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          * sk->sk_send_head has to have one skb with
 103          * DCCP_SKB_CB(skb)->dccpd_type set to one of the retransmittable DCCP
 104          * packet types. The only packets eligible for retransmission are:
 105          *      -- Requests in client-REQUEST  state (sec. 8.1.1)
 106          *      -- Acks     in client-PARTOPEN state (sec. 8.1.5)
 107          *      -- CloseReq in server-CLOSEREQ state (sec. 8.3)
 108          *      -- Close    in   node-CLOSING  state (sec. 8.3)                */
 109         BUG_TRAP(sk->sk_send_head != NULL);
 110
 111         /*
 112          * More than than 4MSL (8 minutes) has passed, a RESET(aborted) was
 113          * sent, no need to retransmit, this sock is dead.
 114          */
 115         if (dccp_write_timeout(sk))
 116                 goto out;
 117
 118         /*
 119          * We want to know the number of packets retransmitted, not the
 120          * total number of retransmissions of clones of original packets.
 121          */
 122         if (icsk->icsk_retransmits == 0)
 123                 DCCP_INC_STATS_BH(DCCP_MIB_TIMEOUTS);
 124
 125         if (dccp_retransmit_skb(sk, sk->sk_send_head) < 0) {
 126                 /*
 127                  * Retransmission failed because of local congestion,
 128                  * do not backoff.
 129                  */
 130                 if (icsk->icsk_retransmits == 0)
 131                         icsk->icsk_retransmits = 1;
 132                 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
 133                                           min(icsk->icsk_rto,
 134                                               TCP_RESOURCE_PROBE_INTERVAL),
 135                                           DCCP_RTO_MAX);
 136                 goto out;
 137         }
 138
 139 backoff:
 140         icsk->icsk_backoff++;
 141         icsk->icsk_retransmits++;
 142
 143         icsk->icsk_rto = min(icsk->icsk_rto << 1, DCCP_RTO_MAX);
 144         inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto,
 145                                   DCCP_RTO_MAX);
 146         if (icsk->icsk_retransmits > sysctl_dccp_retries1)
 147                 __sk_dst_reset(sk);
 148 out:;
 149 }
 150
 151 static void dccp_write_timer(unsigned long data)
 152 {
 153         struct sock *sk = (struct sock *)data;
 154         struct inet_connection_sock *icsk = inet_csk(sk);
 155         int event = 0;
 156
 157         bh_lock_sock(sk);
 158         if (sock_owned_by_user(sk)) {
 159                 /* Try again later */
 160                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
 161                                jiffies + (HZ / 20));
 162                 goto out;
 163         }
 164
 165         if (sk->sk_state == DCCP_CLOSED || !icsk->icsk_pending)
 166                 goto out;
 167
 168         if (time_after(icsk->icsk_timeout, jiffies)) {
 169                 sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
 170                                icsk->icsk_timeout);
 171                 goto out;
 172         }
 173
 174         event = icsk->icsk_pending;
 175         icsk->icsk_pending = 0;
 176
 177         switch (event) {
 178         case ICSK_TIME_RETRANS:
 179                 dccp_retransmit_timer(sk);
 180                 break;
 181         }
 182 out:
 183         bh_unlock_sock(sk);
 184         sock_put(sk);
 185 }
 186
 187 /*
 188  *      Timer for listening sockets
 189  */
 190 static void dccp_response_timer(struct sock *sk)
 191 {
 192         inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, DCCP_TIMEOUT_INIT,
 193                                    DCCP_RTO_MAX);
 194 }
 195
 196 static void dccp_keepalive_timer(unsigned long data)
 197 {
 198         struct sock *sk = (struct sock *)data;
 199
 200         /* Only process if socket is not in use. */
 201         bh_lock_sock(sk);
 202         if (sock_owned_by_user(sk)) {
 203                 /* Try again later. */
 204                 inet_csk_reset_keepalive_timer(sk, HZ / 20);
 205                 goto out;
 206         }
 207
 208         if (sk->sk_state == DCCP_LISTEN) {
 209                 dccp_response_timer(sk);
 210                 goto out;
 211         }
 212 out:
 213         bh_unlock_sock(sk);
 214         sock_put(sk);
 215 }
 216
 217 /* This is the same as tcp_delack_timer, sans prequeue & mem_reclaim stuff */
 218 static void dccp_delack_timer(unsigned long data)
 219 {
 220         struct sock *sk = (struct sock *)data;
 221         struct inet_connection_sock *icsk = inet_csk(sk);
 222
 223         bh_lock_sock(sk);
 224         if (sock_owned_by_user(sk)) {
 225                 /* Try again later. */
 226                 icsk->icsk_ack.blocked = 1;
 227                 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOCKED);
 228                 sk_reset_timer(sk, &icsk->icsk_delack_timer,
 229                                jiffies + TCP_DELACK_MIN);
 230                 goto out;
 231         }
 232
 233         if (sk->sk_state == DCCP_CLOSED ||
 234             !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
 235                 goto out;
 236         if (time_after(icsk->icsk_ack.timeout, jiffies)) {
 237                 sk_reset_timer(sk, &icsk->icsk_delack_timer,
 238                                icsk->icsk_ack.timeout);
 239                 goto out;
 240         }
 241
 242         icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
 243
 244         if (inet_csk_ack_scheduled(sk)) {
 245                 if (!icsk->icsk_ack.pingpong) {
 246                         /* Delayed ACK missed: inflate ATO. */
 247                         icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1,
 248                                                  icsk->icsk_rto);
 249                 } else {
 250                         /* Delayed ACK missed: leave pingpong mode and
 251                          * deflate ATO.
 252                          */
 253                         icsk->icsk_ack.pingpong = 0;
 254                         icsk->icsk_ack.ato = TCP_ATO_MIN;
 255                 }
 256                 dccp_send_ack(sk);
 257                 NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKS);
 258         }
 259 out:
 260         bh_unlock_sock(sk);
 261         sock_put(sk);
 262 }
 263
 264 /* Transmit-delay timer: used by the CCIDs to delay actual send time */
 265 static void dccp_write_xmit_timer(unsigned long data)
 266 {
 267         struct sock *sk = (struct sock *)data;
 268         struct dccp_sock *dp = dccp_sk(sk);
 269
 270         bh_lock_sock(sk);
 271         if (sock_owned_by_user(sk))
 272                 sk_reset_timer(sk, &dp->dccps_xmit_timer, jiffies+1);
 273         else
 274                 dccp_write_xmit(sk, 0);
 275         bh_unlock_sock(sk);
 276         sock_put(sk);
 277 }
 278
 279 static void dccp_init_write_xmit_timer(struct sock *sk)
 280 {
 281         struct dccp_sock *dp = dccp_sk(sk);
 282
 283         init_timer(&dp->dccps_xmit_timer);
 284         dp->dccps_xmit_timer.data = (unsigned long)sk;
 285         dp->dccps_xmit_timer.function = dccp_write_xmit_timer;
 286 }
 287
 288 void dccp_init_xmit_timers(struct sock *sk)
 289 {
 290         dccp_init_write_xmit_timer(sk);
 291         inet_csk_init_xmit_timers(sk, &dccp_write_timer, &dccp_delack_timer,
 292                                   &dccp_keepalive_timer);
 293 }
 294
 295 static ktime_t dccp_timestamp_seed;
 296 /**
 297  * dccp_timestamp  -  10s of microseconds time source
 298  * Returns the number of 10s of microseconds since loading DCCP. This is native
 299  * DCCP time difference format (RFC 4340, sec. 13).
 300  * Please note: This will wrap around about circa every 11.9 hours.
 301  */
 302 u32 dccp_timestamp(void)
 303 {
 304         s64 delta = ktime_us_delta(ktime_get_real(), dccp_timestamp_seed);
 305
 306         do_div(delta, 10);
 307         return delta;
 308 }
 309 EXPORT_SYMBOL_GPL(dccp_timestamp);
 310
 311 void __init dccp_timestamping_init(void)
 312 {
 313         dccp_timestamp_seed = ktime_get_real();
 314 }