rf_dagfuncs.c

RAIDFrame是个非常好的磁盘阵列RAID仿真工具

/*
 * Copyright (c) 1995 Carnegie-Mellon University.
 * All rights reserved.
 *
 * Author: Mark Holland, William V. Courtright II
 *
 * Permission to use, copy, modify and distribute this software and
 * its documentation is hereby granted, provided that both the copyright
 * notice and this permission notice appear in all copies of the
 * software, derivative works or modified versions, and any portions
 * thereof, and that both notices appear in supporting documentation.
 *
 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
 * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
 *
 * Carnegie Mellon requests users of this software to return to
 *
 *  Software Distribution Coordinator  or  [e-mail]
 *  School of Computer Science
 *  Carnegie Mellon University
 *  Pittsburgh PA 15213-3890
 *
 * any improvements or extensions that they make and grant Carnegie the
 * rights to redistribute these changes.
 */

/*
 * dagfuncs.c -- DAG node execution routines
 *
 * Rules:
 * 1. Every DAG execution function must eventually cause node->status to
 *    get set to "good" or "bad", and "FinishNode" to be called. In the
 *    case of nodes that complete immediately (xor, NullNodeFunc, etc),
 *    the node execution function can do these two things directly. In
 *    the case of nodes that have to wait for some event (a disk read to
 *    complete, a lock to be released, etc) to occur before they can
 *    complete, this is typically achieved by having whatever module
 *    is doing the operation call GenericWakeupFunc upon completion.
 * 2. DAG execution functions should check the status in the DAG header
 *    and NOP out their operations if the status is not "enable". However,
 *    execution functions that release resources must be sure to release
 *    them even when they NOP out the function that would use them.
 *    Functions that acquire resources should go ahead and acquire them
 *    even when they NOP, so that a downstream release node will not have
 *    to check to find out whether or not the acquire was suppressed.
 */

/* $Locker:  $
 * $Log: rf_dagfuncs.c,v $
 * Revision 1.64  1996/07/31  16:29:26  jimz
 * LONGSHIFT -> RF_LONGSHIFT, defined in rf_types.h
 *
 * Revision 1.63  1996/07/30  04:00:20  jimz
 * define LONGSHIFT for mips
 *
 * Revision 1.62  1996/07/28  20:31:39  jimz
 * i386netbsd port
 * true/false fixup
 *
 * Revision 1.61  1996/07/27  23:36:08  jimz
 * Solaris port of simulator
 *
 * Revision 1.60  1996/07/22  19:52:16  jimz
 * switched node params to RF_DagParam_t, a union of
 * a 64-bit int and a void *, for better portability
 * attempted hpux port, but failed partway through for
 * lack of a single C compiler capable of compiling all
 * source files
 *
 * Revision 1.59  1996/07/18  22:57:14  jimz
 * port simulator to AIX
 *
 * Revision 1.58  1996/07/17  21:00:58  jimz
 * clean up timer interface, tracing
 *
 * Revision 1.57  1996/07/15  17:22:18  jimz
 * nit-pick code cleanup
 * resolve stdlib problems on DEC OSF
 *
 * Revision 1.56  1996/06/11  01:27:50  jimz
 * Fixed bug where diskthread shutdown would crash or hang. This
 * turned out to be two distinct bugs:
 * (1) [crash] The thread shutdown code wasn't properly waiting for
 * all the diskthreads to complete. This caused diskthreads that were
 * exiting+cleaning up to unlock a destroyed mutex.
 * (2) [hang] TerminateDiskQueues wasn't locking, and DiskIODequeue
 * only checked for termination _after_ a wakeup if the queues were
 * empty. This was a race where the termination wakeup could be lost
 * by the dequeueing thread, and the system would hang waiting for the
 * thread to exit, while the thread waited for an I/O or a signal to
 * check the termination flag.
 *
 * Revision 1.55  1996/06/10  22:23:18  wvcii
 * disk and xor funcs now optionally support undo logging
 * for backward error recovery experiments
 *
 * Revision 1.54  1996/06/10  11:55:47  jimz
 * Straightened out some per-array/not-per-array distinctions, fixed
 * a couple bugs related to confusion. Added shutdown lists. Removed
 * layout shutdown function (now subsumed by shutdown lists).
 *
 * Revision 1.53  1996/06/07  21:33:04  jimz
 * begin using consistent types for sector numbers,
 * stripe numbers, row+col numbers, recon unit numbers
 *
 * Revision 1.52  1996/06/06  17:28:44  jimz
 * add new read mirror partition func, rename old read mirror
 * to rf_DiskReadMirrorIdleFunc
 *
 * Revision 1.51  1996/06/03  23:28:26  jimz
 * more bugfixes
 * check in tree to sync for IPDS runs with current bugfixes
 * there still may be a problem with threads in the script test
 * getting I/Os stuck- not trivially reproducible (runs ~50 times
 * in a row without getting stuck)
 *
 * Revision 1.50  1996/06/02  17:31:48  jimz
 * Moved a lot of global stuff into array structure, where it belongs.
 * Fixed up paritylogging, pss modules in this manner. Some general
 * code cleanup. Removed lots of dead code, some dead files.
 *
 * Revision 1.49  1996/05/31  22:26:54  jimz
 * fix a lot of mapping problems, memory allocation problems
 * found some weird lock issues, fixed 'em
 * more code cleanup
 *
 * Revision 1.48  1996/05/30  12:59:18  jimz
 * make etimer happier, more portable
 *
 * Revision 1.47  1996/05/30  11:29:41  jimz
 * Numerous bug fixes. Stripe lock release code disagreed with the taking code
 * about when stripes should be locked (I made it consistent: no parity, no lock)
 * There was a lot of extra serialization of I/Os which I've removed- a lot of
 * it was to calculate values for the cache code, which is no longer with us.
 * More types, function, macro cleanup. Added code to properly quiesce the array
 * on shutdown. Made a lot of stuff array-specific which was (bogusly) general
 * before. Fixed memory allocation, freeing bugs.
 *
 * Revision 1.46  1996/05/24  22:17:04  jimz
 * continue code + namespace cleanup
 * typed a bunch of flags
 *
 * Revision 1.45  1996/05/24  04:28:55  jimz
 * release cleanup ckpt
 *
 * Revision 1.44  1996/05/23  21:46:35  jimz
 * checkpoint in code cleanup (release prep)
 * lots of types, function names have been fixed
 *
 * Revision 1.43  1996/05/23  00:33:23  jimz
 * code cleanup: move all debug decls to rf_options.c, all extern
 * debug decls to rf_options.h, all debug vars preceded by rf_
 *
 * Revision 1.42  1996/05/18  19:51:34  jimz
 * major code cleanup- fix syntax, make some types consistent,
 * add prototypes, clean out dead code, et cetera
 *
 * Revision 1.41  1996/05/08  21:01:24  jimz
 * fixed up enum type names that were conflicting with other
 * enums and function names (ie, "panic")
 * future naming trends will be towards RF_ and rf_ for
 * everything raidframe-related
 *
 * Revision 1.40  1996/05/08  15:24:14  wvcii
 * modified GenericWakeupFunc to use recover, undone, and panic node states
 *
 * Revision 1.39  1996/05/02  17:18:01  jimz
 * fix up headers for user-land, following ccmn cleanup
 *
 * Revision 1.38  1996/05/01  16:26:51  jimz
 * don't include rf_ccmn.h (get ready to phase out)
 *
 * Revision 1.37  1995/12/12  18:10:06  jimz
 * MIN -> RF_MIN, MAX -> RF_MAX, ASSERT -> RF_ASSERT
 * fix 80-column brain damage in comments
 *
 * Revision 1.36  1995/12/04  19:19:09  wvcii
 * modified DiskReadMirrorFunc
 *  - added fifth parameter, physical disk address of mirror copy
 *  - SelectIdleDisk conditionally swaps parameters 0 & 4
 *
 * Revision 1.35  1995/12/01  15:58:33  root
 * added copyright info
 *
 * Revision 1.34  1995/11/17  18:12:17  amiri
 * Changed DiskReadMirrorFunc to use the generic mapping routines
 * to find the mirror of the data, function was assuming RAID level 1.
 *
 * Revision 1.33  1995/11/17  15:15:59  wvcii
 * changes in DiskReadMirrorFunc
 *   - added ASSERTs
 *   - added call to MapParityRAID1
 *
 * Revision 1.32  1995/11/07  16:25:50  wvcii
 * added DiskUnlockFuncForThreads
 * general debugging of undo functions (first time they were used)
 *
 * Revision 1.31  1995/09/06  19:23:36  wvcii
 * fixed tracing for parity logging nodes
 *
 * Revision 1.30  95/07/07  00:13:01  wvcii
 * added 4th parameter to ParityLogAppend
 * 
 */

#ifndef KERNEL
#include <errno.h>
#endif /* !KERNEL */

#include <sys/ioctl.h>
#include <sys/param.h>

#include "rf_archs.h"
#include "rf_raid.h"
#include "rf_dag.h"
#include "rf_layout.h"
#include "rf_etimer.h"
#include "rf_acctrace.h"
#include "rf_diskqueue.h"
#include "rf_dagfuncs.h"
#include "rf_general.h"
#include "rf_engine.h"
#include "rf_dagutils.h"

#ifdef KERNEL
#include "rf_kintf.h"
#endif /* KERNEL */

#if RF_INCLUDE_PARITYLOGGING > 0
#include "rf_paritylog.h"
#endif /* RF_INCLUDE_PARITYLOGGING > 0 */

int (*rf_DiskReadFunc)(), (*rf_DiskWriteFunc)();
int (*rf_DiskReadUndoFunc)(), (*rf_DiskWriteUndoFunc)();
int (*rf_DiskUnlockFunc)(), (*rf_DiskUnlockUndoFunc)();
int (*rf_RegularXorUndoFunc)();
int (*rf_SimpleXorUndoFunc)();
int (*rf_RecoveryXorUndoFunc)();

/*****************************************************************************************
 * main (only) configuration routine for this module
 ****************************************************************************************/
int rf_ConfigureDAGFuncs(listp)
  RF_ShutdownList_t  **listp;
{
  RF_ASSERT( ((sizeof(long)==8) && RF_LONGSHIFT==3) || ((sizeof(long)==4)  && RF_LONGSHIFT==2) );
  rf_DiskReadFunc  = rf_DiskReadFuncForThreads;
  rf_DiskReadUndoFunc = rf_DiskUndoFunc;
  rf_DiskWriteFunc = rf_DiskWriteFuncForThreads;
  rf_DiskWriteUndoFunc = rf_DiskUndoFunc;
  rf_DiskUnlockFunc = rf_DiskUnlockFuncForThreads;
  rf_DiskUnlockUndoFunc = rf_NullNodeUndoFunc;
  rf_RegularXorUndoFunc = rf_NullNodeUndoFunc;
  rf_SimpleXorUndoFunc = rf_NullNodeUndoFunc;
  rf_RecoveryXorUndoFunc = rf_NullNodeUndoFunc;
  return(0);
}



/*****************************************************************************************
 * the execution function associated with a terminate node
 ****************************************************************************************/
int rf_TerminateFunc(node)
  RF_DagNode_t  *node;
{
  RF_ASSERT(node->dagHdr->numCommits == node->dagHdr->numCommitNodes);
  node->status = rf_good;
  return(rf_FinishNode(node, RF_THREAD_CONTEXT));
}

int rf_TerminateUndoFunc(node)
  RF_DagNode_t  *node;
{
  return(0);
}


/*****************************************************************************************
 * execution functions associated with a mirror node
 *
 * parameters:
 *
 * 0 - physical disk addres of data
 * 1 - buffer for holding read data
 * 2 - parity stripe ID
 * 3 - flags
 * 4 - physical disk address of mirror (parity)
 *
 ****************************************************************************************/

int rf_DiskReadMirrorIdleFunc(node)
  RF_DagNode_t  *node;
{
  /* select the mirror copy with the shortest queue and fill in node parameters
     with physical disk address */

  rf_SelectMirrorDiskIdle(node);
  return(rf_DiskReadFunc(node));
}

int rf_DiskReadMirrorPartitionFunc(node)
  RF_DagNode_t  *node;
{
  /* select the mirror copy with the shortest queue and fill in node parameters
     with physical disk address */

  rf_SelectMirrorDiskPartition(node);
  return(rf_DiskReadFunc(node));
}

int rf_DiskReadMirrorUndoFunc(node)
  RF_DagNode_t  *node;
{
  return(0);
}



#if RF_INCLUDE_PARITYLOGGING > 0
/*****************************************************************************************
 * the execution function associated with a parity log update node
 ****************************************************************************************/
int rf_ParityLogUpdateFunc(node)
  RF_DagNode_t  *node;
{
  RF_PhysDiskAddr_t *pda = (RF_PhysDiskAddr_t *) node->params[0].p;
  caddr_t buf = (caddr_t) node->params[1].p;
  RF_ParityLogData_t *logData;
  RF_AccTraceEntry_t *tracerec = node->dagHdr->tracerec;
  RF_Etimer_t timer;

  if (node->dagHdr->status == rf_enable)
    {
      RF_ETIMER_START(timer);
      logData = rf_CreateParityLogData(RF_UPDATE, pda, buf, (RF_Raid_t *) (node->dagHdr->raidPtr),
				    node->wakeFunc, (void *) node, node->dagHdr->tracerec, timer);
      if (logData)