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->sk_dst_cache);
  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->sk_dst_cache);
  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         /* retransmit timer is used for feature negotiation throughout
  91          * connection.  In this case, no packet is re-transmitted, but rather an
  92          * ack is generated and pending changes are placed into its options.
  93          */
  94         if (sk->sk_send_head == NULL) {
  95                 dccp_pr_debug("feat negotiation retransmit timeout %p\n", sk);
  96                 if (sk->sk_state == DCCP_OPEN)
  97                         dccp_send_ack(sk);
  98                 goto backoff;
  99         }
 100
 101         /*
 102          * More than than 4MSL (8 minutes) has passed, a RESET(aborted) was
 103          * sent, no need to retransmit, this sock is dead.
 104          */
 105         if (dccp_write_timeout(sk))
 106                 return;
 107
 108         /*
 109          * We want to know the number of packets retransmitted, not the
 110          * total number of retransmissions of clones of original packets.
 111          */
 112         if (icsk->icsk_retransmits == 0)
 113                 DCCP_INC_STATS_BH(DCCP_MIB_TIMEOUTS);
 114
 115         if (dccp_retransmit_skb(sk) != 0) {
 116                 /*
 117                  * Retransmission failed because of local congestion,
 118                  * do not backoff.
 119                  */
 120                 if (--icsk->icsk_retransmits == 0)
 121                         icsk->icsk_retransmits = 1;
 122                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 123                                           min(icsk->icsk_rto,
 124                                               TCP_RESOURCE_PROBE_INTERVAL),
 125                                           DCCP_RTO_MAX);
 126                 return;
 127         }
 128
 129 backoff:
 130         icsk->icsk_backoff++;
 131
 132         icsk->icsk_rto = min(icsk->icsk_rto << 1, DCCP_RTO_MAX);
 133         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto,
 134                                   DCCP_RTO_MAX);
 135         if (icsk->icsk_retransmits > sysctl_dccp_retries1)
 136                 __sk_dst_reset(sk);
 137 }
 138
 139 static void dccp_write_timer(unsigned long data)
 140 {
 141         struct sock *sk = (struct sock *)data;
 142         struct inet_connection_sock *icsk = inet_csk(sk);
 143         int event = 0;
 144
 145         bh_lock_sock(sk);
 146         if (sock_owned_by_user(sk)) {
 147                 /* Try again later */
 148                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
 149                                jiffies + (HZ / 20));
 150                 goto out;
 151         }
 152
 153         if (sk->sk_state == DCCP_CLOSED || !icsk->icsk_pending)
 154                 goto out;
 155
 156         if (time_after(icsk->icsk_timeout, jiffies)) {
 157                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
 158                                icsk->icsk_timeout);
 159                 goto out;
 160         }
 161
 162         event = icsk->icsk_pending;
 163         icsk->icsk_pending = 0;
 164
 165         switch (event) {
 166         case ICSK_TIME_RETRANS:
 167                 dccp_retransmit_timer(sk);
 168                 break;
 169         }
 170 out:
 171         bh_unlock_sock(sk);
 172         sock_put(sk);
 173 }
 174
 175 /*
 176  *      Timer for listening sockets
 177  */
 178 static void dccp_response_timer(struct sock *sk)
 179 {
 180         inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, DCCP_TIMEOUT_INIT,
 181                                    DCCP_RTO_MAX);
 182 }
 183
 184 static void dccp_keepalive_timer(unsigned long data)
 185 {
 186         struct sock *sk = (struct sock *)data;
 187
 188         /* Only process if socket is not in use. */
 189         bh_lock_sock(sk);
 190         if (sock_owned_by_user(sk)) {
 191                 /* Try again later. */
 192                 inet_csk_reset_keepalive_timer(sk, HZ / 20);
 193                 goto out;
 194         }
 195
 196         if (sk->sk_state == DCCP_LISTEN) {
 197                 dccp_response_timer(sk);
 198                 goto out;
 199         }
 200 out:
 201         bh_unlock_sock(sk);
 202         sock_put(sk);
 203 }
 204
 205 /* This is the same as tcp_delack_timer, sans prequeue & mem_reclaim stuff */
 206 static void dccp_delack_timer(unsigned long data)
 207 {
 208         struct sock *sk = (struct sock *)data;
 209         struct inet_connection_sock *icsk = inet_csk(sk);
 210
 211         bh_lock_sock(sk);
 212         if (sock_owned_by_user(sk)) {
 213                 /* Try again later. */
 214                 icsk->icsk_ack.blocked = 1;
 215                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKLOCKED);
 216                 sk_reset_timer(sk, &icsk->icsk_delack_timer,
 217                                jiffies + TCP_DELACK_MIN);
 218                 goto out;
 219         }
 220
 221         if (sk->sk_state == DCCP_CLOSED ||
 222             !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
 223                 goto out;
 224         if (time_after(icsk->icsk_ack.timeout, jiffies)) {
 225                 sk_reset_timer(sk, &icsk->icsk_delack_timer,
 226                                icsk->icsk_ack.timeout);
 227                 goto out;
 228         }
 229
 230         icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
 231
 232         if (inet_csk_ack_scheduled(sk)) {
 233                 if (!icsk->icsk_ack.pingpong) {
 234                         /* Delayed ACK missed: inflate ATO. */
 235                         icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1,
 236                                                  icsk->icsk_rto);
 237                 } else {
 238                         /* Delayed ACK missed: leave pingpong mode and
 239                          * deflate ATO.
 240                          */
 241                         icsk->icsk_ack.pingpong = 0;
 242                         icsk->icsk_ack.ato = TCP_ATO_MIN;
 243                 }
 244                 dccp_send_ack(sk);
 245                 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_DELAYEDACKS);
 246         }
 247 out:
 248         bh_unlock_sock(sk);
 249         sock_put(sk);
 250 }
 251
 252 /* Transmit-delay timer: used by the CCIDs to delay actual send time */
 253 static void dccp_write_xmit_timer(unsigned long data)
 254 {
 255         struct sock *sk = (struct sock *)data;
 256         struct dccp_sock *dp = dccp_sk(sk);
 257
 258         bh_lock_sock(sk);
 259         if (sock_owned_by_user(sk))
 260                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
 261         else
 262                 dccp_write_xmit(sk, 0);
 263         bh_unlock_sock(sk);
 264         sock_put(sk);
 265 }
 266
 267 static void dccp_init_write_xmit_timer(struct sock *sk)
 268 {
 269         struct dccp_sock *dp = dccp_sk(sk);
 270
 271         setup_timer(&dp->dccps_xmit_timer, dccp_write_xmit_timer,
 272                         (unsigned long)sk);
 273 }
 274
 275 void dccp_init_xmit_timers(struct sock *sk)
 276 {
 277         dccp_init_write_xmit_timer(sk);
 278         inet_csk_init_xmit_timers(sk, &dccp_write_timer, &dccp_delack_timer,
 279                                   &dccp_keepalive_timer);
 280 }
 281
 282 static ktime_t dccp_timestamp_seed;
 283 /**
 284  * dccp_timestamp  -  10s of microseconds time source
 285  * Returns the number of 10s of microseconds since loading DCCP. This is native
 286  * DCCP time difference format (RFC 4340, sec. 13).
 287  * Please note: This will wrap around about circa every 11.9 hours.
 288  */
 289 u32 dccp_timestamp(void)
 290 {
 291         s64 delta = ktime_us_delta(ktime_get_real(), dccp_timestamp_seed);
 292
 293         do_div(delta, 10);
 294         return delta;
 295 }
 296 EXPORT_SYMBOL_GPL(dccp_timestamp);
 297
 298 void __init dccp_timestamping_init(void)
 299 {
 300         dccp_timestamp_seed = ktime_get_real();
 301 }