rf_parityscan.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
 *
 * 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.
 */

/*****************************************************************************
 *
 * rf_parityscan.c -- misc utilities related to parity verification
 *
 *****************************************************************************/

/*
 * $Locker:  $
 * $Log: rf_parityscan.c,v $
 * Revision 1.47  1996/08/20  20:35:01  jimz
 * change diagnostic string in rewrite
 *
 * Revision 1.46  1996/08/20  20:03:19  jimz
 * fixed parity rewrite to actually use arch-specific parity stuff
 * (this ever worked... how?)
 *
 * Revision 1.45  1996/08/16  17:41:25  jimz
 * allow rewrite parity on any fault-tolerant arch
 *
 * Revision 1.44  1996/07/28  20:31:39  jimz
 * i386netbsd port
 * true/false fixup
 *
 * Revision 1.43  1996/07/27  23:36:08  jimz
 * Solaris port of simulator
 *
 * Revision 1.42  1996/07/22  21:12:01  jimz
 * clean up parity scan status printing
 *
 * Revision 1.41  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 1.40  1996/07/13  00:00:59  jimz
 * sanitized generalized reconstruction architecture
 * cleaned up head sep, rbuf problems
 *
 * Revision 1.39  1996/07/09  21:44:26  jimz
 * fix bogus return code in VerifyParityBasic when a stripe can't be corrected
 *
 * Revision 1.38  1996/06/20  17:56:57  jimz
 * update VerifyParity to check complete AccessStripeMaps
 *
 * Revision 1.37  1996/06/19  22:23:01  jimz
 * parity verification is now a layout-configurable thing
 * not all layouts currently support it (correctly, anyway)
 *
 * Revision 1.36  1996/06/09  02:36:46  jimz
 * lots of little crufty cleanup- fixup whitespace
 * issues, comment #ifdefs, improve typing in some
 * places (esp size-related)
 *
 * Revision 1.35  1996/06/07  22:26:27  jimz
 * type-ify which_ru (RF_ReconUnitNum_t)
 *
 * Revision 1.34  1996/06/07  21:33:04  jimz
 * begin using consistent types for sector numbers,
 * stripe numbers, row+col numbers, recon unit numbers
 *
 * Revision 1.33  1996/06/05  18:06:02  jimz
 * Major code cleanup. The Great Renaming is now done.
 * Better modularity. Better typing. Fixed a bunch of
 * synchronization bugs. Made a lot of global stuff
 * per-desc or per-array. Removed dead code.
 *
 * Revision 1.32  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 1.31  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 1.30  1996/05/30  23:22:16  jimz
 * bugfixes of serialization, timing problems
 * more cleanup
 *
 * Revision 1.29  1996/05/30  12:59:18  jimz
 * make etimer happier, more portable
 *
 * Revision 1.28  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 1.27  1996/05/27  18:56:37  jimz
 * more code cleanup
 * better typing
 * compiles in all 3 environments
 *
 * Revision 1.26  1996/05/24  22:17:04  jimz
 * continue code + namespace cleanup
 * typed a bunch of flags
 *
 * Revision 1.25  1996/05/24  04:28:55  jimz
 * release cleanup ckpt
 *
 * Revision 1.24  1996/05/23  21:46:35  jimz
 * checkpoint in code cleanup (release prep)
 * lots of types, function names have been fixed
 *
 * Revision 1.23  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 1.22  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 1.21  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 1.20  1995/12/12  18:10:06  jimz
 * MIN -> RF_MIN, MAX -> RF_MAX, ASSERT -> RF_ASSERT
 * fix 80-column brain damage in comments
 *
 * Revision 1.19  1995/11/30  16:16:49  wvcii
 * added copyright info
 *
 * Revision 1.18  1995/11/19  16:32:19  wvcii
 * eliminated initialization of dag header fields which no longer exist
 * (numDags, numDagsDone, firstHdr)
 *
 * Revision 1.17  1995/11/07  16:23:36  wvcii
 * added comments, asserts, and prototypes
 * encoded commit point nodes, barrier, and antecedents types into dags
 *
 */

#include "rf_types.h"
#include "rf_raid.h"
#include "rf_dag.h"
#include "rf_dagfuncs.h"
#include "rf_dagutils.h"
#include "rf_mcpair.h"
#include "rf_general.h"
#include "rf_engine.h"
#include "rf_parityscan.h"
#include "rf_map.h"
#include "rf_sys.h"

/*****************************************************************************************
 *
 * walk through the entire arry and write new parity.
 * This works by creating two DAGs, one to read a stripe of data and one to
 * write new parity.  The first is executed, the data is xored together, and
 * then the second is executed.  To avoid constantly building and tearing down
 * the DAGs, we create them a priori and fill them in with the mapping
 * information as we go along.
 *
 * there should never be more than one thread running this.
 *
 ****************************************************************************************/

int rf_RewriteParity(raidPtr)
  RF_Raid_t  *raidPtr;
{
  RF_RaidLayout_t *layoutPtr = &raidPtr->Layout;
  RF_AccessStripeMapHeader_t *asm_h;
  int old_pctg, new_pctg, rc;
  RF_PhysDiskAddr_t pda;
  RF_SectorNum_t i;

  pda.startSector = 0;
  pda.numSector   = raidPtr->Layout.sectorsPerStripeUnit;
  old_pctg = -1;
  for (i=0; i<raidPtr->totalSectors; i+=layoutPtr->dataSectorsPerStripe) {
    asm_h = rf_MapAccess(raidPtr, i, layoutPtr->dataSectorsPerStripe, NULL, RF_DONT_REMAP);
    rc = rf_VerifyParity(raidPtr, asm_h->stripeMap, 1, 0);
    switch (rc) {
      case RF_PARITY_OKAY:
      case RF_PARITY_CORRECTED:
        break;
      case RF_PARITY_BAD:
        printf("Parity bad during correction\n");
        RF_PANIC();
        break;
      case RF_PARITY_COULD_NOT_CORRECT:
        printf("Could not correct bad parity\n");
        RF_PANIC();
        break;
      case RF_PARITY_COULD_NOT_VERIFY:
        printf("Could not verify parity\n");
        RF_PANIC();
        break;
      default:
        printf("Bad rc=%d from VerifyParity in RewriteParity\n", rc);
        RF_PANIC();
    }
    rf_FreeAccessStripeMap(asm_h);
    new_pctg = i*1000/raidPtr->totalSectors;
    if (new_pctg != old_pctg) {
#ifndef KERNEL
      fprintf(stderr,"\rParity rewrite: %d.%d%% complete",
        new_pctg/10, new_pctg%10);
      fflush(stderr);
#endif /* !KERNEL */
    }
    old_pctg = new_pctg;
  }
#ifndef KERNEL
  fprintf(stderr,"\rParity rewrite: 100.0%% complete\n");
#endif /* !KERNEL */
}

/*****************************************************************************************
 *
 * verify that the parity in a particular stripe is correct.
 * we validate only the range of parity defined by parityPDA, since
 * this is all we have locked.  The way we do this is to create an asm
 * that maps the whole stripe and then range-restrict it to the parity
 * region defined by the parityPDA.
 *
 ****************************************************************************************/
int rf_VerifyParity(raidPtr, aasm, correct_it, flags)
  RF_Raid_t             *raidPtr;
  RF_AccessStripeMap_t  *aasm;
  int                    correct_it;
  RF_RaidAccessFlags_t   flags;
{
  RF_PhysDiskAddr_t *parityPDA;
  RF_AccessStripeMap_t *doasm;
  RF_LayoutSW_t *lp;
  int lrc, rc;

  lp = raidPtr->Layout.map;
  if (lp->faultsTolerated == 0) {
    /*
     * There isn't any parity. Call it "okay."