xnu-2050.48.11.tar.gz

[apple/xnu.git] / bsd / netinet / tcp_timer.h

diff --git a/bsd/netinet/tcp_timer.h b/bsd/netinet/tcp_timer.h
index dea6ce3c0ee38f02b8fbd16bf9403d277afce4e7..213e87a395f241ca9960ee70324097ff4def978b 100644 (file)
--- a/bsd/netinet/tcp_timer.h
+++ b/bsd/netinet/tcp_timer.h
@@ -1,16 +1,19 @@
 /*
 /*

- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2000-2010 Apple Computer, Inc. All rights reserved.

  *
  *

- * @APPLE_LICENSE_HEADER_START@
- * 
- * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@

  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * 
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in

- * compliance with the License. Please obtain a copy of the License at
- * http://www.opensource.apple.com/apsl/ and read it before using this
- * file.
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ * 
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.

  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * 
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER


@@ -20,7 +23,7 @@
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 
  * Please see the License for the specific language governing rights and
  * limitations under the License.
  * 

- * @APPLE_LICENSE_HEADER_END@
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@

  */
 /*
  * Copyright (c) 1982, 1986, 1993
  */
 /*
  * Copyright (c) 1982, 1986, 1993


@@ -62,16 +65,25 @@
 #define _NETINET_TCP_TIMER_H_
 #include <sys/appleapiopts.h>
 
 #define _NETINET_TCP_TIMER_H_
 #include <sys/appleapiopts.h>
 

+#ifdef KERNEL
+#include <kern/thread_call.h>
+#endif /* KERNEL */
+

 /*
 /*

- * Definitions of the TCP timers.  These timers are counted
- * down PR_SLOWHZ times a second.
+ * Definitions of the TCP timers.

  */
  */

-#define        TCPT_NTIMERS    4
+#define        TCPT_NTIMERS    5
+
+/* Keep the external definition the same for binary compatibility */
+#define TCPT_NTIMERS_EXT       4

 
 #define        TCPT_REXMT      0               /* retransmit */
 #define        TCPT_PERSIST    1               /* retransmit persistence */
 #define        TCPT_KEEP       2               /* keep alive */
 #define        TCPT_2MSL       3               /* 2*msl quiet time timer */
 
 #define        TCPT_REXMT      0               /* retransmit */
 #define        TCPT_PERSIST    1               /* retransmit persistence */
 #define        TCPT_KEEP       2               /* keep alive */
 #define        TCPT_2MSL       3               /* 2*msl quiet time timer */

+#define        TCPT_DELACK     4               /* delayed ack timer */
+#define        TCPT_MAX        4
+#define        TCPT_NONE       (TCPT_MAX + 1)  

 
 /*
  * The TCPT_REXMT timer is used to force retransmissions.
 
 /*
  * The TCPT_REXMT timer is used to force retransmissions.


@@ -109,26 +121,38 @@
  * amount of time probing, then we drop the connection.
  */
 
  * amount of time probing, then we drop the connection.
  */
 

+#ifdef PRIVATE
+

 /*
  * Time constants.
  */
 /*
  * Time constants.
  */

-#ifdef __APPLE_API_PRIVATE
-#define        TCPTV_MSL       ( 30*PR_SLOWHZ)         /* max seg lifetime (hah!) */
+#define        TCPTV_MSL       ( 15*TCP_RETRANSHZ)             /* max seg lifetime (hah!) */

 #define        TCPTV_SRTTBASE  0                       /* base roundtrip time;
 #define        TCPTV_SRTTBASE  0                       /* base roundtrip time;

-                                                  if 0, no idea yet */
-#define        TCPTV_RTOBASE   (  3*PR_SLOWHZ)         /* assumed RTO if no info */
-#define        TCPTV_SRTTDFLT  (  3*PR_SLOWHZ)         /* assumed RTT if no info */
+                                                  if  0, no idea yet */
+#define        TCPTV_RTOBASE   (  1*TCP_RETRANSHZ)             /* assumed RTO if no info */
+#define        TCPTV_SRTTDFLT  (  1*TCP_RETRANSHZ)             /* assumed RTT if no info */

 
 

-#define        TCPTV_PERSMIN   (  5*PR_SLOWHZ)         /* retransmit persistence */
-#define        TCPTV_PERSMAX   ( 60*PR_SLOWHZ)         /* maximum persist interval */
+#define        TCPTV_PERSMIN   (  5*TCP_RETRANSHZ)             /* retransmit persistence */
+#define        TCPTV_PERSMAX   ( 60*TCP_RETRANSHZ)             /* maximum persist interval */

 
 

-#define        TCPTV_KEEP_INIT ( 75*PR_SLOWHZ)         /* initial connect keep alive */
-#define        TCPTV_KEEP_IDLE (120*60*PR_SLOWHZ)      /* dflt time before probing */
-#define        TCPTV_KEEPINTVL ( 75*PR_SLOWHZ)         /* default probe interval */
+#define        TCPTV_KEEP_INIT ( 75*TCP_RETRANSHZ)             /* initial connect keep alive */
+#define        TCPTV_KEEP_IDLE (120*60*TCP_RETRANSHZ)  /* dflt time before probing */
+#define        TCPTV_KEEPINTVL ( 75*TCP_RETRANSHZ)             /* default probe interval */

 #define        TCPTV_KEEPCNT   8                       /* max probes before drop */
 
 #define        TCPTV_KEEPCNT   8                       /* max probes before drop */
 

-#define        TCPTV_MIN       (  1*PR_SLOWHZ)         /* minimum allowable value */
-#define        TCPTV_REXMTMAX  ( 64*PR_SLOWHZ)         /* max allowable REXMT value */
+#define        TCPTV_REXMTMAX  ( 64*TCP_RETRANSHZ )    /* max allowable REXMT value */
+#define        TCPTV_REXMTMIN  ( TCP_RETRANSHZ/33 )    /* min REXMT for non-local connections */
+#define TCPTV_UNACKWIN ( TCP_RETRANSHZ/10 )    /* Window for counting rcv bytes to see if 
+                                                  ack-stretching can start (default 100 ms) */
+#define TCPTV_MAXRCVIDLE (TCP_RETRANSHZ/5 )    /* Receiver idle time, avoid ack-stretching after that*/
+#define TCPTV_RCVBUFIDLE (TCP_RETRANSHZ/2)     /* Receiver idle time, for rcv socket buffer resizing */
+
+/* No ack stretching during slow-start, until we see some packets.
+ * By the time the receiver gets 512 packets, the senders cwnd 
+ * should open by a few hundred packets consdering the 
+ * slow-start progression.
+ */
+#define TCP_RCV_SS_PKTCOUNT     512

 
 #define TCPTV_TWTRUNC  8                       /* RTO factor to truncate TW */
 
 
 #define TCPTV_TWTRUNC  8                       /* RTO factor to truncate TW */
 


@@ -136,42 +160,115 @@
 
 #define        TCP_MAXRXTSHIFT 12                      /* maximum retransmits */
 
 
 #define        TCP_MAXRXTSHIFT 12                      /* maximum retransmits */
 

-#define        TCPTV_DELACK    PR_FASTHZ               /* 125ms timeout */
-

 #ifdef TCPTIMERS
 static char *tcptimers[] =
 #ifdef TCPTIMERS
 static char *tcptimers[] =

-    { "REXMT", "PERSIST", "KEEP", "2MSL" };
+    { "REXMT", "PERSIST", "KEEP", "2MSL" , "DELACK"};

 #endif
 
 #endif
 

+#ifdef KERNEL
+
+/* We consider persist, keep and 2msl as slow timers which can be coalesced
+ * at a higher granularity (500 ms). Rexmt and delayed ack are considered fast
+ * timers which fire in the order of 100ms.
+ *
+ * The following conditional is to check if a timer is one of the slow timers. This 
+ * is fast and works well for now. If we add more slow timers for any reason, 
+ * we may need to change this.
+ */
+#define IS_TIMER_SLOW(ind) ((ind & 0x3) != 0)
+
+struct tcptimerlist;
+
+struct tcptimerentry {
+        LIST_ENTRY(tcptimerentry) le;  /* links for timer list */ 
+        uint32_t timer_start;          /* tcp clock when the timer was started */
+       uint16_t index;                 /* index of lowest timer that needs to run first */
+       uint32_t runtime;               /* deadline at which the first timer has to fire */
+};
+
+LIST_HEAD(timerlisthead, tcptimerentry);
+
+struct tcptimerlist {
+       struct timerlisthead lhead;     /* head of the list of timer entries */
+       lck_mtx_t *mtx;                 /* lock to protect the list */
+       lck_attr_t *mtx_attr;           /* mutex attributes */
+       lck_grp_t *mtx_grp;             /* mutex group definition */
+       lck_grp_attr_t *mtx_grp_attr;   /* mutex group attributes */
+       uint32_t fast_quantum;          /* minimum time quantum to coalesce fast timers */
+       uint32_t slow_quantum;          /* minimum time quantum to coalesce slow timers */
+       thread_call_t call;             /* call entry */
+       uint32_t runtime;               /* time at which this list is going to run */
+       uint32_t entries;               /* Number of entries on the list */
+       uint32_t maxentries;            /* Max number of entries at any time */
+
+       /* Set desired mode when timer list running */
+       boolean_t running;              /* Set when timer list is being processed */
+#define TCP_TIMERLIST_FASTMODE 0x1
+#define TCP_TIMERLIST_SLOWMODE 0x2
+       uint32_t mode;                  /* Current mode, fast or slow */
+       uint32_t pref_mode;             /* Preferred mode set by a connection, fast or slow */
+       uint32_t pref_offset;           /* Preferred offset set by a connection */
+       uint32_t idlegen;               /* Number of times the list has been idle in fast mode */
+       struct tcptimerentry *next_te;  /* Store the next timer entry pointer to process */
+
+};
+
+#define TCP_FASTMODE_IDLEGEN_MAX 20    /* Approximately 2 seconds */
+

 /*
 /*

- * Force a time value to be in a certain range.
+ * Minimum retransmit timeout is set to 30ms. We add a slop of 
+ * 200 ms to the retransmit value to account for processing 
+ * variance and delayed ack. This extra 200ms will help to avoid 
+ * spurious retransmits by taking into consideration the receivers 
+ * that wait for delayed ack timer instead of generating an ack 
+ * for every two packets.
+ *
+ * On a local link, the minimum retransmit timeout is 100ms and
+ * variance is set to 0. This will make the sender a little bit more
+ * aggressive on local link. When the connection is not established yet,
+ * there is no need to add an extra 200ms to retransmit timeout because
+ * the initial value is high (1s) and delayed ack is not a problem in 
+ * that case.

  */
  */

-#define        TCPT_RANGESET(tv, value, tvmin, tvmax) do { \
-       (tv) = (value); \
-       if ((u_long)(tv) < (u_long)(tvmin)) \
+#define TCPTV_REXMTSLOP ( TCP_RETRANSHZ/5 )    /* rexmt slop allowed (200 ms) */
+
+/* macro to decide when retransmit slop (described above) should be added */
+#define TCP_ADD_REXMTSLOP(tp) (tp->t_state >= TCPS_ESTABLISHED) 
+
+#define        TCPT_RANGESET(tv, value, tvmin, tvmax, addslop) do { \
+       (tv) = ((addslop) ? tcp_rexmt_slop : 0) + (value); \
+       if ((uint32_t)(tv) < (uint32_t)(tvmin)) \

                (tv) = (tvmin); \
                (tv) = (tvmin); \

-       else if ((u_long)(tv) > (u_long)(tvmax)) \
+       else if ((uint32_t)(tv) > (uint32_t)(tvmax)) \

                (tv) = (tvmax); \
 } while(0)
 
                (tv) = (tvmax); \
 } while(0)
 

-#ifdef KERNEL
+#define TCP_KEEPIDLE(tp) \
+       (tp->t_keepidle && (tp->t_inpcb->inp_socket->so_options & SO_KEEPALIVE) ? \
+               tp->t_keepidle : tcp_keepidle)
+
+/* Since we did not add rexmt slop for local connections, we should add
+ * it to idle timeout. Otherwise local connections will reach idle state
+ * quickly
+ */
+#define TCP_IDLETIMEOUT(tp) \
+       (((TCP_ADD_REXMTSLOP(tp)) ? 0 : tcp_rexmt_slop) + tp->t_rxtcur)
+

 extern int tcp_keepinit;               /* time to establish connection */
 extern int tcp_keepidle;               /* time before keepalive probes begin */
 extern int tcp_keepintvl;              /* time between keepalive probes */
 extern int tcp_maxidle;                        /* time to drop after starting probes */
 extern int tcp_keepinit;               /* time to establish connection */
 extern int tcp_keepidle;               /* time before keepalive probes begin */
 extern int tcp_keepintvl;              /* time between keepalive probes */
 extern int tcp_maxidle;                        /* time to drop after starting probes */

-extern int tcp_delacktime;             /* time before sending a delayed ACK */
+extern int tcp_delack;                 /* delayed ack timer */

 extern int tcp_maxpersistidle;
 extern int tcp_msl;
 extern int tcp_ttl;                    /* time to live for TCP segs */
 extern int tcp_backoff[];
 extern int tcp_maxpersistidle;
 extern int tcp_msl;
 extern int tcp_ttl;                    /* time to live for TCP segs */
 extern int tcp_backoff[];

+extern int tcp_rexmt_slop;
+extern u_int32_t tcp_max_persist_timeout;      /* Maximum persistence for Zero Window Probes */

 
 

-void   tcp_timer_2msl __P((void *xtp));
-void   tcp_timer_keep __P((void *xtp));
-void   tcp_timer_persist __P((void *xtp));
-void   tcp_timer_rexmt __P((void *xtp));
-void   tcp_timer_delack __P((void *xtp));
+#define OFFSET_FROM_START(tp, off) ((tcp_now + (off)) - (tp)->tentry.timer_start)

 
 #endif /* KERNEL */
 
 #endif /* KERNEL */

-#endif /* __APPLE_API_PRIVATE */
-
+#endif /* PRIVATE */

 #endif /* !_NETINET_TCP_TIMER_H_ */
 #endif /* !_NETINET_TCP_TIMER_H_ */

+