Delayed ACK missed: leave pingpong mode and
 227                          * deflate ATO.
 228                          */
 229                         icsk->icsk_ack.pingpong = 0;
 230                         icsk->icsk_ack.ato = TCP_ATO_MIN;
 231                 }
 232                 dccp_send_ack(sk);
 233                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
 234         }
 235 out:
 236         bh_unlock_sock(sk);
 237         sock_put(sk);
 238 }
 239
 240 /* Transmit-delay timer: used by the CCIDs to delay actual send time */
 241 static void dccp_write_xmit_timer(unsigned long data)
 242 {
 243         struct sock *sk = (struct sock *)data;
 244         struct dccp_sock *dp = dccp_sk(sk);
 245
 246         bh_lock_sock(sk);
 247         if (sock_owned_by_user(sk))
 248                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
 249         else
 250                 dccp_write_xmit(sk, 0);
 251         bh_unlock_sock(sk);
 252         sock_put(sk);
 253 }
 254
 255 static void dccp_init_write_xmit_timer(struct sock *sk)
 256 {
 257         struct dccp_sock *dp = dccp_sk(sk);
 258
 259         setup_timer(&dp->dccps_xmit_timer, dccp_write_xmit_timer,
 260                         (unsigned long)sk);
 261 }
 262
 263 void dccp_init_xmit_timers(struct sock *sk)
 264 {
 265         dccp_init_write_xmit_timer(sk);
 266         inet_csk_init_xmit_timers(sk, &dccp_write_timer, &dccp_delack_timer,
 267                                   &dccp_keepalive_timer);
 268 }
 269
 270 static ktime_t dccp_timestamp_seed;
 271 /**
 272  * dccp_timestamp  -  10s of microseconds time source
 273  * Returns the number of 10s of microseconds since loading DCCP. This is native
 274  * DCCP time difference format (RFC 4340, sec. 13).
 275  * Please note: This will wrap around about circa every 11.9 hours.
 276  */
 277 u32 dccp_timestamp(void)
 278 {
 279         s64 delta = ktime_us_delta(ktime_get_real(), dccp_timestamp_seed);
 280
 281         do_div(delta, 10);
 282         return delta;
 283 }
 284 EXPORT_SYMBOL_GPL(dccp_timestamp);
 285
 286 void __init dccp_timestamping_init(void)
 287 {
 288         dccp_timestamp_seed = ktime_get_real();
 289 }