rf_decluster.c

RAIDFrame是个非常好的磁盘阵列RAID仿真工具

    /* compute the size of each disk, and the number of tables in the last fulltable (which
     * need not be complete)
     */
    if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
	
	PUsPerDisk = layoutPtr->stripeUnitsPerDisk / layoutPtr->SUsPerPU;
	spareRegionDepthInPUs = (info->TablesPerSpareRegion * info->TableDepthInPUs +
				 (info->TablesPerSpareRegion * info->TableDepthInPUs) / (v-1));
	info->SpareRegionDepthInSUs = spareRegionDepthInPUs * layoutPtr->SUsPerPU;
	
	numCompleteSpareRegionsPerDisk = PUsPerDisk / spareRegionDepthInPUs;
	info->NumCompleteSRs = numCompleteSpareRegionsPerDisk;
	extraPUsPerDisk = PUsPerDisk % spareRegionDepthInPUs;

	/* assume conservatively that we need the full amount of spare space in one region in order
	 * to provide spares for the partial spare region at the end of the array.  We set "i" to
	 * the number of tables in the partial spare region.  This may actually include some fulltables.
	 */
	extraPUsPerDisk -= (info->SpareSpaceDepthPerRegionInSUs / layoutPtr->SUsPerPU);
	if (extraPUsPerDisk <= 0) i = 0;
	else i = extraPUsPerDisk/info->TableDepthInPUs;

	complete_FT_count = raidPtr->numRow * (numCompleteSpareRegionsPerDisk * (info->TablesPerSpareRegion/k) + i/k);
        info->FullTableLimitSUID = complete_FT_count * info->SUsPerFullTable;
	info->ExtraTablesPerDisk = i % k;

	/* note that in the last spare region, the spare space is complete even though data/parity space is not */
	totSparePUsPerDisk = (numCompleteSpareRegionsPerDisk+1) * (info->SpareSpaceDepthPerRegionInSUs / layoutPtr->SUsPerPU);
	info->TotSparePUsPerDisk = totSparePUsPerDisk;
	
	layoutPtr->stripeUnitsPerDisk =
	    ((complete_FT_count/raidPtr->numRow) * info->FullTableDepthInPUs +	 	/* data & parity space */
	     info->ExtraTablesPerDisk * info->TableDepthInPUs +
	     totSparePUsPerDisk								/* spare space */
	    ) * layoutPtr->SUsPerPU;
	layoutPtr->dataStripeUnitsPerDisk = 
	    (complete_FT_count * info->FullTableDepthInPUs + info->ExtraTablesPerDisk * info->TableDepthInPUs)
	    * layoutPtr->SUsPerPU * (k-1) / k;

    } else {
        /* non-dist spare case:  force each disk to contain an integral number of tables */
        layoutPtr->stripeUnitsPerDisk /= (info->TableDepthInPUs * layoutPtr->SUsPerPU);
        layoutPtr->stripeUnitsPerDisk *= (info->TableDepthInPUs * layoutPtr->SUsPerPU);

	/* compute the number of tables in the last fulltable, which need not be complete */
        complete_FT_count =
            ((layoutPtr->stripeUnitsPerDisk/layoutPtr->SUsPerPU) / info->FullTableDepthInPUs) * raidPtr->numRow;
	    
        info->FullTableLimitSUID = complete_FT_count * info->SUsPerFullTable;
        info->ExtraTablesPerDisk =
		((layoutPtr->stripeUnitsPerDisk/layoutPtr->SUsPerPU) / info->TableDepthInPUs) % k;
    }

    raidPtr->sectorsPerDisk = layoutPtr->stripeUnitsPerDisk * layoutPtr->sectorsPerStripeUnit;

    /* find the disk offset of the stripe unit where the last fulltable starts */
    numCompleteFullTablesPerDisk = complete_FT_count / raidPtr->numRow;
    diskOffsetOfLastFullTableInSUs = numCompleteFullTablesPerDisk * info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
    if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
        SpareSpaceInSUs  = numCompleteSpareRegionsPerDisk * info->SpareSpaceDepthPerRegionInSUs;
        diskOffsetOfLastFullTableInSUs += SpareSpaceInSUs;
        info->DiskOffsetOfLastSpareSpaceChunkInSUs =
	    diskOffsetOfLastFullTableInSUs + info->ExtraTablesPerDisk * info->TableDepthInPUs * layoutPtr->SUsPerPU;
    }
    info->DiskOffsetOfLastFullTableInSUs = diskOffsetOfLastFullTableInSUs;
    info->numCompleteFullTablesPerDisk = numCompleteFullTablesPerDisk;
	
    /* 4.  create and initialize the lookup tables */
    info->LayoutTable = rf_make_2d_array(b, k, raidPtr->cleanupList);
    if (info->LayoutTable == NULL)
      return(ENOMEM);
    info->OffsetTable = rf_make_2d_array(b, k, raidPtr->cleanupList);
    if (info->OffsetTable == NULL)
      return(ENOMEM);
    info->BlockTable  =	rf_make_2d_array(info->TableDepthInPUs*layoutPtr->SUsPerPU, raidPtr->numCol, raidPtr->cleanupList);
    if (info->BlockTable == NULL)
      return(ENOMEM);

    first_avail_slot = rf_make_1d_array(v, NULL);
    if (first_avail_slot == NULL)
      return(ENOMEM);

    for (i=0; i<b; i++)
      for (j=0; j<k; j++)
        info->LayoutTable[i][j] = *cfgBuf++;

    /* initialize offset table */
    for (i=0; i<b; i++) for (j=0; j<k; j++) {
        info->OffsetTable[i][j] = first_avail_slot[ info->LayoutTable[i][j] ];
        first_avail_slot[ info->LayoutTable[i][j] ]++;
    }

    /* initialize block table */
    for (SUID=l=0; l<layoutPtr->SUsPerPU; l++) {
        for (i=0; i<b; i++) {
            for (j=0; j<k; j++) {
                info->BlockTable[ (info->OffsetTable[i][j] * layoutPtr->SUsPerPU) + l ]
		                [ info->LayoutTable[i][j] ] = SUID;
            }
            SUID++;
        }
    }

    rf_free_1d_array(first_avail_slot, v);

    /* 5.  set up the remaining redundant-but-useful parameters */

    raidPtr->totalSectors = (k*complete_FT_count + raidPtr->numRow*info->ExtraTablesPerDisk) *
    			  info->SUsPerTable * layoutPtr->sectorsPerStripeUnit;
    layoutPtr->numStripe = (raidPtr->totalSectors / layoutPtr->sectorsPerStripeUnit) / (k-1);

    /* strange evaluation order below to try and minimize overflow problems */
    
    layoutPtr->dataSectorsPerStripe = (k-1) * layoutPtr->sectorsPerStripeUnit;
    layoutPtr->bytesPerStripeUnit = layoutPtr->sectorsPerStripeUnit << raidPtr->logBytesPerSector;
    layoutPtr->numDataCol = k-1;
    layoutPtr->numParityCol = 1;

    return(0);
}

/* declustering with distributed sparing */
static void rf_ShutdownDeclusteredDS(arg)
  RF_ThreadArg_t  arg;
{
  RF_DeclusteredConfigInfo_t *info;
  RF_Raid_t *raidPtr;

  raidPtr = (RF_Raid_t *)arg;
  info = (RF_DeclusteredConfigInfo_t *) raidPtr->Layout.layoutSpecificInfo;
  if (info->SpareTable)
    rf_FreeSpareTable(raidPtr);
}

int rf_ConfigureDeclusteredDS(
  RF_ShutdownList_t  **listp,
  RF_Raid_t           *raidPtr,
  RF_Config_t         *cfgPtr)
{
  int rc;

  rc = rf_ConfigureDeclustered(listp, raidPtr, cfgPtr);
  if (rc)
    return(rc);
  rc = rf_ShutdownCreate(listp, rf_ShutdownDeclusteredDS, raidPtr);
  if (rc) {
    RF_ERRORMSG1("Got %d adding shutdown event for DeclusteredDS\n", rc);
    rf_ShutdownDeclusteredDS(raidPtr);
    return(rc);
  }
  return(0);
}

void rf_MapSectorDeclustered(raidPtr, raidSector, row, col, diskSector, remap)
  RF_Raid_t       *raidPtr;
  RF_RaidAddr_t    raidSector;
  RF_RowCol_t     *row;
  RF_RowCol_t     *col;
  RF_SectorNum_t  *diskSector;
  int              remap;
{
    RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
    RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
    RF_StripeNum_t SUID = raidSector / layoutPtr->sectorsPerStripeUnit;
    RF_StripeNum_t FullTableID, FullTableOffset, TableID, TableOffset; 
    RF_StripeNum_t BlockID, BlockOffset, RepIndex;
    RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
    RF_StripeCount_t fulltable_depth  = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
    RF_StripeNum_t base_suid = 0, outSU, SpareRegion=0, SpareSpace=0;

    rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
	    
    FullTableID     = SUID / sus_per_fulltable;		/* fulltable ID within array (across rows) */
    if (raidPtr->numRow == 1) *row = 0;                 /* avoid a mod and a div in the common case */
    else {
      *row            = FullTableID % raidPtr->numRow;
      FullTableID    /= raidPtr->numRow;			/* convert to fulltable ID on this disk */
    }
    if (raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE) {
	SpareRegion = FullTableID / info->FullTablesPerSpareRegion;
        SpareSpace  = SpareRegion * info->SpareSpaceDepthPerRegionInSUs;
    }
    FullTableOffset = SUID % sus_per_fulltable;
    TableID         = FullTableOffset / info->SUsPerTable;
    TableOffset     = FullTableOffset - TableID * info->SUsPerTable;
    BlockID         = TableOffset / info->PUsPerBlock;
    BlockOffset     = TableOffset - BlockID * info->PUsPerBlock;
    BlockID        %= info->BlocksPerTable;
    RepIndex        = info->PUsPerBlock - TableID;
    if (!raidPtr->noRotate) BlockOffset    += ((BlockOffset >= RepIndex) ? 1 : 0);
    *col            = info->LayoutTable[BlockID][BlockOffset];

    /* remap to distributed spare space if indicated */
    if (remap) {
      RF_ASSERT( raidPtr->Disks[*row][*col].status == rf_ds_reconstructing || raidPtr->Disks[*row][*col].status == rf_ds_dist_spared ||
	     (rf_copyback_in_progress && raidPtr->Disks[*row][*col].status == rf_ds_optimal));
      rf_remap_to_spare_space(layoutPtr, info, *row, FullTableID, TableID, BlockID, (base_suid) ? 1 : 0, SpareRegion, col, &outSU);
    } else {
    
        outSU	    = base_suid;
        outSU      += FullTableID * fulltable_depth;  				        /* offs to strt of FT */
        outSU	   += SpareSpace;						        /* skip rsvd spare space */
        outSU      += TableID * info->TableDepthInPUs * layoutPtr->SUsPerPU;   	        /* offs to strt of tble */
        outSU      += info->OffsetTable[BlockID][BlockOffset] * layoutPtr->SUsPerPU;	/* offs to the PU */
    }
    outSU          += TableOffset / (info->BlocksPerTable * info->PUsPerBlock);	        /* offs to the SU within a PU */
    
    /* convert SUs to sectors, and, if not aligned to SU boundary, add in offset to sector.  */
    *diskSector     = outSU*layoutPtr->sectorsPerStripeUnit + (raidSector % layoutPtr->sectorsPerStripeUnit);

    RF_ASSERT( *col != -1 );
}


/* prototyping this inexplicably causes the compile of the layout table (rf_layout.c) to fail */
void rf_MapParityDeclustered(
  RF_Raid_t       *raidPtr,
  RF_RaidAddr_t    raidSector,
  RF_RowCol_t     *row,
  RF_RowCol_t     *col,
  RF_SectorNum_t  *diskSector,
  int              remap)
{
    RF_RaidLayout_t *layoutPtr = &(raidPtr->Layout);
    RF_DeclusteredConfigInfo_t *info = (RF_DeclusteredConfigInfo_t *) layoutPtr->layoutSpecificInfo;
    RF_StripeNum_t SUID = raidSector / layoutPtr->sectorsPerStripeUnit;
    RF_StripeNum_t FullTableID, FullTableOffset, TableID, TableOffset; 
    RF_StripeNum_t BlockID, BlockOffset, RepIndex;
    RF_StripeCount_t sus_per_fulltable = info->SUsPerFullTable;
    RF_StripeCount_t fulltable_depth  = info->FullTableDepthInPUs * layoutPtr->SUsPerPU;
    RF_StripeNum_t base_suid = 0, outSU, SpareRegion=0, SpareSpace=0;

    rf_decluster_adjust_params(layoutPtr, &SUID, &sus_per_fulltable, &fulltable_depth, &base_suid);
    
    /* compute row & (possibly) spare space exactly as before */
    FullTableID     = SUID / sus_per_fulltable;
    if (raidPtr->numRow == 1) *row = 0;                         /* avoid a mod and a div in the common case */
    else {
      *row            = FullTableID % raidPtr->numRow;
      FullTableID    /= raidPtr->numRow;			/* convert to fulltable ID on this disk */
    }
    if ((raidPtr->Layout.map->flags & RF_DISTRIBUTE_SPARE)) {
	SpareRegion = FullTableID / info->FullTablesPerSpareRegion;
        SpareSpace  = SpareRegion * info->SpareSpaceDepthPerRegionInSUs;
    }

    /* compute BlockID and RepIndex exactly as before */
    FullTableOffset = SUID % sus_per_fulltable;
    TableID         = FullTableOffset / info->SUsPerTable;
    TableOffset     = FullTableOffset - TableID * info->SUsPerTable;
    /*TableOffset     = FullTableOffset % info->SUsPerTable;*/
    /*BlockID         = (TableOffset / info->PUsPerBlock) % info->BlocksPerTable;*/
    BlockID         = TableOffset / info->PUsPerBlock;
    /*BlockOffset     = TableOffset % info->PUsPerBlock;*/
    BlockOffset     = TableOffset - BlockID * info->PUsPerBlock;
    BlockID        %= info->BlocksPerTable;

    /* the parity block is in the position indicated by RepIndex */
    RepIndex        = (raidPtr->noRotate) ? info->PUsPerBlock : info->PUsPerBlock - TableID;
    *col	    = info->LayoutTable[BlockID][RepIndex];
    
    if (remap) {
      RF_ASSERT( raidPtr->Disks[*row][*col].status == rf_ds_reconstructing || raidPtr->Disks[*row][*col].status == rf_ds_dist_spared ||
	     (rf_copyback_in_progress && raidPtr->Disks[*row][*col].status == rf_ds_optimal));
      rf_remap_to_spare_space(layoutPtr, info, *row, FullTableID, TableID, BlockID, (base_suid) ? 1 : 0, SpareRegion, col, &outSU);
    } else {

        /* compute sector as before, except use RepIndex instead of BlockOffset */
        outSU        = base_suid;
        outSU       += FullTableID * fulltable_depth;
        outSU	    += SpareSpace;						/* skip rsvd spare space */
        outSU       += TableID * info->TableDepthInPUs * layoutPtr->SUsPerPU;
        outSU       += info->OffsetTable[BlockID][RepIndex] * layoutPtr->SUsPerPU;
    }
    
    outSU       += TableOffset / (info->BlocksPerTable * info->PUsPerBlock);
    *diskSector  = outSU*layoutPtr->sectorsPerStripeUnit + (raidSector % layoutPtr->sectorsPerStripeUnit);