rf_parityloggingdags.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: 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.
 */

/*
 * $Log: rf_parityloggingdags.c,v $
 * Revision 1.27  1996/07/28  20:31:39  jimz
 * i386netbsd port
 * true/false fixup
 *
 * Revision 1.26  1996/07/27  23:36:08  jimz
 * Solaris port of simulator
 *
 * Revision 1.25  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 1.24  1996/06/11  13:47:21  jimz
 * fix up for in-kernel compilation
 *
 * Revision 1.23  1996/06/07  22:26:27  jimz
 * type-ify which_ru (RF_ReconUnitNum_t)
 *
 * Revision 1.22  1996/06/07  21:33:04  jimz
 * begin using consistent types for sector numbers,
 * stripe numbers, row+col numbers, recon unit numbers
 *
 * Revision 1.21  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 1.20  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 1.19  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 1.18  1996/05/27  18:56:37  jimz
 * more code cleanup
 * better typing
 * compiles in all 3 environments
 *
 * Revision 1.17  1996/05/24  22:17:04  jimz
 * continue code + namespace cleanup
 * typed a bunch of flags
 *
 * Revision 1.16  1996/05/24  04:28:55  jimz
 * release cleanup ckpt
 *
 * Revision 1.15  1996/05/23  21:46:35  jimz
 * checkpoint in code cleanup (release prep)
 * lots of types, function names have been fixed
 *
 * Revision 1.14  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 1.13  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 1.12  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 1.11  1996/05/03  19:42:02  wvcii
 * added includes for dag library
 *
 * Revision 1.10  1995/12/12  18:10:06  jimz
 * MIN -> RF_MIN, MAX -> RF_MAX, ASSERT -> RF_ASSERT
 * fix 80-column brain damage in comments
 *
 * Revision 1.9  1995/12/06  20:55:24  wvcii
 * added prototyping
 * fixed bug in dag header numSuccedents count for both small and large dags
 *
 * Revision 1.8  1995/11/30  16:08:01  wvcii
 * added copyright info
 *
 * Revision 1.7  1995/11/07  15:29:05  wvcii
 * reorganized code, adding comments and asserts
 * dag creation routines now generate term node
 * encoded commit point, barrier, and antecedence types into dags
 *
 * Revision 1.6  1995/09/07  15:52:06  jimz
 * noop compile when INCLUDE_PARITYLOGGING not defined
 *
 * Revision 1.5  1995/06/15  13:51:53  robby
 * updated some wrong prototypes (after prototyping rf_dagutils.h)
 *
 * Revision 1.4  1995/06/09  13:15:05  wvcii
 * code is now nonblocking
 *
 * Revision 1.3  95/05/31  13:09:14  wvcii
 * code debug
 * 
 * Revision 1.2  1995/05/21  15:34:14  wvcii
 * code debug
 *
 * Revision 1.1  95/05/16  14:36:53  wvcii
 * Initial revision
 * 
 *
 */

#include "rf_archs.h"

#if RF_INCLUDE_PARITYLOGGING > 0

/*
  DAGs specific to parity logging are created here
 */

#include "rf_types.h"
#include "rf_raid.h"
#include "rf_dag.h"
#include "rf_dagutils.h"
#include "rf_dagfuncs.h"
#include "rf_threadid.h"
#include "rf_debugMem.h"
#include "rf_paritylog.h"
#include "rf_memchunk.h"
#include "rf_general.h"

/******************************************************************************
 *
 * creates a DAG to perform a large-write operation:
 *
 *         / Rod \     / Wnd \
 * H -- NIL- Rod - NIL - Wnd ------ NIL - T
 *         \ Rod /     \ Xor - Lpo /
 *
 * The writes are not done until the reads complete because if they were done in
 * parallel, a failure on one of the reads could leave the parity in an inconsistent
 * state, so that the retry with a new DAG would produce erroneous parity.
 *
 * Note:  this DAG has the nasty property that none of the buffers allocated for reading
 *        old data can be freed until the XOR node fires.  Need to fix this.
 *
 * The last two arguments are the number of faults tolerated, and function for the
 * redundancy calculation. The undo for the redundancy calc is assumed to be null
 *
 *****************************************************************************/

void rf_CommonCreateParityLoggingLargeWriteDAG(
  RF_Raid_t              *raidPtr,
  RF_AccessStripeMap_t   *asmap,
  RF_DagHeader_t         *dag_h,
  void                   *bp,
  RF_RaidAccessFlags_t    flags,
  RF_AllocListElem_t     *allocList,
  int                     nfaults,
  int                   (*redFunc)())
{
  RF_DagNode_t *nodes, *wndNodes, *rodNodes, *syncNode, *xorNode, *lpoNode, *blockNode, *unblockNode, *termNode;
  int nWndNodes, nRodNodes, i;
  RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
  RF_AccessStripeMapHeader_t *new_asm_h[2];
  int nodeNum, asmNum;
  RF_ReconUnitNum_t which_ru;
  char *sosBuffer, *eosBuffer;
  RF_PhysDiskAddr_t *pda;
  RF_StripeNum_t parityStripeID = rf_RaidAddressToParityStripeID(&(raidPtr->Layout), asmap->raidAddress, &which_ru);

  if (rf_dagDebug)
    printf("[Creating parity-logging large-write DAG]\n");
  RF_ASSERT(nfaults == 1); /* this arch only single fault tolerant */
  dag_h->creator = "ParityLoggingLargeWriteDAG";

  /* alloc the Wnd nodes, the xor node, and the Lpo node */
  nWndNodes = asmap->numStripeUnitsAccessed;
  RF_CallocAndAdd(nodes, nWndNodes + 6, sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList);
  i = 0;
  wndNodes    = &nodes[i]; i += nWndNodes;
  xorNode     = &nodes[i]; i += 1;
  lpoNode     = &nodes[i]; i += 1;
  blockNode   = &nodes[i]; i += 1;
  syncNode    = &nodes[i]; i += 1;
  unblockNode = &nodes[i]; i += 1;
  termNode    = &nodes[i]; i += 1;

  dag_h->numCommitNodes = nWndNodes + 1;
  dag_h->numCommits = 0;
  dag_h->numSuccedents = 1;

  rf_MapUnaccessedPortionOfStripe(raidPtr, layoutPtr, asmap, dag_h, new_asm_h, &nRodNodes, &sosBuffer, &eosBuffer, allocList);
  if (nRodNodes > 0)
    RF_CallocAndAdd(rodNodes, nRodNodes, sizeof(RF_DagNode_t), (RF_DagNode_t *), allocList);

  /* begin node initialization */
  rf_InitNode(blockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nRodNodes + 1, 0, 0, 0, dag_h, "Nil", allocList);
  rf_InitNode(unblockNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, 1, nWndNodes + 1, 0, 0, dag_h, "Nil", allocList);
  rf_InitNode(syncNode, rf_wait, RF_FALSE, rf_NullNodeFunc, rf_NullNodeUndoFunc, NULL, nWndNodes + 1, nRodNodes + 1, 0, 0, dag_h, "Nil", allocList);
  rf_InitNode(termNode, rf_wait, RF_FALSE, rf_TerminateFunc, rf_TerminateUndoFunc, NULL, 0, 1, 0, 0, dag_h, "Trm", allocList);

  /* initialize the Rod nodes */
  for (nodeNum = asmNum = 0; asmNum < 2; asmNum++) {
    if (new_asm_h[asmNum]) {
      pda = new_asm_h[asmNum]->stripeMap->physInfo;
      while (pda) {
	rf_InitNode(&rodNodes[nodeNum], rf_wait, RF_FALSE, rf_DiskReadFunc,rf_DiskReadUndoFunc,rf_GenericWakeupFunc,1,1,4,0, dag_h, "Rod", allocList);
	rodNodes[nodeNum].params[0].p = pda;
	rodNodes[nodeNum].params[1].p = pda->bufPtr;
	rodNodes[nodeNum].params[2].v = parityStripeID;
	rodNodes[nodeNum].params[3].v = RF_CREATE_PARAM3(RF_IO_NORMAL_PRIORITY, 0, 0, which_ru);
	nodeNum++;
	pda=pda->next;
      }
    }
  }
  RF_ASSERT(nodeNum == nRodNodes);

  /* initialize the wnd nodes */