This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / net / dccp / timer.c
diff --git a/net/dccp/timer.c b/net/dccp/timer.c
new file mode 100644 (file)
index 0000000..5244415
--- /dev/null
@@ -0,0 +1,267 @@
+/*
+ *  net/dccp/timer.c
+ * 
+ *  An implementation of the DCCP protocol
+ *  Arnaldo Carvalho de Melo <acme@conectiva.com.br>
+ *
+ *     This program is free software; you can redistribute it and/or
+ *     modify it under the terms of the GNU General Public License
+ *     as published by the Free Software Foundation; either version
+ *     2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/dccp.h>
+#include <linux/skbuff.h>
+
+#include "dccp.h"
+
+static void dccp_write_timer(unsigned long data);
+static void dccp_keepalive_timer(unsigned long data);
+static void dccp_delack_timer(unsigned long data);
+
+void dccp_init_xmit_timers(struct sock *sk)
+{
+       inet_csk_init_xmit_timers(sk, &dccp_write_timer, &dccp_delack_timer,
+                                 &dccp_keepalive_timer);
+}
+
+static void dccp_write_err(struct sock *sk)
+{
+       sk->sk_err = sk->sk_err_soft ? : ETIMEDOUT;
+       sk->sk_error_report(sk);
+
+       dccp_send_reset(sk, DCCP_RESET_CODE_ABORTED);
+       dccp_done(sk);
+       DCCP_INC_STATS_BH(DCCP_MIB_ABORTONTIMEOUT);
+}
+
+/* A write timeout has occurred. Process the after effects. */
+static int dccp_write_timeout(struct sock *sk)
+{
+       const struct inet_connection_sock *icsk = inet_csk(sk);
+       int retry_until;
+
+       if (sk->sk_state == DCCP_REQUESTING || sk->sk_state == DCCP_PARTOPEN) {
+               if (icsk->icsk_retransmits != 0)
+                       dst_negative_advice(&sk->sk_dst_cache);
+               retry_until = icsk->icsk_syn_retries ? :
+                           /* FIXME! */ 3 /* FIXME! sysctl_tcp_syn_retries */;
+       } else {
+               if (icsk->icsk_retransmits >=
+                    /* FIXME! sysctl_tcp_retries1 */ 5 /* FIXME! */) {
+                       /* NOTE. draft-ietf-tcpimpl-pmtud-01.txt requires pmtu
+                          black hole detection. :-(
+
+                          It is place to make it. It is not made. I do not want
+                          to make it. It is disguisting. It does not work in any
+                          case. Let me to cite the same draft, which requires for
+                          us to implement this:
+
+   "The one security concern raised by this memo is that ICMP black holes
+   are often caused by over-zealous security administrators who block
+   all ICMP messages.  It is vitally important that those who design and
+   deploy security systems understand the impact of strict filtering on
+   upper-layer protocols.  The safest web site in the world is worthless
+   if most TCP implementations cannot transfer data from it.  It would
+   be far nicer to have all of the black holes fixed rather than fixing
+   all of the TCP implementations."
+
+                           Golden words :-).
+                  */
+
+                       dst_negative_advice(&sk->sk_dst_cache);
+               }
+
+               retry_until = /* FIXME! */ 15 /* FIXME! sysctl_tcp_retries2 */;
+               /*
+                * FIXME: see tcp_write_timout and tcp_out_of_resources
+                */
+       }
+
+       if (icsk->icsk_retransmits >= retry_until) {
+               /* Has it gone just too far? */
+               dccp_write_err(sk);
+               return 1;
+       }
+       return 0;
+}
+
+/* This is the same as tcp_delack_timer, sans prequeue & mem_reclaim stuff */
+static void dccp_delack_timer(unsigned long data)
+{
+       struct sock *sk = (struct sock *)data;
+       struct inet_connection_sock *icsk = inet_csk(sk);
+
+       bh_lock_sock(sk);
+       if (sock_owned_by_user(sk)) {
+               /* Try again later. */
+               icsk->icsk_ack.blocked = 1;
+               NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKLOCKED);
+               sk_reset_timer(sk, &icsk->icsk_delack_timer,
+                              jiffies + TCP_DELACK_MIN);
+               goto out;
+       }
+
+       if (sk->sk_state == DCCP_CLOSED ||
+           !(icsk->icsk_ack.pending & ICSK_ACK_TIMER))
+               goto out;
+       if (time_after(icsk->icsk_ack.timeout, jiffies)) {
+               sk_reset_timer(sk, &icsk->icsk_delack_timer,
+                              icsk->icsk_ack.timeout);
+               goto out;
+       }
+
+       icsk->icsk_ack.pending &= ~ICSK_ACK_TIMER;
+
+       if (inet_csk_ack_scheduled(sk)) {
+               if (!icsk->icsk_ack.pingpong) {
+                       /* Delayed ACK missed: inflate ATO. */
+                       icsk->icsk_ack.ato = min(icsk->icsk_ack.ato << 1,
+                                                icsk->icsk_rto);
+               } else {
+                       /* Delayed ACK missed: leave pingpong mode and
+                        * deflate ATO.
+                        */
+                       icsk->icsk_ack.pingpong = 0;
+                       icsk->icsk_ack.ato = TCP_ATO_MIN;
+               }
+               dccp_send_ack(sk);
+               NET_INC_STATS_BH(LINUX_MIB_DELAYEDACKS);
+       }
+out:
+       bh_unlock_sock(sk);
+       sock_put(sk);
+}
+
+/*
+ *     The DCCP retransmit timer.
+ */
+static void dccp_retransmit_timer(struct sock *sk)
+{
+       struct inet_connection_sock *icsk = inet_csk(sk);
+
+       /* retransmit timer is used for feature negotiation throughout
+        * connection.  In this case, no packet is re-transmitted, but rather an
+        * ack is generated and pending changes are splaced into its options.
+        */
+       if (sk->sk_send_head == NULL) {
+               dccp_pr_debug("feat negotiation retransmit timeout %p\n", sk);
+               if (sk->sk_state == DCCP_OPEN)
+                       dccp_send_ack(sk);
+               goto backoff;
+       }
+
+       /*
+        * sk->sk_send_head has to have one skb with
+        * DCCP_SKB_CB(skb)->dccpd_type set to one of the retransmittable DCCP
+        * packet types (REQUEST, RESPONSE, the ACK in the 3way handshake
+        * (PARTOPEN timer), etc).
+        */
+       BUG_TRAP(sk->sk_send_head != NULL);
+
+       /* 
+        * More than than 4MSL (8 minutes) has passed, a RESET(aborted) was
+        * sent, no need to retransmit, this sock is dead.
+        */
+       if (dccp_write_timeout(sk))
+               goto out;
+
+       /*
+        * We want to know the number of packets retransmitted, not the
+        * total number of retransmissions of clones of original packets.
+        */
+       if (icsk->icsk_retransmits == 0)
+               DCCP_INC_STATS_BH(DCCP_MIB_TIMEOUTS);
+
+       if (dccp_retransmit_skb(sk, sk->sk_send_head) < 0) {
+               /*
+                * Retransmission failed because of local congestion,
+                * do not backoff.
+                */
+               if (icsk->icsk_retransmits == 0)
+                       icsk->icsk_retransmits = 1;
+               inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
+                                         min(icsk->icsk_rto,
+                                             TCP_RESOURCE_PROBE_INTERVAL),
+                                         DCCP_RTO_MAX);
+               goto out;
+       }
+
+backoff:
+       icsk->icsk_backoff++;
+       icsk->icsk_retransmits++;
+
+       icsk->icsk_rto = min(icsk->icsk_rto << 1, DCCP_RTO_MAX);
+       inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS, icsk->icsk_rto,
+                                 DCCP_RTO_MAX);
+       if (icsk->icsk_retransmits > 3 /* FIXME: sysctl_dccp_retries1 */)
+               __sk_dst_reset(sk);
+out:;
+}
+
+static void dccp_write_timer(unsigned long data)
+{
+       struct sock *sk = (struct sock *)data;
+       struct inet_connection_sock *icsk = inet_csk(sk);
+       int event = 0;
+
+       bh_lock_sock(sk);
+       if (sock_owned_by_user(sk)) {
+               /* Try again later */
+               sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
+                              jiffies + (HZ / 20));
+               goto out;
+       }
+
+       if (sk->sk_state == DCCP_CLOSED || !icsk->icsk_pending)
+               goto out;
+
+       if (time_after(icsk->icsk_timeout, jiffies)) {
+               sk_reset_timer(sk, &icsk->icsk_retransmit_timer,
+                              icsk->icsk_timeout);
+               goto out;
+       }
+
+       event = icsk->icsk_pending;
+       icsk->icsk_pending = 0;
+
+       switch (event) {
+       case ICSK_TIME_RETRANS:
+               dccp_retransmit_timer(sk);
+               break;
+       }
+out:
+       bh_unlock_sock(sk);
+       sock_put(sk);
+}
+
+/*
+ *     Timer for listening sockets
+ */
+static void dccp_response_timer(struct sock *sk)
+{
+       inet_csk_reqsk_queue_prune(sk, TCP_SYNQ_INTERVAL, DCCP_TIMEOUT_INIT,
+                                  DCCP_RTO_MAX);
+}
+
+static void dccp_keepalive_timer(unsigned long data)
+{
+       struct sock *sk = (struct sock *)data;
+
+       /* Only process if socket is not in use. */
+       bh_lock_sock(sk);
+       if (sock_owned_by_user(sk)) {
+               /* Try again later. */ 
+               inet_csk_reset_keepalive_timer(sk, HZ / 20);
+               goto out;
+       }
+
+       if (sk->sk_state == DCCP_LISTEN) {
+               dccp_response_timer(sk);
+               goto out;
+       }
+out:
+       bh_unlock_sock(sk);
+       sock_put(sk);
+}